| AFFILIATIONS | - Université de Tunis El Manar, ENIT, Tunis, Tunisia
- University of Twente, ITC, Enschede, Netherlands
- Université de Tunis El Manar, ENIT, Tunis, Tunisia
|
| ABSTRACT | Five satellite rainfall estimation algorithms are evaluated against rainfall extreme events over Northern Tunisia. Evaluations are implemented for 77 daily heavy rainfall events observed during the study period from 2007 to 2009. We mean by heavy event the rain exceeding 50 mm/day for at least one station of the study area. Daily rainfall observations are derived from an average of 318 rain gauges interpolated using inverse distance method. Three statistical indices (correlation coefficient (R), amounts ratio bias (RB), normalized root mean square error (NRMSE)), as well as a contingency table (probability of detection (POD) and false alarm ratio (FAR)), are quantified to evaluate the satellite rainfall estimates quality. In general, the product that is most close to observed rainfall according to POD, RB, and R is raw Climate Prediction Center (CPC) Morphing Technique (CMORPH), followed by adjusted CMORPH, Tropical Rainfall Measuring Mission (TRMM 3B42), Precipitation Estimation from RemotelySensedInformationusingArtificial Neural Networks (PERSIANN), and Multisensor Precipitation Estimate (MPE). In terms of FAR, TRMM 3B42 product shows most potential followed by raw CMORPH, adjusted CMORPH, MPE, and PERSIANN. Additionally, TRMM 3B42 product shows the best skills in term of NRMSE followed by PERSIANN, raw CMORPH, MPE, and adjusted CMORPH. In terms of the POD, all the products perform better during the wet season (from November to April). An overestimation is noticed according to the RB coefficient for TRMM 3B42, PERSIANN, and MPE for both dry and wet seasons. However raw CMORPH and adjusted CMORPH showed an overestimation foremost during the dry season (from May to October). In term of R, all the products perform better during the wet season except MPE. For the NRMSE the two CMORPH products get a bit higher NRMSE coefficients in comparison with the other products for both seasons. We conclude that differences between the products are large. Thus, the evaluation of any satellite data before any application is recommended. In this case study, based on the average of all the evaluation metrics, the adjusted CMORPH is the product the most adequate. |