Search

Mitigating methane (CH4) emissions is of uttermost importance to meet the targets of the Paris Agreement as the second-most important greenhouse gas in terms of radiative forcing. The atmospheric CH4 burden is rising, and has been doing so since the industrial times predominantly due to human activities. Novel airborne and space-based instruments have the potential to address the challenges towards detecting and quantifying localised emissions. The ultimate goal of this thesis is to use these state-of-art remote sensing measurements to improve inventories currently informed by modelled and process-based data. CH4 point source emissions require a high spatial resolution to distinguish between different sources and a high spectral resolution to characterise the CH4 component of the atmosphere. This thesis uses the hyperspectral imager AisaFENIX, with a spectral range capable of measuring CH4 in the SWIR around 2300 nm and a spatial resolution of about 2.5 m2, in the first tailored flight campaign with the BAS Twin Otter aircraft over UK CH4 point sources. Based on a fast data-driven retrieval algorithm to infer atmospheric concentrations of CH4 from spectroscopic measurements, it successfully explores the detection limits using real and simulated CH4 plumes. The detection limit suggests that assuming a constant rate of emissions throughout the year, even the highest reported UK emitter for 2018 (Drax Power Station) would not be detected. The retrieval has applications beyond the flight campaign and can be easily adapted for the next generation of hyperspectral satellites. This thesis also investigates the use of the Sentinel-5P satellite, with unprecedented daily global coverage and spatial resolution of around 7 km2, to quantify CH4 emissions over areas with limited observations. By identifying persistent CH4 enhancements, this thesis has successfully quantified emissions in South African coal mines and oil and gas facilities in Egypt. The estimates in this thesis for South Africa are in good agreement with the EDGAR v4.3.2 inventory, which underestimates emissions in Egypt by a third probably due to uncaptured fugitive emissions.

The biosphere plays a significant role in climate mitigation. In the course of photosynthesis, plants sequester approximately a third of the carbon dioxide (CO2) that humans release into the atmosphere each year. The protection of terrestrial ecosystems is therefore essential for the global carbon, water, and energy cycles, as well as for offsetting atmospheric carbon dioxide emissions. In order to better preserve and restore global ecosystems for endangered biodiversity, species, and human well-being, it is imperative to accurately measure the ecosystem-atmosphere mass (e.g., carbon) and energy exchanges - i.e., fluxes. The limitations of current observational and/or estimation techniques, however, make it difficult to quantify fluxes for terrestrial ecosystems, particularly those under management. Accordingly, it remains unclear what the exact values of global ecosystem carbon fluxes are and how they vary over time. Eddy covariance (EC) is a micrometeorological method that provides continuous and direct flux measurements. However, these measurements are only representative of the tower footprint, which ranges in radius from hundreds to several kilometres. It is common to upscale or extrapolate eddy covariance fluxes by integrating satellite remote sensing data in order to determine fluxes over a larger area. It should be noted, however, that many current flux upscaling studies focus on gross primary productivity rather than the net exchange, which necessitates deducting ecosystem respiration from the estimates. In addition, present flux upscaling estimations have a relatively coarse spatial resolution for the purpose of monitoring field-scale fluxes and assessing management activities. The goal of this thesis is to fill the knowledge gaps and develop continuous field-scale flux estimates for ecosystems under management practices such as livestock grazing. In order to accomplish this goal, the thesis will answer three major questions: 1) What factors determine the performance of flux upscaling; 2) How can flux gaps of different lengths be filled reliably; 3) Can field-scale estimates accurately reflect livestock grazing's effects on ecosystem fluxes? Chapter 2 examines the feasibility of consistently estimating global carbon, water, and energy fluxes with the flux upscaling technique. This 'consistent upscaling' means using the same machine learning algorithm as well as satellite and meteorological data over the course of the flux upscaling. According to the results, the consistent flux upscaling explains 70 % of the daily variability of carbon, water, and energy fluxes. Furthermore, this chapter analyses the effects of algorithms, predictor sets, and eddy covariance itself on flux upscaling performance, concluding that the spatiotemporal sampling density of eddy covariance towers is critical to the flux upscaling accuracy. The eXtreme Gradient Boosting (Xgboost) outperforms other machine learning algorithms with at least 14 % higher accuracy and an average of 90 % shorter run time. Using different predictors results in a difference in flux upscaling accuracy of less than 4 %. The spatial sampling density of eddy covariance towers, however, can affect the flux upscaling accuracy to a degree larger than 50 %. This chapter lays the technical foundation for the entire thesis. Chapter 3 improves the method for filling in eddy covariance data gaps since incomplete data may compromise the accuracy of both eddy covariance and flux upscaling. In this chapter, a study at more than a hundred eddy covariance towers around the world demonstrates that the random forest algorithm can fill gaps even longer than a month. Depending on how many environmental drivers are used, random forest has a 15 % to 30 % improvement in gap-filling accuracy over the research standard method. Random forest exhibits greater stability in filling long gaps, as the gap length increases from one day to a month, its accuracy drop is 81 % less than the standard research method. In Chapter 4, the gap-filling study from the previous chapter is extended to methane fluxes and ecosystems under various management practices (e.g., livestock grazing). The random forest method is 29 % to 54 % more accurate than the standard method when it comes to gap-filling methane fluxes. When using modelled environmental drivers, random forest maintains 90 % of gap-filling accuracy compared to using measured drivers. By doing so, it is possible to promote the use of eddy covariance applications when measured environmental drivers are not available. When gaps exceed three months, the R2 of random forest gap-filling remains 0.7. In addition, the random forest is found to preserve environment-flux responses well in the filled gaps. Chapters 3 and 4 contribute to the aim of this thesis - i.e., reliable field-scale flux estimation - by providing complete eddy covariance time series. Chapter 5 examines how satellite spatial resolution affects the extrapolation of eddy covariance fluxes from tower footprints to the entire field. It also evaluates the proposed field-scale flux estimates in light of the performance of reproducing flux spatial variability and recognising grazing periods. According to the results, satellite spatial resolution has little impact on the estimated flux sum of the whole field (R2 difference ≪ 0.1), but it has a great impact on the estimated flux spatial variability. It is observed that the flux estimates are highly consistent with measurements (R2 = 0.7 and annual bias < 2 Mg ha-1 yr-1 for carbon fluxes), and they clearly distinguish between fluxes occurring during grazing periods and those occurring during non-grazing periods. Since Sentinel-2's imagery textiles are of greater clarity, it is superior to other satellite platforms in extrapolating tower-based fluxes (in southwest England). In conclusion, this thesis accomplishes its goal of filling the knowledge gap in reliably estimating ecosystem fluxes with multi-source observation and machine learning. The results confirm that eddy-covariance carbon, water, and energy fluxes can be scaled up in a consistent manner. Specifically, this research emphasises the importance of spatially well-distributed towers for upscaling eddy covariance fluxes, particularly in tropical regions and/or ecosystem types such as evergreen broadleaf forests. This research also identifies the validity of the random forest algorithm in gap-filling - random forest is effective in filling varying-length gaps while maintaining relationships between fluxes and environmental variables. Furthermore, this research highlights the importance of satellite platform selection in quantifying the spatial variability of field-scale flux estimates. As a final note, it validates machine learning algorithms for obtaining fluxes that are resistant to disturbances caused by management activities such as livestock grazing. There are important implications of this research for climate mitigation efforts that are supported by the management and protection of terrestrial ecosystems. Findings from this research are beneficial for pinpointing carbon uptake capacity of global land ecosystems, for determining where to install future eddy covariance towers, and for evaluating the effects of other management activities on ecosystem fluxes.

Coastal flooding presents an increasing threat to many urban centres. Current flood models are often based upon low-density topographic data and do not include information regarding underground structures, which are high-risk areas and act as sinks in flood scenarios, affecting the flow of water through the streetscape. The background study details flood modelling and topographic surveying approaches. Comparison finds that mobile LiDAR provides the most promising solution for accurate, efficient and repeatable modelling of urban environments. The methodology describes a novel and repeatable process for surveying a cityscape, using Belfast city centre as the primary study area. The GeoSLAM Zeb Horizon handheld LiDAR scanner is the primary surveying instrument as it is cost effective (£30-35k) and highly manoeuvrable. Results indicate that the digital city model created has improved point density and accuracy compared to the previous highest resolution model. Relative point accuracy ranges from 1-2cm, and point density for targeted features exceeds 3000pts/m². The merging and georeferencing process introduces a negligible amount of error. Comparison with high-resolution aerial laser scanning data collected in Dublin (2015) highlights the strengths and weaknesses of the handheld approach. The GeoSLAM provides improved point density and accuracy, especially on building facades, but provides poor coverage of rooftops and gated areas, due to the angle of data capture. Comparison with tripod mounted terrestrial laser scanning (TLS) collected in Belfast highlights a disparity in accuracy and resolution, which the GeoSLAM makes up for with surveying efficiency. The relationship between geographical information systems (GIS) and LiDAR data is explored and showcases the importance of GIS based datasets in organising and producing city scale LiDAR datasets. GIS based approaches were developed to locate underground spaces within the point cloud, to promote their repetition and widespread use throughout relevant organisations. ArcGIS Pro has proven to be a suitable platform for displaying the location of basements scanned directly, indirectly and for which façade features indicate their presence. Handheld mobile LiDAR scanning proves to fill a niche within the remote sensing market. The hardware provides cost and time efficient data capture, inclusive of street level and underground features with resolution and accuracy great enough to enhance flood simulations. The easy deployment of the technology promotes consistent updates of digital models as the streetscape inevitably changes over time. Handheld scanning is suitable for creating a relatively inexpensive digital model or as a source of supplementary data for pre-existing topographic datasets.

Since the last century, glaciers have been retreating due to climate change, especially in the Arctic, which is causing the sea level to rise, has the potential to change global weather and climate patterns. Therefore, it is important to understand climate change impacts on glaciers and to develop new automated approaches that can accurately quantify glacier area changes. In this study, we developed a new Object-based image analysis approach in Google Earth Engine, utilizing Landsat satellite imagery and mapped a total of 2,203 (≈32,894 km2) glaciers in four regions in the Arctic and high latitudes: Novaya Zemlya, Russia, Baffin Island, Canada, Disko Island, Greenland and Kenai, Alaska at three different time periods; 1985-89, 2000-02, and 2019-21. The results showed a clear reduction in the total glacier area in each region. 480 glaciers in Novaya Zemlya that cover an area of 22,990±301 km2 experienced a 1,319±419 km2 (5.7%) decline in total glacier area. 523 glaciers on Baffin Island covering 7,211±158 km2 lost a total area of 452±227 Km2 (6.6%). 748 glaciers on Disko Island covering 1,929±127 km2 lost 457±168 km2 (23.6%), and 452 glaciers in Kenai covering 764±60 km2 lost 196±84 km2 (25.7%). Seventy-three glaciers have completely retreated including sixty-nine on Disko Island, three on Novaya Zemlya, and one in Kenai. The results also show that glaciers area loss was greater in 2000-02 to 2019-21, compared to 1985-89 to 2000-02 in all four regions. The accuracy of this method was evaluated by comparing manually corrected outlines, resulting in an overall accuracy estimate between 93% and 98% with the reference data based on random sampling approach. This demonstrates that our methodology in GEE is a powerful and robust tool for glacier mapping that reduces time required for manual correction and has the potential to be applied to other glacierised regions.

Covering roughly a third of the Earth's surface, agricultural land is central to livelihoods, food security, biodiversity and climate. The need for food and materials, declining soil fertility in agricultural systems, and climate change have all led to wide-scale agricultural expansion into natural habitats. More than 90% of deforestation across the tropics is estimated to be driven by agriculture. Deforestation and degradation damage the ecosystem services that many rural and forest-proximate people rely on. Agriculture, forestry and other land use account for nearly a quarter of anthropogenic greenhouse gas emissions. How we use land is critical to the future trajectories of biodiversity, climate change and poverty alleviation. The global agricultural system needs transformation, and trees on farms have been identified as an important tool in this transformation. Trees on farms can improve biodiversity, sequester carbon, provide ecosystem services and improve livelihoods in agricultural landscapes. To understand the state of, and changes in, biodiversity in these agricultural landscapes with trees, they must be monitored in a consistent and timely manner. However, efficient wide-scale monitoring methods are lacking, and it is not yet clear how Earth Observation (EO) data can be used for the landscape-level analyses needed to monitor biodiversity. In chapter 2, I test a novel approach to mapping tree species assemblages by mapping ordination axes of floristic composition using EO data. Existing methods for modelling floristic gradients have been confined to more homogeneous landscapes or using hyperspectral imagery and have limited wider utility. The models were tested in three complex agricultural-forest landscapes (in Uganda, Rwanda and Honduras) using a fusion of optical and radar imagery alongside other geospatial datasets. Nonmetric Multidimensional Scaling ordination scores describing floristic composition were modelled using random forest regression (with the remote sensing data as predictors) and mapped across the study sites, testing the approach's applicability in multiple contexts. EO data were able to predict some of the variations in floristic composition: model fits varied from R2=0.56 - 0.77 and RMSE from 9 - 19% across sites. The resultant maps capture the main landscape features of tree floristic gradients. The results show that this novel approach using a fusion of optical and radar EO data alongside geospatial data in a machine learning model can map the tree floristic gradients in complex agricultural systems. The floristic gradient map provides more detailed spatial assessments of floristic composition for understanding biodiversity in agricultural landscapes than were previously possible with satellite data and is a step towards monitoring biodiversity in these systems. EO data can be used to scale up field data to assess aspects of biodiversity at landscape scales, but this is not feasible at national scales. A lack of systematic data on the biodiversity in agricultural land at national scales means monitoring global targets is difficult. There is a need for indicators of agricultural biodiversity applicable at widescale and across different landscapes. In chapter 3, I develop and present the proof of concept for an indicator of the biodiversity value of agricultural landscapes by assessing the properties of their trees. The tool uses freely available satellite data products to estimate wooded area, structural diversity and spectral diversity of agricultural lands. It combines them to ascribe a score that can be mapped at national scales. Qualitative photointerpretation validation shows promising results in four case studies in various agricultural contexts. Ideas for developing the indicator to ensure indicator continuity are discussed and include improvements in data that can come from upcoming satellite products, further qualitative and quantitative validation from those with on-site expertise, and testing the applicability of the indicator for change detection and quantification. The indicator should be a valuable tool for planners and decision-makers to monitor agricultural land, report on biodiversity, and plan informed conservation strategies. It has the potential to be a much-needed indicator for the post-2020 agenda for measuring and monitoring agricultural biodiversity. In order to realise the potential that trees on farms have, it must be adopted widely by farmers. Promoting agroforestry for all its benefits requires an understanding of the determinants of adoption. We know the adoption of agroforestry depends on many factors, including a number of socioeconomic and biophysical conditions. Current research in understanding these determinants is focused on context-specific case studies and is inconsistent between studies. More generalisable information is needed to ensure effective and informed policy and action. Chapter 4 takes a regional approach to exploring these determinants to see how they vary from region to region across Uganda. The results show that, on average, across all regions, travel time to cities was the most important factor, but there is significant regional disparity in which factors are most important as well as inconsistent directions of the relationships. This spatially explicit information can help improve agroforestry adoption through better extension services and interventions tailored to regional circumstances, tackling the most important barriers in each region.

