科研之友
微信
新浪微博
Facebook
分享链接
摘要
Remote sensing of forest condition is typically based on broadband vegetation indices to quantify coarse categories of canopy condition. More detailed and accurate assessments have been demonstrated using narrowband sensors, although with more limited image availability. While differences in sensor capabilities are obvious, I hypothesized that multispectral imagery may be able to detect more subtle canopy stress symptoms if a new calibration approach was considered. This involves three major changes to traditional decline assessments: (1) calibration with more detailed field measurements, (2) consideration of narrowband derived indices adapted for broadband calculation, and (3) a multivariate calibration model. Testing this approach on Landsat-5 (TM) imagery in the Catskills, NY, USA, a five-term linear regression model (r(2)=0.621, RMSE 0.403) based on a unique combination of vegetation indices sensitive to canopy chlorophyll, carotenoids, green leaf area, and water content was able to quantify a broad range of forest condition across species. When rounded to a class-based system for comparison to more traditional methods, this equation predicted decline across 42 mixed-species plots with 65% accuracy (10-classes), and 100% accuracy (5-classes). This approach was a significant improvement over commonly used vegetation indices such as NDVI (r(2)=0.351, RMSE=0.500, 10-class accuracy=60%, and 5-class accuracy=74%). These results suggest that relying solely on a single common vegetation index to assess forest condition may artificially limit the accuracy and detail possible with multispectral imagery. I recommend that future efforts to monitor forest decline consider this three-pronged approach to decline predictions in order to maximize the information and accuracy obtainable with broadband sensors so widely available at this time.
- 出版日期2014-5-3
