Floating point data sets are growing more rapidly than storage and communication capabilities. Compression is an important technique for managing the data, yet floating point data are notoriously insubmissive to standard methods. Adaptive coarsening, an extension of adaptive subsampling (Belfor et al., 1994), solves the problem of compressing smoothly varying, gridded floating-point data while guaranteeing user-specified error bounds and allowing for nontrivial processing of the compressed product. Baden and Shafaat developed the idea in one dimension in 2004. Recently, adaptive coarsening in two and three dimensions has been developed and evaluated using data from fluid dynamics simulations. The results have been impressive. Adaptive coarsening achieves compression factors of 32 in 2D and 45 in 3D, for a relative error bound of one part in one thousand. For comparison, the adaptive subsampling method achieves a factor of 15 in 2D for the same data and error bound.