Delivering large data science projects remotely: lessons learnt
Doing data science working from home is challenging: here is what we have learnt...
We are delighted to finally produce the final technical report on water demand insights from summer 2018. The full report has been provided to the collaborating organisations who funded this work. This extended blog includes an edited version of the project executive summary, and presents the main findings.
The publication of this milestone report comes at a time when water companies have been experiencing another period of exceptional household consumption. This has been driven by the extremely dry and sunny weather in April and May 2020, combined with the effects of the unprecedented COVID-19 lockdown.
The report also comes whilst the planning guidance for the 2024 water resources planning round is being finalised. The report makes a number of recommendations for water resources planning, which should be pertinent to this guidance.
This has been a true team effort from the Artesia team, and thanks also to the collaborating organisations!
After a wet winter and spring, the summer of 2018 was the joint hottest on record for the UK. The South East of England experienced 58 days of total rainfall < 0.5 mm. Affinity Water saw peak demand increase by more than 11% for 39 consecutive days in summer 2018. Interestingly though, the favourable water resource position at the start of the summer meant that demand was unconstrained by temporary use bans.
The industry was keen to understand the drivers for the exceptional demand for water and 15 water companies and the Environment Agency pooled their data for the study. This included long term time series data on system input through to shorter term detailed data on water use in homes and businesses.
The infographic below provides a summary of the results from this study. Read on for some further commentary on these results. Domestic metering reduces the impact of peak consumption.
Our analysis demonstrates that both metered and unmetered homes increased their water use in summer 2018. However, homes who are charged for their water services based on use, only increased their water use by 20% (and from a lower baseline) compared to ‘unmeasured homes’ which increased their water use by 30%. The increased peak use in homes was largely due to water use outside in gardens, for both metered and unmeasured homes.
We need to improve how we communicate with customers about water use and drought
Water companies engaged extensively with household customers during summer 2018, working hard to get the message about unprecedented demand out via a wide range of traditional and social media channels. This was reactive activity, calling for restraint to ensure the taps kept flowing.
This activity exemplifies what companies currently do during an emerging drought. However, the discussion with companies on this topic was focused on how the awareness of water resources can be increased amongst the public, by taking a more proactive approach outside times of supply-demand stress. The options that were discussed included:
• Always having some level of water scarcity status, which is consistent across the country and enables each company to broadcast their status clearly and unambiguously at any time of the year. These levels could be integrated with current drought plan status.
• Having a regular television slot – for example as part of the BBC Countryfile weather forecast, or as a sponsored section on commercial channels – which provides summary information on the water resources situation.
The extent and frequency of communications should be managed to reflect the prevailing water resources situation and kept under constant review to respond to heatwaves and other events that result in peak demands. Communication could continue at a basic level even if there is no resource issue, for example highlighting what the public can do to reduce consumption.
Project collaborators and other stakeholders should consider co-ordinating communications at a regional or national level, with water companies, regulators and other trusted partners (like the “Beat the Drought” campaign in 2006). It is recommended that further work be carried out to assess this approach, engaging with social scientists, customer communications teams and other relevant subject specialists.
What would the effect of demand restrictions be in the 2020s?
The weather-demand modelling in this project has enabled the effect of the 2006 “Beat the Drought” campaign to be evaluated. However, this occurred 14 years ago, and it is very uncertain whether a similar campaign now would result in a similar outcome. Customer and wider public attitudes are likely to have changed and the customer base itself is now different, with more metered households and new uses of water (e.g. larger demountable pools). In addition, social media has changed the customer communication landscape and will expose customers to the different messaging around the UK.
Each drought will be unique and the effectiveness of restrictions on demand will depend on many variables including time of year, population affected and the weather itself at the time of the restriction. These factors mean that it will always be challenging to predict the effect of TUBs on customer demand. Further research on customer attitudes, including differences between metered and unmetered households is recommended to improve how this is predicted.
Should demand restrictions still be applied?
This study suggests that water companies will always need ways of managing customer demand, as part of the ‘toolbox’ for maintaining resilient, sustainable and cost-effective supplies, in the face of climate change and other uncertainties. Temporary Use Bans (TUBs – which have replaced ‘hosepipe bans’) are a legitimate step in water resources management. Regulators and others should not penalise water companies for applying such restrictions but recognise that TUBs ensure the supply of water to customers in the most resilient and sustainable way.
