Earth observation based monitoring of urbanization and environmental impact in Kigali, Rwanda

Abstract

Urbanization is one of the great challenges in the 21st century. Despite being an engine for the global economy, urban areas consume 78% of World,s energy and emit more than 60% of greenhouse gas emission. Sub-Saharan African cities, e.g. Kigali, are characterized by rapid population growth and accelerated land use/land cover change. Yet, the implementation of policies and regulations catalyzing sustainable urbanization is constrained by scarce and fragmented data related to land use/land cover spatial patterns and changes in population. Collected statistics are most of the time outdated, or geographically aggregated to large heterogeneous administrative entities, which is judged meaningless for informed decision making. Therefore, there is a need for timely and reliable data, and tools to monitor the spatio-temporal patterns of urbanization and its environmental impact for informed and sustainable decision making. The objectives of this thesis are i)to investigate the use of multi-temporal and multi-resolution Earth observation data for mapping and monitoring urbanization patterns and trends in Kigali, Rwanda, a complex urban area characterized by a subtropical highland climate; and ii)to analyze the environmental impacts of urbanization using the integration of land cover information classified from Earth observation data with landscape metrics and ecosystem services. Using satellite imagery from 1984 to 2021, spatial patterns and temporal trends of urbanization in Kigali were investigated and analyzed. Specifically, optical satellite imagery at medium to very high resolution, i.e. Landsat TM/ETM+/OLI at 30m resolution, Sentinel2 MSI at 10-20m resolution and WorldView-2 at 2m spatial resolution were used for land use/land cover mapping and change analysis. Diverse image processing techniques, including texture feature analysis using Gray level Cooccurrence matrix, pan- sharpening and derivation of various biophysical indices, were applied to enhance land use/land cover classification and analysis. Various land use/land cover classification methods were used, including pixel- and object-based support vector machine classification, Google Earth Engine-LandTrendr cloud computing, and a hybrid framework combining intermediate classification results derived from both random forest classification, and U-Net deep neural networks. The land use/land cover classes were then used not only to derive indices characterizing spatio-temporal changes in urban landscape composition and configuration, but also to analyze the impacts of land use/land cover change on ecosystem services. Areas which provide ecosystem services were evaluated in terms of changes in spatial attributes and structure of landscape patches. The most prominent ecosystem services in the study area divided into three groups - provisioning, regulating and supporting services - were further analyzed using a matrix spatially linking landscape units with service supply and demand budgets. In one of the studies, a monetary based valuation approach was performed for assessing spatio-temporal change in value of selected ecosystem services. Using multi-temporal, multi-resolution Earth observation data, five to twelve land use/land cover classes were derived with an overall accuracy exceeding 83% and with Kappa coefficients above 0.8. The most prominent change was the conversion of croplands into built-up areas. As a result, the built-up areas increased from 2.13km2 to100.17km2 between 1984 and 2016. The results revealed that the urbanization between 1987 to 1998 was characterized by slow development, with an annual growth rate less than 2%. The post-conflict period (1995 on-wards) was characterized by accelerated urbanization, with a 4.5% annual growth rate. From 2004, urbanization was promoted due to migration pressure and investment promotion in the construction sector. The five-year interval analysis from 1990 to 2019 revealed that impervious surfaces increased from 4233.5 to 12116 hectares,with a 3.7% average annual growth rate. In order to map urban land use/land cover at fine scale, very high resolution WorldView-2 imagery was acquired and analyzed using object- and rule-based classification. Urban land cover at fine scale could be mapped with an overall accuracy exceeding 85% (kappa above 0.8). Multi-temporal Sentinel-2 MSI data were found advantageous for monitoring spatio-temporal trends of urban development, and producing reliable baseline data for the analysis of urban landscape changes at entire city scale with sufficient details. During the 37 years study period, landscape fragmentation could be observed, in particular in forest and cropland. The landscape configuration indices demonstrate that, in general, the land cover pattern remained stable for cropland, but that it was highly changed for built-up areas. Estimated changes in ecosystem services amount to a loss of 69 million US dollars because of cropland degradation in favour of urban areas and in a gain of 52.5 million within urban systems between 1984 and 2016. Most of the ecosystem services bundles show that built-up areas have a high demand on ecosystem services, whereas green and blue space are strong contributors in supplying bundles of ecosystem services. The study demonstrated that multi-temporal multi-resolution Earth observation data and advanced image processing offer great opportunities for quantifying urbanization, and analyzing its environmental impacts using landscape metrics and ecosystem services variables. Medium resolution data, Landsat and Sentinel-2 MSI, were found useful for global annual urban growth and environmental impact analysis at entire city scale. Very- high-resolution satellite data are still only available at high cost. Therefore, land use/land cover mapping based on very high resolution data should be produced only at special occasion based on cost-benefit analysis. Meanwhile,open data policy and free access to cloud computing systems such as Google Earth Engine were also found cost-effective and useful for continuous monitoring of the complex dynamics of urban land use/land cover, especially in areas where the cost of Earth observation data is restricting due to budget reasons, and in data-scarce regions. The thesis contributes to the development of approaches for mapping and monitoring urban development and associated environmental impact in Sub-Saharan through the exploration of potential and limitations of multi resolution remote sensing data. Methodological frameworks for urban land cover production based on state-of-the-art machine learning, deep learning, and Earth observation big data analytics were implemented and tested. The research output compiled in this thesis demonstrated that the open-access Earth observation data are cost-effective data source for monitoring urbanization and for investigating the impact of spatial structure changes on the distributions and patterns of ecosystem service bundles. The frameworks developed in this research can easily be transferred to other Sub-Saharan African cities. Future research will explore the integration of multiple-source data, i.e., Earth observation data, population statistics and other types of data to detect and map urban deprivation and environmentally sensitive areas. Finally, the combination of optical and radar remote sensing data, the use of machine learning and deep learning methods in a cloud computing environment will be further investigated to develop a dynamic framework for continuous urban land use/land cover change monitoring.