The Assessment of Rainfall Prediction Using Climate Models Results and Projections under Future Scenarios: the Machangara Tropical Andean Basin Case

Abstract

Rainfall is vital in the biosphere and predicting it is essential under the possible adverse effects of climate change. Rainfall behavior is linked to the availability of fresh water and the development of almost all the activities necessary for human subsistence. Therefore, knowing their patterns under future scenarios could help decision-makers to plan water use policies. This study used the random forest algorithm to predict rainfall in Chanlud and El Labrado stations, located in the tropical Machangara high mountain basin in Ecuador. Data from the Ecuador project's third national communication (TNC) were used to train the prediction models. First, those models' performance was analyzed to know which climate model results of the TNC provide more information to learn observed rainfall patterns. Then, the rainfall signal was projected under the RCP 4.5 and 8.5 scenarios. Among the most important results obtained, it stands out that the assembly results of the TNC provided the best information to learn rainfall patterns in the present. The performance is the best from January to July, but from August to December it is lower. Rainfall projections under RCP 8.5 are, in general, lower than under RCP 4.5. No significant trends were found in the future. However, a very slight increase (decrease) of rainfall was observed for the driest (wettest) months in both stations, although slightly more accentuated in El Labrado.