An Assessment of the Vegetative Health Surrounding the Aliso Canyon Gas Leak Using Remote Sensing Techniques

Abstract

On October 23, 2015, a large underground natural gas leak was detected from Well SS-25 of the Aliso Canyon natural gas storage facility located outside Porter Ranch, California. The leak was an environmental disaster that affected thousands of people, increased greenhouse gas emissions, and costed millions of dollars in damages. The leak was the largest release of anthropogenic methane into the atmosphere in the history of the United States. Engineers and emergency personnel were quickly tasked with stopping the leak and treating the health issues related to the exposure of chemicals found in the emissions. While the effects of the natural gas exposure on the human population were closely monitored, vegetative health was overlooked. Natural gas exposure affects the properties and processes of soils and root systems. These effects can cause spectral responses in vegetation that are measurable with remote sensing techniques. This study uses remotely sensed data to assess the level of stress imposed on surrounding vegetation by the Aliso Canyon gas leak. This includes an analysis of space-borne multi- and hyperspectral datasets. The multispectral analysis assesses changes in the Modified Soil Adjusted Vegetation Index (MSAVI) applied to Enhanced Thematic Mapper Plus (ETM+) data surrounding Well SS-25. Average MSAVI composite images of the study area are compared to MSAVI images collected during the period of the leak. Furthermore, the hyperspectral analysis assesses the health of vegetation using two different approaches to calculate the red-edge position. The red-edge positions are calculated using a first derivative and a linear interpolation approach, which are applied to NASA’s space-based Hyperion data. The study also explores how preprocessing can affect Hyperion data and red-edge position algorithms. Specifically, preprocessing methodologies including cross-track illumination correction and spectral smoothing are applied to the Hyperion data prior to calculating the red-edge position. The multispectral image analysis was able to detect regions of stressed vegetation in the study area; however, it was not possible to discern whether the stress was caused by the leak or low levels of precipitation. The amount of vegetative stress in the study area closely followed the amount of precipitation i.e., less precipitation resulted in more stress. There were also issues inherent to the ETM+ data that had to be mitigated. The scan line problem resulted in missing information and provided additional challenges with processing, which could have affected the results of the analysis. Despite the scan line limitations, the ETM+ data still proved useful to this study. The hyperspectral image analysis served more as a comparison of preprocessing methods and red-edge position algorithms than an assessment of vegetative health because it was not possible to assess the accuracy without ground-truth measurements. This does not assert that the analysis was meaningless; rather, the study outlined how preprocessing and algorithm selection can achieve differing results, which is important to understand in vegetation monitoring with remotely sensed data. The results showed that the red-edge position calculated using the first derivative approach was more sensitive to spectral smoothing than the approach calculated with interpolation. Excluding the data that were smoothed, red-edge positions calculated using linear interpolation were generally at longer wavelengths than the positions calculated with the first derivative method, suggesting that the results of each algorithm are not directly comparable. It was also shown that cross track illumination correction did not have a major influence on these data, which may be due to the size and location of the study area in the scene. Similar to the multispectral analysis, it was also observed that shifts in the red-edge position indicative of stress were also related the amount of precipitation. Furthermore, the results indicated that the interpolation approach may be more sensitive to background moisture content than the first derivative approach.