How to improve water management as the resource becomes more unpredictable? An interview with Upmanu Lall from Columbia University, one of the authors of a work published in Water Resources Research exploring a water allocation model based on probabilistic forecasts.
How is the information present in the fine structure of forecasts and user requirements currently being used in the usual water allocation process?
At present, the water allocation process in most places does not usually use high resolution forecasts. In the US, NOAA (the National Oceanic and Atmospheric Administration) issues some categorical forecasts (Above, Average, Below) with some probabilities. Most water managers are not keen to use these since the uncertainty is high and their reward structure is not set up to take the risk of following a forecast. Helen Ingram and Steve Rayner wrote a report a few years ago documenting this and commenting on how water managers will not use forecasts given that they are risk averse. The argument is that if the forecast is “right” and the manager acts, he or she does not get any rewards from the institution. However, if the forecast is “wrong” and there was an action and adverse impacts result, then they could be fired. This adverse set of incentives governs behavior, and many of the public sector managers will respond with “I want the forecast to be 99% accurate before I use it,” without even being precise about what that means.
However, there are some examples where hydropower generation agencies like Hydro-Quebec and others, who have the capacity for hedging risk by taking some measures, will use either these forecasts or their own custom generated forecasts to decide on releases or allocation. The South Florida Water Management District uses a procedure to adjust rule curves for their reservoirs using the NOAA or related forecasts. They also do their own forecasts, which are arguably much better than the NOAA forecasts, but don’t use them officially. The reason is that if they do, then the liability associated with any impacts falls on them, whereas if they use the US Government’s product (that is, NOAA’s) they cannot be held liable. It’s a very interesting dichotomy.
How does the water allocation process you tested differ from the current one? What does it require from — and what would it offer to — water managers and users?
The key idea we had in designing the new process was that we needed to manage the risk that the manager and the user faced through the use of the forecasts. This could then facilitate their use and also improve the overall risk management strategy. Many farmers and users argue that they should not have to pay for water. Surely, they have the option to wait for it to rain and be subject to its variance and its effect on their productivity (average yields from rain-fed agriculture are one third to one fifth of those from irrigated agriculture where the water supply is reliable). So the paradigm we are pushing is that you pay for reliability and for delivery. How much is reliability worth? We try to address this is in the participatory framework by pricing and getting feedback at different levels of projected reliability. This is the first idea, and it is one that has been used in other places.
The second key idea is to tie the allocation to the forecast with reliability assessed in a probabilistic way, and then the contracts are offered so that the price includes an insurance-like rider so if you asked for low reliability and you don’t get water you get some compensation. If you asked for and paid for higher reliability and you don’t get water, you get a much higher level of compensation — effectively if you paid for 50% reliability and you paid $1 for 1 unit of water, and you did not get it, you may be compensated $2, whereas if you paid for 1 unit of water at 95% reliability, and you don’t get water then you would get compensated $20. Of course, if you wanted 95% reliability you would pay much more than the 50% reliability guy and not the same amount of money, but hopefully it’s clear that the fair premium you are charged relates in some way to the likely probability of failure (50% and 5% respectively in the above example).
So, what this effectively does is removing the risk from the decision maker, and managing it financially. In fact, the contracts with reliability we offer based on the forecast allow a pre-determination of how much water can be supplied with different levels of reliability, and the charges are then guaranteeing a financial compensation in case there is a loss of service. The lower reliability contract buyers take a hit first, then those with higher reliability contracts, and so on, so there is also a prescribed priority for failure. Since this is all laid out in the contract, both the buyer and the manager are insured from the process. Reinsurance companies could in turn insure the entire package since the probabilities of the outcomes and the total package insured are clear.
What is needed to make it work are well calibrated probabilistic forecasts. They don’t need to be perfectly accurate, but realistically they do need to be better than the historical probability distribution most of the time. The way we do it, there are years where the forecast distribution is very tight (a high precision forecast) and years in which the forecast distribution is quite spread out (a diffuse forecast). Potentially one could price the uncertainty differently each year, but we are not specifically doing that so far. The second ingredient that is needed is the capability to simulate the reservoir system to determine the combination of releases for different users at different reliabilities. This requires simulation software and one could use ours or others. This is not really a big deal. The third ingredient is the legal and institutional structure to enforce the contracts and to make sure water deliveries take place. This applies more generally to any systems where users are paying. Finally, there needs to be a way to determine prices, either set by the state as a function of reliability and perhaps with graduated tariffs, or voted for in an auction-like environment by users. If trading of contracts is considered, then that system needs to be set up as well.
So, the infrastructure required is a bit more than the traditional water allocation system where the managers follow a rule curve and are not as active. The gains are that the chance of spill and shortage is reduced and the users planting or usage decisions may be improved since they have a better idea of how much they may get and what the compensation is if they don’t get it. Yes, the users need to get a bit more sophisticated to make such decisions, and we visualize that the users will be aggregate users, i.e. irrigation districts and not individual farmers. The irrigation district could in turn disaggregate the water to farmers within it, and the same for urban water supply companies and urban users.
Do you think similar tools for risk management would be useful in other resource allocation problems?
Yes, I think so. The basic idea to develop a probabilistic forecast, and then to manage the risk in the allocation using user inputs as to price and reliability with the insurance mechanism. This is an idea that could be used by others.
Related articles:
- And You Thought High Finance Is Complex
