Spatio-temporal analysis of remote sensing images provides early warning signals of forest mortality

Abstract

Ecosystems are under unprecedented pressures, reflected in rapid changes in the regime of disturbances that may cause negative impacts on them. This highlights the importance of characterizing the state of an ecosystem and its response to disturbances, which is known as a notoriously difficult task. The state-of-the-art knowledge has been tested rarely in real ecosystems for a number of reasons such as mismatches between the time scale of ecosystem processes and data accessibility as well as weaknesses in the performance of available methods. This study aims to use remotely sensed spatio-temporal data to identify early warning signals of forest mortality using satellite images. For this purpose, I propose a new approach that measures local spatial autocorrelation (using local Morans I and local Gearys c method) at each time, which proved to produce robust results in multiple different study sites examined in this article. This new approach successfully generates early warning signals from time series of local spatial autocorrelation values in unhealthy study sites 2 years prior to forest mortality occurrence. Furthermore, I develop a new R package, called "stew", that enables users to explore spatio-temporal analysis of ecosystem state changes. This work corroborates the suggestion that spatio-temporal indicators have the potential to diagnose early warning signals to identify upcoming climate-induced forest mortality, up to two years before its occurrence.