Adaptive lossy compression of climate model data based on hierarchical tensor with Adaptive-HGFDR (v1.0)

Abstract. Lossy compression has been applied to large-scale experimental model data compression due to its advantages of a high compression ratio. However, few methods consider the uneven distribution of compression errors affecting compression quality. Here we develop an adaptive lossy compression method with the stable compression error for earth system model data based on Hierarchical Geospatial Field Data Representation (HGFDR). We extended the original HGFDR by firstly dividing the original data into a series of the local block according to the exploratory experiment to maximize the local correlations of the data. After that, from the mathematical model of the HGFDR, the relationship between the compression parameter and compression error in HGFDR for each block is analyzed and calculated. Using optimal compression parameter selection rule and an adaptive compression algorithm, our method, the Adaptive-HGFDR, achieved the data compression under the constraints that the compression error is as stable as possible through each dimension. Experiments concerning model data compression are carried out based on the Community Earth System Model (CESM) data. The results show that our method has higher compression ratio and more uniform error distributions, compared with other commonly used lossy compression methods, such as the Fixed-Rate Compressed Floating-Point Arrays method.