Rainfall around Mount Kenya is of high importance for the people in its surrounding. The mountain is the water tower for a wide-spread region. People on the slopes of Mt. Kenya mainly depend on rain-fed agriculture, whereas for people in the semi-arid lowlands the river water is important for pastoralism. The high amounts of precipitation around the mountain due to orographic enhancement recharge the main river and provide a livelihood for both upstream farmers and downstream pastoralists. The characteristic topography shows a major influence on local to regional rainfall variation. The study area located in the tropics exhibits a bimodal annual rainfall cycle with areas comparatively dry for their latitude. Prolonged dry seasons or failing rains trigger conflicts between upstream and downstream users and highlight the need to develop water resource allocation strategies.
Continuous spatial information of the rainfall pattern may support decisions in water resource allocations between upstream and downstream users or may improve hydrological studies. In order to provide such continuous spatial information a reliable and sufficiently dense network of rain data is necessary. For the study area in Central Kenya such a dense network of rain gauge with reliable long-term records is available.
The aim of this study is to estimate precipitation at unsampled locations using topography and to produce gridded precipitation maps of seasonal and annual precipitation for a period from 1977 to 2014. A multiple linear regression model is evaluated for the seasonal and annual precipitation totals for geographic and topographic predictors. In addition, residual precipitation is interpolated using ordinary kriging and is added to the maps predicted by the multiple linear regression model. Topographic predictors, such as slope, aspect, altitude or combinations of these variables, are calculated at different scales, as topography might influence precipitation distinctively at different scales. Further, a predictor to account for the barrier effect of Mt. Kenya is constructed. It consists of an angle around the mountain and a distance function. The angle is used to predict on what expositions of the mountain highest precipitation occurs. The distance function accounts for a decreasing effect of the mountain on precipitation with distance from the mountain and for decreasing precipitation at high altitudes. Geographic predictors are latitude and longitude. They are used in order to model the gradient between the semi-arid north and the humid southern part of the study area.
The best regression model consists of predictors of slope, a combination of slope and aspect, two variables at different frequencies to account for Mt. Kenya, and of the geographic predictors latitude and longitude. The regression model is the same for the two rainy seasons and annual total precipitation. Slope calculated at a scale of 27 km shows higher correlations with precipitation totals than calculated at lower scales. This scale is therefore used in the regression model. Altitude is not used, as a lot of years show unrealistic negative correlations with precipitation. However, correlations with altitude are higher calculated for lower scales and for annual totals compared to the two rainy seasons. The impact of the Mt. Kenya variables on predictive R2 shows that constructing
such a predictor is a suitable way to model the precipitation pattern around the mountain. Mean best predictive skills for the entire time period are found for the boreal autumn rains with a predictive R2 of 0.656. Values for boreal spring rain totals and annual totals are lower and of 0.464, respectively of 0.454. This indicates that there is an influence of topography on the spatial pattern of precipitation and that this influence is more pronounced for the boreal autumn rains, than for the boreal spring rains and annual precipitation totals.
Ordinary Kriging of the residual precipitation allows to add additional valuable information that the multiple linear regression model is not able to predict. Thus, this combined approach of multiple linear regression and Ordinary Kriging seems a reasonable way to model the precipitation pattern in the area around Mt. Kenya.