Thursday, August 5, 2010

PS 79-108: Impact of typhoon disturbance on vegetation dynamics in Taiwan assessed using satellite images and logistic regression models

Chung-Te Chang1, Teng-Chiu Lin1, and Su-Feng Wang2. (1) National Taiwan Normal University, (2) National Changhua University of Education

Background/Question/Methods

Tropical cyclones are known to have major impact on ecosystem structure and function.  Taiwan experiences three to six typhoons annually.  Although the impact of typhoon disturbance on ecosystems has been reported for several forest ecosystems, no studies have examined spatial pattern of typhoon damages for the entire island.  We used the MODIS satellite images to evaluate the island-wide patterns of typhoon impacts on vegetation dynamics in Taiwan between 2003 and 2006.  Our objectives are to (1) understand the spatial distribution of NDVI reduction following typhoons with different types of tracks, and (2) characterize typhoon impacts in relation topography, land use and typhoon characteristics.

The paired images within one-month before and after six typhoons were selected for analysis between 2003 and 2006. We calculate the maximum value composites of NDVI and the NDVI reductions of paired images.  Only NDVI reduction > 0.1 was considered to indicate typhoon damages to account for vegetation growth between the period of image acquisition.  The landuse types, precipitation quantity, distance to typhoon track and topographical parameters are extracted from typhoon damaged pixels and used for logistic regression analysis to explore their relation to typhoon damage.

Results/Conclusions

The NDVI reduction, ranging from 0.98% to 4.7%, was proportional to intensities of typhoons when they landed in Taiwan (R2= 0.91, p < 0.001).  The path of the typhoons affects the distribution of typhoon damaged areas.  If a typhoon headed westerly or northwesterly toward Taiwan, NDVI reduction concentrated on the windward locations.  On the other hand, if a typhoon headed northerly and parallel with Central Mountain Range (with an average of >1500 m), the greenness reduction showed no clear pattern.  The χ2 for the six logistical models, one model for each typhoon, were all significant (ps < 0.001), but the importance of the six parameters varied among typhoons.   The Nagelkerke R2 were only 0.11 and 0.23 for two northerly typhoons indicating low power of the models in predicting typhoon damages. The Nagelkerke R2 were much higher, ranging from 0.43 to 0.83, for westerly and northwesterly typhoons that were not disrupted by the Central Mountain Range before arriving Taiwan.  The result indicates that if the structure of a typhoon is relative unchanged before arriving in Taiwan the damages to NDVI can be reasonably predicted from meteorological, vegetational and topographical factors.  On the other hand, once a typhoon is disrupted by the Central Mountain Range, its impacts will be more difficult to predict.