Advances in validation methods for remote sensing products of land surface parameters

Abstract

The important function of land surface remote sensing products in scientific research and quantitative application lies in their capability to record spatial-temporal earth surface features at a relatively enhanced performance. Validating quantitative remote sensing products involves evaluating their accuracy, stability, and consistency to show the performance of these products. This study investigates the validation method at a global scale according to the features of land surface parameters. The validation method is categorized into five main types: validation based on a single-point ground measurement, validation based on multipoint ground measurement, validation based on high-resolution remote sensing data, cross validation, and indirect validation. The characteristics and applicability of these methods are expected to support developments in validation techniques and widen the application of remote sensing products. In situ data, as the basis of validation data sets, have been shown to influence validation method development. Given the differences in spatial resolution between ground measurement and satellite measurement, validation is adjusted according to land surface features and the parameter scale effect. For parameters that do not show obvious scale effects, a direct point-pixel comparison can be performed. However, most of the land surface parameters show scale effects when the land surface is heterogeneous. Therefore, multiple-point measurements within a pixel are necessary, with the average value of these points used to compare with satellite pixel values. If the land surface is heterogeneous and even the multiple points cannot capture the intra-pixel variation in the parameter features, a multi-scale validation strategy based on high-resolution imagery should be used to obtain unbiased pixel scale values. The satellite value matches well with the in situ value when land surface is homogeneous. However, a scale mismatch is observed between the ground-based measurement and coarse-scale satellite measurement in land surface heterogeneity. Using multiple-point measurement is necessary to capture the variance within a larger region or use a fine-scale map as a bridge between the ground-based value and coarse-scale remote sensing value. The average value of multiple points within a pixel scale can represent the pixel scale “ground truth” when the land surface is not considerably heterogeneous. Otherwise, the aggregate value of high-resolution imagery is closer to the pixel scale “ground truth” relative to multi-point measurement. This study has demonstrated that remote sensing product validation, including in situ sampling, scale effect, and precision assessment, is a significant and necessary step before remote sensing products are applied. The five main validation methods are validation based on a single-point ground measurement, validation based on multi-point ground measurement, validation based high-resolution remote sensing data, cross validation, and indirect validation. These methods could be used according to the heterogeneity of the land surface and the scale effects of parameters. For relatively homogeneous land surfaces, ground-based measurements are representative enough for the sample plot, and the scale effects can be ignored. For heterogeneous land surfaces, multi-point measurement observation, or multi-scale validation strategy based on high-resolution imagery are recommended to represent the pixel scale as “relative truth”.