There is a discrepancy in the scale at which ecological data is collected and the scale at which we need to understand the dynamics of populations and communities. As environmental conditions and population responses vary over different over time and space, models parameterised with small-scale data can fail to accurately capture the breadth of large-scale dynamics. There is therefore a need to increase the scale of ecological monitoring to provide adequate data for informative modelling.
A central problem is the trade-off between quality and quantity when collecting data, detailed counting of individuals within a population is expensive, time consuming, and limits the scale of ecological research. In this thesis I examine novel methods of generating and analysing data over large scales. The first of these are density-structured models, which forgo counting abundances in favour of simple categorisation of a population at a particular site, and model dynamics as a function of the probability of transitions between categories. In Chapter 2 I develop these models to account for population structure by incorporating hierarchical effects in the parameterisation of transition probability matrices, which define dynamics in these models. I show that hierarchical models provide considerable improvement over non-hierarchical models and suggest several useful parameterisations for future applications. Models that incorporate field-level effects into the cut-point parameters of ordered category logistic regressions demonstrate the best performance.
In Chapter 3 I apply the models developed in Chapter 2 to a national-scale density-structured agricultural weed data set, to examine the effects of spatial heterogeneity and management on weed dynamics. The weed in question, black-grass, is wide-spread, economically damaging, and difficult to control. Through transient analysis, two-step ahead projections and stochastic simulations, I demonstrate that an essential part of cultural weed control- crop rotation- does decrease the severity of weed infestations. Using break crops in rotations dominated by wheat can reduce weed densities and inter-year variability, with some break crops providing greater reductions than others. However, variance decompositions show that field-specific effects and the initial densities of weeds are greater contributors to the change in weed density than any specific crop-rotation. This suggests that site level factors may obscure or overwhelm the effect of interventions, highlighting the need for large-scale studies of population dynamics such as the one I have undertaken here.
Spatial structure is a major driver of population dynamics, and as such Chapter 4 investigates methods of expanding density-structured models to incorporate spatial information. I developed model parameterisations that incorporate spatial information into density-structured models through inclusion of a simple spatial covariate in the linear predictor of transition probabilities. I show that spatial models perform better than non-spatial counterparts, and the formulations I develop provide modest improvements to the ability of density structured models to capture field-scale spatial structure. Despite relatively modest improvements to model performance, these models demonstrate different dynamics in response to crop-rotation than spatially naïve models, with the contribution to system variances between field-specific factors and managements being far more comparable. This suggests that future predictive applications of density-structured approaches should only consider spatially explicit models when modelling the effects of crop-rotation.
The second technique I investigated in this thesis was metabarcoding, which involves using high-throughput sequencing and molecular taxonomy to simultaneously identify organisms across entire assemblages. This technology has particular promise for arthropod surveys, where traditional methods rely on the dwindling abundance of expert taxonomists, making biodiversity surveys slow, expensive, and often reliant on numerous different individuals. DNA-based identification using metabarcoding can be conducted using environmental samples or bulked samples of the organisms themselves and has been demonstrated to show fast and accurate identification of multiple organisms simultaneously. There has been little critical analysis of the limits of metabarcoding in terms of the number and type of organisms it can detect from bulked samples. I show that using standard pipelines and methodologies metabarcoding can produce taxonomic bias towards certain taxa and exclude others. I go on to demonstrate that pooling (or bulking) samples during DNA extraction reduces community coverage, and PCR produces stochasticity in species detection. I then conduct diversity analyses to assess the impact of landscape features on agricultural arthropod assemblages, however, given that there are obvious issues with barcoding in this manner it is difficult to draw robust ecological conclusions from these data.
Overall this thesis further develops methods for large scale monitoring and modelling of populations and communities. I develop density-structured models to incorporate hierarchical population effects and spatial structure and provide a demonstration of their utility. I also highlight potential problem areas for future metabarcoding studies of abundant and diverse arthropod communities.
