Performance Optimization of Operational WRF Model Configured for Indian Monsoon Region

Providing timely weather forecasts from operational weather forecast centers is extremely critical as many sectors rely on accurate predictions provided by numerical weather models. Weather research and forecasting (WRF) is one of primary tools used in generating weather predictions at operational weather centers, and an optimal domain configuration of WRF along with suitable combination of robust computational resources and high-speed network are required to achieve the maximum performance on high-performance computing cluster (HPC) for the WRF model to deliver the timely forecasts. In this study, we have analyzed the number of methods to optimize WRF model to reduce computational time taking for the operational weather forecasts tested on HPC available at University Grants Commission center for mesosphere stratosphere troposphere radar applications, Sri Venkateswara University (SVU). To do this exercise, we have prepared a benchmark dataset by configuring WRF model for the Indian monsoon region as similar to real-time weather forecasting system model configuration. We have first carried out a series of scalability tests by increasing the number of computational nodes till it reaches a scalable point using the prepared benchmark dataset. Our node scalability results indicate the WRF model is scalable up to 65 nodes for the benchmark dataset and configured model domain on HPC available at UGC SVU center. As the total time taken for generating the model forecasts is the sum of computational time taken for predicting weather and the input/output (IO) time for writing into the storage disks. Further, we have performed several tests to optimize the time taken for IO by the weather model, and the results of IO tests clearly indicate that the WRF configured with parallel IO is highly beneficial method to reduce the total time taken for the generation of weather forecasts by the WRF model.