Forest suitability mapping and land use change analysis in eThekwini Municipality: leveraging remote sensing and machine learning for forest restoration and rehabilitation efforts.

Doctoral Degree. University of KwaZulu-Natal, Pietermaritzburg.

Ecosystem services are vital in environmental policy, emphasising the crucial role of forests in providing these services. However, urbanisation, deforestation, forest degradation, and climate change continue to threaten forests globally. Moreover, despite international efforts like deforestation and forest degradation in developing countries (REDD+) and growing recognition of their importance, forests continue to face significant threats from deforestation, degradation, and climate change. While regulatory frameworks and incentives are important, they may fall short without advancements in forest restoration processes. As such, this study aimed to streamline forest restoration and rehabilitation by employing remote sensing and machine learning techniques to enhance natural forest monitoring and regional forest suitability modelling. Objectives included reviewing and analysing global publications on forest rehabilitation and restoration efforts to investigate trends, explore various practices, and identify opportunities for enhancing the success of these initiatives. This was achieved through the usage of a systematic review methodology. Findings for this objective revealed an increasing research activity in recent years, indicating growing interest in forest rehabilitation and restoration. Geographic analysis highlighted regional disparities, with Asian countries leading in research frequency. Policy recommendations underscored the importance of community participation, efficient fire control, and government support in forest rehabilitation efforts. The second objective reviewed and analysed publications on the utilisation of forest suitability models and remote sensing techniques for identifying areas suitable for forest vegetation, also using a systematic review methodology. The findings also indicate a notable increase in research output. Furthermore, the analysis of reviewed articles revealed a preference for medium to high-resolution remote sensing data, with Landsat being the predominant sensor used for forest suitability assessments. Maximum entropy (MaxEnt) was identified as the most utilised model, followed by the increasingly popular random forests (RF). However, the research revealed a significant geographical disparity, with a heavy concentration of publications in the Americas and Asia. The third objective explored mapping land use and land cover (LULC) changes within the eThekwini Municipality from 2002 to 2022 using remote sensing data from the three most recent Landsat sensors and machine learning algorithms. It utilised RF, support vector machine (SVM), and extreme gradient boosting (XGBoost) to conduct LULC classifications. The generated maps revealed a significant decrease in cropland and an increase in impervious surfaces. As such, this research established a framework for continuous LULC mapping and highlighted Landsat 9's potential in LULC classifications. The fourth objective assessed land degradation within the eThekwini Municipality by focusing on land cover change and soil organic carbon (SOC) stock using medium-resolution remote sensing data and machine learning algorithms. Variables for land cover change and SOC stock prediction were extracted and analysed using XGBoost, light gradient boosting (LightGBM), RF, and SVM models. LightGBM outperformed other models, revealing a notable land cover shift, with forests and shrubland being converted to cropland and urban areas within the municipality. The fifth objective sought to provide a framework for monitoring natural forests at a municipal scale using the last three Landsat Missions, focusing on the eThekwini Municipality, to facilitate forest rehabilitation and restoration. Classifications based on Landsat 7 significantly underestimated the extent of natural forests within the study area, whereas Landsat 8 and Landsat 9 data revealed an increase in natural forests from 2015 to 2023. The final objective aimed to model the suitability of areas for forest species within the eThekwini Municipality using species distribution modelling (SDM)/environmental niche modelling (ENM) methodology. The study modelled current forest suitability (2023) using bioclimatic variables from the WorldClim dataset, and elevation and slope data from the Shuttle Radar Topography Mission (SRTM). Remote sensing data was obtained from Landsat 9 and Sentinel-2A. For future forest suitability (2021–2040), bioclimatic variables from two Global Climate Models (GCMs) under four WorldClim shared socioeconomic pathway (SSP)-based representative concentration pathway (RCP) scenarios were used. The models employed were RF, LightGBM, and artificial neural networks (ANN), with data processing conducted via Google Earth Engine (GEE), QGIS, and Python. Currently, 30% of the municipality's land is deemed suitable, primarily concentrated in the central region. Future projections highlight the mountainous north-western region as most suitable, notably under the SSP370 scenario with a projected suitable area of 63%. Overall, findings from this study highlight the potential of remote sensing and machine learning in supporting forest restoration and rehabilitation efforts, with significant implications for informing policy and prioritising areas for future interventions. Ultimately, this research provides a comprehensive framework for leveraging modern technological advancements to streamline forest restoration initiatives, ensuring sustainable management and conservation of forest ecosystems amidst escalating environmental challenges.