Driving small-scale agricultural and hydrological models using globally available stochastic climate datasets that include point-scale precipitation

Authors: FULLHART, A. - University Of Arizona; GAO, S. - University Of Arizona; HERNANDEZ, M. - University Of Arizona; WANG, W. - Beijing Normal University; Armendariz, Gerardo; Goodrich, David - Dave

Submitted to: Journal of Soil and Water Conservation
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/17/2024
Publication Date: N/A
Citation: N/A

Interpretive Summary: Climate datasets are needed for many USDA computer models that simulate agricultural and hydrological systems. Traditionally, such models have used observed climate records from meteorological stations, though global climate models offer an alternative source of climate data that is increasingly applied in modeling. However, the coverages of global climate models often represent spatial and temporal scales that are too coarse in their resolutions for application to small-scale model domains, e.g., agricultural fields or hillslopes. In particular, this issue results in underestimation of precipitation intensity, runoff, and subsequent soil erosion. A new globally available collection of climate inputs for small-scale USDA models addresses this issue and provides a variety of required weather variables. Using this dataset, analysis was done that demonstrates international modeling case studies and identifies model outputs affected by issues with scale. The models applied were the Water Erosion Prediction Project (WEPP) model, the Rangeland Hydrology and Erosion Model (RHEM), and the Hydrus 1-D model. Results showed that runoff and erosion are better estimated by applying the new small-scale climate dataset compared to existing global climate datasets, while other model outputs such as crop yield and evapotranspiration were found to be consistent between the different climate datasets. The new small-scale climate dataset is available for all agricultural regions and provides a solution to numerous data limitations commonly faced in the application of agricultural and hydrological models.

Technical Abstract: Stochastic weather generators (SWGs) are a portable solution for obtaining site-specific climate time series needed in many modeling applications. The SWG called CLImate GENerator (CLIGEN) provides simulated daily sequences of basic weather variables in a stationary climate, including intra- and inter-annual variability, and statistical information related to long-term normals. Required input parameter sets for CLIGEN are available in the form of spatial grids and ground-based networks that cover most global land area. Crucially, CLIGEN outputs contain point-scale precipitation time series that make it a compatible driver for small-scale model domains (e.g., individual fields or hillslopes). In such applications, precipitation drivers acting at a single point are more likely to accurately force runoff, erosion, and other processes than comparable grid-scale or mean areal precipitation drivers. In order to validate simulated precipitation drivers from gridded CLIGEN datasets and identify model outcomes that are sensitive to the scale of precipitation, the present analysis contrasts simulated point-scale precipitation drivers against both observed precipitation and the popular daily, grid-scale Climate Hazards Ground InfraRed Precipitation with Stations (CHIRPS) dataset. Case studies for the climate drivers involved model simulations at selected international sites using the following models: the Water Erosion Prediction Project (WEPP) model, the Rangeland Hydrology and Erosion Model (RHEM), and the Hydrus 1-D model. The analysis concluded that simulated precipitation drivers produce runoff and erosion rates consistent with outcomes from observation-based drivers. The stochastic data is demonstrated to meet model requirements in a number of contexts, and given its global availability, there are a wide variety of potential applications, particularly when dealing with data limitations in small-scale modeling.