Skip to main content
ARS Home » Plains Area » El Reno, Oklahoma » Grazinglands Research Laboratory » Agroclimate and Natural Resources Research » Research » Publications at this Location » Publication #355305

Research Project: Towards Resilient Agricultural Systems to Enhance Water Availability, Quality, and Other Ecosystem Services under Changing Climate and Land Use

Location: Agroclimate and Natural Resources Research

Title: A parallel computation tool for dynamic sensitivity and model performance analysis of APEX: Evapotranspiration modeling

Author
item Talebizadeh, Mansour - Orise Fellow
item Moriasi, Daniel
item Steiner, Jean
item Gowda, Prasanna
item Tadesse, Haile
item Nelson, Amanda
item Starks, Patrick - Pat

Submitted to: Journal of the American Water Resources Association
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/4/2019
Publication Date: 5/28/2019
Citation: Talebizadeh, M., Moriasi, D.N., Steiner, J.L., Gowda, P.H., Tadesse, H.K., Nelson, A.M., Starks, P.J. 2019. A parallel computation tool for dynamic sensitivity and model performance analysis of APEX: Evapotranspiration modeling. Journal of the American Water Resources Association. https://doi.org/10.1111/1752-1688.12758.
DOI: https://doi.org/10.1111/1752-1688.12758

Interpretive Summary: Hydrologic and water quality models are increasingly used to evaluate the impacts of climate, land use, and land management practices on quantity and quality of land and water resources. Despite the utility of these models, there are situations such as dynamic sensitivity analysis and multiple scenarios where users need to run models repeatedly using different settings to obtain optimal solutions, which result in a time-consuming process, making it impossible to render results within a reasonable computational time. Therefore, an open source software packaged named PARAPEX was developed in R language to perform sensitivity and uncertainty analysis for the APEX model using parallel computation. The software was applied to a research field at USDA-ARS Conservation and Production Research Laboratory in Bushland, Texas. Compared to sequential framework, the parallel computation framework significantly reduced the computation time from 5939 seconds to 379 seconds. The dynamic sensitivity analysis results, revealed the importance of different model parameters during the simulation period. In addition, dynamic model performance analysis identified potentials for improving crop-growth model components. The improved computational performance provided by PARAPEX help users to analyze a large number of scenarios within a reasonable time. The same framework and be extended and applied to other hydrologic/agronomic models for performing a comprehensive parameter sensitivity and scenario analysis.

Technical Abstract: In this study, an open-source software package named PARAPEX was developed in R to perform dynamic sensitivity and uncertainty analysis for the APEX model using parallel computation. The PARAPEX package was used to conduct dynamic sensitivity analysis of evapotranspiration (ET) predictions using lysimetric data from a research filed managed by the USDA-ARS Conservation and Production Research Laboratory in Bushland, Texas. Compared to the sequential computation that uses a single computation node, the parallel computation framework provided by PARAPEX with 20 computation nodes significantly reduced the computation time from 5939 seconds to 379 seconds. The sensitivity analysis results indicated the model input parameter Soil_evap_plant_cover (accounting for the reducing effect of plant cover on evaporation from the soil surface), Root_growth_soil (accounting for the impact of soil on the plant root growth), and Microbial_top_soil_coeff (impacting the cycling and transformation of organic matter at the top soil layer) as top sensitive parameters based on the mean daily simulated ET and the Nash-Sutcliffe (NS) model performance measure. The dynamic performance analysis indicated poor performance by the APEX model during the growing seasons. The time series of median ET corresponding to the ensemble of simulations resulted in an NS value of 0.7, suggesting the utility of ensemble simulations especially in situations with limited or no data.