Cloud computing random forest classification for major agricultural crops in the Nile Delta of Egypt

Abstract

Accurate land cover and cropland statistics are essential for science, economics, and governance, and time-series satellite images outperform single-date views. Google Earth Engine (GEE) improves classification accuracy by effectively processing massive time-series datasets that span multiple years or seasons. This study uses Sentinel-2 data to look at how compositional techniques and input images influence land cover map classification in Egypt's Nile Delta. The Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Water Index (NDWI), Modified Chlorophyll Absorption Ratio Index (MCARI), and Normalized Difference Building Index NDBI were all used to conduct time series analyses. Monthly medians of NDVI, EVI, SAVI, GNDVI, NDWI, MCRI, and NDBI decreased cloud interference. Wheat and clover dominate the winter (October to May), but maize and rice dominate the summer (May to October). These two seasons account for more than 70% of all seasonal field crop production. Potatoes, sugar beets, tomatoes, onions, faba beans, peanuts, and cotton account for less than 30% of total field crop production in the region. The analysis eliminated urban neighborhoods, bare soil, water bodies, and permanent vegetation. The Random Forest (RF) classifier on the GEE platform achieved an overall classification accuracy for the winter 2022/2023 season reached 92.79%, with a Kappa coefficient of 0.889, indicating strong agreement beyond chance. The summer 2023 season achieved an even higher overall accuracy of 95.94% and a Kappa coefficient of 0.941, reflecting excellent classification reliability. The study also addresses categorization challenges caused by overlapping cultivation seasons and similar seasonal crop phenological stages.