Raster-based topographic maps are commonly used in geoinformation systems to overlay geographic entities on top of digital terrain models. Using compressed texture formats for encoding topographic maps allows reducing latency times while visualizing large geographic datasets. Topographic maps encompass high-frequency content with large uniform regions, making current compressed texture formats inappropriate for encoding them. In this paper we present a method for locally-adaptive compression of topographic maps. Key elements include a Hilbert scan to maximize spatial coherence, efficient encoding of homogeneous image regions through arbitrarily-sized texel runs, a cumulative run-length encoding supporting fast random-access, and a compression algorithm supporting lossless and lossy compression. Our scheme can be easily implemented on current programmable graphics hardware allowing real-time GPU decompression and rendering of bilinear-filtered topographic maps.