On the predictability of monthly precipitation across the U.S. Midwest

Monthly precipitation amount often influences governments and stakeholders in making decisions about water management. The ability of anticipating harmful events is therefore crucial to mitigate economic and social damages. In this study we propose simple linear regression models correlating the total monthly precipitation (i.e., the predictand) to temperature (i.e., the predictor) which can be potentially used to predict future precipitation amounts through forecasts of temperature. We define three different temperature predictors, considering the average of one, two or three months prior the analyzed month. We use raster maps of precipitation and temperature at ~30 km resolution from 1895 to 2017 of the U.S. Midwest to calibrate and validate our models. We find, overall, low prediction skill. For the summer months (June-July-August), the skill of the models in reproducing the observed precipitation slightly increases when considering a shorter period among which the average temperature is referred. The best skill is obtained for the months of June and July when using the previous month's average temperature as predictor.