Consistency analysis to evaluate accuracy of commonly-used non-homologous LULC products in Erhai Lake Basin

Land Use and Land Cover (LULC) is the direct and concentrated reflection of the interaction between human activities and natural geographical environment. High-precision LULC information can also depend mainly on the global climate change, material cycle and dynamic balance of water and heat. Current environmental challenges were remained with the rapid increase in remote sensing observation platforms, the free disclosure of high-resolution satellite remote sensing data and the advancement of LULC mapping technology. Freely available medium and high-resolution land cover products are emerging for the open source. The medium and high resolution LULC datasets have also been constructed worldwide. However, there are different degrees of uncertainty in the multi-source data. It is a high demand for the suitable land cover products at the regional scale in various fields. Therefore, it is very important to evaluate the accuracy of the current commonly-used land cover data at the regional scale. Taking the Erhai Lake Basin as the study area, the consistency analysis was carried out to evaluate accuracy of commonly-used non-homologous LULC products. 2,947 validation samples were collected using Third National Land Survey data, the kilometer grid sampling, field surveys, and high-resolution image interpretation. Seven commonly-used heterogeneous LULC data products were evaluated, in terms of area, spatial consistency, confusion levels, and accuracy. The influence of LULC product mapping accuracy was quantitatively analyzed from four aspects: shrub forest proportion, landscape pattern index, elevation standard deviation and average patch area. The applicability of each dataset was also evaluated. The results reveal that the high, moderate, and low consistency areas were represented by 64.13%, 34.00%, and 1.87% of the total area, respectively, among the eight datasets of land cover. Notable confusion and misclassification occurred in shrub land and grassland, indicating the significant differences in the various products to represent different regions and land cover types. The overall accuracy of the LULC products was ranged from 69.5% to 81.1%. Notably, ESA_WC was offered the best data quality and spatial detail, especially for the cultivated land in fragmented landscapes. Additionally, the Shannon Diversity Index (SHDI) was found to share the most considerable impact on spatial consistency of land cover in the Erhai Basin, followed by the proportion of shrub land. In contrast, there was the less effect of some factors, such as elevation standard deviation, patch size, and cloud cover frequency. In all features of land cover, 10 m resolution data should prioritize the highest overall accuracy provided by ESA_WC. Among them, CRLC data was better performed, if the shrubland and grassland were not subdivided. While for 30 m resolution data, CLCD demonstrated the relatively high accuracy. The impermeable surface area was significantly underestimated, compared with the rest products unsuitable for urban expansion. This finding can serve as a valuable reference to assess the classification accuracy, strengths and weaknesses of the seven LULC products. Targeted data selection was facilitated for the specific applications. The accuracy of LULC products was directly evaluated for the applicability and adaptation of the data. The finding can provide the scientific basis for ecological environment protection, rational utilization of resources and sustainable development in plateau mountainous areas.