Mapping a rice region in south America using geo big data and sentinel 2

Abstract

Geo Big Data and Sentinel-2 have been extensively employed to map rice paddies and general land use/land cover. Focusing on the environmental value and relevance of rice production in Cuenca de la Laguna Merín in Uruguay, the research aimed to 1) map rice paddies and other land use/land cover classes, 2) compare the capabilities of Random Forest and Support Vector Machine for classifying two different Sentinel-2 time series stacks, and 3) identify the most important features according to Random Forest. In addition to quoted imagery and classifiers, the materials include Google Earth Engine, GEEMAP, and Python's Scikit-learn GridSearchCV. The main methods comprised hyperparameter tuning, supervised classification, and accuracy assessment. Quoted assessment revealed that the four maps performed well. The feature importance analysis highlighted the Near-Infrared and Shortwave Infrared as the most relevant features. Future research should focus on integrating diverse data sources and comparing different time series than those employed here.