2022 CSDMS meeting-068

From CSDMS



(if you haven't already)




Log in (or create account for non-CSDMS members)
Forgot username? Search or email:CSDMSweb@colorado.edu


Browse  abstracts


Sensitivity of high-resolution WRF model to initialization starting point

Mahdad Talebpour, University of Maryland Baltimore County Baltimore Maryland, United States. mahdad1@umbc.edu
Claire Welty, UMBC Baltimore Maryland, United States. weltyc@umbc.edu
Elie Bou-Zeid, Princeton University Princeton New Jersey, United States. ebouzeid@princeton.edu



Soil moisture state has a critical role on subsurface-land surface-atmosphere energy and water balance. Yet, there is still no consensus on how to initialize atmospheric-hydrologic models to improve the representation of soil moisture content. Lack of accurate observational soil moisture data is the root of this issue. Although there has been progress in providing remotely sensed soil moisture data (e.g., Soil Moisture Active Passive (SMAP) data), their resolution is not adequate for high-resolution simulations. As an alternative approach, many atmospheric-hydrological simulations use various spin-up periods prior to the start of their analysis to perturb and improve the low-resolution soil moisture with precipitation. It has been shown that such method can improve soil moisture distribution in some studies in comparison to observational data. However, starting simulations from earlier times can cause divergence from accurate initial atmospheric conditions, which were obtained from observational data when simulation reaches the analysis period of interest. Therefore, there is a tradeoff between starting several days or hours before the analysis period in accurate representation of atmospheric data versus soil moisture input. In this study, we evaluated the sensitivity of a high-resolution (150-m) Weather Research and Forecasting (WRF) model to initialization starting point. We ran five nested domains with 12150-, 4050-, 1350-, 450-, and 150-m resolutions to downscale NCEP North American Regional Reanalysis (NARR) to our domain of interest encompassing Baltimore-Washington metropolitan area. The five domains were run in three scenarios starting 4, 7, and 14 days before the analysis period. Land surface temperature (LST) output was compared to LandSat data to investigate the impact of initialization starting point on model’s LST predictability. Results indicate that while the three scenarios underperformed in prediction of the urban heat island, there was no significant difference among the three scenarios. We determined that one of WRF’s thermal roughness parameterizations, which improves LST simulation over nonurban areas, caused significant errors in LST prediction over urban areas. Further simulations and analysis are underway to improve urban LST prediction. The three case scenarios will be compared against LandSat again when urban LST prediction is improved.