This thesis develops analytical frameworks for evaluating the validity of MBES data with comparable ground truthing and hydrodynamic data with respect to geomorphology and marine biodiversity in two Special Areas of Conservation (SAC) in the Malin-Hebrides Sea off the north-Irish coast. First, a modelling framework based on Generalised Linear Modelling (GLM) is developed to test the validity of multi-frequency MBES backscatter data (30, 95, 300 kHz) to characterise sediment grain size in the Hempton's Turbot Bank (HTB) SAC. The results demonstrate that the single-frequency sources have a marginal gain on the multi-frequency model, with the 30 kHz model driving the significance of the multi-frequency model, and the inclusion of the higher frequencies diminishes the level of agreement. Secondly, a combination of K-Means unsupervised classification and GLM based on MBES backscatter (95, 300 kHz) and bathymetry analysis are successfully used to predict different levels of sandeel Hyperoplus lanceolatus densities with respect to geomorphology in the HTB SAC. The study demonstrates that the lower frequency source is more adept at capturing the variety inherent in shallow sub-surface sedimentary environments, which this species prefers. Finally, multidisciplinary modelling and analysis approaches using MBES data, hydrodynamic data and theodolite tracking observations are used to assess the spatial dynamics of harbour porpoise Phocoena phocoena in the Skerries and Causeway SAC. Generalised Additive Modelling (GAM) identifies that slope, aspect and backscatter intensity are the most statistically significant variables accounting for the highest deviance in porpoise sighting density. Models predict a high probability (> 0.6) of porpoise encounters in nearshore areas, particularly concentrated around headlands where local flow acceleration results in coarser beds. The robust workflows developed in this thesis, provides a proof of concept for developing robust monitoring strategies for Marine Protected Areas and associated species. Respective recommendations will inform marine policy, marine spatial planning and management.

Brazilian ecosystems are facing major threats to their conservation and functioning from human induced disturbances in combination with climate change. The two largest Brazilian ecosystems, the Amazon Forest and Cerrado, lost together approximately 15,690 km2 year-1 of their intact vegetation cover between 2010 and 2020 (equivalent to 12% year-1 of England area), with an intensification of vegetation loss from 2019. Brazil is responsible for the largest carbon dioxide emissions from land use and land cover changes (LULCC). Still, uncertainties arise in global carbon budget assessments on the magnitude and trends of these disturbance fluxes and its impacts on the carbon balance. In this thesis, I investigate the components of human disturbance and how they impact the contemporary carbon cycle of Brazilian ecosystems, such as LULCC, as well the relationship between fire occurrence and landscape fragmentation. The thesis ends with a synthesis of the contemporary net land carbon balance of the Amazon using a set of state-of-the-art estimates from different methodological approaches. In chapter 2 I attempt to reconcile estimates of LULCC emissions for Brazil. I first evaluated new global LULCC maps used in global bookkeeping and Dynamic Global Vegetation Models (DGVMs) from the Global Carbon Budget (GCB) assessments for Brazil. These new maps were based on Food Agriculture Organization (FAO) statistics combined with multi-year land cover data from the European Space Agency. I found good spatial agreement between the global LULCC maps with MapBiomas, the Brazilian remote sensing based LULCC maps. However, the global LULCC dataset still failed to capture the magnitude and recent increase in deforestation in Brazil. This is due to the method used to infer deforestation from agricultural area change reported to FAO in addition the use of a simple linear trend to extrapolate beyond the final census year. The findings of this study have been welcomed by the global land modelling community and led to the subsequent adoption of MapBiomas in the Global Carbon Budget assessment for 2022 (GCB2022). Conversion of intact vegetation cover to other land uses induce landscape fragmentation and fire spread. In chapter 3, I test the relationship between fire occurrence within intact vegetation fragments and the degree of landscape fragmentation in the two major and contrasting Brazilian ecosystems (Amazon and Cerrado). I found that there is an ecosystem-specific relationship between burned area fraction within intact vegetation fragments and landscape fragmentation induced by human activity. I find that fire occurrence within intact vegetation fragments associated to landscape fragmentation has an opposite relationship when comparing Amazon and Cerrado. These human-induced disturbances cause large losses of carbon to the atmosphere, thus impacting the carbon balance of tropical areas. Building on all the findings of chapters 2 and 3, in chapter 4, I use the latest data from a set of regional and global models to attribute and synthesise the contemporary net land carbon balance of Amazon. The different approaches agree that the south-eastern Amazonia is a net carbon source to the atmosphere. Bottom-up approaches suggest that Brazilian Amazonia was a net land carbon source in 2020 due to an increase in deforestation and fire emissions. Given the large uncertainties from the models and remaining knowledge gaps, there is insufficient data to reject the hypothesis that the Amazon was not neutral between 2010 and 2018. In conclusion, this thesis reconciles diverging estimates of land use and land cover emissions of Brazil in global carbon budget assessments. Moreover, this thesis brings together state-of-the art estimates of the Amazon carbon budget using bottom-up and top-down models and the knowledge gaps to reconcile them. Finally, I highlight the importance of the use of remote sensing data to constrain global estimates of land use and land cover change datasets and emissions. This thesis also demonstrates divergent fragmentation-fire relationships in the two major and contrasting Brazilian biomes. The results present in this thesis have important implications for the mitigation potential of Brazilian ecosystems within the goals of Paris Agreement.

This thesis develops the technique of millimetre-wave radar at 94 GHz for close-range remote sensing of glaciers and terrestrial snow cover (the cryosphere). The capabilities of 94 GHz radar for cryosphere mapping are demonstrated using the 2ⁿᵈ generation All-weather Volcano Topography Imaging Sensor (AVTIS2), which maps 3D terrain from real-beam scanning. AVTIS2 acquires 3D point clouds of terrain and a comparison to co-located high density point clouds derived from Terrestrial Laser Scanner (TLS) data showed that AVTIS2 point cloud uncertainties were 1.5 m at 1.5 km and 3 m at 3 km. These values are smaller than other close-range radar systems used to map cryospheric terrain in 3D. Next, the distribution of Normalised Radar Cross Section (σ⁰) values over glacier ice at 94 GHz was found to be −17.0 < σ⁰ < −3.4; σ⁰ₘₑₐₙ = −9.9 dB and followed a log-normal distribution. These values are comparable to other terrain types at 94 GHz such as refrozen snow, wet snow and wet soil, hence glacier surfaces were found to be suitable targets for terrain mapping at 94 GHz. These fundamental results were used to apply the AVTIS2 94 GHz radar for monitoring snow cover and glacier calving. Field trials in Scotland showed that the transition of wet snow to a refrozen snowpack in response to reductions in air temperature could be detected from changes in 94 GHz σ⁰. Snow hazard features could also be identified, such as post-avalanche debris accumulations which manifest as localised increases in radar backscatter. Next, 3D AVTIS2 data was used to quantify glacier calving rates at the Hansbreen tidewater glacier in Svalbard. A time series of ice cliff morphology derived from the AVTIS2 3D data sets exposed the role of melt undercutting and demonstrated the critical role of the ocean on calving processes in Svalbard. The high accuracy 3D data acquired from AVTIS2 and the sensitivity of radar backscatter to surface changes has demonstrated the unique capabilities of 94 GHz radar for cryosphere mapping, paving the way forward for new applications and opportunities for close-range remote sensing of the cryosphere.

Polarimetric synthetic aperture radar (PolSAR) has been widely used in ocean and cryospheric applications. This is because, PolSAR can be used in all-day operations and in areas of cloud cover, and therefore can provide valuable large-scale monitoring in polar regions, which is very helpful to shipping and offshore maritime operations. In the last decades, attention has turned to the potential of PolSAR to detect icebergs in the Arctic since they are a major hazard to vessels. However, there is a substantial lack of literature exploring the potentialities of PolSAR and the understanding of iceberg scattering mechanisms. Additionally, it is not known if high resolution PolSAR can be used to detect icebergs smaller than 120 metres. This thesis aims to improve the knowledge of the use of PolSAR scattering mechanisms of icebergs, and detection of small icebergs. First, an introduction to PolSAR is outlined in chapter two, and monitoring of icebergs is presented in chapter three. The first data chapter (Chapter 4) is focused on developing a multi-scale analysis of icebergs using parameters from the Cloude-Pottier and the Yamaguchi decompositions, the polarimetric span and the Pauli scattering vector. This method is carried out using ALOS-2 PALSAR quad polarimetric L-band SAR on icebergs in Greenland. This approach outlines the good potential for using PolSAR for future iceberg classification. One of the main important outcomes is that icebergs are composed by a combination of single targets, which therefore may require a more complex way of processing SAR data to properly extract physical information. In chapter five, the problem of detecting icebergs is addressed by introducing six state-of-the-art detectors previously applied to vessel monitoring. These detectors are the Dual Intensity Polarisation Ratio Anomaly Detector (iDPolRAD), Polarimetric Notch Filter (PNF), Polarimetric Matched Filter (PMF), reflection symmetry (sym), Optimal Polarimetric Detector (OPD) and the Polarimetric Whitening Filter (PWF). Cloude-Pottier entropy, and first and third eigenvalues (eig1 and eig3) of the coherency matrix are also utilised as parameters for comparison. This approach uses the same ALOS-2 dataset, but also evaluates detection performance in two scenarios: icebergs in open ocean, and in sea ice. Polarimetric modes (quad-pol, dual-pol, and single intensities) are also considered for comparison. Currently it is very difficult to detect icebergs less than 120 metres in length using this approach, due to the scattering mechanisms of icebergs and sea ice being very similar. However, it was possible to obtain detection performances of the OPD and PWF, which both showed a Probability of Detection (PF) of 0.99 when the Probability of False Alarms (PF) was set to 10-5 in open ocean. Similarly, in dual pol images, the PWF gave the best performance with a PD of 0.90. Results in sea ice found eig3 to be the best detector with a PD of 0.90 while in dual-pol mode, iDPolRAD gave a PD of 0.978. Single intensity detector performance found the HV channel gave the best detection with a PD of 0.99 in open ocean and 0.87 in sea ice. In the previous two approaches, only satellite data is used. However, in chapter six, data from a ground-based Ku-band Gamma Portable Radio Interferometer (GPRI) instrument is introduced, providing images that are synchronised with the satellite acquisitions. In this approach, the same six detectors are applied to three multitemporal RADARSAT-2 quad pol C-band SAR images on icebergs in Kongsfjorden, Svalbard to evaluate the detection performance within a changing fjord environment. As before, we also make use of Cloude-Pottier entropy, eig1 and eig3. Finally, we evaluate the target-to-clutter ratio (TCR) of the icebergs and check for correlation between the backscattering coefficients and the iceberg dimension. The results obtained from this thesis present original additions to the literature that contributes to the understanding of PolSAR in cryospheric applications. Although these methods are applied to PolSAR and ground-based radar on vessels, they have been applied for the first time on icebergs in this thesis. To summarise, the main findings are that icebergs cannot be represented as single or partial targets, but they do exhibit a collection of single targets clustered together. This result leads to the fact that entropy is not sufficient as a parameter to detect icebergs. Detection results show that the OPD and PWF detectors perform best in an open ocean setting and using quad-pol mode. These results are degraded in dual-pol mode, while single intensity detection is best in the HV cross polarisation channel. When these detectors are applied to the RADARSAT-2 in Svalbard, the OPD and PWF detectors also perform best with PD values ranging between 0.5-0.75 for a PF of 0.01-0.05. However, the sea ice present in the fjord degrades performance across all detectors. Correlation plots with iceberg size show that a regression is not straightforward and Computer Vision methodologies may work best for this.

The planet is heating up. Rapid-paced human change in the earth systems is increasing frequencies of catastrophic events, ranging from tropical cyclones to ecological collapse, oceanic increase to coastal floods, unprecedented wildfires to lethal droughts. In a 'compound state', as recent scientific reports suggest, climate change might render the planet uninhabitable. In regions exposed to the increased impacts of compound climate-related events, risks concerning human existence or the so-called 'existential risks' run high. 'Existential risk' is the possible chance of human extinction. Extinction means the irreversible termination of a species class and its genetic heritage. It is also known as an 'existential catastrophe'. 'Existential risk studies research' is a 'scientific discipline'. It estimates probabilities of human extinction using statistical and rational analysis of incoming evidence. In the climate context, the present literature analyses independent risks rather than geographic hotspots where multiple risks overlap and create novel catastrophic bifurcations across spatial and temporal scales. 'Ex-risk Architecture' integrates practices of computational architecture, remote-sensing earth science and design informatics into an expanded interface and an open-ended research process of information synthesis, form-making and design speculation--in order to anticipate global catastrophic risks of climate change in forensic-grade architectural images. I foreground architecture as an augmented form of risk perception in order to anticipate spaces of extinction rupturing from extreme climate change. Though oft-used to eliminate risks, I enact 'anticipation' through design to perceptualise risks in a spatiotemporal form and galvanise preparedness as catastrophic anthropogenic change accrues. I develop research through practice, using design to manufacture possible spatiotemporal experiences of catastrophic scenarios in order to shape global social imaginaries on climate change. I use architecture representation as a perceptual interface to compress geo-environmental information, indexing material legacies of human change in perceptual artefacts. I construct extreme spatiotemporal projections of the known impact of human change at smaller scales in computational models. I use site-specific empirical evidence from the planet's most exposed regions to climate change in the design process, ranging from glacier melting (Warren Cave) and environmental pollution (New Orleans) to atmospheric depletion (Mauna Kea). I develop algorithms through advanced programming languages to synthesise remote sensed geospatial datasets--revealing spatial and temporal accounts of the existential catastrophe at multiple scales in a broad range of architectural diagrams and digital animations. The outcomes form a design portfolio, prised apart into three-site specific projects. 1. Warren Cave (2018-2019) 2. New Orleans (2018-2019) 3. Mauna Kea (2019-2021) The work appraises climate change through architecture as an existential risk par excellence. The dissertation contributes to an advanced computational design workflow--integrating planet- scale contingencies into architectural concerns. It appraises architecture as a methodological parallel to existential risk research through original design investigation.

Detailed knowledge of the ecology and environmental conditions suitable for individual species across the landscape is vital for effective conservation measures. Similarly, understanding demographic factors that influence the structure of animal populations is crucial for understanding species trends. In this thesis, I explore one of the most interesting aspects that distinguishes the hazel dormouse (Muscardinus avellanarius) from other woodland small mammals - its ability to hibernate. Hibernation is a complex strategy with marked trade-offs that shapes the demography and structure of hazel dormouse populations and yet it is one of the least studied facets of their life cycle. Firstly, I introduce relevant background to the thesis. I evaluate different methods to locate hibernacula, investigate dormouse movements before hibernation, their behaviour as they prepare to face months of low activity at low temperatures and fewer foraging opportunities, to the point where they find a suitable place to build a nest to hibernate on the ground. I then examine population structure and estimate overwinter survival of different hazel dormouse populations. Using telemetry, I found that hazel dormice select hibernation sites within their autumnal home range. I investigate the impact of hibernation on body weight of hazel dormice and quantify rates of weight loss in wild animals. With the use of high-resolution airborne LiDAR derived canopy structure and topography, I develop novel models to characterise hazel dormouse hibernaculum locations and predict suitable locations across the landscape. I demonstrate that topography, sky view and canopy height can influence hibernaculum location selection. At the hibernaculum location, I demonstrate how hazel dormouse hibernation nests are built in a similar fashion to their summer nests and that they utilise a range of materials that are available in the immediate vicinity of the selected hibernation site. I quantify hazel dormouse overwinter survival of different populations and find that on average 0.36 (0.29-0.44, 95% Confidence Intervals (CI)) of the population survives. My findings, based on the existing literature and evidence I collected in the field, suggests that hazel dormice are resourceful, able to cope with diverse habitat characteristics and resources. Conservation efforts should therefore focus on creating, managing and/or enhancing diversity within their habitat by promoting a varied canopy structure that is well connected and made up of assorted tree and shrub species of value to the hazel dormice in order to increase nesting and foraging opportunities through the seasons.

