Added value by a regional climate model: precipitation and extreme temperature

Abstract:

Evaluations of RCM downscaling skills in temperature extremes and understanding the extreme temperature changes are among the greatest desire because temperature extremes are expected to change in severity, frequency, and duration as a result of anthropogenic global warming. Here we develop a “borrowing strength” GEV model to evaluate the Weather Research and Forecast (WRF) downscaling skill in extreme temperature. A spatial-varying long-term trend is assumed for the location parameter of the GEV model, which estimates the trend of extreme temperature. The performance of WRF in extreme maximum temperature is evaluated by comparing with a reanalysis data set from North American Regional Reanalysis (NARR). The WRF model’s ability to retain and add value beyond the driving conditions is investigated by comparing with National Centers for Environmental Prediction-U.S. Department of Energy Reanalysis II (NCEP-R2), which provides the initial and boundary conditions to the WRF. Generally, despite of certain biases for different parameters over different parts of US, there is a good agreement in spatial pattern and magnitude of three parameters of GEV distribution between the WRF and NARR. The WRF retains some value from its driver — NCEP-R2, but more importantly, it adds small-scale features beyond NCEP-R2, which is close to the NARR and essential to reproduce the extreme events.

On the other hand, we will show useful approaches for identifying added value by the WRF beyond the driving data in terms of precipitation (light and heavy). The approaches include (1) spatial correlation for a range of spatial lags using total monthly precipitation and non-seasonal precipitation components; and (2) spatio-temporal correlation for a wide range of distances, directions, and time lags using daily precipitation occurrence. The WRF provides realistic details of spatial and spatio-temporal correlations at finer scales than the NCEP-R2, and interestingly, the WRF also provides more accurate correlations than the driver (NCEP-R2) at the resolution of the NCEP-R2.