3. Methodology
Time series datasets, microwave data from Sentinel-1 and optical data from Sentinel-2 were taken. The microwave data was splitted. Calibrate and Deburst as a part of pre-processing. Thereafter the backscatter parameters σ0, β0 and γ0 were retrieved during calibration. Thereafter, the data was geocoded using terrain correction from SRTM (Shuttle RADAR Topography Mission) plugin. Thereafter, the data was multilooked to generate square pixels and remove radiometric distortions.
The optical data of Sentinel-2, of the same dates as microwave data was taken and layer stacked to put together all 13 bands as a single imagery. Since the study area was not covered in one image tile, the different image tiles were mosaiced together. Thereafter, the study area of Rupnagar was clipped out from its district shape file. From the clipped imageries, band rationing was done to derive Normalised Differential Salinity Index (NDSI) which is an indicator of soil salinity from optical satellite data.
Since optical data does not have an all-weather availability and microwave data suffers from speckle noise, this study was aimed at retrieval of salinity sensitive parameters from both the datasets and put them together into one salinity estimation model. The model also uses field data collected using an instrument from FieldScout Ltd, USA. Surface Soil moisture, soil temperature and Electrical conductivity were recorded from this instrument at surface as well as at 60 cm below the surface in early stages of wheat crop growth (November 20th, 2019, December 27th, 2019, and January 20th, 2020).
The SAR Backscatter coefficients, NDSI, surface soil moisture, soil electrical conductivity (EC) and soil temperature were fed into an Ordinary Least Squares Model. R2 statistics, F-test and Durbin-Watson test were carried out to assess the model accuracy. The detailed methodology flow diagram is mentioned in Figure 2.
Prior to model development, correlative plots were made to check the sensitivities of all other parameters used in the model with the most indicative of Soil Salinity – EC and NDSI. Also, the three backscatter parameters were plotted for correlation with Sigma Nought σ0 for any major variation since use of excess parameters could lead to model overfitting.