*3.3. Sentinel-2*

We used a total of 118 Sentinel-2 (S2) images with level 1C processing to surface reflectance as part of this study. S2 is a wide-swath multi-spectral earth observation mission with spatial resolution varying from 10 to 60 m. The multi-spectral data include 13 bands in the visible, near-infrared (NIR), and shortwave spectra, revisiting every 10 days under the same viewing angle. The level 1-C products within GEE are orthorectified and radiometrically corrected, providing top-of-atmosphere (TOA) reflectance values. We adopted an automatic cloud masking procedure using the QA60 band of the S2 1C product to mask the opaque and cirrus clouds. We also set the cloud coverage within S2 scenes to a maximum of 10 percent over the time of data acquisition. Due to frequent cloud coverage over the study area, we used a median of 5 months (May to October 2016) of the reflectance values. We used four bands of S2 (blue, green, red, and near-infrared) with a spatial resolution of 10 m to create the band compositions for supervised classifications using machine learning algorithms. We used median values of S2 temporal images to be used in the multisensory band composite. Additionally, we calculated the normalized difference vegetation index (NDVI) [33] and normalized difference water index (NDWI) [34] using the four bands of S2, and used them as predictors in the classification process (Figure 3). Figure 4 shows the variation of NDWI over two potholes in the study area, showing periods of inundation and drought. Typically, NDWI > 0.3 and <0.3 indicates the presence and absence of detectable surface water [35]

**Figure 3.** Sentinel-2 derived NDWI (**A**); NDVI (**B**); Sentinel-1 VV (**C**); Sentinel-2 RGB (**D**).
