Catchment Management Modelling Platform

Case Study 5

Uncertainty in ecological responses to water quality control measures at the river basin scale

Participants

Richard Williams1; Andrew Wade2; Peter Daldorph3; Mike Hutchins1.

1Centre for Ecology and Hydrology; 2Geography and Environmental Science, University of Reading; 3Atkins Global.

Stakeholder Representatives

Jo-Anne Pitt, Environment Agency; Matt Charlton, Environment Agency; Sian Davies, Environment Agency; Rachel Cassidy, Agri-Food and Biosciences Institute.

Forum issues addressed by case study

How can I know if land management will be effective?

  • What is the effect of different land management interventions on water quality? 

Does it matter what data and models I use in planning /assessing catchment management? Can the answers they provide be transferred to other catchments?

  • Is the answer sensitive to the quality and quantity of input data?
  • Is the answer sensitive to different models?
  • How different are the answers from different models?

How can I look at sensitivity / impacts on different catchment functions / services?

  • Is freshwater biology more or less sensitive than chemistry?
What did we find?

QUESTOR modelling shows that all the mitigation methods applied can reduce the amount of Chlorophyll-a predicted to occur in the river. The extent of the reduction is different depending on the data used to drive the QUESTOR model. The effect of increasing the shading (from 20-60%) declines downstream as the river becomes too wide for trees to provide shade.

cs5_01.jpg

Effect of increasing riparian tree coverage from 20 to 60%: using three different sets of driving phosphorus data. Lines only refer to 20% shading (current condition) and circles show 60% shading.  Red lines are observed driving data, blue lines are derived from P data from SAGIS and green data are P and flow data from INCA. Values of Chlorophyll-a above 0.03 mg/L are considered undesirable.

 

How did we do it? (Synopsis)

We investigated the effectiveness of a range of possible measures for the mitigation of algal blooms in the River Thames using a combination of models and observed data. Three mitigation options were employed: Increased river shading, reduced diffuse inputs of P and reduced P inputs from sewage works.  The QUESTOR model was used to link changes in light and river P concentrations to changed algal growth along the river Thames. SAGIS and INCA-P were used to calculate the inputs to QUESTOR for the P reduction mitigations. The QUESTOR and SAGIS models had already been calibrated for the Thames catchment in previous studies. INCA-P was a new model application and was calibrated for P species and river flow using observed data. Model outputs allowed us to investigate how sensitive the results of the mitigation methods were to the model used to implement that mitigation.

How did we do it? (Full)

This study used models to assess the effects of interventions designed to reduce periods of unacceptable river water quality that arise from the growth of algae. A particular focus was on driving the river model (QUESTOR) with observed data and data from the models INCA-P and SAGIS to investigate the effects of such different data sources. The QUESTOR and SAGIS models had already been calibrated for the Thames catchment in previous studies. INCA-P was a new model application and was calibrated for P species and river flow using observed data

The study was carried out on the River Thames Catchment in South East UK, a river known to have high loadings of nutrients and subsequent high productivity leading to nuisance algal growth and on occasions low dissolved oxygen (DO) levels. The interventions investigated were:

Increased river shading provided by growing tree on the river bank (only feasible down to Wallingford where the river width is small enough). The shading was applied to the baseline QUESTOR calibrated model run driven by observed data, the QUESTOR model driven by INCA-P data (as calibrated) and the QUESTOR model driven by SAGIS data (as calibrated).

Changes in farm practice. A national case study has looked at sets of farm practices to reduce N and P loss from catchments. The percentage change in loss rates associated with various levels of uptake in the farming community have been calculated using the Farmscoper model. The values for the Thames basin were extracted from this assessment and applied to the SAGIS input loads. The resulting SAGIS outputs were used to drive the QUESTOR model to determine the effects on river P concentrations and algal growth relative to current practice. Two uptake levels of uptake were assessed: 100% uptake of best practice is taken up voluntary (22.6% P reduction) and all Farmscoper measures are taken up (~ 37.3% P reduction).

The effect of imposing stricter discharge consents for SRP on all sewage works irrespective of size (80% reduction in discharge loads). Catchment loads estimated from the calibrated versions of SAGIS and INCA-P were reduced by reducing the point source discharges that were included in them by 80% and then each was used to drive the QUESTOR model. The Questor model itself received direct sewage works discharges and these were also reduced by 80%.

Data from these different model runs were analysed to look at sensitivities in model output resulting from different sources of driving data and how these differences compared with differences that resulted from the effects of the mitigations themselves. 

