Validation of Satellite-based Rainfall Estimation over the Limpopo Basin

Abstract

The main objective of this study was to validate satellite-based rainfall estimation algorithms over the Limpopo Basin. The satellite rainfall estimation was done using four algorithms which combine infrared and passive microwave data. These are Climate Prediction Centre (CPC) Morphing (CMORPH), Multiple Precipitation Analysis (MPA), Precipitation Estimation Remotely Sensed Information using Neural Network (PERSIANN), and Naval Research Laboratory Blended (NRLB) methods. The validation was done by comparing satellite rainfall estimates with daily gauge data collected from Botswana, Mozambique, South Africa and Zimbabwe and three-daily moving area average rainfall with three-daily gauge data during 2005/2006 rainfall season. The gauge data were averaged into grid boxes of 0.25o x 0.25o resolution, using the inverse weighting interpolation method and the satellite estimates were developed using pixel by pixel at resolution of 0.25o. A surface mask was used over the Limpopo Basin.

A variety of validation statistics were used to measure different aspects of each algorithm quality, based on contingency tables, and threshold rain of 1 mm/day. All the algorithms showed some skill in estimating rainfall with coefficients of determination ranging from 0.528 to 0.69. Both CMORPH and MPA algorithms exhibited high values of coefficient of determination. The values ranged from 0.5495 to 0.7767 by moving the daily area average rainfall to every three days. The statistical results showed that the CMORPH algorithm performed better than the other three methods. All satellite estimation methods overestimated rainfall with a positive bias which ranged from 0.2 to 0.3. The mean absolute error and root-mean square error ranged from 2.5 – 5.2 and 5.7 – 8.9 mm/day, respectively. The errors were caused by the sparse rain gauge network quality of rain gauge data and inadequacy in the satellite estimation algorithms. The Heidke skill score ranged from 0.19 to 0.28.

The study concluded that CMORPH performed better than the other techniques although all methods overestimate rainfall in the region. The satellite rainfall estimation algorithms can perform better if there is improvement of rainfall measurement infrastructure and data exchange within the Limpopo Basin between the four countries.