Identifying Optimal Spatial Resolutions For Trend Detection

One of the key challenges facing the ecological community lies in understanding the impacts of forecast climate change on the structure and function of ecosystems both through time and across large spatial extents. There are two main obstacles that currently hinder the ability to quantify ecosystem change in this manner within the context of a shifting climate. First, with respect to key ecosystem responses, there are limited data that have sufficient spatial-temporal resolution and extent. Second, methods for quantifying long term changes in ecosystem responses with complex spatial and temporal structure driven by climatic forcing are not well developed.

In this work we focus on the question “What is the most appropriate spatial resolution for detecting a process level trend?” We address this question through the development and application of a simulation framework that allows for the parametric specification of the following model components: measurement error, the functional form of the link between climate drivers and the ecosystem response, and annual process variability dealing with non-separable space-time covariance structures. We consider varying spatial resolution and the corresponding impact of the covariance structure associated with the data model. This is parameterized in the simulation framework through an additive measurement error term. As the spatial resolution becomes coarser, sub-pixel heterogeneity becomes absorbed by the measurement error term as opposed to attribution to parameters in the process model. The results of this study characterize an “envelope” of values for these parameters that allow for the detection of trend in process model parameter. This study outlines a quantitative method of choosing a spatial resolution that allows for the detection of trend given a collection of prior distributions associated with other model parameters.