Fig 1.
Schematic summarising steps to evaluate and compare radar-based waterbody maps with the optical-based JRC global surface water product.
Fig 2.
Map showing the location of the three focal districts (Sindhudurg, Shivamogga and Wayanad), India.
Source of Administrative boundaries: https://github.com/HindustanTimesLabs/shapefiles/tree/master/india/district.
Table 1.
Description of the three focal districts.
Fig 3.
Raw and pre-processed Sentinel-1A SAR VV backscatter image of Shivamogga of January 2018.
Fig 4.
(a) Example Sentinel-1A SAR VV backscatter image for October 2017, Sindhudurg. Source of imagery: Copernicus Sentinel-1 data 2017 accessible via https://dataspace.copernicus.eu/explore-data/data-collections/sentinel-data/sentinel-1 or through GEE using ee.ImageCollection("COPERNICUS/S1_GRD")); (b) Unimodal backscatter histogram for the whole image covering Sindhudurg district; (c) Bimodal backscatter histogram for the highlighted area in the Sentinel-1A SAR image. Red, blue and dashed blue vertical lines are threshold values obtained using manual, valley and Otsu, respectively. (d) Bimodal backscatter histogram on a log scale with the Bayesian inferred likelihood of a backscattering coefficient value being a suitable threshold (purple dashed line) and the most likely threshold (vertical solid purple line). We decompose the likelihood into prior probability (blue dashed line) and posteriori probability (red dashed line) (see S1 Appendix). (e-f) Resultant waterbody area (in blue) using threshold derived from the whole image shown for the highlighted area using the (e) manual thresholding (threshold: -17.58 dB); (f) Otsu thresholding (threshold: -8.76 dB), (g) Valley Emphasis approach (-15.28 dB). (h) The probability of being a waterbody using the Bayesian approach. Areas in blue are most likely waterbodies, while areas in red are least likely waterbodies.
Table 2.
For Sindhudurg Sentinel-1A SAR images, the most likely thresholds identified using valley emphasis (bin25) and Bayesian inference compared with our corresponding manual thresholds.
Table 3.
Manual thresholding: The district’s average backscatter threshold values (and standard deviation) for 2017 and 2018.
Fig 5.
Distribution of ground reference points (GFP) displayed with the S1A water body map and JRC cloud cover layer for the month matching the timing when reference points were collected in situ:A. Shivamogga, B. Sindhudurg and C. Wayanad. Source of Administrative boundaries: The Global Administrative Unit Layers (GAUL) dataset, implemented by FAO within the CountrySTAT and Agricultural Market Information System (AMIS) projects.
Fig 6.
Surface water area and cloud cover impact in Shivamogga, Sindhudurg and Wayanad for January 2018.
S1A (left) and JRC (middle) surface water area (shown in black); and for JRC (right), the impact of cloud cover (shown in grey) and scan line correction gaps. White represents areas within the districts where there is no surface water.
Table 4.
Accuracy assessment of waterbodies in Shivamogga, Sindhudurg and Wayanad showing average user and producer accuracy and standard error values for S1A and JRC-derived surface water maps and the number of water (W) and non-water (NW) reference points.
Fig 7.
Seasonal variation in the total surface water and cloud cover area derived from the S1A and JRC waterbody maps for Shivamogga, Sindhudurg and Wayanad, 2017 and 2018.
Across the districts, the annual average is 25% for 2017 and 23% for 2018, while the average maximum cloud cover in July or August is 100% for both 2017 and 2018.
Fig 8.
District surface water area patch number, mean patch area and patch density derived from the S1A and JRC maps for months when JRC has the least cloud cover (<20%).