Automated Image Classification For Environmental Monitoring Using Google Earth Engine Python API: A Case Study Of Kajiado County, Kenya

Abstract

In response to the growing challenges of global warming and environmental degradation, image classification has emerged as a critical tool for environmental monitoring. Traditional methods are often labor-intensive and prone to errors, necessitating more efficient and accurate approaches. This study aims to automate land cover classification for environmental monitoring in Kajiado County, Kenya. We leveraged the power of Google Earth Engine Python API and machine learning to introduce an efficient, reproducible workflow using Sentinel-2 imagery, spectral indices (NDVI, NDBI, MNDWI), and ESA land cover data. In this study, the model was validated using ground truth data from European Space Agency’s Global Land Cover Map. Major LULC classified were Tree cover/Forest, Shrubland, Grassland, Cropland, Built-up, Bare/sparse vegetation, Water bodies. A stratified sampling approach and automated feature extraction were used to train and validate the model. The classifier achieved an accuracy of over 70% and a Kappa score of 0.65, indicating substantial agreement. Compared to traditional manual classification methods, this automated approach reduced processing time by over 50%. The resulting land cover maps revealed key environmental changes such as deforestation, urban expansion, and water body shrinkage. These findings demonstrate the potential of integrating cloud-based geospatial tools for real-time monitoring and data-driven decision-making in environmental policy.