Models of upland vegetation communities at multiple spatial scales

Abstract

The composition and occurrence of vegetation communities changes across multiple spatial scales in response to both environment and human management. Key drivers at small spatial scales (<1m or quadrat-scale) include patch structure between individual species, at intermediate scales (1ha or field-scale) local environmental conditions, whereas at large scales (km or national-scale) broad climate and soil characteristics. This research takes advantage of vegetation data collected via contrasting methods across these multiple spatial scales to quantify the role of these drivers. Data from 167 1m2 quadrats in an upland 96ha sheep-grazed heft at Ashtrees Dipper, Northumberland, was used to understand the relationship between vegetation patch patterns and environmental drivers at sub-quadrat (10cm) and quadrat (1m) scales. The numbers, areas and shapes of vegetation patches were primarily determined by soil characteristics, especially pH and water content, and proximity of sheep tracks (distance and length of sheep tracks). The resulting species patch patterns were then interpolated to field scale across the whole 96ha grazing area. Many countries have developed formal systems to classify vegetation communities, but no single generalisable method exists to allocate vegetation quadrats to community classes. Using the National Vegetation Classification (NVC) as an example, a novel generalisable method was developed to allocate vegetation quadrats to any classification via the computational generation of sets of “pseudoquadrats” for each NVC community at Ashtrees Dipper. These pseudoquadrats were summarised via detrended correspondence analysis (DCA) and new field quadrats placed within the ordination as passive samples. This then allowed a probability score to be calculated for each of the 167 field quadrats for its NVC community membership, which could then be interpolated across the whole 96ha grazing area. The NVC provides detailed information on the national distribution and characteristics of vegetation in Great Britain. Species distribution models (SDMs) were derived from data in the NVC handbooks, and geographic information system (GIS) predictor layers were used as SDM inputs. Predictions of NVC communities occurring in the protected Biodiversity Action Plan (BAP) habitats in England and Wales were made at 1km spatial resolution. Five SDM ii models were tested: generalised linear models (GLM), support vector machines (SVM), random forests (RF), maximum entropy (MaxEnt) and maximum likelihood (MaxLike). The distribution of individual species at 1km scale was then derived from the NVC community predictions. These species predictions were compared to records of species recorded in the National Biodiversity Network Atlas (NBN Atlas), using the catchment of the River Rede, Northumberland (~40km2) as a case study. GLMs, RF and MaxEnt produced robust predictions of the species distributions, with RF the most accurate. Overall, this research has demonstrated that the role of environment and management on individual plant species and their communities is best understood at multiple spatial scales, from the influence of sheep grazing in small-scale vegetation patches through to large-scale spatial distributions of species in BAP habitats.