Illigal irrigation Mapping in Oases Using Optical and radar Time-Series Data and Machine-Learning Classifiers

The scarcity of water and the need for the irrigation of the oases leading to a significant overexploitation of the groundwater. The detection of irrigated areas based on illicit drilling is a well-resolved problem in Tunisia. The goal of thus paper is to explore the potential of optical and radar remote sensing data and machine learning in the identification and quantification of oases areas irrigated from illicit boreholes in the governorate of Kebili identification. We used a synergy of time-series optic Sentinel2(S2) and radar Sentinel1 (S1) data acquired from February to June 20022. The retrieved Oases land cover classes included three categories of stratums: 1 stratum (date Palm), 2 stratums (date palm and associated crops), 3 stratums (date palm, fruit trees and crops). In the first time, a hierarchical object-based classification approach was processed using the combined data derived from the normalized vegetation index (NDVI) and the 3 principal components (PCA1, 2 and 4) of the S2 time-series dataset. Three classifiers tested the random forest (R F), support vector machine (SVM) and k-nearest neighbor (KNN) classifiers, respectively using 40 samples of ground control data. The RF classifier performs better than SVM and KNN, with highest overall accuracy of 92% and a Kappa coefficient of 0.87. Obtained The Rf result map were crossed to the soil moisture map derived from S1 data to obtain irrigated crops outside the legal irrigated areas The result retained following this approach was compared to illegal drillings identified in the field survey and legal drilling and confirm more than 90% of illicit drillings. Furtherer more illegal extensions represent five times the area of the legal Public Irrigated Perimeters (PPI). The findings of this study might be useful for policy makers as well as planners.