An efficient method for change detection of soil, vegetation and water in the Northern Gulf of Mexico wetland ecosystem

Abstract

Mapping and monitoring wetland ecosystems over large geographic areas based on remote sensing is challenging because of the spatial and spectral complexities of the inherent ecosystem dynamics. The main objective of this research was to develop and evaluate a new method for detecting and quantifying wetland changes in the Northern Gulf of Mexico (NGOM) region using multitemporal, multispectral, and multisensor remotely sensed data. The abundance of three land- cover types (water, vegetation, and soil) was quantified for each Landsat 30 m pixel for 1987, 2004, 2005, and 2006 using a regression tree algorithm. The performance of the algorithm was evaluated using an independent reference data set derived from a high-resolution QuickBird image, and several statistics including average error (AE), relative error (RE), and the Pearson correlation coefficient (r). For per-pixel percentage estimation, the AE is under 10% for water prediction, 9.5–11.4% for vegetation, and 9–11.1% for soil. The correlation coefficients between predicted and reference data range from 0.90 to 0.96 for water, from 0.80 to 0.89 for vegetation, and from 0.79 to 0.86 for soil. The high accuracy achieved by this method is attributed to the high quality of training data and the rigorous calibrations applied to multisensor and multitemporal satellite imagery. Based on the multitemporal estimation of the three land-cover components, spatial and temporal changes of the land-cover types from 1987 to 2006 were quantified and analysed. The study demonstrates that the method provided useful information on the abundance and changes of the key land-cover types in the NGOM region where long-term disturbances and episodic events occurred. Such information is valuable for monitoring land and vegetation loss and recovery processes, and for understanding possible drivers of the coastal wetland evolution in the region.

Acknowledgements

The research was partially supported by the USGS NGOM Ecosystem Change and Hazard Susceptibility project. Special thanks go to the USGS NGOM Ecosystem Change and Hazard Susceptibility project team members: Brady Couvillion, John Barras, Monica Palaseanu-Lovejoy, and Christine Kranenburg for their support and collaborations during this study. We also thank Lei Ji and anonymous reviewers for their constructive comments.

Notes