Scaling Correction of Remotely Sensed Leaf Area Index for Farmland Landscape Pattern With Multitype Spatial Heterogeneities Using Fractal Dimension and Contextural Parameters

High-accuracy retrieval of the crop leaf area index (LAI) in farmlands via remote sensing is the premise of reflecting the true growth condition of the crop. This paper aimed at scaling correction of LAI retrieval and developed an LAI scaling transfer model for farmland landscape pattern with multitype spatial heterogeneities according to the multiple types of farmland underlying surfaces in China. The interclass heterogeneity (caused by the alternate distribution of different cover types) and intraclass heterogeneity (caused by the difference in growth conditions within the same crop) both exist in the farmland landscape. The contextural parameters (fractions of components) and fractal dimension of the up-scaling pixel were used to quantitatively describe and correct the scaling effect caused by the two types of spatial heterogeneity, respectively. A scaling transfer model of inversed LAI was built by comprehensively considering intraclass and interclass heterogeneities. Results indicated that the LAI scaling bias of the up-scaling mixed pixel was mainly caused by the interclass heterogeneity even when the areal proportion of the noncrop component was low. The scaling transfer model corrected the scaling effect of LAI, with the root-mean-square error and mean absolute percentage error decreasing from 0.599 and 10.00% to 0.077 and 1.11%, respectively. The developed method based on fractal theory and contextural parameters effectively weakened the influence of the scaling effect on the accuracy of LAI retrieval.