< Back to previous page

Project

Modeling vegetation dynamics driven by climate variability and land use changes in Rwanda

Tropical ecosystems are confronted with the impacts of climate change and increased climate variability. Therefore, mitigation and adaptation strategies are being developed to limit further exacerbation of climate change and to preserve the provision of ecosystem services to societies. Tropical forests are particularly vulnerable to climate variability, yet are also shaped by deforestation, fragmentation, degradation and restoration processes. This thesis investigates to what extent observed changes in tropical ecosystems can be attributed to climatic effects or to local land use impacts. Thereto, it presents concepts and methods for the use of satellite image time series, climate data and field observations for modeling these drivers with a focus on East African ecosystems.

First, a time series processing tool was developed to detect interannual changes in vegetation cover from satellite imagery via the Normalized Difference Vegetation Index (NDVI). Thereto the Ensemble Empirical Mode Decomposition (EEMD) method was implemented for automated detection of an interannual NDVI signal, and validated with simulated time series modeled from real NDVI data. Sensitivity analysis of signal and noise characteristics showed that detectability of the interannual signal is primarily determined by its amplitude relative to seasonal changes and by fluctuations in the onset of annual growing seasons by precipitation variability.

Vegetation response to precipitation variability differs across East Africa, and the biogeographical factors determining this response were studied with a spatial explanatory model. Vegetation type, mean annual precipitation, elevation and ocean-coupled climate indices explained up to 43% of the observed variability in vegetation response to precipitation. Local effect models per ecological zone showed that precipitation sensitivity is highest in ecosystems dominated by herbaceous cover and in the transition zones from woody savanna towards tropical evergreen forest, also beyond the expected 900 mm per year limit of subhumid conditions. Evaluation of the spatial and temporal consistency of precipitation and NDVI datasets indicated that spatial heterogeneity in the quality of precipitation estimations over observation-scarce areas and rugged topography is a principal limitation for modeling this process.

Land use changes in heterogeneous landscapes were studied in Rwanda, a small country consisting of densely populated mosaic landscapes of cropland and agroforestry, wooded savanna systems and sparse Afromontane forest remnants. Historical trajectories of land cover (1984-2013) were mapped from Landsat satellite imagery using land cover inventory data collected from 185 plots across 51 ecological strata. Via image classification, mature forest, young forest and shrubs, tree-rich and tree-poor cropland were detected with 68% overall accuracy. The detection results were evaluated with a set of documented cases of historic land use change. Forest clearing, large scale reforestations and abrupt local land occupation were detected, whereas more subtle changes in tree cover changes by agroforestry and local afforestation could only be partially detected with the limited set of land cover classes. The aggregated effect of local land cover changes was used as a model for the direct human impact on ecosystems on a regional scale.

Finally, an explanatory spatiotemporal model was formulated to model observed interannual NDVI over Rwanda in terms of precipitation variability and aggregated local land use changes. The independent and explicit modeling of the local land use driver was proposed as a fundamental improvement over implicitly deriving it from the non-climatic residual in a vegetation-precipitation model. Overall, the model explained 50% to 56% of the observed interannual NDVI variability. The precipitation effect dominates across Rwanda and highest precipitation sensitivity was detected in the rainfed shrub cropping systems in Central and East Rwanda. The detected effect of land use change explained only 5% to 10% of observed vegetation changes, and was limited to areas with forested land use types.

This spatiotemporal model with an explicit land use change driver offers an approach to attribute vegetation changes to climatic effects and land use change effects, and to evaluate the potential impacts of climate change scenarios and mitigation measures. Its most important limitations are posed by the detectability of land cover classes from Landsat imagery, by the spatial scale of analysis and by missing drivers in the model. The scope for application includes quantification of climatic baseline scenarios for the evaluation of the impacts of local mitigation measures and landscape restoration efforts. Further research is recommended to refine the spatiotemporal attribution model with novel satellite data sources, and to further develop the EEMD method for detection of seasonal variability in crop monitoring applications.

Date:12 Dec 2011 →  15 Mar 2019
Keywords:vegetation dynamics, climate variability, land use changes, remote sensing, carbon stocks, time series decomposition
Disciplines:Forestry sciences, Physical geography and environmental geoscience, Communications technology, Geomatic engineering, Landscape architecture, Art studies and sciences, Ecology, Environmental science and management, Other environmental sciences
Project type:PhD project