干旱区生态环境知识资源中心

Integration of multisource remote sensing data is one of the methods to invert temporal high-spatial-resolution (time-continuous and with the resolution in 10-m scale) leaf area index (LAI). However, a few studies are related to addressing the uncertainty of data sources in the inversion algorithm and investigating the relationship between the uncertainty of data sources and the credibility of inversion results. This research is designed to retrieve temporal high-resolution LAI using an improved dynamic Bayesian network approach to fuse the dynamic change information of coarse-resolution historical data with the spatial information of high-resolution remote sensing observations. In this process, the focus was on handling the uncertainty of data sources that is mainly derived from the uncertainty of high-resolution remote sensing observations. On the basis of retrieving the temporal high-resolution LAI, the credibility of the inversion results was calculated and the influence of data source uncertainty on inversion results was investigated. To implement the work framework, this study takes the Xiaoman irrigation area in the arid middle reaches of the Heihe region as the study area, the uncertainty generated during the Advanced Space borne Thermal Emission and Reflection Radiometer (ASTER) atmospheric correction process as the uncertainty in the data sources and the ASTER images as the remote sensing information, and uses the Moderate-resolution Imaging Spectroradiometer MCD15A2 historical LAI data to construct the dynamic LAI information. By constructing an improved dynamic Bayesian network, the LAI products with 15-m spatial resolution and 8-day time-series resolution were produced. The validation results revealed that the determination coefficient R-2 between LAI inversion results and actual measured values is 0.85, and the root-mean-square error (RMSE0) is 0.40 m(2)/m(2). It was also observed that the high-resolution observation information can be severed to gradually correct the dynamic growth information during the time series inversion. This finding is manifested by the fact that with the addition of high-resolution remote sensing observation data, the reliability of the inversion results gradually increases. Meanwhile, the uncertainty of the data sources has a relatively impact on the reliability of the inversion results. When the uncertainty level of data sources is lower than 0.24, the reliability of the inversion results is high. It is concluded that the reliability of LAI will increase with the decreasing of the uncertainty level of remotely sensed data source.