Cyber-enabled autocalibration of hydrologic models to support Open Science

Abstract

Automatic calibration (autocalibration) of models is a standard practice in hydrologic sciences. However, hydrologic modelers, while performing autocalibrations, spend considerable amount of time in data pre-processing, coding, and running simulations rather than focusing on science questions. Such inefficiency, as this paper outlines, stems from: (i) platform dependence, (ii) limited computational resource, (iii) limited programming literacy, (iv) limited model structure and source code literacy, and (v) lack of data-model interoperability in the so-called autocalibration process. By expanding and enhancing an existing web-based modeling platform SWATShare, developed for the Soil and Water Assessment Tool (SWAT) hydrologic model, this paper demonstrates a generalizable pathway to making autocalibration efficient via cyberinfrastructure (CI) solutions. SWATShare is a collaborative platform for sharing and visualization of SWAT models, model results, and metadata online. This paper describes the front and back end architectures of SWATShare for enabling efficient SWAT model autocalibration on the web. In addition, this paper also demonstrates three implementation case studies to validate the autocalibration workflow and results. Results from these implementations show that SWATShare autocalibration can produce streamflow hydrograph and parameters that are comparable with commonly used offline SWATCUP calibration outputs. In some instances, the parameter values from SWATShare calibration are more physically relevant than those from SWATCUP. Although the discussion in this paper is in the context of SWAT and SWATShare, the conceptual and technical design presented here can be used as an Open Science blueprint for similar CI-enabled developments in other hydrologic models, and more importantly, in other domains of Earth system sciences.