Utilization of NWP model outputs for short-range weather forecast (Case study: Pontianak, Pekanbaru, Semarang, Surabaya, and Palu)

Abstract

Numerical weather prediction (NWP) is an operational daily weather forecasting using mathematical equations by super-computer. The skill of NWP has improved significantly since it was first generated fourty years ago. The urgent needs for daily weather forecasting in Indonesia has led to the utilization of NWP model outputs to be incorporated into the forecasting scheme. NWP provides the basic guidance for weather forecasting beyond the first few hours. Unfortunately, spatial resolution of NWP model is still too rough to describe local conditions. Model output statistics (MOS) technique may be used to reduce error and increase accuracy of NWP model ouputs. The objective of this study includes the selection of the best combination of atmospheric predictors for daily precipitation, relative humidity, maximum temperature, and minimum temperature in five locations, namely, Pontianak, Pekanbaru, Semarang, Surabaya, and Palu which characterized three dominant rainfall regions in Indonesia, for the wet and dry seasons. The subsequent objective is to identify locations where NWP model outputs resemble observational data. Model outputs patterns are compared with observational data to analyze consistency, and whenever necessary some forms of correction factor is applied. The results indicate that humidity at the surface (at z=2m) and at mid-tropospheric levels (rh0 and rh8), and pressure vertical velocity at mid-tropospheric levels (vv7 and vv5) are the relevant variables for daily precipitation in all of the locations and all seasons analyzed. The most potential location to utilize NWP model is Palu. In Palu, surface air temperature (T0) and surface dew point temperature (DP0) are the relevant variables for relative humidity; humidity at the surface (rh0) is the relevant variable for maximum temperature; while surface dew point temperature (DP0) is the relevant variable for minimum temperature. Closeness of model outputs with observational data is inferred using error indicators, namely, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE).