Distributed quantitative precipitation forecasts combining information from radar and numerical weather prediction model outputs

Abstract

Applications of distributed Quantitative Precipitation Forecasts (QPF) range from flood forecasting to transportation. Obtaining QPF is acknowledged to be one of the most challenging areas in hydrology and meteorology. Recent advances in precipitation physics, Numerical Weather Prediction (NWP) models, availability of high quality remotely sensed measurements, and data dictated forecasting tools, offer the opportunity of improvements in this area. Investigative studies were performed to quantify the value of available tools and data, which indicated the promise and the pitfalls of emerging ideas. Our studies suggested that an intelligent combination of NWP model outputs and remotely sensed radar measurements, that uses process physics and data dictated tools, could improve distributed QPF. Radar measurements have distributed structure, while NWP-QPF incorporate large scale physics. Localized precipitation processes are not well handled by NWP models, and grid average NWP-QPF are not too useful for distributed QPF owing to the spatial variability of rainfall. However, forecasts for atmospheric variables from NWP have information relevant for modeling localized processes and improving distributed QPF, especially in the Summer. Certain precipitation processes like advection and large scale processes could be modeled using physically based algorithms. The physics for other processes like localized convection or residual structures are not too well understood, and data dictated tools like traditional statistical models or Artificial Neural Networks (ANN) are often more applicable.

(cont.) A new strategy for distributed QPF has been proposed that utilizes information from radar and NWP. This strategy decomposes the QPF problem into component processes, and models these processes using precipitation physics and data dictated tools, as appropriate and applicable. The proposed strategy improves distributed QPF over existing techniques like radar extrapolation alone, NWP-QPF with or without statistical error correction, hybrid models that combine radar extrapolation with NWP-QPF, parameterized physically based methods, and data dictated tools alone. New insights are obtained on the component processes of distributed precipitation, the information content in radar and NWP, and the achievable precipitation predictability.