Evaluation of Soil Mapping Methods for SWAT Hydrological Modeling through Multi-Objective Calibration

Soil data is a crucial component for hydrological models operating at the catchment scale, such as the Soil and Water Assessment Tool (SWAT). Nevertheless, the reliability of these models is heavily contingent upon the quality and spatial resolution of the soil information employed. This study addresses the pressing need for robust soil data in SWAT modeling by evaluating various soil mapping techniques.

The first objective was to prove different mapping techniques, such as textural class combination and clustering approach using soil grid data to have different soil maps. The performance of those maps, together with WRB, WSR, Zobler, HWSD v2.0 and DSOLMAPS, was evaluated to improve the accuracy and reliability of the SWAT hydrological model. To achieve this, we conduct a comprehensive investigation involving multi-objective calibration, utilizing both flow data and soil moisture data to calibrate the model. Finally we incorporate a  pedotransfer function to include the landcover effect on Saturated Hydraulic Conductivity to improve the reliability of soil hydrological processes in the SWAT model.

The study area, situated within the Cauquenes River Catchment, presents a complex hydrological system characterized by substantial spatial heterogeneity in soil properties. The soil mapping techniques under evaluation encompass traditional soil survey data integrated with remotely sensed soil information and machine learning-based soil mapping methodologies. These methods are compared in their ability to enhance the SWAT model's representation of the catchment's hydrological dynamics.

In the case of the kmeans clustering approach the results of soil clusters are equivalent to soil units. A number of clusters from 3 to 100 were evaluated with the lowest DB index. Clusters from 3 to 16 presented an optimal range. The SWAT model calibration was performed under multi-objective evaluation, with kmeans soil cluster and DSoilMaps with better result for daily simulations.

The work to correct the application of soil data, including in situ observation, satellite data and machine learning approach, provides a valuable approach to improve the calibration and validation processes of hydrological models in semi-arid regions, important for cacthment management and decision making processes, and to correctly assess the impacts of land use changes, climate variability and extreme events on water resources.