Flood forecast based on numerical weather prediction and distributed runoff model

Abstract

The accuracy of quantitative precipitation forecast from numerical weather prediction (NWP) grows as higher resolutions are achieved by the computation capacity of supercomputers. Various distributed dataset, globally and locally, with better spatial and temporal resolution has been rapidly developed. The objective of this paper is to have a flood forecast model that utilises these great benefits and gives reliable accuracy. Another asset of a flood forecast model is a conceptual-distributed runoff model; it has been chosen because of its simplicity. Calibration and validation of the model has been met by good agreement for events in 2002 in the case of a 237 km operational scale basin. These results are based on input Grid Point Value (GPV) precipitation of Japan radar observation. Forecasted Precipitation was based on a GPV Mesoscale Model of Japan NWP. It had an 18 hour lead time and updated four times a day. Flood forecasting based on input from forecasted precipitation shows that the accuracy decreases as lead-time increased. It is clear that flood forecasting depends on precipitation forecast accuracy. Observed precipitation and discharge are used for model updating to determine initial data for flood forecasting. The simplicity of the runoff model gives advantage on water content estimation in soil storage. It is necessary because the runoff model might have basic errors and it needs to have better initial data. Updating calculated discharge with observed discharge approximates the estimation. By estimating more correctly, the model shows to be more reliable.