The Use of Machine Learning and Remote Sensing Data for Land Cover and Moisture Classification

Severe water scarcity felt in Tunisia during the last years, means that water resources must be urgently managed sparingly in order to ensure good production and water sustainability. The use of field data to determine the crop water use is a hard work with uncertainties due to the lack of representativeness of measures. We can improve our analyses by using remote sensing data concomitantly with infield measures. Recent studies experienced the use of the neural networks trained by using synthetic data generated by the modified Integral Equation Model (IEM) model and validated with actual soil data (SM). This study was carried out at Kairouan, center of Tunisia, where arid climate prevails. It proposes: (1) an empirical modeling approach based on the ‘ER’ formulae for soil water status estimation and (2) two proposed colored composition images to control the changes of soil humidity and crops (3) monitoring the impact of climate variations on land uses by unsupervised classification: ISODATA and K-MEANS. Data are uploaded from Sentinel-1 and 2 satellites (10 to 60 meters) from the Copernicus hub platform. Sentinel-2's MSI sensor provides multi-spectral images while Sentinel-1's synthetic aperture radar antenna provides polarimetric data at both polarizations VV and VH. Data validation is performed by using: (i) the measured average deviations of precipitation over successive months (EP) considering three EP cases: (1) no rainfall EP < =0, (2) a little rainfall 0 < EP < =20(mm),(3) a lot of rainfall EP > 20 (mm). Results showed that, by considering VH polarization, we have ER < 0 for LC3 and ER=0 for land crops LC1 and LC2 when EP < =0; the ER > 0 for LC3 and ER=0 for land crops LC1 and LC2 when 0 < EP < =20(mm) and ER > 0 For heavy rainfall, ER > 0 for all LC. These ‘ER’ results were validated by using composite images and classified images and shows that it is possible to follow the soil water reserves considering the synergy between the optical and radar satellite data and the potential use of Sentinel data to analyze the rainfall variations over a certain period.