Search

Mapping vegetation properties globally is crucial to understand the role of terrestrial ecosystems in the global carbon cycle. Spatially explicit, high-resolution data are needed to manage terrestrial ecosystems so that climate change can be mitigated and biodiversity loss prevented. Since no current single data source can provide such data with global coverage and high spatial resolution, new solutions must be found. This thesis aims to develop novel data-driven tools based on state-of-the-art deep learning to advance the mapping of vegetation properties, in particular canopy height, at global scale. Two ongoing space missions, namely the Copernicus Sentinel-2 mission and NASA’s GEDI LIDAR mission, deliver publicly available data that form the basis of the methods presented in this thesis. While GEDI is a key climate mission that provides sparse vegetation structure measurements at global scale (between 51.6° N & S), Sentinel-2 delivers dense optical images with global coverage, but cannot directly measure vertical vegetation structure. The presented work is a holistic approach based on gradually extended methods towards the large-scale fusion of Sentinel-2 and GEDI for the global mapping of canopy top height with high spatial resolution. Furthermore, since transparency of the modelling limitations is critical to build trust and to inform downstream applications about the reliability of the estimates, probabilistic deep learning techniques are integrated to quantify the predictive uncertainty. In a first step, a novel approach based on deep convolutional neural networks (CNNs) was developed to estimate dense canopy height maps from Sentinel-2 optical images by training with local dense reference data from airborne measurement campaigns (LIDAR and photogrammetry) in Gabon and Switzerland. By exploiting textural image features, the model achieved low error, even for canopies up to 50 m height. However, its applicability is limited to regions represented by the available training data. The launch of the spaceborne GEDI full waveform LIDAR in December 2018 promised to provide sparse reference data of vegetation structure measurements at global scale. Since interpreting on-orbit GEDI LIDAR waveforms proved to be a difficult task due to unknown noise in the data, a novel probabilistic deep learning approach was developed to retrieve canopy top height globally and quantify the predictive uncertainty from GEDI. Given these footprint-level estimates, the Sentinel-2 based canopy height mapping approach could be extended to be trained with sparse supervision. After demonstrating that this approach allows to estimate canopy top height suitable to map indicative high carbon stocks in tropical Southeast Asia, a global probabilistic model was developed to retrieve canopy top height anywhere on Earth. Ultimately, the first global, wall-to-wall canopy top height map at 10 m ground sampling distance was computed for the year 2020.

Increasing interest in characterising carbon dynamics in the agricultural landscape has been driven by the clear need for greenhouse gas mitigation strategies in response to global climate change. However, little quantitative information is available about carbon stocks and processes in agro-ecosystems. Better understanding of carbon stocks and processes is of particular importance in New Zealand, where farmland occupies large areas across the country and ongoing change in farming practices, land management and intensification are affecting carbon stocks and processes and ultimately, agricultural sustainability. This study aimed to quantify and understand variation in above- and below-ground carbon stocks and processes in shelterbelts and paddocks at multiple scales within the agro-ecosystem of Canterbury, New Zealand. In an initial pilot study across four farms showed that variability in above- and below-ground carbon stocks among-farms was smaller than between shelterbelt and paddocks. Using this knowledge to determine appropriate sampling intensity, I quantified carbon stocks and processes within shelterbelts across 34 farms, focussing particularly on understanding how within-shelterbelt-scale (e.g. soil moisture, pH), shelterbelt-scale (e.g. shelterbelt type, age, and tree biomass), farm-scale (e.g. land use) and landscape-scale (e.g. soil type) factors was related to variability of these variables. Total above- and below-ground carbon pools in shelterbelts were similar for exotic and native shelterbelts (152.6 ± 131 t C ha-1 and 110.8 ± 128 t C ha-1, respectively [mean ± standard errors]), although native shelterbelts were on average half the age of exotic shelterbelts (16 and 27 years, respectively). The above-ground biomass carbon pool represented about 62% and soil carbon pool 33% in both native and exotic shelterbelts. Most variability in carbon stocks was explained at the shelterbelt scale by shelterbelt type, age and size, whereas farm scale played a minor role and soil type was unimportant. Carbon cycle processes varied primarily at the shelterbelt and paddock level. Leaf litter decomposition and microbial activity were twice as high and invertebrates were more active in native shelterbelts compared to exotic shelterbelts, and variability in these indicators were best explained by within-shelterbelt factors (e.g., soil moisture, pH, organic and labile carbon) and shelterbelt characteristics (age, biomass, and vegetation type). Therefore, above-ground shelterbelt-scale factors drive carbon cycle process indicator rates and biological activity in the Canterbury agro-ecosystem, rather than land use or landscape factors. These field data were used in GIS-based spatial carbon model which expored and evaluated a number of agricultural land use, cover and management scenarios and their impact on soil carbon stocks over the enxt 10 years (2014 – 2024) in combination with quantifying the amount of shelterbelt area needed to offset the potential soil carbon loss. While the model default settings predicted an increase in soil carbon stocks, the literature based alternative scenarios, commonly predicted a decrease of 10%, 6%, 34% in soil carbon stocks in the alternative dairy, sheep and beef and arable scenarios. This study shows that native shelterbelts in agro-ecosystems have considerable potential for increasing carbon pools and enhancing carbon cycle process, particularly as they age through time. Contrary to current assessments, this study was not able to provide definate evidence on the effect of further intensification and land clearing on carbon stocks in Canterbury. Future studies should include long-term investigation of carbon stocks and processes across a wider variety of shelterbelt and farm types.