What did this case study show? (Synopsis)

The simulations with SAGIS and QUESTOR showed that improved on-farm practice as describe by Farmscoper had a very small impact on the Chlorophyll-a concentrations predicted for the river Thames (Figure 1). Both increased shading (driven by observed data or by outputs from SAGIS and INCA-P, Figure 2) and the 80% sewage P reduction (inputs from INCA-P, Figure 3) had a significant effect on the Chlorophyll-a concentrations. Figure 3 also shows that the 80% effluent reduction is a more effective mitigation measure along the whole river than shading. However, note that shading is only effective down to Wallingford. It is encouraging to note that the results obtained for shading with all three models are consistent for Chlorophyll-a. This is also shown in the dissolved oxygen concentrations that are predicted to change as a result of the changes in Chlorophyll-a concentration (Figure 4). QUESTOR simulated an afternoon dissolved oxygen concentration. 

cs5_02.jpg

Days exceeding WFD good status values (90th percentile > 0.03 mg/L) in the period 2010-2012 for Chlorophyll-a under two different levels of uptake of farm scale mitigation measures implemented in SAGIS and used to provide P inputs to QUESTOR. Blue bar is 100% of best practice and green Bar is all possible measures in the Farmscoper model. 

cs5_03.jpg

Effect of increasing riparian tree coverage from 20 to 60%: using three different sets of driving phosphorus data. Lines only refer to 20% shading (current condition) and circles show 60% shading.  Red lines are observed driving data, blue lines are derived from P data from SAGIS and green data are P and flow data from INCA. Values of Chlorophyll-a above 0.03 mg/L are considered undesirable.

cs5_04.jpg

Effect of two different mitigation options on the Chlorophyll-a concentrations in the River Thames. In all cases the QUESTOR model is driven by output data from INCA-P. Red line is the current situation (20% shading). The green line shows the effects of 60% shading and the blue line an 80% cut in the P effluent from all sewage treatment works.

cs5_05.jpg

Effect of increasing riparian tree coverage from 20 to 60% on Dissolved Oxygen using three different sets of driving phosphorus data. Lines only refer to 20% shading (current condition) and circles show 60% shading.  Red lines are observed driving data, blue lines are derived from P data from SAGIS and green data are P and flow data from INCA. Values of Chlorophyll-a above 0.03 mg/L are considered undesirable.

What did this case study show? (Full)

Model Performance: The QUESTOR model was calibrated against observed data for the case where the model inputs are based on interpolated observed values. In this study the QUESTOR model was also driven with outputs from SAGIS and from INCA-P. SAGIS provided only P data while INCA-P provided both P and flow data. The model performance against observed flow at Caversham is poorer when the QUESTOR model uses the flow data from INCA-P (Nash-Sutcliffe = 0.81) than when the observed flows are used (Nash-Sutcliffe = 0.90, Figure 1). INCA-P flow data leads to an over prediction of summer low flows and to over prediction of some high flows in winter compared to the observed data. For the simulation of dissolved P a comparison can be made for all three sets of driving data (Figure 2). The differences between the simulations are not so clear cut. In fact the simulation with the observed data at Sonning is not particularly impressive and give the same in goodness of fit measures to the results when the model is driven by data from SAGIS (% error in means = -7.4 and RMSE =  0.083). The values for the model simulations when driven by INCA data are inferior (% error in means = 47.8 and RMSE = 0.134).

Results of mitigation measures: The locations of the sites referred to in this section are shown in Figure 3. Implementing the two Farmscoper scenarios in SAGIS to drive QUESTOR, gave a small reduction in P concentrations in the River Thames and therefore very little change in the number of days where Chlorophyll-a concentrations were over WFD limits (Tables 1 and 2). These differences are essentially negligible in the context of the likely simulation errors.

Reducing light reaching the River by shading from trees had a significant impact in the top half of the river dopwn to Wallingford - the section where it is narrow enough to have an effect. Tables 3-5 show the days where unacceptable conditions were exceeded for Chlorophyll-a and Dissolved Oxygen for 20% (present conditions) and 60% shading. The direction of change of all the model runs is the same although the magnitude of the change is different. The results from the INCA-P/QUESTOR model pair are somewhat different from the other two model runs. This is most likely due to the overestimation of the summer low flows, which are critical for algal growth in the River Thames.

An 80 percent reduction in P from all sewage treatment plants was also very effective in reducing Chlorophyll-a along all the river system (Figure 4). Figure 4 shows a comparison between the effects of shading and P reduction from sewage works for the INCA-P/ Questor model pair. This implies that a combination of upstream shading and tighter downstream sewage works discharge consents would significantly improve water quality.

