Modeling the Distribution of Forest Understory Plants Using Remote Sensing and Soil Chemical Predictors

Abstract:

Despite their diminutive stature, understory plants have been recognized for playing a disproportionately large role in forest nutrient cycling, successional dynamics, and diversity - representing on average 80% of species in forests of eastern North America. These species remain understudied, including their distribution and niche, limiting our ability to determine their fate in a changing environment.

Our study aims to determine what predictor variables are necessary to model the distribution of understory species – particularly whether remote sensing variables alone can create useful predictive models. To accomplish this, we sampled 200 plots across the White Mountain National Forest, New Hampshire, USA. Species presence-absence data were collected for 214 understory plant species. Soils were chemically analyzed by genetic horizon. A 1-meter resolution lidar-derived digital elevation model was used to calculate predictor variables related to topography, and Sentinel-2A spectral data and related spectral indices were calculated. Statistical analyses were performed using ensemble models and evaluated using Cohen’s kappa.

Preliminary results suggest that a combination of variables derived from soil chemistry, topographic indices, and Sentinel-2A spectral data result in the strongest predictive models, but that remotely sensed data alone are not sufficient to accurately predict the distribution of understory species. Soil chemical variables related to soil fertility, including base cations and carbon to nitrogen ratio, are frequently important model parameters suggesting that remotely sensed data are unable to account for a soil fertility gradient driving understory species’ distributions.

These models will serve to better understand the distribution of an important functional group in forest ecosystems. In addition, this work addresses current limitations to accurate modeling of plant species’ distributions. Our results suggest that soil chemistry is an important parameter in plant distribution models and currently limits the extrapolative ability of models built otherwise using remotely sensed data. Remotely sensed data capable of accounting for soil fertility in forested environments will be important for the future of modeling forest species’ distributions.