Leveraging machine learning for spatio-temporal monitoring of carbon sequestration in urban reforested landscapes in eThekwini Municipality, South Africa.

Abstract

Urbanization and expanding cities have emerged as major drivers of deforestation, forest degradation, and associated carbon emissions. While urban areas account for a disproportionate share of global greenhouse gas emissions, urban reforestation initiatives provide significant opportunities for climate change mitigation. This study investigates the utility of remote sensing and machine learning techniques for quantifying, mapping, and tracking carbon accumulation and storage by reforested urban landscapes. Through a systematic literature review, multisource satellite data integration (i.e. Sentinel-1, Sentinel-2, Rapideye and Planetscope spectral data), time-series image analysis, and future climate scenario modelling (i.e., CMIP6 models with 3 representative climate pathways in shared socio-economic pathways RCP-SSP), the research demonstrates the capabilities of geospatial analytics for robust carbon monitoring to strengthen localized climate action planning. Key findings showcase the ability of remote sensing spectral data, derived from the multisource satellite integration, to map heterogeneous reforested biomass and model spatiotemporal variations in aboveground carbon stocks across an urban landscape in South Africa, over seven years. Comparative assessment of reforestation carbon accumulation based on ecological history informs management prioritization for optimized climate benefits. Furthermore, simulation of carbon stocks using remote sensing data, climate models and machine learning indicate a shift in carbon accumulation under the three shared socioeconomic pathways, with the low emissions pathway showing an increase in carbon overtime. The methodological framework in this study delivers an adaptable tool for continuous near real-time quantification of urban forest carbon accumulation to support evidence-based decision-making through targeted monitoring, strategic expansions, and adaptive management. While underscoring the immense potential of urban reforestation for mitigation, the research highlights the need to refine an understanding of realized carbon capture rates, utilize emerging data sources such as Planetscope, Sentinel-1 and 2, and Rapid-Eye, expand spatio-temporal assessments, and integrate economic valuations to maximize climate policy impacts. This contributes to sustainable forest management and climate change mitigation.