FOREST CARBON MAPPING AND SPATIAL UNCERTAINTY ANALYSIS: COMBINING NATIONAL FOREST INVENTORY DATA AND LANDSAT TM IMAGES

摘要

Being able to accurately map forest carbon is a critical step in the global carbon cycle modeling and management process. This project is aimed at enhancing the current methodologies used for forest carbon mapping, and applying a method to account for any errors produced. By doing so, more accurate decisions can be made based on the knowledge gained from forest carbon maps; such as policy decisions on how to manage forests, or how to mitigate climate change. The use of remotely sensed images, in combination with Forest Inventory and Analysis (FIA) data, is one such way of doing this. This study compared three different methods; including linear regression, cosimulation, and up-scaled cosimulation to interpolate forest carbon based on a defined relationship between sample plots of national FIA data and satellite images. An uncertainty analysis was completed in an effort to quantify, and separate the different sources of error produced within a cosimulation mapping effort. The results indicated that the band ratio of TM4 / TM5 + TM4 / TM7 had the highest correlation coefficient, around 0.56, with the FIA forest carbon values. At a resolution of 90 m ×by 90 m, co-simulation predicted carbon values from about 14 Mg/ha, to 135 Mg/ha. The regression model, at the same resolution, estimated carbon values from about -17 Mg/ha, to 2,400 Mg/ha. Up-scaled cosimulation at a resolution of 990 m x× 990 m, predicted carbon values of ranging from 16 Mg/ha, to 133 Mg/ha. The uncertainty analysis was unable to produce any statistically significant results, with all R2 values below 0.1. These results showed that using a linear regression produced some impossible estimates, while cosimulation led to more realistic values. However, no conclusion can be made when comparing the methods based on the map validation techniques used. Although limited validation of the results was conducted, using both the FIA data and some independent sampling data; further work that focuses on validation is recommended.