Abstract
The aim of this study is to compare and discuss the validation of rainfall estimates from five methods, at various space and time scales. The studied areas are the Sahelian Africa where a network of about 580 raingauges is available and a small region covering Burkina-Faso where the raingauge network is denser, with 58 raingauges in a 2.5°×3° area. The reference estimation (i.e., the `ground-truth' rainfall) is computed by kriging the raingauge data. Rainfall is estimated with three satellite-based operational algorithms which provide 10-day cumulated rainfall using Meteosat infrared channel data. Rainfall is also estimated from two ground-based methods, using either climatologic data or the rain gauge data of the synoptic network which are available in real-time. One of the satellite methods is calibrated using the synoptic raingauge data and is thus a satellite–gauge combined method. Comparisons are performed for 10-day and 30-day integration periods and for spatial scales from 0.20 to 1 degree. A set of validation criteria is used for quality assessment. An interesting result is that the satellite-based methods and the ground-based methods lead to similar scores. The satellite–gauge combined method leads to slightly better scores. For this method, the estimation error for the 10-day cumulated rainfall and a 0.5°×0.5° spatial resolution is about 35% of the mean rainfall amount. Other results concern the general methodology of method intercomparison. It is shown that for accurate performance assessment a single criterion is seldom sufficient and that the performance criteria are sensible to dataset partitioning.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have