PNNL

Abstract

Watershed models such as the Soil and Water Assessment Tool (\texttt{SWAT}) consist of high-dimensional physical and empirical parameters. These parameters must be calibrated for models to produce reliable predictions on hydrological fluxes and states. Existing parameter estimation methods can be time consuming, inefficient, and computationally expensive, with reduced accuracy when estimating high-dimensional parameters. In this paper, we present a fast, accurate, and reliable methodology to calibrate the \texttt{SWAT} model (i.e., 20 parameters) using scalable deep learning (DL). We developed DL-enabled inverse models based on convolutional neural networks to assimilate observed streamflow data and estimate the \texttt{SWAT} model parameters. Scalable hyperparameter tuning is performed using high-performance computing resources to identify the top 50 optimal neural network architectures. We used ensemble \texttt{SWAT} simulations to train, validate, and test the above DL models. We estimated the parameters of the \texttt{SWAT} model using observed streamflow data and assessed the impact of measurement errors on SWAT model calibration. We tested and validated the proposed scalable DL methodology on the American River Watershed, located in the Pacific Northwest-based Yakima River basin. Our results show that the DL model-based calibration is better than two popular parameter estimation methods (i.e., generalized likelihood uncertainty estimation [GLUE] and dynamically dimensioned search [DDS], a global optimization algorithm). The sensitive parameter sets estimated by DL yield narrower ranges than GLUE but broader than DDS produced values within the sampling range even under high relative observational errors. This narrow range of parameters shows the reliability of the proposed workflow to estimate sensitive parameters accurately even under high observed flow errors. The \texttt{SWAT} model calibration performance using DL, GLUE, and DDS are estimated using $R^2$ and a set of efficiency metrics including Nash-Sutcliffe, logarithmic Nash-Sutcliffe, Kling-Gupta, modified Kling-Gupta, and non-parametric Kling-Gupta scores. The best DL-based calibrated set has scores of 0.71, 0.75, 0.85, 0.85, 0.86, and 0.91. The best DDS-based calibrated set has scores of 0.62, 0.69, 0.8, 0.77, 0.79, and 0.82. The best GLUE-based calibrated set has scores of 0.56, 0.58, 0.71, 0.7, 0.71, and 0.8. The above scores show that the DL calibration provides more accurate low and high streamflow predictions than the GLUE and DDS sets. For its fast and reasonably accurate estimations of process parameters, the proposed DL workflow is attractive for calibrating integrated hydrologic models for large spatial-scale applications.