Table 1 Median dissolved inorganic P concentration  (mg/L) in the period 2010-2012 under two different levels of uptake of farm scale mitigation measures implemented in SAGIS and used to provide P inputs to QUESTOR. S1 is 100% of best practice and S2 is all possible measures in the Farmscoper model. 

cs5tab1.jpg

Table 2 Days exceeding WFD good status values (90th percentile > 0.03 mg/L) in the period 2010-2012 for Chlorophyll-a under two different levels of uptake of farm scale mitigation measures implemented in SAGIS and used to provide P inputs to QUESTOR. S1 is 100% of best practice and S2 is all possible measures in the Farmscoper model. 

cs5tab2.jpg

Table 3 Days exceeding WFD good status values (90th percentile > 0.03 mg/L) in the period 2010-2012 for Chlorophyll-a under two shading scenarios (20% current conditions) and 60% for three different sets of QUESTOR driving data. Observed Flow and P, Observed Flow and P from SAGIS and Flow and P from INCA-P. Note that increasing shading is only practical down to Wallingford.

cs5tab3.jpg

Table 4 Days exceeding WFD good status values (10th percentile < 6 mg/L) in the period 2010-2012 for Dissolved Oxygen under two shading scenarios (20% current conditions) and 60% for three different sets of QUESTOR driving data. Observed Flow and P, Observed Flow and P from SAGIS and Flow and P from INCA-P. Note that increasing shading is only practical down to Wallingford.

cs5tab4.jpg

 

cs5_06.jpg

 

Figure 1: Comparison of the modelled flow (m3/s) data with observed values at Caversham on the River Thames (note the log10 scale). The black line is the observed data. The red line is the QUESTOR output driven by observed data. The green line show the simulation when the observed data is replaced by modelled data from INCA-P. 

cs5_07.jpg

Figure 2: Comparison of the modelled output ortho-P (mg/L) data with observed values at Sonning on the River Thames. The black dots are the observed data. The red line is the QUESTOR output driven by observed data. The blue (SAGIS) and the green (INCA-P) lines show the simulation when the observed data is replaced by modelled data. 

cs5_08.jpg

Figure 3: Location of sites on the River Thames mentioned in figures. Site 2 is Eynsham, Site 4 is Wallingford, Site 5 is Caversham, Site 6 is Sonning and Site 9 is Runnymeade. Major tributary inputs and the Tidal limit are also shown.

cs5_10.jpg

 

Figure 4: Effect of two different mitigation options on the Chlorophyll-a concentrations in the River Thames. In all cases the QUESTOR model is driven by output data from INCA-P. Red line is the current situation (20% shading). The green line shows the effects of 60% shading and the blue line an 80% cut in the P effluent from all sewage treatment works.

What were the benefits of using more than one model?

QUESTOR can predict the effects of nutrients on algal growth, but cannot predict future nutrient changes resulting from the mitigation methods. We needed another model to derive the impacts of nutrient reduction mitigation options to provide revised loads from which QUESTOR could calculate changes in algal biomass.

Using one model for the assessment of mitigation options leads to one outcome that depends on the model and the data used to drive it. Using multiple models in combination provides multiple outcomes for mitigation methods allowing assessment of how sensitive were the outcomes to driving data and models.

What were the lessons learned about how to apply the models?

When joining models off-line, it is very important to ensure that the model output and input points are properly paired up, which can be a slow process. The three models operate at different time steps, but this was simple to accommodate in the QUESTOR programme structure.

 

What pre- and post-processing was done on the input and output data?

All models require considerable pre-processing on inputs to provide data in specific formats for the models. Outputs are extensive and available in files compatible with data analysis packages for the production of graphical output. SAGIS can provide show outputs directly in a GIS based environment.

 

W​hat datasets were used in the case study?

Flow data from the National River Flow Archive

Effluent Flow Data from a number of sources

Environment Agency routine water quality monitoring (GQA) data (including Temperature) from the WIMS database

Weekly Nutrient and Chlorophyll-a data from the CEH Thames Catchment Initiative Sampling

Solar Radiation data from BADC

QUESTOR:

  • Location of weirs from WIKIPedia
  • Extent of tree shading from visual inspection of satellite data on Google Earth®

INCA – P:

  • Land Cover Data
  • Rainfall and soil moisture deficit across the catchment

SAGIS:

  • Flow data from LowFlowEnterprise
  • N and P loss data from Farmscoper Model
Videos
CaMMP_Case_Study_5_-_Project_view-0.jpg
Project view
Stakeholder view
Location

Thames catchment, England

Issues
  • Management outcome
  • Confidence levels
  • Other services

Pollutants
  • P Species
  • Algal blooms (Cholophyll-a)
Scale

Catchment

 Models
  • QUESTOR
  • INCA-P
  • SAGIS
CaMMP

Improved access to and integration among data and models to address key questions in catchment management for water quality and wider ecosystem services, providing a more holistic view to inform scientific understanding and policy development.

LEARN MORE

Featured video
CaMMP_Community_Forum-0.jpg