Invasive Species Spread Mapping Using Multi-resolution Remote Sensing Data

The invasions of non-native species of vegetation pose significant threats to natural environments at all geographical scales. Saltcedar has been commonly treated as one of the several most threatening invasive species in U.S. in the next ten years. The spatial extent and density of infestation by saltcedar in the Rio Grande floodplain has been poorly understood in the past. Remote sensing provides a unique tool to map and monitor invasive species and provides a mean to detect major land cover changes and quantify the rate of change. To date, remote sensing has been mainly applied for mapping some canopy dominant invasive species. Accurate mapping of saltcedar distribution and abundance in a timely manner plays a central role to assist the undertaking of an effective control. Current studies have largely concentrated on the large-area detection with coarse resolution remote sensing data. Nevertheless, it is lacking of studies that systematically evaluate the respective potential of high spatial resolution satellite imagery, airborne hyperspectral imagery, as well as moderate resolution imagery for mapping saltcedar’s extents, distribution and monitoring its spread over time. Such cost and benefit analysis will be particularly invaluable to the regional or national scale study, in which selection of an appropriate image type to maximize the outcome plays an important part. In this study, a comprehensive test was designed and carried out to examine the ability to integrate multi-temporal and multi-resolution imagery: including very-high spatial resolution (QuickBird), hyperspectral resolution imagery (AISA), and moderate resolution satellite imagery (Landsat TM), in differentiating saltcedar from other riparian vegetation types in the Rio Grande river basin. Two types of analyses were fulfilled: first, five pixel-based classification methods were adopted for assessing effectiveness of QuickBird and AISA, respectively, i.e., the Maximum Likelihood Classifier (MLC), Neural Network Classifier (NNC), Support Vector Machine (SVM), Spectral Angle Mapper (SAM), and Maximum Matching Feature (MMF); Second, Landsat TM imagery was synthesized from AISA and tested for mapping abundance of saltcedar with four linear spectral unmixing methods and three back-propagation neural network methods. Results indicated that AISA outperformed QuickBird imagery in differentiating saltcedar from other riparian vegetation species. SVM achieves the highest classification accuracy among all the five classifiers. Linear spectral unmixing method exhibited similar mapping accuracy with neural network methods in estimating abundance of saltcedar at a spatial resolution of 30 by 30 square meters, but with significantly better computing efficiency. Overall, this study reflects the maximum capability of contemporary remote sensing in assisting reconnaissance of saltcedar, the most threatening invasive species in southwest United States.