Assessing Disagreement and Tolerance of Misclassification of Satellite-derived Land Cover Products Used in WRF Model Applications

As more satellite-derived land cover products used in the study of global change, especially climate modeling, assessing their quality has become vitally important. In this study, we developed a distance metric based on the parameters used in weather research and forecasting (WRF) to characterize the degree of disagreement among land cover products and to identify the tolerance for misclassification within the International Geosphere Biosphere Programme (IGBP) classification scheme. We determined the spatial degree of disagreement and then created maps of misclassification of Moderate Resolution Imaging Spectoradiometer (MODIS) products, and we calculated overall and class-specific accuracy and fuzzy agreement in a WRF model. Our results show a high level of agreement and high tolerance of misclassification in the WRF model between large-scale homogeneous landscapes, while a low level of agreement and tolerance of misclassification appeared in heterogeneous landscapes. The degree of disagreement varied significantly among seven regions of China. The class-specific accuracy and fuzzy agreement in MODIS Collection 4 and 5 products varied significantly. High accuracy and fuzzy agreement occurred in the following classes: water, grassland, cropland, and barren or sparsely vegetated. Misclassification mainly occurred among specific classes with similar plant functional types and low discriminative spectro-temporal signals. Some classes need to be improved further; the quality of MODIS land cover products across China still does not meet the common requirements of climate modeling. Our findings may have important implications for improving land surface parameterization for simulating climate and for better understanding the influence of the land cover change on climate.