|
Abstract:
|
Satellite data provide an unprecedented opportunity to quantify land cover change across multiple decades at a global extent. The increasing availability of satellite imagery necessitates the development of statistical models to analyze these big data. Our goal is to understand the ecological processes that drive population recovery after wildfires. We focus on quantifying the revegetation rate within a 1988 wildfire in Utah using Landsat data. We began by fitting an ARIMA model to the data. We found relatively high predictive accuracy, including 5% Mean Absolute Percentage Error (MAPE). We then fit an ecological process model, the Beverton-Holt model, to the data for comparison. Although both models showed a good fit to the data, the Beverton-Holt model embeds mechanistic relationships such as density-dependence, that traditionally require laborious field or experimental data. In both models, the nature of satellite data challenged statistical modeling, including the large size of the dataset and multiple sources of measurement error. The next steps in our analysis will include adding a spatial component to account for the dispersal and diffusion of populations in space and time.
|