Aluminium (Al) toxicity is a key factor limiting pasture production in the high country of New Zealand. Many high country soils are acidic, with a pH less than 5.5. Low soil pH makes Al soluble, leading to high concentations of exchangeable Al in soils and plant Al toxicity. Aluminium toxicity reduces plant vigour and yields; resticting root growth and limiting nutrient and water uptake. The harsh climatic conditions experienced in the high country could also play a role in the exacerbation of Al toxicity. Literature on the influence of low moisture extremes on exchangeable Al concentrations is scarce. Nomad white clover was grown for 10 months under glasshouse conditions at Lincoln University in two acidic high country soils collected from Omarama and Glenmore Stations, Central Otago. The trial was a full factorial design of two soil types, three soil moisture levels and four fertiliser treatments. Lime (100% CaCO3) was applied at three rates (0,2 and 8 t lime/ha) with one treatment of P and lime (2 t lime/ha and 150 mg P/kg). Sulphur was applied to all treatments at 120 kg S/ha (Gypsum). Plants were harvested four times during the experiment and yields were determined. Herbage was analysed for macro and micro nutrient concentrations by ICP-OES and NIRS. Soils were analysed for pH and exchangeable Al, with bulk soil samples also analysed for P and S contents at completion of the experiment. A significant interaction between soil moisture and exchangeable soil aluminium was observed for the Glenmore soil. The medium moisture treatment (23% VWC) had lower exchangeable Al, 4 mg/kg compared to the low (17% VWC) and high (33% VWC) moisture treatments which contained 5 and 5.6 mg exchangeable Al/kg respectively, for the zero lime treatment. The low Al Omarama soil in comparison, only contained Al concentrations of 1.3, 1.1 and 0.8 mg Al/kg in the same treatments at the completion of the experiment. This experiment did show an interesting interaction and strongly suggests that soil moisture influences exchangeable Al, but only when Al concentrations are greater than 4 mg Al/kg. Similar experiments are required using high Al soils to confirm this result. The Glenmore soil contained higher initial concentrations of exchangeable Al (6.6 mg Al/kg), had greater carbon stocks (6.9 kg C/m3) and produced higher shoot yields (10.6 g DM/pot) than the Omarama soil (3.8 mg Al/kg , 3.5 kg C/m3 and 7.8 g DM/pot). The high P fertility of the Glenmore soil (Olsen P 18) probably supressed soil Al affects on yield. High soil carbon may have also masked the effects of soil moisture on exchangeable Al for the Glenmore soil. Moisture treatments caused a significant growth limitation on the low moisture treatment (17% VWC), with a yield of 8.7 g DM/pot compared to 9.5 and 9.4 g DM/pot for the medium (23% VWC) and high (33% VWC) moisture treatments. Herbage nutrient concentrations were effected predicatably by lime and P addition. Shoot concentrations of P and Mo increased with lime addition, while shoot concentrations of Zn, Mn, B and Al decreased. Shoot nutrient uptake followed the same trends as concentration.

Opportunity costs of conserving forest for the purpose of greenhouse gas emissions reductions underpin economic sustainability of ‘reducing emissions from deforestation and forest degradation (REDD+)’ and are essential to ascertain before embarking into this newly proposed mechanism to mitigate climate change. Two major determinants of REDD+ opportunity costs are the drivers of deforestation and forest degradation and the carbon stock. These factors differ from place to place. Use of global land use models and of forest carbon stock estimates to design REDD+ at sub-national or national scale may be misleading if they do not reflect local socio-economic and agro-ecological conditions. This study examined the potential costs of emissions reductions in the Terai Arc Landscape of Nepal, one of the biodiversity hotspots of the world. The first part of the thesis examines the underlying causes of deforestation and forest degradation. A sythesized econometric model of deforestation is prepared including rarely integrated part of forest degradation. A newly assembled panel data sets of 15 districts of Nepal over a 19-year period were used for the analysis. The results highlight that increased agricultural yield and promotion of community-based forest management regimes reduce deforestation. Fuel wood, the main source of energy used for cooking in the landscape, is the proximate cause of forest degradation along with timber extraction. Alternative energy sources like solar and biogas can be substituted for fuel wood to reduce forest degradation. The second part of the study deals with carbon stock in the forests. The research estimated the distribution of C stock across the different pools and management regimes of tropical Sal forest. It applied a field measurement approach and collected biomass data from 113 sample plots. The estimated average total carbon stock was 228.76±19.61 Mg ha⁻¹. The value of total C stock, however, varied according to management regimes. The estimated carbon stocks differed from all earlier estimates based on biome-average dataset. Evidence of strong association of C stock with management regime provides valuable information for policy makers to make informed choice of management regime for the landscape. The third part of the study is focused on estimation of opportunity costs of emissions reductions through avoided deforestation and forest degradation. A bottom-up approach is applied using time series data of agriculture, timber and fuel wood production and prices to estimate the opportunity costs. The estimated mean opportunity cost of emissions reductions from avoided deforestation on the Terai Arc Landscape is found to be US$ 8.95 per Mg CO₂e. The study reveals that emission reduction from avoided forest degradation is cheaper than emission reduction through avoided deforestation. The opportunity cost estimates are higher than those reported in earlier global studies and are attributed to higher agricultural returns and lower carbon density in the forest of the Terai Arc Landscape. The study suggests that the levels of funding needed for REDD+ based on the earlier global estimates may be insufficient for effective emissions reductions. Policy makers to be cautious when using global models and values to design any sub-national REDD+ scheme.