This research focuses on the influence of discontinuities on the origin and development of landslide instability mechanisms at different scales. Various remote sensing techniques have been used for data acquisition to characterize discontinuity and landslide-related features. The remotely captured data was subsequently interrogated using a variety of analytical methods and numerical modelling to investigate the role of discontinuities at different scales in the origin and development of rock slides through analysis of selected case studies. The investigation has been carried out through landslide susceptibility mapping (LSM) with the inclusion of discontinuities, the use and application of 3D distinct element method (DEM) modelling of the Hell's Mouth landslide in the UK, 2D DEM modelling of a catastrophic rock avalanche in Italy, and the analysis of slope instability in an open pit mine (more than 200m height) in South America using both a 2D finite/discrete element method (FDEM) approach and a 3D DEM approach. The result of LSM demonstrates that integration of discontinuity orientation through GIS based kinematic analysis can effectively improve landslide prediction using machine learning (ML) modelling. In Hell's Mouth case study, numerical modelling and field observations both suggest that the cliff instability was characterised by a combination of planar sliding, wedge sliding, and toppling modes of failure controlled by the discrete fracture network geometry. For the Italian pale landslide, it was controlled by translational sliding along a folded bedding plane, with toe removal induced by river erosion resulting in daylighting of the bedding, creating kinematic freedom for the landslide. In addition, due to the presence of an anticline, the landslide region was constrained in the middle-lower section of the slope where the relatively high inclination of the bedding plane was detected. With respect to the inter-ramp deformation in South America, modelling results indicate the potential control and influence of a rock bridge, constrained by two faults at the toe of the slope. It was also demonstrated that blasting-induced weak zones and sequential excavations were potential key triggers of the observed slope deformation.

Vegetation ecosystems are increasingly under pressure from both direct human influence and indirect anthropogenically-driven climate change. Increasing amounts of data are made available from satellite systems which can image these ecosystems from afar. The work in this thesis provides several examples of the utility of remotely sensed data from satellites to assess the resilience of ecosystems. This notion of resilience is measured by considering the return rate following a perturbation, with statistical metrics such as AR(1) and variance providing an indication of system resilience and the proximity to a potential tipping point. The first focus of this work is on direct human environmental intervention through community-based agroforestry groups in Kenya. These results show that the efforts of these groups can be detected with satellite data as a greening trend which occurs both within designated tree planting groves and in the surrounding landscape. These groups provide a case study for the power of positive social tipping points to achieve environmental improvement. Following this, the potential of high-resolution satellite data from Sentinel-2 to quantify patterned vegetation in the Sahel is explored. These striking patterns have often been associated with vegetation resilience in drylands. No correlation is found between pattern morphology and resilience, contrary to a previously held hypothesis from the literature. Precipitation is also identified as a key driver of these patterns. Moving beyond drylands, satellite data is utilised at a global scale to assess the link between vegetation resilience and climatic variables across the world. There is a clear relationship between average resilience, as measured by AR(1), and precipitation, which is evident at three spatial scales; the local (pixel), ecoregion and biome. There is also a temperature component, with hotter, drier locations displaying lower levels of resilience. This thesis finishes with a discussion of the potential for a resilience sensing framework constructed by combining remote sensing data with new cloud computing technologies. This will enable the monitoring of resilience change across the world and the identification of regions which require further investigation and intervention.

Savannah biomes are home to a great diversity of flora and fauna species whose survival depend on the inherent heterogeneity within the savannah landscape. In southern African savannahs, grazing lawn patches play a central role in ecosystem dynamics via their influence on herbivory and fire regimes. The extent and distribution of grazing lawns therefore have important cascading impacts on habitat heterogeneity, biodiversity and important ecosystem processes such as nutrient cycling. Despite the keystone nature of grazing lawns in savannah ecosystems, there is limited knowledge on their extent and distribution. There is substantial empirical evidence of the factors that drive the formation and persistence of grazing lawns. However, no broad-scale approach exist to monitor grazing lawns and enable investigations into changes in their distribution and the impacts on broader ecosystem dynamics. Ground-based monitoring of grazing lawns is laborious and inefficient over large spatial and temporal scales. This research uses high-resolution satellite remote sensing to characterize grazing lawns and investigate dynamics in their cover and structure in southern African savannahs. This is achieved through a series of analysis that addresses three main objectives including (i) developing methods for accurate detection and mapping of savannah grazing lawn distribution using high-resolution satellite imagery and machine learning techniques; (ii) identifying changes in cover and structural distribution of grazing lawns over space and time; and (iii) Identifying the dominant drivers of change in grazing lawn cover and structure. For the analysis, a robust machine learning workflow is developed to identify grazing lawns in mesic and semi-arid savannah landscapes usingWorldView-3 imagery; a cost effective approach for high resolution grazing lawn monitoring is developed via fusion of open access Planet and Sentinel-2 imagery; multi-temporal high-resolution satellite images from 2002, 2014 and 2019 are used to identify changes in grazing lawn cover and structure under different savannah landscape conditions, including the effect of drought stress; and lastly, spatio-temporal analysis of grazing lawn occurrence and change trajectory is used to identify the dominant drivers of grazing lawn dynamics. High grazing lawn detection accuracies were achieved with all machine learning algorithms. Random Forest, Multilayer Perceptron and Support Vector Machine algorithms had similar accuracies and marginally outperformed Classification and Regression Trees algorithm. The results demonstrate the utility of high-resolution satellite images for overcoming savannah heterogeneity challenges to classification, leading to accurate grazing lawn detection. WorldView-3 with its high spatial resolution and broad array of vegetation sensitive spectral bands is particularly ideal for grazing lawn monitoring, and could be used for targeted investigations due to acquisition cost. Alternatively, fusion of open-access planet and Sentinel-2 images provide a cost effective option for operational management applications. In the absence of drought stress, grazing lawn extent increase uniformly, signaling to possible grazer population increase between 2002 and 2014. Most of the increase tend to occur as an expansion to existing patches or in close proximity to existing patches. The impact of drought stress on grazing lawns depends on local landscape characteristics, particularly related to water availability. Gains and losses in grazing lawn cover largely occur as transitions to and from tall grass swards, further highlighting tall grass vegetation as the main competitor for space. Woody encroachment was found not to be an immediate threat to grazing lawn cover, but could gain more significance with projected increase in drought frequency and intensity. In terms of the dominant drivers of grazing lawn dynamics, this study found the presence of water points as resource hot-spots to be an important determining factor of grazing lawn spatial distribution within the savannah landscape. Overall, grazing lawn dynamics was inferred to be primarily driven by grazers, the pattern and nature of which depends on factors that act to alter grazer population and behavior. This research makes vital methodological contributions to grazing lawn and overall savannah vegetation monitoring. A pioneering remote sensing based methodology for monitoring grazing lawn dynamics is developed. Additionally this research contributes to literature on grazing lawn ecology. A greater knowledge of grazing dynamics has been achieved, contributing to a better understanding of habitat heterogeneity in southern African savannahs. Overall, this research provides important tools and novel ecological insights to guide conservation management in savannah ecosystems.

Modelling and monitoring seasonal snow is critical for water resource management, flood forecasting and avalanche risk prediction. Snowmelt processes are of particular importance. The behaviour of liquid water in snow has a big influence on melting processes, but is difficult to measure and monitor non-invasively. Recent work has shown the promise of using electrical self potential and electrical resistivity measurements as snow hydrology sensors. Self potential magnitudes can be used to infer both liquid water content of snow and bulk meltwater runoff, and electrical resistivity is affected by liquid water content. In autumn 2018, a prototype geophysical monitoring array was installed at Col de Porte in the French Alps, alongside full hydrological and meteorological measurements made routinely at the site. Self potential measurements were taken throughout the following two winters, with manual snow pit data obtained in spring 2019. Electrical resistivity measurements were unsuccessful due to problems with power and control units. Observed self potential peaks preceded measured basal runoff peaks, indicating that self potential measurements are sensitive to water dynamics within the snowpack, most clearly during spring melting and rain-on-snow events. A physically-based snow hydrology model (Flexible Snow Model 2.0) was evaluated at Col de Porte against observations in order to select a best-performing configuration, by utilising the ability to easily change model parameters. Three different hydrology and two density configurations were tested, as well as investigating the effect of varying the irreducible water saturation and saturated hydraulic conductivity. It was found that an irreducible water saturation of 0.03 performed best, and that changing the saturated hydraulic conductivity had little effect on performance. This snow model was then coupled to an electrical model of liquid water in snow to create a synthetic set of self potential observations. These synthetic observations were compared to the observed self potential magnitudes to evaluate the effectiveness of the model and investigate the possibility of using the self potential array as part of a coupled geophysical monitoring and modelling system. It was found that modelled self potential magnitudes are extremely sensitive to small changes in prescribed snow properties, giving large uncertainties. Timings of modelled self potential peaks were able to be related to meteorological and hydrological observations, meaning self potential measurements could be used to improve liquid water flow representation in snow models. An empirical relationship between measured self potential and modelled internal water flow was trialled, which highlighted the potential for future empirical methods to exploit self potential observations. The combination of self potential, and meteorological and hydrological measurements has highlighted the value of combining observations with models both to guide future observation networks, and improve modelling capabilities.

Phytoplankton constitute the bottom of the aquatic food web, produce half of Earth's oxygen and are part of the global carbon cycle. A measure of aquatic phytoplankton biomass therefore functions as a biological indicator of water status and quality. The abundance of phytoplankton in most lakes on Earth is low because they are weakly nourished (i.e., oligotrophic). It is practically infeasible to measure the millions of oligotrophic lakes on Earth through field sampling. Fortunately, phytoplankton universally contain the optically active pigment chlorophyll-a, which can be detected by optical sensors. Earth-orbiting satellite missions carry optical sensors that provide unparalleled high spatial coverage and temporal revisit frequency of lakes. However, when compared to waters with high nutrient loading (i.e., eutrophic), the remote sensing estimation of phytoplankton biomass in oligotrophic lakes is prone to high estimation uncertainties. Accurate retrieval of phytoplankton biomass is severely constrained by imperfect atmospheric correction, complicated inherent optical property (IOP) compositions, and limited model applicability. In order to address and reduce the current estimation uncertainties in phytoplankton remote sensing of low - moderate biomass lakes, machine learning is used in this thesis. In the first chapter the chlorophyll-a concentration (chla) estimation uncertainty from 13 chla algorithms is characterised. The uncertainty characterisation follows a two-step procedure: 1. estimation of chla from a representative dataset of field measurements and quantification of estimation uncertainty, 2. characterisation of chla estimation uncertainty. The results of this study show that estimation uncertainty across the dataset used in this chapter is high, whereby chla is both systematically under- and overestimated by the tested algorithms. Further, the characterisation reveals algorithm-specific causes of estimation uncertainty. The uncertainty sources for each of the tested algorithms are discussed and recommendations provided to improve the estimation capabilities. In the second chapter a novel machine learning algorithm for chla estimation is developed by combining Bayesian theory with Neural Networks (NNs). The resulting Bayesian Neural Networks (BNNs) are designed for the Ocean and Land Cover Instrument (OLCI) and MultiSpectral Imager (MSI) sensors aboard the Sentinel-3 and Sentinel-2 satellites, respectively. Unlike established chla algorithms, the BNNs provide a per-pixel uncertainty associated with estimated chla. Compared to reference chla algorithms, gains in chla estimation accuracy > 15% are achieved. Moreover, the quality of the provided BNN chla uncertainty is analysed. For most observations (> 75%) the BNN uncertainty estimate covers the reference in situ chla value, but the uncertainty calibration is not constantly accurate across several assessment strategies. The BNNs are applied to OLCI and MSI products to generate chla and uncertainty estimates in lakes from Africa, Canada, Europe and New Zealand. The BNN uncertainty estimate is furthermore used to deal with uncertainty introduced by prior atmospheric correction algorithms, adjacency affects and complex optical property compositions. The third chapter focuses on the estimation of lake biomass in terms of trophic status (TS). TS is conventionally estimated through chla. However, the remote sensing of chla, as shown in the two previous chapters, can be prone to high uncertainty. Therefore, in this chapter an algorithm for the direct classification of TS is designed. Instead of using a single algorithm for TS estimation, multiple individual algorithms are ensembled through stacking, whose estimates are evaluated by a higher-level meta-learner. The results of this ensemble scheme are compared to conventional switching of reference chla algorithms through optical water types (OWTs). The results show that estimation of TS is increased through direct classification rather than indirect estimation through chla. The designed meta-learning algorithm outperforms OWT switching of chla algorithms by 5-12%. Highest TS estimation accuracy is achieved for high biomass waters, whereas for low biomass waters extremely turbid waters produced high TS estimation uncertainty. Combining an ensemble of algorithms through a meta-learner represents a solution for the problem of algorithm selection across the large variation of global lake constituent concentrations and optical properties.

Forests play a vital role in the wellbeing of our planet. Large and small scale deforestation across the globe is threatening the stability of our climate, forest biodiversity, and therefore the preservation of fragile ecosystems and our natural habitat as a whole. With increasing public interest in climate change issues and forest preservation, a large demand for carbon offsetting, carbon footprint ratings, and environmental impact assessments is emerging. Satellite remote sensing is the only method that can provide global coverage at frequent revisit times and is therefore the standard method for global forest monitoring. Most often, deforestation maps are created from optical data such as Landsat and MODIS. Although such maps are of generally good quality, they cannot quantify biomass and are not typically available at less than annual intervals due to persistent cloud cover in many parts of the world, especially the tropics where most of the world's forest biomass is concentrated. Synthetic Aperture Radar (SAR) can fill this gap as it penetrates clouds and interacts with the three-dimensional structure on the ground in a way that scales with volume and therefore biomass. While longer wavelengths are better for deeper penetration of the canopy and therefore biomass estimation, one of the most readily available data sources is Sentinel-1, a shorter wavelength C-band radar. In this thesis, the theory behind SAR is discussed to demonstrate its usefulness for forest monitoring. The potential for C-band SAR data to distinguish forest and non-forest is then assessed empirically in different regions of the world and existing change detection algorithms for deforestation are reviewed. The effect of spatial and temporal context on change detection accuracy is investigated, leading to the development of a robust method for deforestation detection in the absence of reliable reference data which often constitutes the largest practical hurdle. This method achieves a change detection sensitivity (producer's accuracy) above 99%, although false positives lead to a low user's accuracy of about 60%. The mean change detection delay amounts to about two to three months. While further work is required to reduce the false positive rate, improve detection delay, and validate this method in different biomes, the results show that Sentinel-1 data have the potential to advance global deforestation monitoring.

