Monitoring Insect Outbreaks Using MODIS Tasseled Cap Transformation Indices

Forests play a central role in the global carbon cycle. Natural disturbances such as fires or insect outbreaks affect the carbon sink capacity of forest ecosystems. A warmer climate and stressed vegetation under more frequent extreme climate events are expected to increase insect population growth and change insect disturbance regimes. Therefore it is crucial to develop methods to quantify consistently the global area affected by insect outbreaks.

Currently, neither a globally consistent dataset nor satellite-based global detection methods for insect outbreaks exist. Distinguishing between spectral signals of distinct forest disturbances is, however, challenging since insect outbreaks and other disturbances affect the spectral signature of forests in similar ways. The U.S. Forest Service of the USDA offers well documented aerial detection surveys (ADS) that cover the continental USA and several decades. This dataset offers a unique reference to test detection methods based on satellite imagery. In a previous study we have analyzed commonly used vegetation greenness indices from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor over a 6-yr period in the early 2000s in a small region in northwestern USA. Although greenness-based indices were able to capture the overall decrease in photosynthesis in response to insect disturbance, results were strongly contaminated by other signals.

Here we focus on the same region (USDA R4) and explore more sophisticated multi-spectral indices to evaluate the response of forests to different types of insect disturbance. The Tasseled Cap Transformation (TCT) was developed for Landsat satellites, but can also be adapted to MODIS. It uses seven reflectance bands in the visible and infra-red spectra and has high potential for application in temporal analysis and to investigate plant phenology. We compare the commonly-used Normalized Difference Vegetation Index (NDVI) with the TCT indices (brightness, wetness and greenness) and the conjunction of all three indices for areas affected by distinct insect types (defoliators and bark-beetles). We test different temporal (anomalies, trend breaks) and spatiotemporal (non-supervised clustering) methods to evaluate their potential to detect insect outbreaks without prior-information.