Comparing partial least squares (PLS) discriminant analysis and sparse PLS discriminant analysis in detecting and mapping Solanum mauritianum in commercial forest plantations using image texture

Abstract Solanum mauritianum is a highly destructive and resourceful plant invader, resulting in severe economic and ecological damage. Detecting and mapping the spatial distribution of S. mauritianum is a priority for effective management of commercial forest plantations. Therefore, image texture computed from a 2 m WorldView-2 image with sparse partial least squares discriminant analysis (SPLS-DA) and partial least squares discriminant analysis (PLS-DA) were developed and applied to detect and map co-occurring S. mauritianum within a commercial forest plantation. The results indicated that SPLS-DA successfully performed simultaneous variable selection and dimension reduction to yield an overall classification accuracy of 77%. In contrast, the PLS-DA model in conjunction with variable importance in the projection (VIP) yielded an overall classification accuracy of 67%. The most significant texture parameters selected by the SPLS-DA model were correlation, homogeneity and second moment, which were predominantly computed from the red, red edge and NIR bands. Overall, this study validates the potential of image texture integrated with SPLS-DA to effectively detect and map co-occurring S. mauritianum in a commercial forest plantation.