The scale issue in quantitative remote sensing is a significant challenge that comprises three major problems that need to be addressed:（1） the forward modeling of remote sensing signals for heterogeneous land surfaces,（2） the parameter inversion for heterogeneous land surfaces, and（3） the upscaling of in situ observation for the validation of remote sensing products. This study focuses on the third problem by reviewing the progress of upscaling research in the Heihe Watershed Allied Telemetry Experimental Research（Hi WATER）. First,we define several basic concepts associated with scaling on the basis of the probability space and data assimilation theory. These concepts include spatial average, spatial upscaling, footprint scale, pixel scale, point scale, representativeness error, observation truth, and validation threshold. Second, we introduce the multiscale observation platform of Hi WATER and multiscale observation data, which covers the scales from point to pixels, sub-basins, and the whole river basin. Third, we describe several new developments in the sampling design based on geostatistics and temporal stability analysis. Specifically, a hybrid model-based sampling method without any spatial autocorrelation assumptions is developed to optimize the distribution of eco-hydrological wireless sensor network nodes, a universal co Kriging model is proposed to optimize multivariate sampling design, and a stratified block kriging is used to optimize the sampling locations in a spatial heterogeneous area. The temporal stability analysis is improved for the selection of the representative sampling points of the albedo and leaf area index. The stratified temporal stability analysis is proposed to identify the representative sampling points for monitoring long-term soil moisture at the pixel scale in high-intensity irrigated agricultural landscapes. Fourth, the representativeness of the in situ observation of solar radiation, carbon flux, soil moisture, and land surface temperature is evaluated. Results showed that the uncertainty of the validation for remote sensing products in heterogeneous areas mainly comes from the spatial and temporal representativeness of in situ measurements. Fifth, several upscaling methods are developed. The Kriging method is extended to block regression Kriging, area-to-area regression Kriging, spatiotemporal regression block Kriging, and unequal accuracy block Kriging for upscaling the in situ observation from the point-scale or footprint-scale to the pixel scale. Additionally, several case studies show that the Bayesian maximum entropy, a nonlinear method, is capable of providing a generalized theory framework to fuse general knowledge（such as that obtained from a model） and specific knowledge（such as that obtained from direct and indirect observations）. The usefulness of high-resolution remote sensing data as auxiliary information in improving the accuracy of upscaling is verified in this work. Overall, the multi-scale observation data collected in Hi WATER are helpful in improving our understanding of remote sensing scale problems.

pixel-scale;representativeness error;true value;sampling design;validation;Heihe river basin