|
Abstract:
|
The USDA Forest Service aims to monitor and predict change in forest-conditions over the time across large geographic regions within the country. Whereas the data collected from the field is few and sparse in both spatial and temporal scales, auxiliary information derived from satellite imagery can be made available at a monthly basis in a dense spatial resolution. At a specific location, how these measurements vary over the course of a year, can be informative about the forest condition at that location. In this talk, we present a predictive model based on functional regression. The variable of interest is basal area per unit acre at a specific month. We model the response as a function of auxiliary variables observed over a fixed window around that month. We developed the hierarchical model in two stages - the first layer for presence/absence of any forested land in a cell and the second one for the variation in response within a cell conditional on the presence of forest. We treat the covariate effects as functions over the time window and month, and use Elastic Net prior to shrink the parameter size. Post-estimation, we present maps and tables to summarize the outcome.
|
