Mapping invasive Fallopia Japonica in the Upper Delaware River from multi-spectral and multi-temporal aerial imagery

Keywords: Remote Sensing, Invasive Species, OBIA, Classification
Abstract Type: Paper Abstract

Abstract

Invasive plants can cause ecosystem disruption and pose a significant threat to natural habitats and agricultural lands in North America. Mapping the extent of invasive plants is crucial for conservation practitioners and land managers to prioritize efforts to control and eradicate invasive plants. Japanese Knotweed (Fallopia japonica) is considered among the worst invasives because of its rapid spread. Numerous attempts to map knotweed have been carried that define the general lines of methodology. However, there is still a range of aspects that must be considered when specifying an approach for local conditions. We analyze the cost-efficiency of mapping knotweed with high-resolution imagery in three sections of the Upper Delaware River. Accuracy of alternative Random Forest classifications using combinations of temporal, spectral, segmentation, and structural characteristics is assessed, as well as the prospect of scaling the results to broader areas.
Like other studies, the inclusion of the NIR and the use of multitemporal data has been determinant to raise accuracy in detecting knotweed. Independent image classifications based on different variable combinations and segmentation show complementary errors. Combining those classifications increases performance reaching accuracies close to 90%. Unfortunately, despite the good results obtained locally, the maps and models obtained present a serious problem of scalability, suggesting that the local context and the data acquisition procedures impact the degree of which the variables interact.