Research On The Acquisition Of Pixel Scale Ground Truth Over Heterogeneous Land Surfaces

With the rapid development of remote sensing technology,numerous satellite products have been generated for the study of ecological environment monitoring and climate change research.Nevertheless,due to the errors of satellite observations and the limitations of the retrieval algorithm,quantitative satellite products inevitably suffer from errors.Only by fully validating these products can the industrial applications be better supported.The core issue of validation is to obtain pixel scale ground truth.However,the spatial representativeness of in situ observations is limited due to the widely distributed surface heterogeneity and the spatial scale mismatch between satellite pixel and in situ site-based data,it is necessary to upscale in situ observations to generate the pixel scale ground truth.Challenged by the main problems in the validation of optical and passive microwave remote sensing products,this paper aims to obtain high-quality and spatiotemporal continuous pixel scale ground truth,conduct research on machine learning-based scaling methods,and comprehensively evaluate the performance of several error correction models based on the obtained pixel scale ground truth for the first time.The main research contents and relevant conclusions of the paper are as follows:(1)Research on validation methods of optical remote sensing products.In response to the limited spatial coverage of pixel scale ground truth obtained using traditional upscaling methods,this study developed a machine learning-based upscaling model and explored corresponding uncertainty factors.Three machine learning models,including random forest,k-nearest neighbor,and Cubist models,were selected to upscale single site in situ-based albedo to the coarse pixel scale.The upscaled results were carefully assessed through comparison with pixel scale reference data.The results indicate that the accuracy of upscaled results depends on the machine learning models,the inclusion of key variables related to albedo,the dataset selection of these variables,the amount of training data,and the sensitivity of machine learning models to these factors.(2)Research on validation methods of passive microwave remote sensing products.Aiming at the significant spatial scale mismatch between the in situ observations and coarse-resolution satellite products(ie.,the passive microwave soil moisture(SM)products),this study first generated a high-resolution dataset based on the machine learning-based downscaling model and then evaluated three mainstream SM products including SMOS-IC,SMAP L3,and AMSR2 LPRM within the multi-scale validation framework(ie.,from in situ to high resolution,and then to coarse resolution).First,a coarse resolution better-performing and independent SM dataset was produced by combining three SM products mentioned above;Second,the high resolution SM dataset was derived from the coarse resolution SM dataset with a random forest-based downscaling model and the high-resolution datasets of other variables.Finally,the high resolution SM was evaluated using in situ SM measurements,which was then aggregated to a coarse pixel scale for the assessment of the three coarse-resolution satellite SM products.It was found that the quality of the product shows a certain relationship with the indicators selected for validation.When the median values of the accuracy indicators were selected,SMAP L3 and AMSR2 LPRM performs best regarding the correlation and deviation from the pixel scale ground truth in the spatial domain,respectively,and SMOS-IC is always the worst.However,when the pixel-based evaluation results were focused,AMSR2LPRM performs best in most cases,followed by SMAP L3 and SMOS-IC.The accuracy of satellite SM products shows more dependence on slope than elevation,land cover types,and land surface temperature.(3)The evaluation of the error correction models based on the matched pixel scale ground truth.In this study,three typical models,namely random forests(RF),cumulative distribution function(CDF),and Kalman Filter(KF)were comprehensively evaluated based on the pixel scale ground truth regarding their ability to correct errors of coarse-resolution satellite albedo products(ie.,MCD43A3).These three models all show significant improvements regarding the accuracy of the corrected MCD43A3.RF shows the best overall performance,followed by CDF and KF.These three models are more adept at handling the bias of MCD43A3 than their consistency with respect to the pixel scale ground truth,and the improvement is the most significant at the sites with large errors.Regarding the stability of their performance,RF performs better in reducing RMSE while CDF performs better in reducing Bias and improving R~2.This study focuses on cutting-edge scientific issues in quantitative remote sensing validation and conducts systematic research work,which has improved the accuracy and spatiotemporal continuity of pixel scale ground truth to a certain extent.The results not only promote the systematization of the validation work but are also conducive to further improving the accuracy of remote sensing products.It has played a certain role in promoting the level of remote sensing quantification in China.