It is concerning that many stakeholders, public and media regard the implementation of demand restrictions as a failure of management. This view is incorrect, and the misperception should be addressed by government and regulators. We recommend that the improved communication described above will help address this.
Who should be subject to demand restrictions?
At present, TUBs would be applied across a water company region, even though this study shows that only a relatively small percentage of household customers use large volumes of water in the summer. A more equitable solution would be to only limit certain users or uses of water. However, without better and more consistent data on individual household consumption – e.g. from smart metering – it is difficult to see how restrictions could be targeted to the ‘appropriate’ properties.
How else can we reduce peak demand?
Evidence presented in this report shows that reducing baseline demand, via leakage reduction, metering or water efficiency will bring down the total value of peak demand. This approach is a low-risk way to reduce peak demand which will reduce the need for demand management.
How should we measure peak demands in the future?
The summer of 2018 clearly demonstrated that a single metric, such as a peak factor, is unable to capture the range of dry year conditions which are likely to test water resources systems. The duration and total volume of additional water demand in the peak needs to be accounted for.
It is recommended that water companies test how their 2018 demands would affect the supply-demand balance in other drought years like 1976, 1995 and 2006. This is a relatively easy task in the behavioural water resources models that most companies now use. It is harder to assign a probability or return period to summer 2018 demands in a changing climate. The Met Office suggests that summers like 2018 will become more frequent in a climate changed future. And the impact of such summers will be much greater when preceded by one or more dry winters.
These issues will become easier to incorporate as water resource planning moves away from the deterministic ‘dry year’ and ‘critical period’ planning scenarios used over the last 25 years. Weather-demand models linked to supply-side weather sequences will be able to incorporate year-to-year variability in the weather that drive summer peak demand. It is recommended that this more dynamic approach to demand analysis be included in the guidance on assessing the impact of variable peak demand in water resources plans.
The other key observation from this project is the clear division between those regions where peaks are weather driven, and those that are population driven. The UKWIR Peak Demand Forecasting Methodology from 2006 recognised this and suggested alternative methods for analysing these different kinds of peak. It is recommended that ‘population-based peak demand’ methods should be explored and evaluated further.
The consultation conducted as part of this project also highlighted other factors influencing peak demand including from non-households. This project was unable to discern any quantitative relationships between non-household demand and explanatory variables, based on the data provided in 2018. Further consultation indicates that there is probably sufficient non-household consumption data available to explore this further. It is recommended that stakeholders consider a collaborative study to collate and analyse available evidence and specify data requirements for further assessing non-household peak demand responses.
Regulation and peak demands
Water companies have been set Outcome Delivery Incentives (ODIs) by Ofwat as part of their business plans for the 2019 Periodic Review (PR19). The most relevant of these to this project is the ODI for PCC. This is based on the three-year rolling average of reported annual PCC, with no correction or normalisation for the effect of weather on demand. This means the starting point for water companies in PR19 will be slightly higher than it would have been without the summer of 2018.
It is recommended that Ofwat and water companies look at options for applying a normalisation factor to PCC, to remove the effect of weather on this key metric.
This report highlights the large variation in household night use between normal and dry years. This is a critical consideration in ensuring companies estimate leakage correctly. Without an adjustment for seasonal night use it is likely that companies will over-report leakage and under-report consumption. Improved estimates of seasonal night use will also enable better leakage targeting.
It is therefore recommended that companies include seasonal night use adjustments in their leakage calculations.
The benefits of collaboration
This project was a genuine collaboration between 15 water companies and the Environment Agency. The project grew organically and has been run largely informally and with minimal management overhead. This model has enabled the scope and timescales of the project to be flexible, which has been (mostly) refreshing and liberating in terms of delivery.
Collaboration has also meant that we have been able to deliver significant, valuable outputs based on relatively small contributions from each participating organisation.
Collaborators have shared the data they had available quickly and openly. This has enabled the project to glean greater understanding and insights than would have been the case if the project had been delivered for separate organisations. Certain data sets have given us strong insights into how and why the demand was high and all participants have benefitted from this data.