Organism-environment interactions take place through a multitude of processes that generate patterns across scales in space and time, but our understanding of pattern and processes is traditionally constrained by observational limitations. Contemporary technological advances in remote sensing, explored in this thesis, are extending the power and capability of ecological investigation. Three-dimensional (3D) ecosystem structure can now be analysed across scales from millimetres to kilometres and from minutes to decades, providing insight into scale-dependent patterns and their driving processes in complex and dynamic systems like temperate reefs. Remote sensing technologies are available for 3D mapping and recent years have seen a rapid expansion in their use in field ecology. In chapter 2, I reviewed the current state of the art in high-resolution 3D ecosystem mapping technologies and their applications, highlighting the emerging era of 3D spatial ecology and identifying potential barriers to widespread uptake. I addressed a paucity of information on the accuracy and practicality of emerging optical remote sensing tools in ecological contexts by testing structure-from-motion photogrammetry and terrestrial laser scanning, in three coastal habitats, over three spatial scales. The accuracy of structure-from-motion photogrammetry, compared to terrestrial laser scanning models, was greatest at fine spatial scales (25 m², < 1 cm resolution) on more stable substrates like rock, with mean ± sd absolute difference of 4 mm ± 14 mm. Accuracy decreased with increasing spatial scale and in less stable vegetated scenes, with a maximum difference of 56 mm ± 111 mm in saltmarsh at a scale of 2500 m² extent and < 2 cm resolution. Structure-from-motion photogrammetry was more portable, faster, flexible and lower-cost than terrestrial laser scanning, but was more vulnerable to error propagation. Capturing sufficient ecologically relevant spatial and temporal variation in 3D structure is challenging in complex, dynamic habitats like intertidal temperate reefs. In chapter 3 I used the tools tested in chapter 2 to investigate spatial and temporal patterns in the structure of biogenic Sabellaria alveolata reef across scales. At a habitat scale (~35,000 m² extent, 10 cm horizontal resolution) most of the variation in reef structural change was explained by a combination of systematic trends with shore height and positive spatial autocorrelation up to the scale of colonies (1.5 m) or patches (4 m). Plot-scale mapping (2500 m² extent, 10 cm horizontal resolution) over five years (2014-2019, 6-month intervals) revealed previously undocumented temporal patterns in reef accretion and erosion. The system was highly dynamic at small spatial and temporal scales (< 4 m, 6 months), but reef accretion and erosion compensated each other, resulting in stable habitat structure over larger scales (>130 m, 5 years). This scale-dependent variability would have been impossible to capture with conventional methods like quadrat, transect or point-based survey using GPS or theodolite, demonstrating the value of modern 3D mapping technologies to enhance our understanding of ecosystem dynamics across scales. Subtidal temperate reefs hosting diverse communities are often found in high-energy waters, but these are understudied compared to lower energy seas, and knowledge of reef distribution is lacking. In chapter 4 I used multiscale 3D seafloor data and hydrodynamic information to predict the spatial distribution of geogenic reef and biogenic Sabellaria spinulosa reef habitats in a high tidal energy region. Random Forest models for reef substrate and S. spinulosa reef had balanced accuracy mean ± 95% CI of 80.7% ± 0.8% and 77% ± 1% respectively. Mean bed shear stress was the most important variable in both models, highlighting the importance of including measures of hydrodynamic energy in predictive mapping of high-energy temperate reef habitats. My research demonstrates the increased power and insight that can be gained with contemporary 3D mapping and monitoring tools in field ecology. I showed that habitat structure in complex systems can be simultaneously highly dynamic and remarkably stable depending on the scale of observation, and that multiscale structural metrics are central to cost-effective mapping of subtidal temperate reef ecosystems. The collective works highlight the need for multiscale and multidisciplinary analysis and the value of embracing technological solutions for ecology in the age of big data. The emerging field of 3D ecosystem mapping and high-resolution remote sensing will have far-reaching implications for research, management and public engagement.

The deterioration of ageing infrastructure assets and the risk of damaging surrounding structures during new construction are major concerns for the transport industry worldwide. Whilst, Structural Health Monitoring (SHM) could assist the lifecycle management of existing assets and the assessment of structures adjacent to large construction projects, the cost of sensors limits the number of structures that can be evaluated. Space-borne Interferometric Synthetic Aperture Radar (InSAR) is a powerful remote-sensing technology which can provide deformation measurements over time for numerous points located on buildings and infrastructure assets. Such measurements are available with an accuracy comparable to traditional in-situ monitoring devices, but at a much lower cost. Furthermore, due to the combination of wide area coverage and high spatial resolution, InSAR can easily shift from local to larger scale, with the potential to provide a cost-effective tool for SHM on a regional level. However, to assess the condition of a given structure, InSAR deformation measurements need to be translated into structural performance indicators. This requires an integration of InSAR displacement measurements and structural information and/or models. Such integration can be particularly demanding for regional analysis, where a vast number of structures needs to be evaluated. The objective of this thesis is to develop a new methodology using InSAR data to monitor and assess the conditions of buildings and infrastructure assets on a regional scale. Such a methodology will allow the understanding of the long-term behaviour of existing infrastructure, and the assessment of damage to existing structures caused by new infrastructure construction. This involved: (i) the definition of performance indicators for the identification of anomalous structural deformations on a large scale; (ii) the integration of InSAR deformation measurements and structural models to interpret the measured deformations in relation to a specific source of movement; (iii) a clear definition of the InSAR challenges that still need to be overcome in order to use the developed methodology as an operational SHM tool. First, a new method using InSAR displacement measurements to evaluate the condition of large transport networks was developed. The proposed method is based on an automated workflow which enables the integration of InSAR deformation measurements and digital databases of roadway and bridge infrastructure to warn of potentially anomalous deformations within a given network. For each asset and roadway segment, the method allows the assessment of monitoring point density, the retrieval of local displacements and velocities, and the identification of anomalous relative movements within the same structure. The developed method provides output maps showing the distribution of PS density, local displacements, velocities and the locations of anomalies. The developed method was tested on the Los Angeles highway and freeway network and on the Italian motorway system, validating its performance from city to national scale. Furthermore, to show the capability of the proposed method of identifying potentially damaging movements, the case of an Italian motorway viaduct that was damaged in 2015 is presented. Second, a new method using InSAR displacement measurements for the structural assessment of buildings adjacent to tunnel excavations was developed. The proposed method is based on an automated workflow which enables the integration of InSAR deformation measurements, digital building databases and structural models of the building response to tunnelling-induced settlements. InSAR displacement measurements were used to estimate the settlement profile for each building. To calculate critical strains for each building, building settlement profiles were analysed through a semi-empirical model. On the basis of the estimated critical strain, a level of damage was assigned to each building. The proposed method provides as outputs damage maps showing the distribution of damage levels for the buildings along the excavation. The developed method was tested on the buildings along the Crossrail tunnel alignment in London, and enabled the identification of structural damage to more than 800 buildings, highlighting its capability as a city-scale assessment tool. The developed integration also provided the first large dataset of field information on building response to tunnelling, enabling the identification of relationships between structural characteristics and building responses. Finally, based on the technical issues encountered during this research and the open problems identified by other researchers, a discussion about the advances needed in InSAR technology to be used as a SHM tool was developed. In this discussion, each InSAR technological limitation was evaluated from a SHM perspective. The challenges that still need to be overcome to use InSAR as an operational tool were defined, with examples showing the practical limitations of InSAR technology. Possible solutions and promising research directions were also identified. The proposed methodology has the potential to inform timely maintenance and prioritisation decisions, and can complement in-situ monitoring instrumentation. Findings can improve current practice for condition monitoring and assessment on large scale, with the possibility to advance the understanding of structural deformation mechanisms.

The transition from foraging to the domestication of plants and animals in the Atacama Desert (ca. 3,000 BP) seemed to spark the emergence of social complexity, sedentarism, and the end for Hunter-Gatherer lifeways. Nonetheless, ethnohistorical evidence suggests the opposite. To challenge current narratives this work proposes an alternative perspective where the key is to ignore simple versus complex dichotomy and instead look deeply at the sources of evidence available. To test this proposed new narrative the basin of the Camarones Valley was the sampling area; a river valley located at the core of the Atacama that covers latitudinally from the Andean mountains to the Pacific. Therefore, an assessment of historical sources, the interpretation of freely available satellite imagery, and a sample of the new sites remotely sensed were surveyed. As a result, approximately 2,000 new archaeological sites in the basin of Camarones shows that agriculture was very limited, small-scale sites constitutes 90% of the record, and that pre-Hispanic people settled and moved around the landscape as a whole deriving in a high-mobility landscape showing limited evidence of sedentarism, disproving the current narrative of 'social complexity' and its close relation with sedentism. With this new information, a reorganization of the current multi-ethnic paradigm was also proposed for the different periods after the Formative. This change reflects that at least two distinct groups of foragers inhabited the western Chilean valleys during the Late Intermediate Period, and possibly prior to this. Finally, together with the proposal of the paradigmatic enhancement, it was possible to identify two types of archaeological sites unpublished for the Camarones Valley, the stoneworks for hunting called chaku and caycu. These have great implications for future research that will allow us to know in more detail the role of hunters in pre-Hispanic times.

This project is about the development of band selection (BS) techniques for better target detection and classification in remote sensing and hyperspectral imaging (HSI). Conventionally, this is achieved just by using the spectral features for guiding the band compression. However, this project develops a BS method which uses both spatial and spectral features to allow a handful of crucial spectral bands to be selected for enhancing the target detection and classification performances. This thesis firstly outlines the fundamental concepts and background of remote sensing and HSI, followed by the theories of different atmospheric correction algorithms - in order to assess the reflectance conversion for band selection - and BS techniques, with a detailed explanation of the Hughes principle, which postulates the fundamental drawback for having high-dimensional data in HSI. Subsequently, the thesis highlights the performances of some advanced BS techniques and to point out their deficiencies. Most of the existing BS work in the field have exhibited maximal classification accuracy when more spectral bands have been utilized for classification; this apparently disagrees with the theoretical model of the Hughes phenomenon. The thesis then presents a spatial spectral mutual information (SSMI) BS scheme which utilizes a spatial feature extraction technique as a pre-processing step, followed by the clustering of the mutual information (MI) of spectral bands for enhancing the BS efficiency. Through this BS scheme, a sharp 'bell'-shaped accuracy-dimensionality characteristic has been observed, peaking at about 20 bands. The performance of the proposed SSMI BS scheme has been validated through 6 HSI datasets, and its classification accuracy is shown to be ~10% better than 7 state-of-the-art BS algorithms. These results confirm that the high efficiency of the BS scheme is essentially important to observe, and to validate, the Hughes phenomenon at band selection through experiments for the first time.

We are in the midst of a new era of Earth Observation for vegetation monitoring. Recent enhancements in the spectral, spatial and temporal resolution, as well as availability, of optical sensors provides opportunities to significantly advance the retrieval of vegetation properties at multiple scales. However, directed research linking optical imagery to field-collected vegetation measurements is required to optimally harness current and future advances in remote sensor technologies. This thesis aims to investigate the optical remote sensing of plant functional traits at four scales: the leaf, unmanned aerial vehicle (UAV), aircraft and satellite level. This thesis has four objectives, each of which are addressed through a research article. The main objective of the first article, entitled 'Multiscale mapping of plant functional groups and plant traits in the High Arctic using field spectroscopy, UAV imagery and Sentinel-2 data', is to evaluate the use of multispectral UAV imagery for mapping biochemical leaf traits in high Arctic tundra. The High Arctic is an ideal environment for establishing a baseline potential for mapping leaf traits using UAV imagery, as species richness is low and vegetation cover lacks the structure and confounding effects of a forest canopy. The article starts by using field spectroscopy to classify plant species and quantify biochemical plant traits (water content, nitrogen (N), phosphorus (P) and C:N ratios). At the UAV level, it classifies three plant functional types (mosses, graminoids and dwarf shrubs) at 72% accuracy and generates maps of plant chemistry. (r2 = 0.44-0.54, RMSEmean = 26%-48%). Lastly, the UAV results were used to explore two methods of upscaling plant water content to the wider landscape using Sentinel-2 imagery. The objective of the second article, entitled 'Mapping the Leaf Economic Spectrum across West African Tropical Forests Using UAV-Acquired Hyperspectral Imagery' is to evaluate the use of hyperspectral UAV imagery for mapping biochemical leaf traits across a species-rich, structurally complex forest canopy in Ghana, West Africa. Similarly, the article starts by using field spectroscopy to estimate six biochemical plant traits (leaf mass per area (LMA), N, P, calcium (Ca) and magnesium (Mg)). At the airborne level, four of the six leaf traits demonstrated weak (p < 0.10) correlations with the UAV-collected hyperspectral imagery: r2 = 0.25 (LMA), r2 = 0.22 (N), r2 = 0.22 (P), and r2 = 0.25 (Ca). Mapping the leaf economic spectrum (LMA, N and P) across a precipitation gradient reflected carbon cycling data, and revealed precipitation and substrate as co-dominant drivers of trait distributions and relationships. The objective of the third article, entitled 'Mapping photosynthetic capacity in a Bornean tropical forest from airborne and satellite imagery: challenges and recommendations', expands focus from mapping biochemical leaf traits to mapping photosynthetic capacity (Vcmax25 and Jmax25) across a forest disturbance gradient in Malaysian Borneo using aircraft-collected hyperspectral imagery and Sentinel-2 imagery. In addition, it explores increased image quality issues associated with expanded spatial coverage. Aircraft-collected hyperspectral imagery was found to exhibit poor radiometric calibration and Sentinel-2 imagery was found to suffer from processing errors related to cloud masking and inconsistent surface reflectance values, especially over humid tropical forests. The strongest model performance (r2=0.36 and 0.43 for Vcmax25 and Jmax25 respectively) was produced by tuning the high spectral and spatial resolution of the aircraft-collected hyperspectral imagery with the improved signal-to-noise ratio of a Sentinel time-series composite, demonstrating the importance of diagnosing and minimising data quality issues. The last article entitled 'Global mapping of the photosynthetic capacity of tropical forests', combines Sentinel-2A imagery, environmental datasets and photosynthesis data from 35 tropical forest GEM plots to create a pan-tropical map of Vcmax25 and Jmax25. To the best of my knowledge, this is the first research that uses Sentinel-2 imagery to map tropical forest Vcmax25 and Jmax25 globally. Models captured 38% and 57% of variation in Vcmax25 and Jmax25 across lowland forest, and 62% and 61% of variation across lowland-and-montane forest. In lowland-only forest, mean climatic water deficit (MCWD) and solar radiation were most important predictors of Vcmax25 and Jmax25 respectively. In lowland-and-montane forest, the strongest predictor of both Vcmax25 and Jmax25 was mean annual temperature (MAT). At intercontinental scales, Vcmax25 and Jmax25 values were mainly controlled by climate. At the regional level, trait distributions were shaped by soil type. At local levels, trait patterns reflected spectral variation arising from topography and canopy structure. Overall, this thesis offers new methodologies to measure and monitor tropical forest functional diversity and productivity. It builds from a simple ecosystem to a floristically and structurally complex tropical forest, and explores the trade-offs of expanding spatial coverage and decreasing spectral resolution for the retrieval of vegetation canopy traits. Alongside, it highlights current limitations of optical remote sensing, and presents best-practice recommendations for future imaging studies. The results are pertinent for high resolution, low cost mapping of tundra and tropical forest ecosystems (and potentially for other terrestrial ecosystems), and for improving the representation of vegetation properties, especially photosynthetic capacity, in the next generation of Earth System Models.

