Empirical modelling


Data Analysis and Model work:

Office-based work will involve data collation and checking. Data will be analysed to determine the impacts of treatments on the cover and growth/vigour of heather, of peat-forming species and of other relevant vegetation types (i.e. sedges). Analysis will specifically include the impact of treatments on the peat water table, which relates to measured and modelled fluxes of GHGs, DOC and POC. The impact of treatments on the discharge stream flow regime will also be established in relation to measured runoff. The GPR surveys will enable an analysis of the impact of treatments on any peat pipes present and changes over time. A particular question to be addressed alongside management impacts is: how do the main investigated parameters (e.g. carbon dynamics) respond to environmental factors? For example, we know that root and microbial respiration respond differently to temperature (e.g. Hartley et al. 2007b) and depend to a large extent on substrate supply via the root system (Hartley et al., 2007a), whith a clear link to above ground (canopy) processes (e.g. Heinemeyer e tal., 2012c); the separated flux components in relation to canopy C fluxes and climate data will provide the necessary data to investigate such relationships (empirically but also at process level).

Before-After-Control-Impact (BACI) approach:

Our proposed catchment scale approach (mowing) will allow a large scale experimental approach to be assessed by statistical means (Stuart-Oaten et al., 1986). This is based on comparing the ‘behaviour’ of investigated parameters before, during and after the application of treatments. Large-scale experiments tend to be much more policy relevant and here we include three of those pairs. Therefore, our extended pre-treatment period (over a full growing season) makes this approach particularly attractive in terms of policy advice and developments. The BACI approach turns a pseudoreplication (replication in time) into a meaningful statistical interpretation. As pointed out by Stuart-Oaten et al. (1986): “Key to the resolution of these issues is the correct identification of the statistical parameter of interest, which is the mean of the underlying probabilistic "process" that produces the abundance, rather than the actual abundance itself”. Accordingly, we will deploy the BACI approach (as suggested in e.g. Lennon, 2011). Basically, this will only use the mean value per control and treatment per site and parameter and then analyse the temporal dynamics of the overall mean (n = 3).

ANOVAs and Empirical Models:

Firstly, we shall deploy common statistical approaches, treating the monitoring sites as independent replicates so enabling application of ANOVA (or General Linear Model or Generalized Linear Model as appropriate to the data collected). This will be applied to the time series data (repeated measurements at the same plot over time) as part of a MANOVA (multivariate ANOVA). Where applicable, we shall further consider measured covariates (e.g. water table depth, peat depth etc.; ANCOVA). Secondly, we will apply multiple regression models to derive site specific empirical models (regression models on how carbon and vegetation interact with hydrology and climate) and include literature data (where appropriate) to test for how our relationships compare to those at other locations. Secondly, we shall develop specific empirical carbon and water flux models, linking environmental data versus observed flux rates: (i) NEE models based on the light response curves, (ii) root and soil decomposition fluxes based on the root cutting treatment, (iii) net carbon uptake (NPP) based on combining measured NEE, estimated gross primary productivity (GPP) based on dark respiration and soil respiration fluxes (see Heinemeyer et al., 2011a), (iv) AET (water) fluxes based on the NEE and soil chamber water fluxes depending on PFT and water table position and (v) derive plant functional type (PFT) climate surfaces based on WTD and climate. The analyses will be done at different temporal time scales, reflecting measurement frequencies and gap-filled datasets (e.g. GHG fluxes derived from empirical models).