Development and accuracy assessment of a 12-digit hydrologic unit code based real-time climate database for hydrologic models in the U.S.

Authors: GAO, JUNGANG - Texas Agrilife Research; BIEGER, KATRIN - Texas Agrilife Research; White, Michael; Arnold, Jeffrey

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/7/2020
Publication Date: 3/10/2020
Publication URL: https://handle.nal.usda.gov/10113/6861004
Citation: Gao, J., Bieger, K., White, M.J., Arnold, J.G. 2020. Development and accuracy assessment of a 12-digit hydrologic unit code based real-time climate database for hydrologic models in the U.S. Journal of Hydrology. 586:124817. https://doi.org/10.1016/j.jhydrol.2020.124817.
DOI: https://doi.org/10.1016/j.jhydrol.2020.124817

Interpretive Summary: Hydrologic models are widely used to predict the effect of human activity on the environment. Accurate daily weather data are critical input but little research on integrating various gridded data by watershed boundary has been conducted. This study evaluated four extracting methods (mean (MN), median (MD), centroid (CT), and area-weighted (AW) approaches) for summarizing weather data for sub-watersheds defined in a hydrologic model. The hydrologic model with the CT weather data performed the best, followed by MN, AW, and then MD. Using this method, a real-time dataset including historical and forecast data at HUC-12 level across the conterminous United States was created. Forecast data often included bias in the extreme weather conditions, such as heavy precipitation and scorching temperature or in the presence of strong frontal systems presumably due to the exact location and speed of the front being difficult to predict. Weather data from this work will be published online to facilitate hydrologic modeling efforts in the U.S. and inform users about the uncertainty in the forecast data.

Technical Abstract: Accurate daily weather data are critical for hydrologic models simulating and predicting hydrologic processes. Many researchers have focused on the impacts of precipitation on hydrologic simulations, but few studies integrated both temperature and precipitation data for historical and forecast periods in hydrologic models and evaluated the weather data accuracy at national scale. This study evaluated four extracting methods (mean (MN), median (MD), centroid (CT), and area-weighted (AW) approaches) for summarizing weather data for sub-watersheds defined in a hydrologic model. Firstly, an optimized extracting method was used to develop a real-time HUC-12 (12-digit Hydrologic Unit Code) level dataset for the Conterminous United States. The hydrologic model with the CT weather data performed the best, followed by MN, AW, and then MD. Secondly, per this method, a real-time dataset including historical and forecast data at HUC-12 level across the conterminous United States was created. Last, continuous daily forecast data at national scale displayed that large forecast overestimations were usually observed in large forecast precipitation events over 20 mm. Simultaneously, there were large underestimations in small forecast precipitation events less than 5 mm. Forecast maximum temperature showed a more substantial bias than that minimum temperature, with the largest underestimation for the lower forecast maximum temperature less than 15 °C. With regard to data stability of the historical observed temperature data, provisional and early temperature data from Parameter-elevation Regressions on Independent Slopes Model (PRISM) in the most recent seven months are as reliable as the stable data that is subject to quality control measures before it replaces the provisional and early data. However, forecast data often included bias in the extreme weather conditions, such as heavy precipitation and scorching temperature. Forecast data, which is updated multiple times each day, was frequently subject to an opposite bias (trending toward less extreme values) in the presence of strong frontal systems presumably due to the exact location and speed of the front being difficult to predict. Fully processed weather data from this work will be published online to facilitate hydrologic modeling efforts in the U.S. and inform users about the uncertainty in the forecast data.