A statistical analysis of the differences between rainfall estimated by Chennai DWR and conventional rainfall data on monthly and seasonal scales during the Indian northeast monsoon season

Abstract

The first Doppler Weather Radar (DWR) of India Meteorological Department has been functional at Chennai since the year 2002 providing various meteorological and hydrological products. Validation and statistical analysis of the DWR estimated rainfall (RERF, x) data with rain gauge measured rainfall (RGRF, y) of 34 land based stations located in the semi-circular land area within 100 km radius of Chennai DWR (CDLR100) has been performed for the northeast monsoon (NEM) season of October-November-December (OND) for the 12 year period 2002-13. The monthly and seasonal data have been derived using more than 1.42 lakh discrete daily RERF values available at a high resolution of 333 m × 333 m. The major objective of the study is to compute the various statistical parameters of x and y including the bias between them on monthly and seasonal scales and to draw certain inferences. The analysis was done using three different types of averaging. The yearly means of x and y for OND over CDLR100 manifested both positive and negative epochs with the mean absolute deviation (MAD) computed as 11 cm (17% of mean). The short term normals over CDLR100 are derived as 274.9, 262.6, 96.5 and 629.8 mm for x and 243.8, 254.6, 128.0 and 627.4 mm for y for October, November, December and OND yielding bias values of -31.2, -8.0, 31.5 and -2.4 mm respectively. The MAD for OND rainfall computed by pooling in all the 12 ´ 34 values is quite substantial at around 19 cm (30% of mean). The RF bias for each month / NEM season is shown to be independent of the geographical locations of the stations using correlation analysis. Based on the raw values of x and a proportional correction technique, estimated values of at the 1.42 lakh grid points of CDLR100 were derived yielding spatial means of 273.3, 262.2, 92.5 and 628.4 mm for x and 243.4, 254.3, 122.9 and 622.1 mm for for the three months and OND respectively. The importance of size of the bias in the correct interpretation of x has been discussed. A few suggestions based on certain statistical considerations have been putforth for decreasing the bias.