A machine learning algorithm for mapping small reservoirs using Sentinel-2 satellite imagery in Google Earth Engine

Abstract

This report outlines an advanced methodology for mapping small reservoirs in Northern Ghana, utilizing Sentinel-2 satellite imagery and Google Earth Engine. Aimed at enhancing mapping accuracy by reducing cloud contamination, the method filters image collections, applies optimal cloud masks, and composes cloudless images. The methodology also included the calculation of spectral indices such as the Normalized Difference Vegetation Index (NDVI) and the Modified Normalized Difference Water Index (MNDWI) to improve classification accuracy, while a Random Forest algorithm classifies water and non-water features based on training samples from satellite imagery. The algorithm, leveraging specific spectral bands and MNDWI, demonstrates high accuracy, with results validated against a test dataset. The process concludes with image cleaning and permanent water masking, exporting the data in raster format for analysis. This methodology supports effective water resource management and the CGIAR Initiative on Aquatic Foods’ goals for food security and sustainable aquaculture in Northern Ghana.