Abstract
Globally, invasive species are identified as one of the most serious threats to ecological stability and biodiversity. Water hyacinth (Eichhornia crassipes), an aggressive invasive aquatic species, has caused severe economic and ecological impacts in the Sacramento-San Joaquin River Delta in California. In the Delta, water hyacinth co-occurs with native pennywort (Hydrocotyle umbellata L.) and non-native water primrose (Ludwigia spp.). All of the species express a wide range of phenotypic variability, making it difficult to map them with remote sensing techniques because their spectral response is highly variable. We present an integrated approach to mapping these floating species using a sequence of hyperspectral methods, such as spectral angle mapper (SAM), linear spectral unmixing (LSU), continuum removal and several indices in a decision tree format. The ensuing tree, based on biophysiological differences between the species, was robust and consistent across three separate years and over multiple flightlines each year, spread across an area of approximately 2500 km2. The most important inputs used to create the tree were reflectance in the short-wave infrared (SWIR), Red Edge Index, near-infrared (NIR) reflectance, LSU fractions and SAM rule values. The floating species were mapped with average accuracy of 88% for water hyacinth, 87% for pennywort and 71% for water primrose.
Acknowledgement
This research was supported by the California Department of Boating and Waterways (CDBW) Agreement 03-105-114. The authors are particularly grateful to Paul J. Haverkamp for a MATLAB program to produce histograms and ANOVA statistics and to Jonathan A. Greenberg for analysing the LiDAR data and writing an IDL batch program to run the decision trees. The authors appreciate the support of M. Carlock of CDBW, the field crews of California Department of Food and Agriculture and the students and staff of the Center for Spatial Technologies and Remote Sensing who helped collect and analyse field data in support of this project. Special thanks go to E. L. Hestir and M. E. Andrew for their assistance in field data collection and many helpful suggestions. Finally, thanks to R. Mcllvaine and G. Scheer for administrative and computational support.