Surface textures of high resolution and quality, either acquired from aerial or satellite imagery or computed using procedural models, are crucial for photorealistic terrain rendering. Procedural models provide a compact representation and can be evaluated at run-time. In this paper we present an extension to an existing, GPU-friendly procedural texturing model, such that it can be fitted semiautomatically to real-world data. In order to increase realism and to account for geographic conditions, we also include temperature, solar radiation and rainfall distributions - simulated ormodeled using measured data from gaging stations - into the reproduction process. The original surface texture is no longer required for rendering: instead a new texture of arbitrary resolution is synthesized at runtime. In addition to the compact procedural model we store elevation data, anyway required for the terrain rendering, and low-resolution geographic data. We show results of our method applied to a comparatively little cultivated region in Central Asia.