Using generalized additive models for interpolating microclimate in dry-site ponderosa pine forests

Abstract

Microclimate in the understory of forests can be an important mediating influence for ecological attributes such a vegetation composition, wildlife habitat, fuel moisture, and downed wood and forest floor decomposition processes. Several tools exist for interpolating data from regional, broad-scale weather stations to derive geospatial climate surface maps representing mesoclimate variations at the earths’ surface. While useful, these tools typically do not provide interpolations that account for variations in forest canopy cover and therefore often fail to provide for higher resolution characterization of near-surface, understory microclimate conditions. In this paper we explore the development of understory microclimate surface maps from a dispersed set of microclimate sensors in a 991-ha managed landscape dominated by ponderosa pine forest patches of various densities in central Oregon, USA. We used generalized additive models (GAM) to interpolate microclimate at 50 m resolution from 2013 to 2016 using elevation, heat load, and normalized difference vegetation index (NDVI) as covariates. Models for all years revealed that air temperature trends were dependent on complex interactions of forest cover, site physiography and time, and produced interpolations closer to observed conditions than freely available coarse-scale models. We analyzed sub-canopy air temperature interpolation maps for trends in the context of differing canopy structures arising from five fuels mitigation treatments implemented between 2011–2014. Variations in NDVI within the treatments had statistically significant and visually discernible influence on sub-canopy temperature as evidenced by spatial temperature trends that were congruent with NDVI maps, and indicates the model accounted for the influence of canopy. The ability to develop higher resolution understory microclimate surface maps is of value to many aspects of ecosystem monitoring and the methods presented here may be of use to researchers and managers who work in forested environments.