Predictability of the biological activity of a sandy grassland under optimal and stress conditions

The biological activity of a sandy pasture was quantified and modeled by integrating multiple categories of attributes and reducing the set of explanatory variables to the smallest subset with a highly statistically significant influence on the response variable. The attribute groups consisted of terrain-related factors (surface heterogeneity, altitudinal difference, topographic position index, etc.), soil properties (soil carbon, soil moisture, etc.), meteorological conditions (e.g., air temperature and precipitation), botanical characteristics (species abundance and diversity metrics), reflectance-based variables (vegetation indices), and physiological activity-related indicators (leaf area index, gross primary production). The datasets were collected from a one-ha spatial grid with a 10 m × 10 m resolution. The data collection spanned 10 occasions during the vegetation periods from autumn 2016 to autumn 2019.

Vegetation biological activity exhibits strong sensitivity to variability in both biotic and abiotic drivers, and species richness represents a key determinant of grassland response capacity. To quantify these processes, we constructed a composite variable that integrated below-ground functioning (derived from soil respiration measurements), above-ground productivity (based on above-ground biomass values), and the diversity of the sandy pasture. This composite metric was termed the biological activity factor (BF). To account for interannual and seasonal variability, all components of BF were rescaled before aggregation.

To gain a deeper insight into the key factors responsible for BF prediction and predictability, the upper and lower quartiles of the BF were modeled separately. This approach enabled the identification of the key drivers determining the biological activity of the vegetation under optimal (upper BF quartile) and stressed (lower BF quartile) conditions. We used linear and generalized additive models (GAMs) to estimate BF quartiles employing a reduced set of explanatory attributes selected through stepwise procedures based on statistical significance.