A general Landsat model to predict canopy defoliation in broadleaf deciduous forests

P. A Townsend, A. Singh, J. R. Foster, N. J. Rehberg, C. C. Kingdon, K. N. Eshleman, S. W. Seagle (2012) A general Landsat model to predict canopy defoliation in broadleaf deciduous forests, Remote Sensing of the Environment 119:255-265.


Philip A. Townsend, Aditya Singh, Jane R. Foster, Nathan J. Rehberg, Clayton C. Kingdon, Keith N. Eshleman, Steven W. Seagle

Defoliation by insect herbivores can be a persistent disturbance affecting ecosystem functioning. We developed an approach to map canopy defoliation due to gypsy moth based on site differences in Landsat vegetation index values between non-defoliation and defoliation dates. Using field data from two study areas in the U.S. central Appalachians and five different years (2000, 2001, 2006, 2007, and 2008), we fit a sigmoidal model predicting defoliation as a function of the difference in the vegetation index. We found that the normalized difference infrared index (NDII, [Band 4 − Band 5] / [Band 4 + Band 5]) and the moisture stress index (Band 5 / Band 4) worked better than visible-near infrared indices such as NDVI for mapping defoliation. We report a global 2-term fixed-effects model using all years that was at least as good as a mixed-effects model that varied the model coefficients by year. The final model was: proportion of foliage retained = 1 / (1 + exp(3.057 − 31.483 ∗ [NDIIbaseyear − NDIIdisturbanceyear]). Cross-validation by dropping each year of data and subsequently refitting the remaining data generated an RMS error estimate of 14.9% defoliation, a mean absolute error of 10.8% and a cross-validation R2of 0.805. The results show that a robust, general model of percent defoliation can be developed to make continuous rather than categorical maps of defoliation across years and study sites based on field data collected using different sampling methods.

 

View this publication online:

http://doi.org/10.1016/j.rse.2011.12.023

Personal tools