Integration of machine learning and remote sensing for assessing the change detection of mangrove forests along the Mumbai coast

Abstract

Mangrove forests, being high-yielding ecosystems, often dominate the intertidal sites along equatorial and subtropical coasts. Despite the known significance of mangroves to the coastal ecology, especially fisheries, deforestation remains a severe danger due to coercion for forest products, ground transformation for aquaculture, and seaside urban growth. Remote sensing is integral in mapping and analysing changes in mangrove forests’ areal extent and spatial patterns due to natural disasters and anthropogenic causes over the last three decades. This work depicts remote sensing analysis for change detection in mangrove forest land use land cover from 2014 to 2019. Indian Remote-Sensing Satellite Resourcesat-2 LISS-IV datasets have been used for analysis. A comparison with the Sentinel-2A dataset and two machine learning models: Random Forest and Classification and Regression Tree, has been performed with 2019 data. This work identifies CART as a suitable choice for supervised landform classification utilising remotely sensed geophysical data that is used to decipher spatial changes concurred over time. An overall growth in the mangrove cover was observed from 2014 to 2019, from 86.26 to 89.63 \({\hbox {km}^{2}}\), along the Mumbai coastline. Spatial comparison over the years shows the growth and loss of land-use cover areas. The performance metrics such as overall accuracy, producer accuracy, Kappa coefficient, and Matthews correlation coefficient are computed. The experiments were conducted using the Google Earth Engine, a powerful cloud computing platform.