This thesis investigates the archaeological landscape of the Yorkshire Wolds as revealed through newly gathered non-invasive survey evidence, with a view to re-evaluating how its rural communities evolved throughout the Late Iron Age and Romano-British period. The research utilises data derived from four large-scale geophysical surveys undertaken across key parts of this landscape, integrated within an extensive aerial photographic study that collates evidence from both new and existing sources. This evidence is contextualised within previous archaeological research, demonstrating the potential of large-scale remote sensing for developing a more nuanced interpretation of past landscapes when integrated with even limited pre-existing excavation evidence. Through the collection, integration and contextualisation of these extensive new datasets, the thesis is able to provide unique insight into both this complex landscape and the processes which underlie the formation of its archaeological record. The first three chapters outline the key research framework, situating it within the current state of knowledge and outlining the methods employed - including details of a bespoke gradiometry cart system designed and built in collaboration with Dominic Powlesland and the Landscape Research Centre during the initial stages of this research. A further five chapters present the results of both the geophysical and aerial remote-sensing surveys, before summarising the implications for both our understanding of this unique archaeological environment and broader approaches to the study of Iron Age and Romano-British landscapes.

Coastal ecosystems around the world are commonly affected by frequent flood events, in particular during winter. This impact is expected to be exacerbated because of climate change. To address this issue, coastal defences are usually built to reduce or mitigate flooding using hard engineering techniques such as sea walls but they are significantly expensive to be created and maintained over time. Alternatively, nature-based strategies, known as soft engineering, are being adopted to recreate or restore natural features. In Europe, including the UK, there is an increasing trend of developing these strategies in coastal management such as managed realignments schemes to restore saltmarshes which are recognised to mainly provide wave attenuation due to their vegetation structure. However, permanent monitoring of this function is challenging due to the extensive data collection needed over time. This thesis takes into account the use of satellite data to retrieve leaf area index (LAI) as an alternative structural parameter of saltmarsh vegetation instead of the traditional frontal surface area (FSA) for wave attenuation purposes. The aim of this research is to assess how optical remote sensing can be used to assess vegetation structure in saltmarshes for wave attenuation. To accomplish this aim, the following objectives are presented: (1) to evaluate the potential of Sentinel-2 MSI imagery for characterising intra-seasonal variation in saltmarsh LAI, (2) to determine whether remotely-sensed LAI may be an alternative for field-based FSA as structural vegetation parameter for wave attenuation, and (3) to assess whether remote sensing-based LAI can effectively be used as an alternative for FSA in wave attenuation modelling. The Brancaster managed realignment scheme was selected as the main study area. The results of this thesis have shown the importance of seasonality to accurately characterise vegetation structure using remote sensing approaches. The remotely-sensed LAI is found to be a suitable descriptor of both saltmarsh vegetation structure and wave attenuation by vegetation. This thesis mainly contributes to improving monitoring of saltmarsh vegetation of managed realignment scheme to assess wave attenuation by developing a brown LAI model for the winter season when storm events are more frequent, highlighting the relationship between satellite-derived LAI and wave attenuation estimations, and incorporating LAI into wave numerical modelling. Remote sensing approaches potentially provide a rapid evaluation of coastal protection service by managed realignment schemes and monitor their performance over time.

This doctoral thesis presents the first work on the extent, spatial dynamics, and long-term trends of the giant kelp (Macrocystis pyrifera) ecosystem in southern South America and sub-Antarctic latitudes. This work was based on an interdisciplinary approach of remote sensing using satellite images, unmanned aerial vehicles (UAV), geographical information systems, underwater exploration, and a collection of historical sources and secondary information from international collaborations. The first research paper of this thesis (Chapter 3) presents a mapping methodology of giant kelp derived from filtering values from the satellite Sentinel-2. This method was validated with secondary sources (previous kelp surveys) and primary sources of data, such as detailed orthomosaics made with UAV. Because of the high overall accuracy of this method, it was possible to map and measure the surface of the complete distribution of the giant kelp ecosystem by ecoregion and province. The second paper (Chapter 4) is based on the mapping methodology from Chapter 3, developing a spatial analysis of the sub-Antarctic kelp from South America, Falkland Islands (Malvinas) and South Georgia to understand the abiotic patterns that could predict the typical size of a giant kelp forest. Additionally, these patterns were compared in a temporal analysis stretching as far back as the 19th century to assess the persistence of this ecosystem. The results of this paper indicate that sub-Antarctic giant kelp is one of the least modified and most persistent marine ecosystems on Earth. Chapter 5 analyses the spatial distribution of marine sea surface temperature (SST) heatwaves and marine cold spells in the Channels and Fjords ecoregion in past decades, with a focus on understanding the climatological trends as they can affect giant kelp ecosystems. This paper established that, in contrast to the main global trend pointing towards SST warming, the frequency of marine cold spells in the nearshore coastlines of southernmost South America has in fact increased in the decade of 2010. We identified two main mechanisms as possible causes for this trend: increased frequency of storms driven by the Antarctic Circumpolar Current and glacier melting in the Patagonian fjords. Cold waters favour the persistence of giant kelp and associated biodiversity, hence the lack of SST warming highlights the region as a potential refugium for the ecosystem. However, more studies are needed to understand these new dynamics, and we conclude the paper with a summary of potential stressors to giant kelp ecosystems in the study area. Finally, Chapter 6 provides an overview of the main findings of this thesis and recommendations for future research, highlighting the need for further interdisciplinary studies on the Magellanic marine province as a climatic refugium for the giant kelp ecosystem.

This thesis presents an interdisciplinary case study of the active Volcán de Fuego in southern Guatemala to examine: (1) the physical behaviours and volcanic hazards that characterize eruptions occurring in recent years, and (2) the factors that influence local residents' decision to evacuate from these eruptions. The thesis presents different answers to these issues depending on the data sources studied: satellite observations, geophysical and gas timeseries, and observations of and interviews with local residents and authorities. The thesis begins by presenting Fuego as an ideal subject for an interdisciplinary volcanology PhD, and establishes the philosophical positions and methodological approaches that were necessary to consider in order to undertake both physical and social science research within this PhD. The first results of this thesis are satellite observations of Fuego's activity between January 2015 and June 2018. These observations, supplemented by other data, identify a new eruptive regime characterized by frequent explosive eruptions ("paroxysms") consistently preceded by lava flow effusion. Thresholds for determining eruption are debated. Physical results are juxtaposed with qualitative narratives of previous eruptions of Fuego and evacuations of communities as told by both authorities and local residents of rural communities around the volcano. These narratives reveal that an eruption at Fuego is not a consistent phenomenon, but is experienced differently by different observers based on their previous experiences, knowledge, resources, and priorities. Finally, quantitative and qualitative data are integrated through analysis of timeframes of eruption and response for several recent eruptions. Quantitative timescales and their qualitative counterpart, timelines, provide detailed chronologies of times and uncertainties involved in forecasting eruptions of Fuego and of deciding, warning, responding to, and evacuating from eruption. Ultimately, this thesis concludes that the lives of local residents cannot be reliably protected from hazards of Fuego without integration of the monitoring and risk mitigation efforts of INSIVUMEH and CONRED. This integration mirrors that of physical and social drivers of volcanic risk explored within this work. This thesis demonstrates the value of integrating physical and social research methods in a single interdisciplinary project, and contributes to volcanological literature with findings that volcanic risk is both spatially and temporally variable around a single volcano.

This thesis uses neural networks and deep learning to address practical, real-world problems in the mining sector. The main focus is on developing novel applications in the area of object detection from remotely sensed data. This area has many potential mining applications and is an important part of moving towards data driven strategic decision making across the mining sector. The scientific contributions of this research are twofold; firstly, each of the three case studies demonstrate new applications which couple remote sensing and neural network based technologies for improved data driven decision making. Secondly, the thesis presents a framework to guide implementation of these technologies in the mining sector, providing a guide for researchers and professionals undertaking further studies of this type. The first case study builds a fully connected neural network method to locate supporting rock bolts from 3D laser scan data. This method combines input features from the remote sensing and mobile robotics research communities, generating accuracy scores up to 22% higher than those found using either feature set in isolation. The neural network approach also is compared to the widely used random forest classifier and is shown to outperform this classifier on the test datasets. Additionally, the algorithms' performance is enhanced by adding a confusion class to the training data and by grouping the output predictions using density based spatial clustering. The method is tested on two datasets, gathered using different laser scanners, in different types of underground mines which have different rock bolting patterns. In both cases the method is found to be highly capable of detecting the rock bolts with recall scores of 0.87-0.96. The second case study investigates modern deep learning for LiDAR data. Here, multiple transfer learning strategies and LiDAR data representations are examined for the task of identifying historic mining remains. A transfer learning approach based on a Lunar crater detection model is used, due to the task similarities between both the underlying data structures and the geometries of the objects to be detected. The relationship between dataset resolution and detection accuracy is also examined, with the results showing that the approach is capable of detecting pits and shafts to a high degree of accuracy with precision and recall scores between 0.80-0.92, provided the input data is of sufficient quality and resolution. Alongside resolution, different LiDAR data representations are explored, showing that the precision-recall balance varies depending on the input LiDAR data representation. The third case study creates a deep convolutional neural network model to detect artisanal scale mining from multispectral satellite data. This model is trained from initialisation without transfer learning and demonstrates that accurate multispectral models can be built from a smaller training dataset when appropriate design and data augmentation strategies are adopted. Alongside the deep learning model, novel mosaicing algorithms are developed both to improve cloud cover penetration and to decrease noise in the final prediction maps. When applied to the study area, the results from this model provide valuable information about the expansion, migration and forest encroachment of artisanal scale mining in southwestern Ghana over the last four years. Finally, this thesis presents an implementation framework for these neural network based object detection models, to generalise the findings from this research to new mining sector deep learning tasks. This framework can be used to identify applications which would benefit from neural network approaches; to build the models; and to apply these algorithms in a real world environment. The case study chapters confirm that the neural network models are capable of interpreting remotely sensed data to a high degree of accuracy on real world mining problems, while the framework guides the development of new models to solve a wide range of related challenges.

Tropical rain forests are important carbon stores and harbours of biodiversity but are being cleared at an unprecedented rate. There is an estimated 2 billion hectares of degraded forest globally, which retains a large proportion of its biodiversity. Restoration of these lands is needed to meet global commitments to combat the interlinked climate and biodiversity crises, and effective, scalable and affordable monitoring of the restoration process is essential. High resolution remote sensing technologies offer the best hope for monitoring at scale. In particular, unoccupied aerial vehicles (UAVs) offer a viable option for high spatial and temporal resolution remote sensing, though methods to guide forest restoration with these are still in their infancy. This thesis introduces approaches for the use of remote sensing data to guide tropical forest management, with particular focus on the use of UAV data in the context of restoration, looking at canopy structure, composition and dynamics. First, I introduce the context of tropical forest restoration, discussing the contribution of remote sensing to monitoring and understanding projects, with a focus on the recent developments around the use of UAVs. I also introduce the main study site of this thesis --- an ecosystem restoration concession of nearly 100 km^2 in Sumatra, Indonesia, known as Hutan Harapan. Next, I introduce a method for delineating individual tree crowns in three dimensions from remote sensing data in the form of point clouds, as created by light detection and ranging (LiDAR) and UAV structure from motion (SfM) approaches. This method, MCGC, makes use of graph cut concepts from mathematics combined with understanding of tree crown geometry and allometric scaling to automatically map tree crowns. I validate this approach using data collected in Borneo, comparing forests with three distinctive structures, showing the power of this approach to both map trees and estimate aboveground biomass. In Chapter 3, I develop a pipeline for automatic mapping of key tree species prevalence at Hutan Harapan from photographs taken from a UAV. I show it is possible to break up imagery over management units into superpixels, and through a combination of spectral and textural patterns in the imagery, train an automatic classifier to detect the species of interest from UAV imagery. I then show the power of this approach to map prevalence of key tree species indicative of the successional stage of forest recovery and demonstrate the utility of this approach for guiding management. I find that using an extra camera to take photographs with additional wavebands only slightly improved mapping accuracy. Finally, I use a combination of a LiDAR survey in 2014 and UAV surveys in 2017 and 2018 to track the effects of the strong El Niño event of 2015-16 on the canopy at Hutan Harapan, looking at 3 sites of varying recovery status spanning 100 ha of forest. I find that early-successional forest was less resistant to the drought than taller secondary forest - with canopy height loss and high mortality. However, in the subsequent high-rainfall period, I observe that early-successional forests recovered strongly. Together, the analyses demonstrate that early-successional stages lost and then regained canopy height to a greater extent that taller forest, highlighting the power of repeat surveys using LiDAR and UAVs to track canopy dynamics. Finally, I critically evaluate the methods developed, highlighting how the insights they provide can be useful for restoration practitioners, underlining the key role that remote sensing, especially with a UAV, can play whilst also needing further development.

High-latitude processes and polar cloud microphysics envelop research areas the furtherance of which in the more accurate description of the arctic energy budget, climate variability, and feedbacks is essential. This study develops and assesses a multifrequency and Doppler-inclusive, ice particle size distribution retrieval technique for mean mass-weighted diameter and ice water content, and additionally proposes a radaronly, moment-based phase partition scheme tailored to polar regions. The technique is applied to datasets collected by the ARM research facility at the North Slope of Alaska and two ARM Mobile Facility 2 campaigns (BAECC, AWARE) that comprise a gamut of ground-based profiling sensors including radars at X, Ka, and W bands, lidar, and microwave radiometer. The microphysical retrieval is a multi-instrument task; firstly, a Ka-band -based, reflectivity-velocity power-law scheme estimates an ice sedimentation velocity corrected for the vertical wind component. An attenuation correction is then applied for the estimation of effective reflectivities. Both reflectivity and sedimentation velocity are integrated in a Bayesian scheme that compares observations against lookup- tables in both backscattering and hydrodynamic ice-particle properties. The radar only, phase partitioning is proven reliable in detecting cloud-top supercooled liquid water layers. Aggregate statistics are computed based on BAECC data. The combined utilization of radar reflectivity in a dual- or triple-frequency forward modeling in conjunction with sedimentation velocity, demonstrates that Doppler observations can assist in microphysical characterizations because connections across observations and simulations of both ice scattering and hydrodynamic properties are less ambiguous compared to scattering-only. Results demonstrate less retrieval spread among X / Ka / W / Doppler or Ka / W / Doppler retrieval scenarios as opposed to X / Ka / W. Lastly, the inclusion of a lower frequency (10-15 GHz) is advocated due to its significance in signal mitigation (attenuation, cross-calibration) and in the retrieval forward modeling ambiguity constraining.

