State-space models to monitor forest dynamics from satellite remote sensing data

Optical satellite remote sensing provides data with unprecedented temporal and spatial coverage, including multidecadal time series at a global scale. These data have great potential for quantifying ecological dynamics, yet also present challenges due to measurement error related to atmospheric conditions, vegetation phenology, and sensor failure. State-space modeling can disentangle measurement error from ecological process and has not yet been widely applied to satellite remote sensing data. We demonstrate how state-space models developed in Stan can accurately forecast the ecological dynamics of forest succession in a tropical landscape. Our modeling approach has potential to integrate multiple types of data with different costs and benefits, such as high resolution lidar data and medium resolution Landsat imagery. More broadly, our work illustrates how publicly-available satellite time series can serve as a nearly unlimited data source to develop novel state-space models.

Presenter biography:
T. Trevor Caughlin

T. Trevor Caughlin is a spatial ecologist working as an Assistant Professor at Boise State University. He is interested in using Bayesian statistics to address applied problems in ecosystem restoration.