The ability to produce high-resolution climate maps is crucial for assessing climate change impacts and mitigating climate disasters and risks in developing countries. Mainstream geostatistical downscaling techniques use spatial interpolation or multi-linear regression models to produce high-resolution climate maps in data-scarce regions. Since global climate evolution is a nonlinear process governed by complex physical principles, these linear downscaling techniques cannot achieve the desired accuracy. Moreover, these techniques cannot utilize different resolution data as model inputs. In this study, we developed a hybrid of multilayer perceptrons that could couple high-resolution topographic data with sparse climate observation data well and then generate high-resolution climate maps. To test the performance of our tool, we generated high-resolution precipitation and air temperature maps using sparse observation data from 21 meteorological stations in Ethiopia. The accuracy of the high-resolution climate maps generated using our hybrid of MLPs clearly outperformed those using a multi-linear regression model or a pure MLP.