Chile's temperate forest is a global biodiversity hotspot due to its levels of endemism and rate of habitat loss. Deforestation and biological invasions are two major threats faced by this ecosystem. Both are produced by the expansion of productive forestry, which fragments the native forest and introduces alien species (from the Pinaceae family and Eucalyptus genus). Despite these threats, broad scale ground-based forest and alien species monitoring is not feasible due to logistical and budgetary limitations. Surveying and mapping alien species are key steps to understand biological invasions. This project uses remote sensing to understand biological invasions by alien trees from productive forestry in Chile's temperate forest at different spatial scales. Sentinel-2 is used to assess alien tree occupancy and model its relationship with fragmentation and landscape configuration at a broad scale. A temporal assessment using Landsat historical archive imagery is used to investigate the dynamics of alien tree spread over the last 40 years. A local scale assessment using very high resolution WorldView-3 imagery and object-based image analysis is used to investigate early invasion and its relationship with land cover configuration. The results demonstrate remote sensing methods are useful for alien tree monitoring at different scales and hold potential for management. The expansion of alien trees at a broad scale is dominated by socioeconomic factors such as cost effectiveness and government subsidies. Alien Pinaceae tend to invade the naturally open Araucaria araucana forests, threatening the long term regeneration of this endemic and culturally relevant tree. Early Pinaceae invasion is linked to fragmentation, distance to seed sources, topography and human intervention. Incidentally, this study found that native forest degradation is occurring, increasing its vulnerability to invasion in the future. Conservation and forest management need to adopt sustainable, evidence-based practices to achieve compatibility between productive forestry and native forest conservation.

This thesis is the culmination of work carried out on a number of different UAS, in order to determine the feasibility of using UAS to deliver sensor payloads to the marine environment. A number of experiments involving said UAS were conducted, and future work detailed. Chapter 2 provides a brief overview of literature in acoustics and UAS, including UAS use around marine mammals. Chapter 3 looks into the noise produced by a gimbal, which relates the noise levels seen to literature and determines if the acoustic signal emitted could interfere with wildlife. Chapter 4 delves into the acoustics generated by the motors and propellers, with a comparison of these results with published literature. Chapter 5 shows the work conducted and implemented on a custom data logging system with in-house designed underwater housing, for use on a Splashdrone UAS, which is used in Chapter 6 for an acoustic survey of the marine environment. Chapter 7 uses on-board accelerometers inside of a flight controller to monitor displacement experienced, for use in monitoring surface heave of a water body. Chapter 8 concludes this thesis with a review on key findings, and a summary of future work that is to be conducted. Overall, this thesis determines that it is feasible to deliver sensor payloads to marine environments to an extent; a number of issues remain, which are detailed throughout this thesis.

Historically, natural history museums have collected and preserved specimens to provide data on the occurrence and distribution of wildlife populations. Zoologists still track animals by recording footprints, collecting dung and spoor and observing, recording and quantifying behaviour from the ground. However, these traditional observational techniques allow only a few populations to be monitored at once at limited spatial scales and disturbance from the ground can disrupt observation of natural behaviour. We are now in a golden age of technological advances and are able to remotely monitor and track wildlife via a variety of electronic sensors. Significant questions remain about how best to methodologically apply these new technologies for the purposes of wildlife monitoring. In this thesis, I consider challenges of using Earth observation satellites and unmanned aerial vehicles (UAVs) to track wildlife and understand movement in relation to the expanding human footprint and anthropogenic risk. Specifically: (i) I collate and analyse spatially explicit data on the distribution of illegal hunting incidences via a systematic map. I show that hunting increases in proximity to roads, water bodies, and human settlement areas and there is a considerable lack of systematically collected quantitative data. (ii) I investigate acoustic disturbance to understand anthrophony from the species perspective. I create a mitigation method applied in the case of UAV noise using species weighted audiograms (iii) I test whether very high-resolution satellite imagery and machine learning can be used to automate the detection of African elephants in vast heterogeneous landscapes. This is achieved presenting a new method to monitor elephants (iv) I record the spatial relationship of African elephants in relation to the human footprint using GPS tracking data and satellite imagery. I show elephants readily adapt their foraging habits and itineraries, spatially and temporally in relation to human settlement. Accurate and up-to-date data is vital for effective wildlife conservation planning. Remote sensing technologies offer enhanced capabilities to understand the spatial relationship between wildlife and the increasing human footprint. This body of work contributes to the global wildlife conservation effort by devising methods that can enable more reliable data collection at larger spatial scales.

The Barents Sea is a highly seasonal and productive shelf ecosystem, representing a gateway between the Atlantic and Arctic Oceans. Anthropogenic climate change is causing rapid changes in temperature, ocean chemistry, water column structure and sea-ice persistence. Sea-ice extents are declining twice as fast in the Barents Sea as other Arctic regions, establishing a mandate to investigate the impact of these environmental changes on productivity in the Seasonal Ice Zone. This is the remit of Arctic 'PRIZE' (PRoductivity in the Seasonal Ice ZonE), and this thesis was conducted under the auspices of work package 3, studying phytoplankton compositional and phenological variability. Remote-sensing is a key means by which ocean ecosystems are monitored and there have been recent efforts in the literature to employ ocean-colour to recognise different assemblages of phytoplankton. It is routine practice in the remote-sensing literature to distinguish functional assemblages from cell-counts and then seek bio-optical signatures to be targeted in remotely-sensed imagery. The analysis herein suggests that switching between different data types in this way introduces a translation error that undermines remote-detection efforts. We therefore recommend that functional assemblages are defined and identified directly from their bio-optical properties. Thereafter, a scheme of bio-optical functional types is defined, based on Barents Sea phytoplankton communities, and a bio-optical algorithm is developed with the aim of retrieving this structure in remotely-sensed ocean-colour. This algorithm is applied to study community compositional change, facilitating the integration of previous in-situ studies towards a more complete understanding of incipient ecosystem change. In addition, the ascendancy of temperate-mode autumnal phytoplankton blooms is investigated and relationships to evolving environmental conditions are explored, producing strong evidence that Atlantic inflow currents regulate the intensity of autumnal blooms in the Barents Sea. In sum, growing Atlantic influence in the Barents Sea is found both to cause community compositional changes and concomitant phenological shifts that may change Barents Sea ecosystem function.

Enhancing resilience to climate change is a key management goal for Mediterranean ecosystems. Typically, these management plans are based on ecological knowledge of species' tolerances derived from local studies limited in time and space. Remote sensing provides opportunities to study resilience over larger scales, but the tools needed to quantify the resilience of forests to drought and evaluate the effectiveness of management plans remain limited. This thesis examines how freely available satellite data can be used to quantify changes in forest canopies in response to climate variability. Using a combination of time-series and break-point analyses of satellite imagery I resolve limitations in forest resilience estimation and show that, for Spanish woodlands, the relative water availability during and following drought events are important in driving the canopy greenness loss and recovery. I show that despite increasing aridity, and examples of localised die-back events, Spanish forests are mostly becoming denser, with only 12% of locations analysed declining in greenness over the 18-year study period. This work demonstrates the importance of large-scale remote sensing analyses for obtaining an objective perspective on drought impacts. The thesis then explores the potential of remote sensing to map tree species in a region of regenerating woodlands near Madrid, providing the information needed for a nuanced understanding of resilience. I found that tree classification using high-resolution airborne hyperspectral imagery was highly accurate, while species maps produced using Sentinel 2 imagery (multispectral data with 10-m spatial resolution) were less successful at identifying species, with average agreement of 64% with the airborne derived map. Following on from this work, I used areas with high classification agreement between the airborne and spaceborne species information to study the effect of species composition on forest responses to droughts. I identify contrasting responses of the canopy greenness and wood production to drought. Specifically, wood production was found to be more sensitive to changes in water availability than canopy greenness. For the oak species, wood production was mirrored by changes in canopy greenness, but black pines reduced their wood production during droughts without substantial reduction in canopy greenness. I investigate the differences between the species and the mixing effects further by studying foliar compositions during a dry summer in Spain. There were strong differences between pines and oaks in the stable isotope ratios of carbon, probably driven by underlying differences in water-use efficiency, and differences in the stable isotope ratios of nitrogen, probably driven by underlying differences in species' investments in the photosynthetic apparatus. I conclude by highlighting the implications of my research for studying the relationships between diversity and ecosystem functioning from space.

Air pollution is a major environmental issue, causing millions of premature deaths worldwide and billions of pounds in economic losses every year. Nitrogen dioxide (NO2) is a key air pollutant and a good proxy for air quality. Reliable NO2 monitoring is key to understanding and tackling air pollution, and can be achieved at a global scale using satellite remote sensing. However, current and upcoming instruments do not provide the necessary spatio-temporal resolution to resolve fine-scale sources. This thesis explores a novel instrument concept, the High-resolution Anthropogenic Pollution Imager (HAPI), for use in a constellation of small satellites as a low-cost solution to the observational data gap. HAPI has only ten narrow spectral bands and uses a novel Discrete-Wavelength variant of the well-established Differential Optical Absorption Spectroscopy (DW-DOAS). This work evaluates a HAPI airborne demonstrator built at the University of Leicester during the OmniSat-HAPI project. Laboratory, field and flight tests were performed which showed consistently poor retrieval results. An investigation was conducted which confirmed that the cause of the poor retrieval performance was a suboptimal choice of wavelengths for NO2 nadir observations in combination with a DOAS-like method. A new set of wavelengths was identified and found to perform remarkably well. The robustness of DW-DOAS with the new wavelengths was evaluated by means of a sensitivity analysis, based on which new instrument performance requirements were defined for the next iteration of HAPI. While there are a few technical challenges to overcome, the OmniSat-HAPI constellation has the potential to provide NO2 observations at unprecedented spatial and temporal resolution, and would be a valuable complement to the existing air quality monitoring landscape.

Volcanic eruptions pose a significant risk to human health and infrastructure. With a growing population living in close proximity to historically active volcanoes, the capacity to mitigate these risks is essential. This is primarily achieved by monitoring volcanoes during eruptions and periods of quiescence. In recent decades, volcanic gas fluxes have been monitored using UV spectrometers and cameras both by short-term campaign measurements and in long-term automated monitoring networks, generating a wealth of expertise, instrumentation and data. The application of such instruments for risk reduction forms the central theme for this thesis, which asks two questions. Firstly, how can we utilise the existing infrastructure and instrumentation designed to measure volcanic gases to also investigate volcanic ash plumes and secondly, how can we improve the quality of the techniques already used to measure these volcanic gases? For the first question, I adapted existing UV cameras and spectrometers to measure volcanic ash through depolarisation imaging and application of spectral analysis techniques developed for UV satellite instruments to ground-based measurements. The results show that existing UV instrumentation can be adapted relatively easily to also measure volcanic ash, providing insights into ash plume dynamics and geometry. These techniques could be used in the future to help inform ash dispersal simulations, helping to mitigate the risks from ash rich eruptions. To improve current volcanic gas measurements, I developed and demonstrated two new tools, named iFit and OpenSO2. The first is a technique to easily retrieve absolute SO2 slant column densities from UV scattered sunlight spectra without a measured reference spectrum. This technique could help to improve the quality and reliability of future gas flux measurements, allowing volcanic processes to be better investigated. The second is an automated UV scanning system controlled by a Raspberry Pi, a low-cost single board computer, designed to be simpler and more adaptable than existing systems to allow volcano observatories more control over their monitoring instruments. The software for both these projects is written in Python and freely available online, making these methods highly accessible for maximum impact. These tools have been well received by the volcanological community. The iFit software has been utilised by several volcano monitoring agencies, including the Montserrat Volcano Observatory, the Istituto Nazionale di Geofisica e Vulcanologia in Italy and the Geological Survey of Japan. The OpenSO2 system has been installed on Montserrat to monitor long term SO2 fluxes from Soufriere Hills Volcano, with plans to install new instruments at La Soufriere in St Vincent and Volcan de Colima in Mexico. The tools and techniques developed in this thesis will help to improve future volcano monitoring of gas fluxes and ash plumes, helping to mitigate volcanic risks in the future.

Spatial information must be collected alongside the data modality of interest in wide variety of sub-sea applications, such as deep sea exploration, environmental monitoring, geological and ecological research, and samples collection. Ocean-bottom seismic surveys are vital for oil and gas exploration, and for productivity enhancement of an existing production facility. Ocean-bottom seismic sensors are deployed on the seabed to acquire those surveys. Node deployment methods used in industry today are costly, time-consuming and unusable in deep oceans. This study proposes the autonomous deployment of ocean-bottom seismic nodes, implemented by a swarm of Autonomous Underwater Vehicles (AUVs). In autonomous deployment of ocean-bottom seismic nodes, a swarm of sensor-equipped AUVs are deployed to achieve ocean-bottom seismic imaging through collaboration and communication. However, the severely limited bandwidth of underwater acoustic communications and the high cost of maritime assets limit the number of AUVs that can be deployed for experiments. A holistic fuzzy-based localisation framework for large underwater robotic swarms (i.e. with hundreds of AUVs) to dynamically fuse multiple position estimates of an autonomous underwater vehicle is proposed. Simplicity, exibility and scalability are the main three advantages inherent in the proposed localisation framework, when compared to other traditional and commonly adopted underwater localisation methods, such as the Extended Kalman Filter. The proposed fuzzy-based localisation algorithm improves the entire swarm mean localisation error and standard deviation (by 16.53% and 35.17% respectively) at a swarm size of 150 AUVs when compared to the Extended Kalman Filter based localisation with round-robin scheduling. The proposed fuzzy based localisation method requires fuzzy rules and fuzzy set parameters tuning, if the deployment scenario is changed. Therefore a cooperative localisation scheme that relies on a scalar localisation confidence value is proposed. A swarm subset is navigationally aided by ultra-short baseline and a swarm subset (i.e. navigation beacons) is configured to broadcast navigation aids (i.e. range-only), once their confidence values are higher than a predetermined confidence threshold. The confidence value and navigation beacons subset size are two key parameters for the proposed algorithm, so that they are optimised using the evolutionary multi-objective optimisation algorithm NSGA-II to enhance its localisation performance. Confidence value-based localisation is proposed to control the cooperation dynamics among the swarm agents, in terms of aiding acoustic exteroceptive sensors. Given the error characteristics of a commercially available ultra-short baseline system and the covariance matrix of a trilaterated underwater vehicle position, dead reckoning navigation - aided by Extended Kalman Filter-based acoustic exteroceptive sensors - is performed and controlled by the vehicle's confidence value. The proposed confidence-based localisation algorithm has significantly improved the entire swarm mean localisation error when compared to the fuzzy-based and round-robin Extended Kalman Filter-based localisation methods (by 67.10% and 59.28% respectively, at a swarm size of 150 AUVs). The proposed fuzzy-based and confidence-based localisation algorithms for cooperative underwater robotic swarms are validated on a co-simulation platform. A physics-based co-simulation platform that considers an environment's hydrodynamics, industrial grade inertial measurement unit and underwater acoustic communications characteristics is implemented for validation and optimisation purposes.

