Test of an Algorithm for the Estimate of Soil Moisture from C-band SAR data with a Hydrological Model
Santi, Emanuele1; Paloscia, Simonetta1; Pettinato, Simone1; Notarnicola, Claudia2; Pasolli, Luca2; Pistocchi, Alberto2
1IFAC - CNR, ITALY; 2EURAC, ITALY
This paper presents a comparison between the soil moisture content (SMC) estimated from C-band SAR data by using a retrieval algorithm based on Artificial Neural Network, and the SMC simulated by a hydrological model. Both SMC values have successively compared to the ground-truth measurements carried simultaneously to the SAR overpasses. This analysis can be considered as a preliminary test for SAR data integration into hydrological models for continuous calibration. In view of the upcoming Sentinel 1 with high revisiting time, the calibration of the hydrological model may be carried out on a weekly basis thus allowing improvement in the model predictions. The study was carried out in an agricultural test site located in North-west Italy, in the Scrivia river basin. The hydrological model considered a single, homogeneous, well-mixed soil layer, for which the daily water balance was computed by accounting for the precipitation, runoff, gravity-driven infiltration and actual evapotraspiration. This model was calibrated on the Scrivia basin, where the necessary input parameters were available. The minimum set of parameters required as inputs included precipitation, mean, minimum and maximum temperature at daily steps, as well as an indication of soil texture. Potential evapotranspiration was estimated by using the well-known Hargreaves-Samani formula. The hydraulic behavior of the top soil layer was described using parameters estimated on the basis of soil texture. The algorithm used for estimating SMC is based on feedforward multilayer perceptron artificial neural networks (MLP-ANN) The algorithm was implemented within the framework of an ESA project devoted to the generation of an operative algorithm for the estimate in a real-time of SMC from Sentinel-1 data (Paloscia et al. 2013). The dataset implemented for the ANN training was obtained by combining experimental measurements of backscattering coefficients, corresponding ground parameters, and data simulated using electromagnetic forward models. The backscattering of the bare rough surfaces was simulated using the AIEM and Oh models, while the contribution of light vegetation was accounted for by using the so-called "Cloud Model", deriving the information on vegetation water content from the NDVI trough a semi-empirical relationship. Six ANNs were defined and trained specifically, in order to deal with the different SAR configurations in terms of polarizations and available ancillary data. If available, the cross- polarized channel was considered instead of NDVI to account for the effect of vegetation cover. The training was carried out by considering the EO data (measured or modeled) as ANN inputs and the SMC as output. The experiment was conducted in an agricultural test site located in North-west Italy, in the Scrivia river basin. It is a flat agricultural plain of about 300 Km2, situated close to the confluence of the Scrivia and the Po Rivers (Lat: 45.0N, Lon: 8.8E), for which six ENVISAT/ASAR images, mainly collected in HH/HV polarizations at an incidence angle of 23° from 2003 to 2009, were available. Simultaneously with satellite acquisitions, ground campaigns were carried out in order to measure all the significant vegetation and soil parameters, such as plant density, leaf and stalk dimensions, the number of leaves per plant, plant water content, SMC, and surface roughness. At least 5-6 samples of SMC (measured with TDR probes) and vegetation were collected for each field considered, while surface roughness was measured with a needle profilometer along and across the rows. The SMC estimated by the SAR algorithm, the SMC estimated by the hydrological model, and the SMC measured on ground resulted to be in a rather good agreement. The hydrological model simulations were performed at two soil depths: 30 and 5 cm and showed that the 30 cm simulations indicated SMC values higher than the satellites estimates. In the 5-cm simulations, instead, the agreement between hydrological simulations, satellite estimates and ground measurements could be considered satisfactory, at least in this preliminary comparison, showing a RMSE ranging from 4,5% and 5,3% (of volumetric SMC) for comparison with ground measurements and SAR estimates respectively.