Recovery of acidified waters in the UK

WP 3: Predicting recovery using dynamic modelling

Recovery of lakes

MAGIC simulations of water chemsitry

Work under this programme represents the most extensive dynamic model assessment of surface waters in the UK, comprising calibrations to 454 sites in eight regions. The MAGIC model has been tested against 15 years of observed water chemistry data at seven sites in the UK AWMN. The simulations effectively match both long-term trends and year-to-year variation at most sites. A poor match of observed and simulated data in the wider dataset indicates a possible influence of S adsorption/desorption dynamics, variation in annual rainfall inputs and climatic effects. Reconstructed ANC concentrations are above zero (except one site which requires further examination) in all regions and are generally above 20 µeq l-1. In all regions, acidity peaked in the early 1970s and has decreased significantly to present day in line with observed reductions in S deposition. Predictions to 2020 under the emission reductions agreed for the Gothenburg Protocol indicate that surface water acidification will remain a problem at a significant (15-50%) number of sites in the Mournes, the S. Pennines and the Lake District. This indicates that further emission reductions will be required.

Invoking nitrogen dynamics in the model (at four regions where appropriate soils data were available) causes model predictions of significantly increased NO3 concentrations and, consequently, decreased ANC by 2050. The changes in ANC, however, are not sufficiently large to cause significant changes in the percentage of sites achieving relevant biological thresholds. It must be considered, however, that the increased NO3 concentration might have an impact with respect to eutrophication. With respect to ANC zero, all regions return to their pre-acidification status by 2030 except for the S. Pennines. For ANC 20 µeq l-1 no regions completely return to their pre-acidification status by 2100. Set in a European context, the predictions for 2016 (i.e. Water Framework Directive deadline) indicate the acidity problems that persist in the Lake District and S. Pennines are comparable with those predicted to occur in S. Sweden, Slovakia and N. Italy. A more significant problem is predicted to remain in the Mournes, comparable to the prediction for S. Norway.

Linking chemical simulations with biological status

Regional MAGIC predictions of water chemistry have been linked to biological status using the predictive models developed in WP 1, Task 1.2, specifically for the sum of acid-sensitive diatoms and invertebrates in six regions; Cairngorms, Galloway, Lake District, Pennines, Dartmoor and Wales. Significant biological change from baseline to present is modeled, with great regional variation. The Cairngorms are the least impacted region while in the most impacted Pennines, the probability of occurrence of acid-sensitive species is currently less than 0.5 at more than 40% of sites. Most sites in the six regions should achieve a probability of occurrence of at least 0.5 for the sum of acid-sensitive diatoms and invertebrates by 2024. Even in the most impacted Pennines region, this biological target will generally be met by 2024. However, in the Lake District, 20% of sites for invertebrates and 10% for diatoms will not meet this biological target within the next 100 years. Using a more stringent biological target of 0.75 probability of occurrence, none of the regions achieve the target entirely by 2024. The greatest biological recovery is predicted for Galloway and Wales. For the Pennines and Lake District, biological targets will still not have been reached in 10-20% of sites by 2100.

Recovery in streams

Testing the PEARLS-MAGIC-biological model in the Conwy region

All streams in the Conwy are subject to episodic ANC decreases. However, biologically-damaging episodic conditions (ANC < 0 or ANC < 20 µeq l-1) are most likely at streams with an already low mean ANC. The PEARLS-MAGIC approach provides an effective means by which to simulate present-day and future stream chemistry at the large-catchment scale based on readily available GIS datasets, and a limited programme of targeted sampling. The models allow lengths of chemically damaged stream to be predicted, both at present day and in the past and future. For the Conwy above Betws-y-Coed, 27% of stream length is currently predicted to be below an acceptable mean chemical threshold (ANC 20 µeq l-1). At high flow, this increases to 46%. Acidification (mean and episodic) is most severe for streams draining peat or conifer forest catchments. In 1850, simulated mean ANC was above 40 µeq l-1 for all landscape types. ANC may have fallen below µeq l-1 during episodes in the most acid-sensitive landscapes, but probably did not fall below zero. By the 1970 S deposition peak, many streams would have been chronically acidic, with mean ANC < 20 µeq l-1 and in much of the catchment negative.

The models predict that some recovery has already occurred; the major acidification problems in the Conwy are now associated with acidic episodes, rather than chronic acidification. Under the Gothenburg Protocol, some further improvement is predicted, but some continued episodic acidification is likely in peat, forest and montane areas of the catchment. For the first time, PEARLS-MAGIC ANC simulations have been used to predict changes in biological status at the catchment scale. This represents an important step towards a full linked chemical-biological model, in which the dynamics of biological recovery may be linked to spatial and temporal patterns of chemical change.

Model uncertainty

Transfer functions for hindcasting ANC from sediment core diatoms

While previous transfer functions have been developed to predict pH from lake sediment diatoms, the MAGIC model more reliably hindcasts ANC. The European Diatom Database (EDDI) created under an EU-funded project was used to develop a new transfer function using diatom data plus alkalinity and DOC to derive ANC. 163 samples from all the main softwater areas of the UK in both acidified and relatively pristine areas had sufficient data to be used in this exercise. Canonical correspondence analysis and Monte Carlo permutation tests were used to ascertain the relationship between ANC and diatom distributions independent of pH effects. While only a small proportion of variance is explained by these two parameters, it is highly significant, and both explain unique components of the diatom data.

A predictive ANC transfer function was developed using weighted-averaging with classical deshrinking, and its performance assessed by comparison with model predicted ANC for training set lakes. Internal cross-validation was also carried out using a bootstrap technique. Compared with other transfer functions the ANC model is relatively weak, accounting for only 49% of variation in the training set ANC compared with 80% for a pH transfer function developed using the same dataset.

Comparison of MAGIC, F-factor and diatom inference models with instrumental data

For a comparison of techniques, seven AWMN sites were selected with both sediment core data and MAGIC hindcasts. It has to be assumed that the training dataset fully encompasses the chemical and biological conditions represented by the fossil assemblages, i.e. contains analogues for the fossil assemblages. Only Round Loch of Glenhead was found to have close analogues for the whole sediment sequence. Other sites (Loch Grannoch, Llyn Llagi, Scoat Tarn, Lochnagar and Loch Tinker) have close modern analogues but not for the whole core, while Blue Lough lacks good analogues for its whole history due to the presence of a diatom species not represented in the training set. Despite these no-analogue problems, both diatom and MAGIC predictions agree well with measured ANC at four sites (Llyn Llagi, Round Loch, Scoat Tarn and Lochnagar).

For 1850, diatom and MAGIC pH hindcasts agree well at Llyn Llagi, Loch Grannoch and Loch Tinker. At Lochnagar, Round Loch and Scoat Tarn, MAGIC pH hindcasts are higher than diatom-based reconstructions by c.0.5 pH units. MAGIC ANC hindcasts are similarly higher than diatom-based reconstructions at all sites except Loch Tinker and Llyn Llagi. A wider dataset was used to compare ANC hindcasts based on MAGIC, diatoms and the F-factor (SSWC) model. There is considerable discrepancy between the three methods, with a systematic bias in the methods towards higher ANC predictions in the diatom, F-factor and MAGIC models respectively. Multiple linear regression of site variables against differences in ANC hindcasts between models found only one significant relationship, in the difference between diatom and MAGIC ANC values and site alkalinity. MAGIC baseline ANC values are increasingly high compared with diatom-based ANC values in the more sensitive, acidified sites with very low current alkalinities. The reasons for these systematic differences are currently unknown.