Using Bayesian Regression Tree Models and remotely sensed data to characterize recent environmental change in Alaska, USA

• Main Author: Harvey, Joann W.
• Format: Text
• Language: unknown
• Published: No Publisher Supplied 2011
• Subjects: Arctic; Bering Sea; Alaska
• Online Access: https://dx.doi.org/10.7282/t31r6pmm; https://rucore.libraries.rutgers.edu/rutgers-lib/36038/

Remotely sensed Advanced Very High Resolution Radiometer (AVHRR) images, collected between 1995 and 2007, and Bayesian Regression Tree Modeling were brought together to characterize growing season environmental (vegetation, temperature, precipitable water, and cloudiness) change in Alaska. This method highlighted general trends and local variation. The method was applied in two stages to reduce the effects of cloudiness upon the results and reveal the temporal distribution of cloudiness conditions. A reversible form of tree model “subtree replacement” was included in the Reversible Jump MCMC algorithm. A sensitivity analysis showed that larger values of some hyperprior parameters could increase the number of subsets delineated by the method. For data collected during 1995 – 2002, the analyses showed local variation and subtle changes. In 2003, conditions of higher precipitable water, higher Normalized Difference Vegetation Index (NDVI), and/or greater cloudiness were highlighted. In 2004, the analyses detected a shift to lower precipitable water and/or lower cloudiness, often accompanied or followed by lower NDVI and higher land surface temperature. In 2007, continued warming was highlighted in the Arctic and northern interior regions, in contrast with a return to earlier conditions and increased cloudiness revealed in regions near the Bering Sea.