Cocoa production plays a significant role in global tree cover and biodiversity loss. Amid rising concerns over its role in environmental degradation, agroforestry has emerged as a potential solution to satisfy growing global demand for cocoa in a climate-smart and sustainable way. As a shade tolerant crop, there has been much discussion over the potential ecosystem service benefits that cocoa agroforestry can provide if adopted in degraded production areas. However, not much is known in the academic literature regarding the amount of cocoa agroforestry that currently exists globally. It is also unclear to what extent cocoa agroforestry can be scaled up to and what the potential scale of impact of such an endeavor would be in terms of carbon sequestration. This study aims to address these questions by developing a spatial model that uses georeferenced data on global cocoa production and percentage of tree cover in agricultural areas to estimate the presence and distribution of cocoa agroforestry globally. Using data from the MapSPAM project on the physical area of cocoa production, this study found that approximately 10-21% of global cocoa production areas were covered by cocoa agroforestry systems in 2010. By applying global average findings on cocoa agroforestry benefits from Niether et al. 2020 to global cocoa monoculture areas in 2010, this study found that scaling agroforestry would contribute to sequestering approximately 175 to 199 million tons of carbon. This study also found that scaled agroforestry adoption is likely to reduce cocoa production by about 1.2 to 1.3 million tons per year. Further research examining the carbon benefits and ecosystem services of agroforestry adoption at different regional levels is needed to improve estimates provided in this global assessment. Moreover, challenges in adopting agroforestry are likely to differ across the multiple cocoa producing regions. Further research is thus needed to highlight the environmental constraints and socio-economic bottlenecks that may limit the benefits of agroforestry adoption in each of these regions.

The research focuses on a novel measurement-based approach to estimate the farm-specific ammonia emission potential (AEP) in the dairy sector. By measuring and evaluating the feed-manure chain, feed management strategies and manure parameters influencing AEP can be identified. Ammonia emissions from dairy farms are not only considered to be an important driver of biodiversity loss, but are also responsible for nutrient losses in the farm cycle. Currently, farm-specific ammonia emissions are calculated using the Kringloopwijzer model, which tends to over- or underestimate actual ammonia emissions. Therefore, the possibilities of a measurement-based approach are evaluated. This study analyses the relationships within the feed-manure-AEP sequence. A comprehensive approach is used, involving 23 manure parameters and 12 feed management parameters. The most important predictors of the AEP include N, TAN, Norg, N90, and the C/N ratio, whilst urea in milk, pH, and DS showed low significance. Silage maize and VEM are identified as feed management parameters with a positive indirect relationship with the AEP, whereas other roughage and fresh grass exhibit a negative indirect relationship. The calculated TAN value plays a central role in the emission calculations of the Kringloopwijzer model. There are concerns about the accuracy of this value as well as the absence of other manure parameters in the calculation, highlighting the need for further research. Currently, it is uncertain whether the AEP measurements will be suitable for an emissions-based policy, due to the incapacity to directly represent actual ammonia emissions and the uncertainty regarding the interpretation of the results caused by the period prior to the measurements. Nonetheless, the measurements are valuable in assessing the influence of the manure composition on the AEP, and how it has been affected by feed management strategies.

The emission of reactive nitrogen species increased rapidly in the twentieth century causing a significant perturbation of the nitrogen cycle. This causes many detrimental effects for both humans and ecosystems such as eutrophication, acidification and biodiversity loss. Ammonia, which is a reactive nitrogen species is difficult to quantify because the compound is hard to observe with measurement equipment such as ground-based measurement stations and remote sensing instruments such as CrIS. Therefore, this thesis aims to improve the monitoring of surface ammonia concentrations with a machine learning technique called random forests (RF). These type of models can detect complex and non-linear relationships between variables and are frequently used in other air pollution studies. In this study, a RF model has been built to estimate the ammonia surface concentration with vertical column density (VCD) datasets, meteorological variables and land-use variables. Different combinations of VCD datasets (either modelled or CrIS VCD data) have been used to train the model and to predict the surface ammonia concentration. The result of the study is that RF models are statistically more accurate at estimating the surface ammonia concentration than the LOTOS-EUROS model when validated by ground-based measurements stations from the MAN and LML network. Especially trained RF models that have been supplied with CrIS VCD data during the prediction phase show strong performance. Moreover, when comparing the RF model that has been trained without VCD data to the ‘complete’ RF models, the complete RF models show better performance, proving the added benefit of incorporating satellite data in RF modelling. Recommendations for further research include performing similar experiments with other input variables and other machine learning algorithms, validating the performance of the RF model in different years and considering using other datasets as the ground-truth variable for RF models – such as ground-based measurement data from MAN and LML.