Predicting shallow surficial failures in the Mississippi River levee system using airborne hyperspectral imagery

Abstract

Shallow surficial failures or levee slides in the Mississippi River levee system are very common. There is currently no system to identify or predict the location of these slides before they occur. Studies of slide occurrence mechanisms suggest that probable slide-affected areas are characterized by anomalous vegetation. Compact Airborne Spectrographic Imager II (CASI II) imagery was analysed for selected levee sites in association with slide inventory data and field observations. Normalized Difference Vegetation Index (NDVI), Red edge Vegetation Stress Index (RVSI) and Red Edge Position Index (REP) were calculated from the acquired CASI II imagery. The vegetation indices were used to locate the stressed or anomalous vegetation and predict levee slides. The statistical significance of the predictors was determined by logistic regression. All predictors were found statistically significant for developing a slide prediction model. The slide prediction model was developed by combining the single predictors, categorized vegetation indices, into a model based on all three predictors. Percentage of Search Area Reduction (PSAR) and Failure Index (FI) were used to evaluate the performance of the slide prediction model. Evaluation of the model performance shows that it achieves a maximum FI of 0.43 and PSAR of 99.5.

Acknowledgements

Thanks are due to the University of Mississippi Geoinformatics Centre (UMGC) for funding the project, and the Remote Sensing Technology Centre (RSTC) of Mississippi State University (MSU) for providing hyperspectral imagery and Mississippi Levee Board for providing levee slide inventory.