Water storage rights: decentralising reservoir operation

In regulated rivers (rivers controlled by large dams) the reservoir storage decision is of vital importance to users exposed to high costs from water shortages. The storage decision can be viewed as a yield reliability trade-off: greater storage reserves decrease the expected frequency and size of shortages at the cost of reduced average supply (due to losses from spills and evaporation). Given incomplete information on the preferences of users it is difficult for central planners to establish optimal storage policies.

This paper presents a model of a regulated river system, in which water users are able to make their own storage decisions (i.e. hold private storage reserves in the reservoir) as well as engaging in a water spot market. The purpose of the model is to quantitatively evaluate approaches to the specification of storage property rights. These approaches vary in how they reflect (internalise) the storage capacity constraint (spills) and evaporation losses. One of the main approaches is `capacity sharing' (Dudley and Musgrave 1988), in which users are assigned percentage shares in storage capacity and inflow.

As with spatial water markets decentralised storage is subject to externality problems. User storage decisions can result in external effects on other users owing to their effect on spills and evaporation losses. While sophisticated approaches like capacity sharing minimise these effects they are not able to remove them completely. Given the presence of externalities our model represents a stochastic game: each user is faced with a stochastic dynamic optimisation problem where the payoff and transition functions are dependent on the behaviour of the other users. We solve this model quantitatively using a relatively novel application of reinforcement learning.