Accuracy of Giovanni and Marksim Software Packages for Generating Daily Rainfall Data in Selected Bimodal Climatic Areas in Tanzania

Abstract

Agricultural adaptation to climate change requires accurate, unbiased, and reliable climate data. Availability of observed climatic data is limited because of inadequate weather stations. Rainfall simulation models are important tools for generating rainfall data in areas with limited or no observed data. Various weather generators have been developed that can produce time series of climate data. Verification of the applicability of the generated data is essential in order to determine their accuracy and reliability for use in areas different from those that were used during models development. Marksim and Giovanni weather generators were compared against 10 years of observed data (1998-2007) for their performance in simulating rainfall in four stations within the northern bimodal areas of Tanzania. The observed and generated data were analyzed using climatic dialog of the INSTAT program. Results indicated that during the long rain season (masika) Giovanni predicted well the rainfall amounts, rainy days, and maximum dry spells compared to Marksim model. The Marksim model estimated seasonal lengths much better than the Giovanni model during masika. During short rain season (vuli), Giovanni was much better than Marksim. All the two software packages had better predictions during masika compared to vuli. The Giovanni model estimated probabilities of occurrence of rainfall much better (RMSE = 0.23, MAE = 0.18, and d =0.75) than Marksim (RMSE = 0.28, MAE = 0.23, and d = 0.63). The Marksim model over-predicted the probabilities of occurrence of dry spells greater than seven days (MBE = 0.17) compared to the Giovanni model (MBE = 0.01). In general the Giovanni model was more accurate than the Marksim model in most of the observed weather variables. The web based Giovanni model is better suited to the northern bimodal areas of Tanzania. The Marksim model produced more accurate climatic data when the long-term average climate data are used as input variables. This study recommends the use of rainfall data generated using Giovanni software over Marksim, for areas receiving bimodal rainfall regimes similar to the northern bimodal areas of Tanzania.

Key words: Bimodal rainfall; long-term average climate data; dry spells; Giovanni; INSTAT; Marksim