Semi-empirical model for retrieval of soil moisture using RISAT-1 C-Band SAR data over a sub-tropical semi-arid area of Rewari district, Haryana (India)

Abstract

We have estimated soil moisture (SM) by using circular horizontal polarization backscattering coefficient ( $$\sigma ^{\mathrm{o}}_{\mathrm{RH}}$$ ), differences of circular vertical and horizontal $$\sigma ^{\mathrm{o}} \, (\sigma ^{\mathrm{o}}_{\mathrm{RV}} {-} \sigma ^{\mathrm{o}}_{\mathrm{RH}})$$ from FRS-1 data of Radar Imaging Satellite (RISAT-1) and surface roughness in terms of RMS height ( $${\hbox {RMS}}_{\mathrm{height}}$$ ). We examined the performance of FRS-1 in retrieving SM under wheat crop at tillering stage. Results revealed that it is possible to develop a good semi-empirical model (SEM) to estimate SM of the upper soil layer using RISAT-1 SAR data rather than using existing empirical model based on only single parameter, i.e., $$\sigma ^{\mathrm{o}}$$ . Near surface SM measurements were related to $$\sigma ^{\mathrm{o}}_{\mathrm{RH}}$$ , $$\sigma ^{\mathrm{o}}_{\mathrm{RV}} {-} \sigma ^{\mathrm{o}}_{\mathrm{RH}}$$ derived using 5.35 GHz (C-band) image of RISAT-1 and $${\hbox {RMS}}_{\mathrm{height}}$$ . The roughness component derived in terms of $${\hbox {RMS}}_{\mathrm{height}}$$ showed a good positive correlation with $$\sigma ^{\mathrm{o}}_{\mathrm{RV}} {-} \sigma ^{\mathrm{o}}_{\mathrm{RH}} \, (R^{2} = 0.65)$$ . By considering all the major influencing factors ( $$\sigma ^{\mathrm{o}}_{\mathrm{RH}}$$ , $$\sigma ^{\mathrm{o}}_{\mathrm{RV}} {-} \sigma ^{\mathrm{o}}_{\mathrm{RH}}$$ , and $${\hbox {RMS}}_{\mathrm{height}}$$ ), an SEM was developed where SM (volumetric) predicted values depend on $$\sigma ^{\mathrm{o}}_{\mathrm{RH}}$$ , $$\sigma ^{\mathrm{o}}_{\mathrm{RV}} {-} \sigma ^{\mathrm{o}}_{\mathrm{RH}}$$ , and $${\hbox {RMS}}_{\mathrm{height}}$$ . This SEM showed $$R^{2}$$ of 0.87 and adjusted $$R^{2}$$ of 0.85, multiple R=0.94 and with standard error of 0.05 at 95% confidence level. Validation of the SM derived from semi-empirical model with observed measurement ( $${\hbox {SM}}_{\mathrm{Observed}}$$ ) showed root mean square error (RMSE) = 0.06, relative-RMSE (R-RMSE) = 0.18, mean absolute error (MAE) = 0.04, normalized RMSE (NRMSE) = 0.17, Nash–Sutcliffe efficiency (NSE) = 0.91 ( $${\approx } 1$$ ), index of agreement (d) = 1, coefficient of determination $$(R^{2}) = 0.87$$ , mean bias error (MBE) = 0.04, standard error of estimate (SEE) = 0.10, volume error (VE) = 0.15, variance of the distribution of differences $$({\hbox {S}}_{\mathrm{d}}^{2}) = 0.004$$ . The developed SEM showed better performance in estimating SM than Topp empirical model which is based only on $$\sigma ^{\mathrm{o}}$$ . By using the developed SEM, top soil SM can be estimated with low mean absolute percent error (MAPE) = 1.39 and can be used for operational applications.