Search

Canopy height is a vital indicator to asses carbon uptake and productivity of forests. However, precise measurements, such as from airborne or spaceborne 3D laser scanning (LiDAR), are expensive and usually cover only small areas. In this work, we propose a novel deep learning model that can generate detailed maps of tree canopy heights. In contrast to previous approaches that use a single image as input, we process multi-temporal data via a an adaptation of the popular U-Net architecture that is based on the EfficientNet and 3D convolution operators. To that end, our model receives multi-spectral Landsat satellite imagery as input and can predict continuous height maps. As labeled data, we resort to spatially sparse LiDAR data from ICESat-2. Thus, with such a model, one can produce dense canopy height maps given only multi-spectral Landsat data. Our experimental evaluation shows that our our model outperforms existing and improved single-temporal models. To test generalizability, we created a non-overlapping dataset to evaluate our approach and further tested the model performance on out-of-distribution data. The results show that our model can successfully learn drastic changes in distribution.

Wind turbines in northern Europe are frequently placed in forests, which sets new wind resource modelling requirements. Accurate mapping of the land surface can be challenging at forested sites due to sudden transitions between patches with very different aerodynamic properties, e.g. tall trees, clearings, and lakes. Tree growth and deforestation can lead to temporal changes of the forest. Global or pan-European land cover data sets fail to resolve these forest properties, aerial lidar campaigns are costly and infrequent, and manual digitization is labour-intensive and subjective. Here, we investigate the potential of using satellite observations to characterize the land surface in connection with wind energy flow modelling using the Wind Atlas Analysis and Application Program (WAsP). Collocated maps of the land cover, tree height, and leaf area index (LAI) are generated based on observations from the Sentinel-1 and Sentinel-2 missions combined with the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2). Three different forest canopy models are applied to convert these maps to roughness lengths and displacement heights. We introduce new functionalities for WAsP, which can process detailed land cover maps containing both roughness lengths and displacement heights. Validation is carried out through cross-prediction analyses at eight well-instrumented sites in various landscapes where measurements at one mast are used to predict wind resources at another nearby mast. The use of novel satellite-based input maps in combination with a canopy model leads to lower cross-prediction errors of the wind power density (rms Combining double low line 10.9 %-11.2 %) than using standard global or pan-European land cover data sets for land surface parameterization (rms Combining double low line 14.2 %-19.7 %). Differences in the cross-predictions resulting from the three different canopy models are minor. The satellite-based maps show cross-prediction errors close to those obtaine

Satellitbaserede datalag kan gøre det lettere at bestemme den årlige energiproduktion for kommende vindmølleparker på land. Det kan føre til en lavere usikkerhed og besparelser i millionklassen.

Tropical environments present a unique challenge for optical time series analysis, primarily owing to fragmented data availability, persistent cloud cover and atmospheric aerosols. Additionally, little is known of whether the performance of time series change detection is affected by diverse forest types found in tropical dry regions. In this paper, we develop a methodology for mapping forest clearing in Southeast Asia using a study region characterised by heterogeneous forest types. Moderate Resolution Imaging Spectroradiometer (MODIS) time series are decomposed using Breaks For Additive Season and Trend (BFAST) and breakpoints, trend, and seasonal components are combined in a binomial probability model to distinguish between cleared and stable forest. We found that the addition of seasonality and trend information improves the change model performance compared to using breakpoints alone. We also demonstrate the value of considering forest type in disturbance mapping in comparison to the more common approach that combines all forest typesinto a single generalised forest class. By taking a generalised forest approach, there is less control over the error distribution in each forest type. Dry-deciduous and evergreen forests are especially sensitive to error imbalances using a generalised forest model i.e., clearances were underestimated in evergreen forest, and overestimated in dry-deciduous forest. This suggests that forest type needs to be considered in time series change mapping, especially in heterogeneous forest regions. Our approach builds towards improving large-area monitoring of forest-diverse regions such as Southeast Asia. The findings of this study should also be transferable across optical sensors and are therefore relevant for the future availability of dense time series for the tropics at higherspatial resolutions.

Despite the importance of tropical forest ecosystems, they continue to be transformed at an alarming rate. In Southeast Asia, the historical deforestation narrative of a growing population gradually encroaching upon forest land is being replaced by the dominating influence of large-scale plantations. In particular, the global demand for natural rubber (Hevea brasiliensis) has been reported as the cause of widespread forest conversion. A critical component of forest conservation strategies, such as Reduced Emission from Deforestation and forest Degradation (REDD+), relies upon the monitoring of forest cover using satellite remote sensing technology. Recently, there has been a shift in data protection policy where rich archives of satellite imagery are now freely available. This has spurred a new era in satellite-based forest monitoring leading to advancements in optical time series processing and open source analysis methods. This thesis explores the utility of optical time series to monitor forest loss in Southeast Asia, with a specific focus on Cambodia. Methodological objectives focus on testing two distinct forest monitoring approaches using 1) annual Landsat time series and LandTrendr, and 2) intra-annual MODIS time series and Breaks for Additive Season and Trend (BFAST). Aspects of data quality and the influence of forest type on time series analysis are explored, as well as working towards a methodological framework for integrating Landsat and MODIS time series for enhanced forest monitoring systems. Thematic objectives of the research focussed on estimating forest loss in Cambodia in the post-2000 era, determining how much of this loss was caused by conversions to natural rubber tree cover, and analysing if there is a link between forest-to-rubber conversion rates and global rubber markets. At the Landsat 30-m resolution, annual time series coupled with linear segmentation usi

