Evaluation of the accuracy of TAMSAT weather data on Rwanda for improved maize yield simulation with APSIM

Abstract

Across Rwanda, the raingauge network is extremely sparse and it often takes many months before raingauge records are accessible to the wider community. However, rainfall information is very critical to predict yields and inform food security early warning system of the country. TAMSAT provides a gridded daily rainfall map at a resolution of 4 km and this gives around 1,646 pixels for Rwanda. Since Rwanda has not such distribution of raingauge, the use of TAMSAT data can be of high importance. TAMSAT rainfall estimates can be used in APSIM to predict maize crop yield. This study attempt to provide information on the reliability of spacio-temporal satellite rainfall data to alleviate the scarcity of rainfall data in Rwanda. In addition, the study highlights the importance of such data in simulating crop yields with APSIM. TAMSAT rainfall data from 30th June 2013 to 30th June 2015 were downloaded for the volcanic highland and Bugesera AEZ using GPS coordinates corresponding to the ground location of weather stations. These data were compared to the recorded data on the stations and were both used in APSIM for maize yield simulation. The simulated maize stover and grain yields were compared to the values observed in the field experiments during season 2014A and 2014B in both the volcanic highland and Bugesera AEZ. Results showed that in Bugesera AEZ, TAMSAT slightly underestimated the total rainfall (1,310.0 mm) as compared to the actual rainfall (1,518.9 mm). This slight difference may be due to lack of one-off local calibration. However, the cumulated daily rainfalls estimated by TAMSAT was strongly correlated (R2 = 99%) to the ones recorded by the weather station in Bugesera. In the humid volcanic highland AEZ, results showed a broad agreement in the trends of both TAMSAT rainfall and weather station records but TAMSAT significantly underestimated the rainfall. In the whole study period, the latter estimated the total rainfall to be 1,418 mm while the actual rainfall received was 2,427.2 mm. This is due to the intricate topography of the volcanic highland which receives complex local rainfall variations and occurrence of non-convective rainfall while TAMSAT mainly predicts the convective rainfall. This is probably the reason why TAMSAT better estimated rainfall in Bugesera AEZ (slightly flat with round hills) than in the volcanic highland AEZ (very hilly). The APSIM-Maize model performed well in the simulation of maize stover and grain yields for both Bugesera and Volcanic Highland AEZ. As expected, the simulated maize stover and grain yields were higher in the volcanic highland than in the Bugesera AEZ. The simulations also showed that there were significantly (p<0.05) higher maize grain yields in season A (2014A) than in season B (2014B) due to differences observed in the rainfalls. Results showed that there were no difference in the outputs of the simulations while using TAMSAT rainfall instead of the station rainfall in the metfile of APSIM module. This obviously shows that APSIM can simulate maize stover and grain yields with almost no noise coming from TAMSAT rainfall data and hence recommendable for places with limited raingauge records. Based on the findings of this study, TAMSAT offers good estimations particularly in the semi-arid regions of Rwanda with less hilly topography. It can be used successfully to identify periods with well below or well above average rainfall even over highland areas, and is therefore useful for providing good APSIM simulations and hence inform food security early warnings.