This thesis explores the use of drone-based data acquisitions for deriving structural plant traits and spectral reflectance of vegetation which are variables of interest for carbon stock estimations and for understanding vegetation functioning. Previous work demonstrated that fine-grained spatio-temporal insights gained from drone-acquired data are critical for understanding local processes but also for interpreting dynamics in coarse scale representations of landscapes from Earth observing satellite data. This work builds on this by assessing uncertainties in data acquisition and processing workflows and demonstrating novel applications of drone acquired data. The findings from the individual chapters presented herein allow conclusions on which metrics can be used with confidence to assess the status and track changes in vegetation structure over time. An example of how the finely resolved spatial information on structural plant traits can be used for simulations of radiation propagation, towards functional landscape representations is demonstrated. The presented work includes experiments conducted in the United Kingdom, Italy and Malaysia arising from international collaborations and addresses three critical questions in the proximal sensing of vegetation: 1. To what extent can drone Structure-from-Motion (SfM) photogrammetry-derived products deliver accurate information about vegetation height parameters? Using SfM photogrammetry for deriving robust vegetation height parameters is of particular interest when seeking to derive growing stock volume and biomass for plantation and forest management purposes and for the assessment of carbon stocks. The associated study focused on an oil palm plantation in Sarawak, Malaysia and examined the quality of SfM-based estimates of palm height and inferred stem height, both metrics which are commonly used for allometric estimates of biomass. Further, the impact of acquisition methodology on point cloud precision was investigated. Results showed that SfM could provide palm height metrics at the individual tree level with mean relative errors between 11.7% and 18.9% dependant on palm age and that for mature palms (>10 years) flight plans favouring coverage over spatial resolution and overlap did not decrease the accuracy. 2. How accurate and consistent are surface reflectance and vegetation index products acquired from drone-based sensors over vegetation canopies? Quantifying the spatial and temporal consistency of surface reflectance and vegetation index data acquired by lightweight sensors mounted on drone platforms is essential for applications in precision agriculture, for species classification and for studying vegetation functioning. This topic was addressed through two studies. The first study compared drone acquired spectral data over a maize field in Grosseto, Italy against reference datasets from near simultaneous airborne and satellite based image acquisitions. While uncertainties in drone acquired surface reflectance were found to be greater than anticipated (5-28% relative errors over the maize field), VIs were highly correlated and comparable across scales. The second study investigated the use of vegetation index data to track phenology related changes over time for mostly deciduous tree species in Cornwall, UK. VIs proved to be sufficiently consistent for both, acquisitions under overcast and cloud free skies to resolve phenological changes with illumination based uncertainties an order of magnitude smaller than the total increase in index values across Spring green-up. 3. Can drone-based data be used to constrain and drive models of radiative transfer for understanding photon-plant interactions in complex heterogeneous canopies? Combining drone acquired canopy height models and vegetation index information to represent vegetation in a 3D radiative transfer model represents a new opportunity for simulating the interaction of light with vegetation at fine spatial scales. Previously, the information required for modelling heterogeneous vegetation canopies could only be acquired through laborious measurements in-situ or financially costly laser-scanning methods. This topic was explored by creating a representation of a local wildlife conservation site in Cornwall, UK within the Discrete Anisotropic Radiative Transfer (DART) model. The model was used for a case study focused on simulating the photosynthetically active radiation (PAR) reaching the understory as hourly fractions and spatially explicit (1 m spatial resolution) daily light integrals across Spring green-up. Results showed that while the drone-data parameterised model could represent the variability across discontinuous vegetation cover, PAR reaching the understory was considerably overestimated at start-of-peak greenness due to uncertainties in modelled plant area density and leaf angular distributions. The primary data acquisitions of all the presented studies were performed exclusively with lightweight multi-rotor drones, trialling relatively low-cost consumer grade and multi-spectral cameras which have since been widely adopted by research groups globally. The presented results therefore represent a timely contribution with relevant insights from appropriate acquisition methodologies to novel applications of drone acquired data for representing vegetation in a radiative transfer modelling context.

Vegetation is a potential indicator of land contaminated by hydrocarbons. Remote sensing holds great potential for detecting stress in vegetation caused by the presence of hydrocarbons, but as yet this potential has not been realised, largely due to a lack of understanding of the biophysiological impacts of hydrocarbons on plant growth, and the specific spectral expression of them. Previous research has demonstrated success in laboratory simulation of hydrocarbon contamination and the analysis of the spectral and biophysical properties in plants, but has tended to always focus on scenarios with high concentrations of hydrocarbons (> 5 g/kg). It was assumed, that vegetation growing in hydrocarbon polluted areas would show a general decrease in plant health condition and show associated changes in reflectance. However, it is not known if lower concentrations of hydrocarbon in soils (< 5 g/kg) have the same effects as high concentrations and if the impacts observed at leaf scale would be observed at a canopy scale. The aim of this research has been to develop understanding at the leaf/plant scale initially, to then scale up to hyperspectral camera and simulated satellite-based scales to develop operational understanding that will benefit industry, regulators and environmental agencies alike. Two experiments were performed, one at plant scale with willow (Salix viminalis var. Tora) and maize (Zea mays var. Lapriora), and the other at canopy scale with turf grasses. At leaf scale, a pot-based experiment was designed with plants growing in pots containing a hydrocarbon (crude oil or refined oil) layer at the base of the pot with different concentrations levels (0.5 g/kg, 5 g/kg and 50 g/kg). At canopy scale, natural oil seep environments were simulated in 1 m by 0.5 m trays with a hydrocarbon layer with two different concentrations (50 g/kg and 0.5 g/kg) present only in the half of each tray and located at the base. Biophysical, reflectance and fluorescence measurements were made in both experiments using different portable spectrometers (MultispeQ, ASD Fieldspec Pro and Specim IQ camera).Biophysical variables (chlorophyll concentration, biomass and growth) decreased in the presence of high concentrations of hydrocarbon in the soil, showing a general reduction in all the studied parameters and in all plant species with respect to the non-polluted plants.26On the other hand, the presence of lower hydrocarbon concentrations (0.5 g/kg and 5 g/kg) revealed an increase in chlorophyll concentration, biomass of the entire plant (even in roots) and the growth rate, with respect to the non-polluted plants. To analyse the effect of different hydrocarbon concentrations on vegetation, changes in spectral response were analysed using continuum removal analysis and vegetation indices. Absorption features were identified related to pigment concentration (chlorophyll, carotenoids and anthocyanin) and biomass (starch, cellulose, lignin and water). Variations in absorption feature characteristics (band depth, band area and band width) depended on the hydrocarbon concentration in the soil, with increases in band depth and band area for low polluted treatments and decreases in the same characteristics for high polluted treatments. Vegetation indices in both plant and canopy scale experiments were able to discriminate between high pollution and low pollution in soil, as did the red edge. The red edge position showed movements towards longer wavelengths (improvements of health status of the plant) with the presence of low pollution in the soil and towards shorter wavelengths (reduction of health status) in high polluted treatments. At canopy scale a greener 'halo' surrounding a high polluted zone was seen, creating a spatial pattern previously observed in field campaigns in natural oil seep areas. To evaluate these outcomes at landscape scale using satellite imagery, reflectance data were convolved to different spectral resolutions related to three different sensors (WorldView- 2, Sentinel-2 and EnMap). The simulations showed that at the moment efforts need to continue to improve the sensor resolutions in order to monitor hydrocarbon polluted areas, or that a combination of sensors with varying resolutions may be required. In conclusion, the results of this study demonstrated biphasic behaviour of plants in hydrocarbon polluted environments and the potential of remote sensing in environmental monitoring and exploration of these areas.

This study aimed to develop a novel end-to-end plant diagnosis model for the analysis of plant health conditions in near real-time to optimize the rate of production on farmlands for an intensive, yet environmentally safe farming production to preserve the natural environment. First, field research was conducted to determine the extent of the problems faced by farmers in agricultural production. This allowed us to refine the research statement and the level of technology involved in the production processes. The advantages of unmanned aerial systems were exploited in the continuous monitoring of farm plantations to develop automated and accurate measures of farm conditions. To this end, this thesis applies the Precision Agricultural technology as a data based management system that takes into account spatial variations by using the Global Positioning System, Geographical Information System, remote sensing, yield monitors, mapping, and guidance system for variable rate applications. An unmanned aerial vehicle embedded with an optic and radiometric sensor was used to obtain high spectral resolution images of plantation status during normal production/growth cycle. Then, an ensemble of classifiers with Convolution Neural Networks (CNN) was used as off the shelf feature extractor to train images to develop an end-to-end feature detection and multiclass classification system for plant overall health's conditions. Whereby previous works have concentrated on using CNN as off the shelf feature extractor and model training to detect only plant diseases from plants. To date, no research has yet been carried out to develop an end-to-end model for the overall plant diagnosis system. Previous studies focused on the detection of diseases at any given time, making it difficult to implement comprehensive real-time PA systems. Applying the pretrained model to the new images showed that the model can accurately predict any plant condition with an average of 97% accuracy.

This DPhil is motivated by the impact that public data can have on the practise of sustainable finance. Disclosure guidance for environmental information is increasingly being used to compel companies to better manage their environmental risk exposure and environmental impacts. Data-driven technologies can 'unravel' company information related to environmental risk and are changing the information asymmetries between company managers, their investors, and society. Three novel and rigorous data science papers have been prepared which provide new access to company environmental risk information. The first paper develops a global inventory of photovoltaic solar energy facilties using machine learning and earth observation and then uses this dataset to assess land cover impacts due to renewable energy siting decisions. The second paper develops asset-level data arrangements of the coal, oil, and gas supply chains, and identifies concentrations of environmental transition risk in the complex network topology. The third paper advances earth observation and computer vision technical methods, assessing the power of these techniques for producing company asset-level data even with limited training data. In developing these papers, this DPhil has identified the opportunities and limits that technology might have in the provision of information for public and private governance of environmental risk and impact. These limits add clarity to the role that policy must play in meeting the diverse information needs of both company shareholders and stakeholders. A discussion chapter provides commentary on these limits; on how the needs of both shareholders and stakeholders can be met with company risk disclosure at the asset-level; and how innovation might be fostered in markets for environmental information. Policy recommendations are summarised and an agenda for further research is provided.

A steadily increasing number of studies have indicated the multidimensional relationships between exposure to greenspace and human health. These previous studies applied diverse, spatially explicit methods and datasets when measuring and modelling greenspace exposure. However, they often did not consider the key underlying principles of geospatial science, such as spatial scale. Potential uncertainties regarding the estimation of greenspace exposure and modelling relationships between greenspace and human health are inherent in such studies. Therefore, research to investigate the spatial dimensions (e.g., scale, data, methods) of exposure to greenspace is warranted. Previous studies measuring greenspace exposure have usually only considered individual exposure types (i.e., availability, accessibility, and visibility). Consequently, there are no existing methodological approaches that standardise and harmonise these three individual metrics to produce a composite estimate of aggregate greenspace exposure. Additional research is essential to develop metrics for greenspace exposure that can reduce the multiplicity of current heterogeneous approaches to greenspace exposure estimation. The four papers foundational to this thesis encompass different methodological approaches and used diverse spatial data to address existing challenges in greenspace exposure assessment. Greater Manchester in the UK was used as the case study area. However, the methods are intended to be transferable to other urban areas. The first paper provides a systematic review of the spatial dimensions considered in previous studies evaluating the associations between greenspace and human health. These preceding studies used a wide range of spatial data, scales of analysis, and buffer distances, resulting in the likelihood that their results were vulnerable to scale effects. Furthermore, these studies represented a heterogeneous collection of study designs. Such heterogeneity of study design may have caused ambiguities in understanding the associations between greenspace and health amongst existing studies. There is a need for new methodological solutions to overcome the existing challenges. The second paper investigated the scale effects inherent in three commonly used metrics of greenspace availability exposure (i.e., NDVI, leaf area index, and land use land cover). Applying lacunarity analysis, this paper identifies the scale sensitivity and the upper limit of the acceptable scale range beyond which these greenspace metrics are vulnerable to scale effects. Using lacunarity weightings, this study also provides a new methodological solution for measuring greenspace availability exposure that combines multiple, properly weighted metrics to reduce the scaling effect. This novel metric offers a unique solution for measuring greenspace availability at high resolutions, independent of scale. The third paper provides an innovative methodological approach to modelling eye-level greenspace visibility using a fusion of digital elevation and land use-land cover data. This study applied a classic viewshed analysis for visibility assessment to more than 86 million observer locations to achieve high resolution and granularity in eye-level exposure assessment. This paper also explores the differences between eye-level and top-down exposure metrics, and the results showed that these two exposure measures are complementary but distinct from one another. The fourth paper develops and evaluates a new methodological approach to measuring greenspaces accessibility at a high spatial resolution. This novel methodological approach allows the availability, accessibility, and visibility metrics of greenspace exposure to be combined to produce a composite greenspace exposure index (CGEI). This CGEI, in turn, introduces the possibility of standardising and harmonising greenspace exposure estimates. The paper also applies the newly developed exposure metrics to investigate variations in greenspace exposure in relation to socioeconomic inequalities and years of potential life lost (YPLL) at the neighbourhood scale. The results of this research indicate the composite greenspace exposure metric is a superior metric for green space exposure estimation and is better at explaining the variability in YPLL compared to any given single exposure metric. It also further supports previous research findings that greenspace exposure spatially co-varies with socioeconomic inequalities and that increased exposure to greenspace is significantly associated with lower YPLL. Overall, this thesis suggests that future studies should critically consider spatial dimensions in the context of exposure assessment. Further studies are required to continue to improve the current understanding of the relationships between greenspace exposure and health.

Coastal communities and land covers are vulnerable receptors of erosion, flooding, or both in combination. The accurate, automated, and wide-scale determination of shoreline position, and its migration at the engineering scale (10-1 - 102 km), is imperative for future coastal risk adaptation and management. The recent increase in the acquisition and availability of Big Datasets, including multispectral remote sensing imagery, is providing new opportunities to monitor engineering scale rates of shoreline change and other constituents of coastal risk, including changes to human coastal population densities. This increase in data availability comes with novel challenges to devise and utilise methods to store, process, analyse and extract information from these Big Datasets. This thesis assesses the suitability of different Big Data approaches, namely Machine Learning (ML) and non-ML based tools, for the automated extraction of the coastal vegetation edge in remote sensing imagery. Compared to the instantaneous waterline, few vegetation edge methods have been developed and analysis of the coastal zone processes that can be detected using the shoreline proxy remain understudied. This thesis initially investigates whether non-ML methods are suitable for the extraction of the coastal vegetation edge from multispectral remote sensing imagery. A novel non-ML tool is introduced and applied, CoasTool, which considers the proximity of the instantaneous water line during vegetation edge extraction. CoasTool performance is compared to the outputs derived from well-established threshold contouring techniques and kernel-based methods as well as one form of ML, Support Vector Machines (SVM). Limitations in the performance of these tools, particularly along shorelines with discontinuous or graded vegetation boundaries, provide justification for the application of a separate form of ML, convolutional neural network (CNN), to this task. A novel CNN, VEdge_Detector, is trained and applied to extract the coastal vegetation edge and its outputs are compared to ground-referenced measurements and manually digitised vertical aerial photographs. VEdge_Detector is applied to a time series of vi images to detect annual to decadal scale shoreline dynamics discernible using the coastal vegetation edge. Shoreline change constitutes one element of coastal risk, and this thesis subsequently investigates the viability of integrating multiple ML-derived datasets, pertaining to different aspects of risk, to calculate relative coastal population exposure to shoreline change. The Guiana coastline, northern South America, is one of the most dynamic stretches of coastline in the world and a region where greater than 90% of its population live below 10 m elevation. The identification of locations where coastal populations are at greatest risk to coastal retreat in this region is thus very important to inform coastal risk management decisions. Accordingly, decadal-scale rates of shoreline change calculated using VEdge_Detector derived shoreline positions are combined with secondary, ML-derived, population datasets (WorldPop). The integration of the two ML-based datasets aids the identification of population exposure hotspot locations and discover, previously unpublished, locations where forced migration due to shoreline change has occurred. In concluding, the relative merits and drawbacks of using ML verses non-ML techniques to detect the coastal vegetation edge are discussed as well as considering the suitability of the coastal vegetation as a proxy of shoreline position. Further discussion is given on the different considerations coastal stakeholders will have when choosing the most suitable tool to use in shoreline detection tasks, including tool performance, speed, transparency, and ease of use. Remaining research gaps and future research requirements are emphasised, including the need for collaboration between different research institutions to suitably train and apply ML tools in the geosciences.