One of the prerequisites of the REDD+ mechanism is to effectively predict business-as-usual (BAU) scenarios for change in forest cover. This would enable estimation of how much carbon emission a project could potentially prevent and thus how much carbon credit should be rewarded. However, different factors like forest degradation and the lack of linearity in forest cover transitions challenge the accuracy of such scenarios. Here we predict and validate such BAU scenarios retrospectively based on forest cover changes at village and district level in North Central Vietnam. With the government’s efforts to increase the forest cover, land use policies led to gradual abandonment of shifting cultivation since the 1990s. We analyzed Landsat images from 1973, 1989, 1998, 2000, and 2011 and found that the policies in the areas studied did lead to increased forest cover after a long period of decline, but that this increase could mainly be attributed to an increase in open forest and shrub areas. We compared Landsat classifications with participatory maps of land cover/use in 1998 and 2012 that indicated more forest degradation than was captured by the Landsat analysis. The BAU scenarios were heavily dependent on which years were chosen for the reference period. This suggests that hypothetical REDD+ activities in the past, when based on the remote sensing data available at that time, would have been unable to correctly estimate changes in carbon stocks and thus produce relevant BAU scenarios.

This chapter summarizes approaches to the detection of dryland vegetation change and methods for observing spatio-temporal trends from space. An overview of suitable long-term Earth Observation (EO) based datasets for assessment of global dryland vegetation trends is provided and a status map of contemporary greening and browning trends for global drylands is presented. The vegetation metrics suitable for per-pixel temporal trend analysis is discussed, including seasonal parameterisation and the appropriate choice of trend indicators. Recent methods designed to overcome assumptions of long-term linearity in time series analysis (Breaks For Additive Season and Trend(BFAST)) are discussed. Finally, the importance of the spatial scale when performing temporal trend analysis is introduced and a method for image downscaling (Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM)) is presented.

This chapter summarizes methods of inferring information about drivers of global dryland vegetation changes observed from remote sensing time series data covering from the 1980s until present time. Earth observation (EO) based time series of vegetation metrics, sea surface temperature (SST) (both from the AVHRR (Advanced Very High Resolution Radiometer) series of instruments) and precipitation data (blended satellite/rain gauge) are used for determining the mechanisms of observed changes. EO-based methods to better distinguish between climate and human induced (land use) vegetation changes are reviewed. The techniques presented include trend analysis based on the Rain-Use Efficiency (RUE) and the Residual Trend Analysis (RESTREND) and the methodological challenges related to the use of these. Finally, teleconnections between global sea surface temperature (SST) anomalies and dryland vegetation productivity are illustrated and the associated predictive capabilities are discussed.

Shifting cultivation has long been a major livelihood for people in the miombo woodlands of southern, central and eastern Africa. However, increasing deforestation and forest degradation throughout the region are resulting in growing pressure on traditional shifting agricultural systems. Indeed, agricultural intensification and expansion itself is considered the primary cause of miombo deterioration, which is driven by both endogenous and exogenous variables operating at various scales. On the basis of data collected in the 1990s and 2010 from two villages in Northern Province, Zambia and two in the Rukwa Region, Tanzania, the paper will document the transition of shifting cultivation towards more intensive land use practices. It is argued that the main drivers influencing miombo degradation, and thereby the transition process of traditional shifting cultivation practices, have been a growing population, government policies, and an increasing commercialization/market integration. Questionnaires, focus group meetings, and in-depth interviews reveal that despite the breakdown of the traditional shifting cultivation practices, a general improvement of livelihoods has taken place. This has happened through adaptation and diversification in both agricultural practices and livelihood activities. However, it is also seen that because of the often rapidly changing external factors (market conditions and policies), life in the shifting cultivation communities involves a continual shift of emphasis among a variety of livelihood strategies.

The I-REDD+ team has opted for a challenging framework – targeting REDD-related problems in mountainous regions of Southeast Asia and putting a focus on forest degradation processes. The combination of • a region with heavy cloud cover during much of the year, • pronounced illumination differences due to steep terrain, • phenological gradients across natural, agricultural and forestry ecosystems including plantations and • the need to serve the REDD-specific context of deforestation and forest degradation across spatial and temporal scales make remote sensing based approaches particularly challenging in I-REDD+. Accordingly, this document serves different purposes: The first part reflects on the regional setting and intends to illuminate the regional specifics related to agriculture, forested land and land use at large. It is an important prerequisite for a deeper understanding of what can be expected from remote sensing imagery and the provided information shall help to better anticipate problems that will be encountered when acquiring, analyzing and interpreting remote sensing data. Beyond remote sensing, it may be a good point of departure for a large group of scientists with a diverse background and different regional expertise. I-REDD+ can luckily draw from the rich expertise of its regional partners including their extensive fieldbased knowledge. For every test region, we provide information collected by local experts on physiogeography, economy, socio-cultural setting, stakeholders and governance, and deforestation and forest degradation processes. The second part summarizes the available literature on remote sensing based good practices for REDD. It largely draws from the documents of the Intergovernmental Panel on Climate Change (IPCC), the United Nations Framework Convention on Climate Change (UNFCCC) and the Global Observation of Forest and Land Cover Dynamics (GOFC-GOLD) methods sourcebook. These documents provide a generic f