A method for constructing a reference dataset for the validation of remote sensing products

The validation of remote sensing products (RSPs) is necessary for the quality assessment of RSPs in order to ensure the reliable and effective application of RSPs. However, the existing validation of RSPs lacks large-scale engineering reference datasets above the regional level. In view of this fact, this study proposed a cross-validation-based method for constructing a reference dataset for RSP validation. First, a reference dataset of China organized by sheet and time was constructed using the Landsat8 OLI data whose accuracy had been verified. Then, the annual optimal reference dataset, which was easy to retrieve and update and enabled large-scale construction, was formed finally. After seven bands of the ZY1E hyperspectrum were matched according to the center wavelength, the reference dataset was used to verify the reflectance of ZY1E images. The calculation of the confusion matrix between ground truth (GT) data and automatic rating results yielded an overall accuracy of 87% and a Kappa coefficient of 0.83, meeting the requirements for engineering applications. The method for constructing a reference dataset proposed in this study provides technical support for large-scale, engineering-oriented RSP validation.