Nitrous oxide (N2O) is a potent greenhouse gas mainly produced by microbial processes in the soil. Anthropogenic N2O is principally emitted from soils after nitrogen fertiliser and manure applications on agricultural land. This thesis focuses on emissions from grazing systems, which are known to be the largest source of uncertainty in global and national N2O emission inventories. Nitrogen-rich excreta deposits from grazing livestock are recognised as hotspots of N losses (N2O emissions in particular). The non-uniform distribution of these emissions hotspots within a typical field contributes significantly to the spatial heterogeneity of emissions often observed in addition to the natural variability of soil properties within the field such as pH, moisture and nutrient availability. However, it is extremely difficult to characterise the spatial and temporal pattern of these grazing inputs other than through the use of demanding and costly approaches such as manual observation or animal based-sensors. Two separate experiments were conducted during this study, in Scotland on sheep grazed grasslands and in Ireland on a dairy cow grazed grassland. Both sites were commercially used and were intensively managed with a nitrogen fertiliser application rate of 225 kg ha-1 yr-1 and 261 kg ha-1 yr-1, respectively. In Scotland, at Easter Bush fields the experiment was conducted during a 9 month campaign of gas, soil and grass sampling over the grazed field to study the spatial and temporal variability of the fluxes and soil properties to improve up-scaling of the fluxes from the plot scale to the field scale. In Ireland, at the Johnstown Castle farm, the experiment was conducted during an 11 month campaign on an experimental plot excluded from grazing. At the Scottish site, gas, soil and grass samples were collected regularly on soil which received different treatments within a randomised block design (e.g. urine deposition, fertiliser application, urine and fertiliser application or no N addition as a control). At both sites, Remotely Piloted Aerial System (RPAS) imagery was collected to study the spatial variability of the grass growth with the aim to map excreta depositions over the whole field. The Scottish site was used as a proof of concept of the method and the method was then used weekly on the Irish site over the entire grazing season. More generally, this thesis details the novel use of remote sensing techniques using high-resolution cameras linked to RPAS to improve our understanding of the spatial and temporal patterns of excreta deposition. This method proved to be repeatable for future studies as it can be automated, is easily deployable in the field, low-cost and the measurements are non-destructive (i.e. has no influence on the soil, vegetation or livestock). Excreta depositions contribute to very high emissions of N2O from relatively small areas of soil and can vary throughout the growing season in response to climatic conditions. Therefore, mapping of the excreta nitrogen inputs to the field facilitated a more accurate estimation of the annual field-scale N2O emission from grazing grasslands. Both experiments conducted in this study showed a high spatial and temporal N2O emissions variability due to the nature of N2O production within the soil and high variability of the soil properties (soil pH, soil moisture content, soil temperature) which influence the microbiological processes. Interaction on N2O emissions between fertiliser application and urine deposition was proved to be statistically significant and the magnitude of the interactions depended on the time of application within the year. The results showed a link between the variability of the emission factors of excreta deposition and fertiliser application and to the variation in weather conditions. This technique can be employed to up-scale emissions to a national level. This study plays a part in the on-going development of precision agricultural tools, based on image analysis of the grass sward to mitigate emissions from grazed grassland. Possible mitigation approaches, based on the methods presented in this thesis, include the use of RPAS technology to deliver nitrification inhibitors to newly deposited excreta within the field to reduce the potential nitrogen losses to the environment. This research indicates the future potential to better adjust fertiliser application using variable-rate fertiliser applications matching the vegetation nitrogen needs and limit nitrogen losses. This thesis identifies opportunities to develop innovative approaches to N2O mitigation by better evaluating emission estimations from agricultural practices, which could then be implemented in the national and global greenhouse gas inventories established by the Intergovernmental Panel on Climate Change.

Rapid advancements in Earth observation, localised data collection, and modelling techniques are revolutionising global understanding of disaster risk. If properly validated and appropriately used, these emerging techniques for high-resolution geospatial analysis can help developing countries overcome challenges pertaining to the first priority of the Sendai Framework for Disaster Risk Reduction 2015-2030 (SFDRR): understanding disaster risk. This thesis focuses on three potential breakthroughs in risk information for developing countries: (i) mapping population exposure; (ii) demonstrating more granular insights into population vulnerability; and, (iii) measuring the extent of flood events. For mapping population exposure, a simplified equation is presented that separates masking variables and density variables. The equation focuses on two distinct objectives: (i) increasing the precision and accuracy of a strong binary mask to limit the spatial extent of settled populations; and, (ii) improving modelling techniques of ancillary variables to account for variations in density within an identified spatial extent. A critical assessment of population modelling techniques that heavily rely on derived characteristics of the built environment as a masking variable is also presented. The assessment confirms that density variables directly relating to the binary masking variable are effective; however, their application requires careful consideration due to its propensity to amplify both positive results and errors. The research provides evidence in support of hybrid techniques towards global population modelling. For demonstrating more granular insights into population vulnerability, a novel methodology is presented for generating spatially referenced synthetic households by combining population synthesis techniques for small area estimation with dasymetric population modelling. The methodology results in an output vector layer of households represented as points with unique structural and socioeconomic attributes which can be applied towards a better understanding of population characteristics, particularly those which are geographically influenced and may not conform to administrative boundaries linked to aggregated census statistics. A case study of Dumuria Upazila, Bangladesh, validates the methodology and provides an example of how georeferenced surveys can be applied towards error assessments in data-scarce regions, which remains a challenge in global population modelling. Lastly, for measuring the extent of flood events, a novel methodology applies freely available synthetic aperture radar (SAR) data for flood detection over large geographic areas immediately following a cyclone hazard event. The methodology results in enhanced precision, demonstrated in coastal districts of Bangladesh heavily affected by 2019 Cyclone Fani. The case study also demonstrates how high-resolution geospatial analysis can supplement locally reported estimates of damages by providing quicker estimates and a means for identifying potential bias from self-reporting. Furthermore, insights into the socioeconomic characteristics of potentially affected households provide a layer of information not covered by post-disaster needs assessments.

This thesis focuses on the seismic hazard of the NE Tibetan Plateau, an area that hosted one of the largest and most deadly earthquakes in China's recorded his- tory. In order to estimate rates of earthquake occurrence from strain rates, it is necessary to use the Gutenberg-Richter relationship and assume a maximum magnitude. The largest recorded earthquake in the NE Tibetan Plateau is the 1920 Haiyuan Earthquake. It has often been quoted as of M 8.7, on a magnitude scale of uncertain meaning, a result of unknown origin and an outlier in empirical magnitude and rupture parameter relationships. Therefore, I first use historical seismograms, modern Pleiades imagery and field-acquired drone data to re-estimate the magnitude of the 1920 Haiyuan Earthquake, which is corrected to MW 7.9±0.2. Literature re- view suggests the possibility that many magnitudes of historical Chinese earthquakes based on the intensity and magnitude scaling relationships could be systematically overestimated. An existing conversion method is highlighted for quick magnitude re- estimations for historical earthquakes, though careful systematic revisits to historical earthquake magnitudes are recommended. Second, 10 ascending and 13 descending frames of Sentinel-1 interferometric synthetic aperture radar (InSAR) data acquired between 2014 to 2019 and covering the NE Tibetan Plateau are analysed to determine crustal velocities in the satellite line of sight. These frame-based line-of-sight velocities are combined with constraints from Global Navigation Satellite System (GNSS) velocities to derive a high-resolution 3-dimensional velocity field for the region in a well-defined frame of reference. The crustal velocities can be used to constrain geodetic slip rates across the Haiyuan, Gulang and East Kunlun Faults, creep time series over the Laohushan Section of the Haiyuan fault, as well as enabling the determination of crustal strain rates that can be used to determine the rates of earthquake occurrence for the region.

This thesis seeks to add to the study of the relationship between land surface temperature (LST) and urban land cover by presenting a method to project Landsat LST data from the satellite overpass time (9:40 am) to a local peak of temperature (estimated to be around 1:15 pm locally), to investigate the impact of the time of image acquisition on modelling the spatial and temporal variations of LST. Additionally, it would also verify the effects of extreme temperature to reach more representative seasonal images. The study uses remote sensing data extracted from Landsat 5 and 8 (30 m resolution) and the Spinning Enhanced Visible and Infrared Imager LST products (SEVIRI 3 km resolution), in addition to LST-based measurements collected from the ground. The study presented a method to convert Landsat images to be estimated during local peaks in LST with an accuracy of: standard error of 1.7°C and an R of 0.82 in comparison with actual ground-based measurements. This allowed an investigation of the effects of time of day on the spatial and temporal variation of LST, where it was found that this factor has clearly affected the relationship between LST and urban land cover. Similarly, the time of day has caused differences in estimating LST change over several years. It is also found that the extreme values of temperature can affect the trend of LST temporal variation, and which can be minimized by using the images in the form of the average of seasonal images for each year rather than images being used in a standalone manner. This study contributes to the improved study of LST by minimizing the uncertainty that can occur because of the angle of the sun and associated factors such as shadows, which has long been a controversial issue among researches due to the lack of appropriate satellite data.

Variation in the rate of meltwater input into the subglacial system of the Greenland Ice Sheet can force dynamic responses on a range of scales from hourly to interannual. Observations of the ice sheet dynamic response are commonly made either through ground-based Global Navigation Satellite System (GNSS) measurements, which can provide continuous and accurate point measurements, or through satellite remote sensing, which can provide regional-scale observations but at coarse temporal resolutions. This thesis investigates the potential of Uncrewed Aerial Vehicles (UAVs) to provide intermediate-level observations of the interactions between ice sheet hydrology and dynamics at a fast-flowing, marine terminating glacier in West Greenland. I first describe the development of a low- cost UAV suitable for deriving ice sheet velocity fields from Structure-from-Motion photogrammetry. In order to geolocate products without using ground control, image locations are determined directly using an on-board L1 GNSS receiver. I validate this method, showing that accuracies are sufficient for producing velocity fields in the ice sheet interior. Next, this method is used, alongside in-situ geophysical observations, to characterise the causes and dynamic influence of a rapid supraglacial lake drainage. I show that rapid drainage can induce a significant dynamic response up to 4 km away from the lake itself, and that fracture history can exert controls on interannual lake drainage behaviour. Finally, I upscale UAV ob- servations using satellite datasets over a ~3,000 km2 area, exploring dynamic controls on crevasse hydrology. I find that in compressive mean stress compressive regimes, crevasses are more likely to display ponding and rapid hydrofracture than in extensional regimes, where continuous slow drainage is typical. Continued high-resolution observations are necessary to further identify key controls on the hydrological influences of Greenland Ice Sheet dynamics.

Forest height is one of the most important forest biophysical parameters influencing light competition, stand productivity and carbon sequestration. This thesis considered investigating the existing and new methods for estimating forest height. The specific focus of the research project concerned the improved estimation of forest height (h) over tropical forest from both single- and multi-source remotely sensed datasets. The potential applications of single-baseline (B) PolInSAR, fusion of multi-baseline PolInSAR and LiDAR, and a combined use of polarimetric SAR and LiDAR were assessed. The single-baseline PolInSAR demonstrated the limitation of the model for estimation of forest height over a heterogeneous forest because of the lack of appropriate values of the vertical wavenumber. For the smallest baseline (B = 20m), height is overestimated (~ 5m) for short stand (h < 10m); and for the largest baseline (B = 120m), height is underestimated (~ 30m) for tall stand (20 < h < 60m). As conventional PolInSAR, small baselines are suitable for estimation of larger stand heights, whereas large baselines are for shorter stands. Therefore, for an improved result over a heterogeneous forest with different height ranges (0-60m), a multi-baseline merging approach was introduced based on a fusion of PolInSAR with LiDAR. The results demonstrated improvement (r2 = 0.81, RMSE = 7.1m) in comparison to PolInSAR alone (r2 = 0.67, RMSE = 9.2m). Whilst this fusion indicated improvement, the model performance is limited by the availability of the PolInSAR data. Therefore, a second data fusion approach was introduced using polarimetric SAR and LiDAR. The obtained results are significant (r2 = 0.70, RMSE = 10m) as the technique relies on polarimetric measurements and not interferometric. Overall, the research results demonstrate the capabilities of SAR and LiDAR data fusions in estimation of tropical forest height, adding value to both scientific understanding and management of forest ecosystems.

In this novel research, the use of optical and SAR images for separate and joint investigations were explored, with the aid of machine learning classifiers for detecting hydrocarbon spill impact on cropland, grassland and dense forested vegetation types. Optical image spectral bands across the Visible, Near-Infrared and Shortwave-Infrared spectrum, various vegetation health indices (including NDVI, NDWI, LAI and SAVI) and SAR derived variables (including backscatter, coherence and textural variables) were used to detect and map oil spill sites within cropland, grassland and TCA vegetation types. Results generally showed that the integration of multi-frequency L, C and X band SAR in the wet (summer) season yielded the best overall classification accuracy in discrimination of polluted and oil-free vegetation types. An overall accuracy (OA) of 82.3%, 66.67% and 70.93% were obtained for Cropland, Grassland and Tree Cover Areas (TCA) vegetation types, respectively. The accuracies recorded were significantly better (P > 0.05) than when the spill was classified using optical imagery only and when integrated optical and SAR image variables are classified. These results were further corroborated by the multi-temporal Sentinel – 1 backscatter analysis, which showed that mean backscatter difference between polluted and oil-free cropland and grassland vegetation are significantly different (P > 0.05) in the wet season than in the dry season. Furthermore, the new fuzzy forest method used for multi-frequency (C and X band) SAR and Optical variable reduction was able to achieve a good result in addressing high dimensionality in cropland and grassland vegetation. This research demonstrates that SAR based monitoring of petroleum hydrocarbon impacts on vegetation is feasible and has high potential for oil pipeline monitoring and facility management. The research also presented a new paradigm into terrestrial oil spill detection, which largely can replace the use of optical data.