Impact of data assimilation on high‐resolution rainfall forecasts: A spatial, seasonal, and category analysis

Abstract

AbstractIn a limited area model (LAM), the impact of data assimilation is likely to depend on the background state through lateral boundary forcing; this may introduce certain seasonality in the impact of data assimilation on rainfall forecasting. It is also likely that the impact of data assimilation on forecasts will have certain spatial variability. Finally, owing to the convective nature of rainfall and the roles of parameterization scheme, the impact of data assimilation may depend on the category (intensity) of rainfall. Here these aspects for rainfall forecasts at high resolution were examined. Using a LAM (An advanced version of Weather Research and Forecasting Model), we have carried out twin simulations with and without data assimilation; the simulations without data assimilation are used as the benchmark for assessing the impact of data assimilation. Analysis of simulations for 40 sample days distributed over the years 2012–2014 over Karnataka (southern state in India) is carried out to estimate impact of data assimilation. Various statistical measures show that data assimilation improved the rainfall prediction in most cases; however, there is also strong seasonality and location dependence in impact of data assimilation. Our results also show that improvement due to data assimilation is higher/lower for lower/higher rainfall categories. Analysis shows that the cases where the initial states with data assimilation depart strongly from the first guess generally result in less or even negative impact. It is pointed out that the results have important implications in design of observation system and assessment of impact of forecasts.