SWAT model parameters correction based on multi-source remote sensing data in saline soil in Ebinur Lake Watershed

Abstract: SWAT model is one of the most widely used hydrological models in the world. The electrical conductivity (EC) is defaulted as 0 or 0.1, which might be not suitable for the soils with high salinity. In this study, we tested the feasibility of SWAT model with default EC values in simulating soil moisture and proposed a method to modify model parameters. The study area was Ebinur Lake Watershed. The watershed was located in Xinjiang with little rainfall and full sunshine. The evaporation was high. In the recent 10 years, the environment around the watershed was deteriorated, threatening sustainable development. The soil EC inversion was obtained by GF-1 16 m WFV hyperspectral remote sensing images. Different bands were used for calculation of vegetation index, soil index, salinity index and saturation. Then, these were used to build EC inversion model by the classification and regression tree method. The inversion values were compared with measured values. Then, the EC values were used to replace those in the Harmonized World Soil Database. Then, the EC distribution in Ebinur Lake Watershed was obtained. Then, the SWAT model driven by soil database, land use database and meteorological database was used for soil moisture simulation. For soil moisture simulation, meteorological database, soil database and land use database were used. The Landsat TM/ETM remote sensing images were used for land use classification. CMADS including temperature, pressure, wind speed, precipitation and radiation was used for meteorological database establishment. Soil EC and moisture were determined in 38 field sampling points. The measurements were used for model accuracy verification. The results showed that the root mean square error was 4.81 and 1.15 dS/m for soil depths of 0-40 and 40-100 cm, respectively. The relative error was 15.2% and 1.66%, respectively. The results showed the EC simulation by the model based on the index such as vegetation index, soil index, salinity index and saturation and EC was well. The surface had higher error since the surface soil had the high variation with coefficient of variation of 1.46. The T_ECE was modified by recalculating parameters in SWAT soil database. Then, soil moisture was calculated. The relative error was 63.04% and 39.20% before and after modification, respectively. The root mean square error was 1.79 and 1.34 mm before and after modification, respectively. It indicated that the modification was effective in improving soil moisture simulation accuracy by the SWAT model. The method proposed here is helpful in SWAT model use in saline soils.