Landscape- and regional-level correlative modeling techniques for the prediction of ecological processes

Abstract

Quantifying landscape patterns and relating them to key biophysical drivers is often of primary interest in ecological research. Knowledge of such relationships can increase our understanding of pattern-process linkages and aid in making predictions across other spatial or temporal extents. With the recent influx of freely available, high resolution and spatially explicit data sources landscape- and regional-level models have gained in popularity. However, certain challenges arise when modeling ecological process at these scales, which are often overlooked. Processes in ecological systems occur at several spatial scales and interactions among processes are many, complex, and often non-linear and can result in highly heterogeneous conditions across spatial scales. The current research represents three separate applications of landscape- and regional- level correlative modeling. The objectives of each application were to develop correlative models to identify key environmental drivers of ecological process of interest, and to predict the process across a large spatial extent. For each application, I used a multi-model approach where a variety of statistical regression and machine learning methods were tested to objectively identify the model or models that best captured the complexity of an ecological process. The first two studies were located in the southern Appalachian mountain region where stream industrial emissions have acidified stream waters for more than a century. The objectives of these studies were to identify main drivers of stream water acid neutralizing capacity (ANC; a metric related to the ability of a stream to buffer against acidic inputs), and base cation weathering (BCw; the level of base cations supplied to the stream by catchment soils) and to predict ANC and BCw across the study region. The third study was located in the Methow Valley region of eastern Washington, which was aimed at identifying the landscape vegetation components most limiting to cavity-nesting bird populations in the area. Each study had different modeling objectives and different associated nuisances related to the data design, the spatial extent of the study region, and the correlative structure of the modeled process in relation to the environmental drivers.