LAND USE LAND COVER CHANGE MAPPING FROM SENTINEL 1B < 2A IMAGERY USING RANDOM FOREST ALGORITHM IN CÔTE D’IVOIRE

• DOI: 10.3846/gac.2024.18724
• Publication Date: 2024-04-15T11:45:27Z
• Authors: Christian Jonatha...; Chen Qian; Dilawar Khan; Lutumba Suika Ach...; Zhang Kebin; James Kehinde Omi...; Tu Ya; Xiaohui Yang
• Keywords: Land Cover

Monitoring crop condition, soil properties, and mapping tillage activities can be used to assess land use, forecast crops, monitor seasonal changes, contribute the implementation of sustainable development policy. Agricultural maps provide independent objective estimates extent crops in a given area or growing season, which support efforts ensure food security vulnerable areas. Satellite data help detect classify different types soil. The evolution satellite remote sensing technologies has transformed techniques for monitoring Earth’s surface over last several decades. European Space Agency (ESA) Union (EU) created Copernicus program, resulted satellites Sentinel-1B (S1B) Sentinel-2A (S2A), allow collection multi-temporal, spatial, highly repeatable data, providing an excellent opportunity study cover, change. goal this is map cover Côte d’Ivoire’s West Central Soubre (5°47′1′′ North, 6°35′38′′ West) between 2014 2020. method based on combination S1B S2A imagery as well three predictors: biophysical indices Normalized Difference Vegetation Index “(NDVI)”, Modified Water “(MNDWI)”, Urbanization “(NDBI)”, “(NDWI)”, spectral bands (B1, B11, B2, B3, B4, B6, B7, B8) polarization coefficients VV. For period 2014–2020, six classifications have been established: Thick_Forest, Clear_Drill, Urban, Water, Palm_Oil, Bareland, Cacao_Land. Random Forest (RF) algorithm with 60 numberOfTrees was primary categorization approach Google Earth Engine (GEE) platform. results show that RF classification performed well, outOfBagErrorEstimates 0.0314 0.0498 2020, respectively. accuracy values kappa were above 95%: 96.42% 95.28% overall 96.97% 96 % Furthermore, User Accuracy (UA) Producer (PA) classes frequently 80%, exception Bareland class achieved 79.20%. backscatter variables had higher GINI significance 2014: VH (70.80) compared (50.37) 2020; VV (57.11) (46.17) Polarization than other predictor variables. During period, Thick_Forest (35.90% ± 1.17), Palm_Oil (57.59% 1.48), (5.90% 0.47) experienced regression area, while Clear_Drill (16.96% 0.80), Urban (2.32% 0.29), (83.54% 1.79), Cacao_Land (35.14% 1.16) increase. regarded obtained.