IMPORTANCE OF UPDATING FOR MONTHLY RAINFALL PREDICTION BASED ON ECMWFs4

Abstract

There is uniqueness in climate services in East Java. Rainfall information is delivered as N-1 analysis for ongoing months N+1, N+2, and N+3 for monthly prediction. This study aims to investigate whether updating monthly predictions improves prediction accuracy. The verification method for this study is based on the percentage accuracy of the rain class category according to SNI 8196: 2015. The data used for this study is ECMWF's monthly rainfall prediction that has three lags system (1, 2, and 3). Rasters of monthly rainfall interpolation from the main rainfall observation (197 locations) in East Java from April 2015 to May 2020 (62 months) are used for the verification process. The temporal and spatial analysis then conducted using R (+ package raster). Studies based on the local governmental zone are also used. In general, the result shows that almost all months need updating, except April-September-October. Verification of ECMWFs4 shows a better verification result (0,56) in the past five years (2016-2020) for March. The regions that need monthly updating are Bawean island, the coast of Gresik, Pasuruan, and Banyuwangi