In arid areas, irrigation water is an essential input into agricultural production. However, rainfall and, correspondingly, surface water supplies, are often highly variable, creating uncertainty over the value of long-term, water-dependent investments in these cropping systems. Moreover, climate change is expected to increase both crop water requirements and the variability of seasonal rainfall, meaning the constraints imposed by variable water supplies are likely to grow in cost as climate change progresses. In this setting, storing water in wet years for use in dry years is valuable. In particular, it would be expected to increase the value of perennial crops, which require large up-front investments that pay off gradually over the life of the tree. We first show, in a simple theoretical model, that given the timing of returns to investments in perennial crops, there is always some level of drought risk above which annual crops will be preferred to perennials. We then demonstrate this effect empirically using a unique institutional setting in which access to a relatively new form of water storage, groundwater banking, effectively created spatial variation drought risk between irrigation districts in Kern County, California. Using a 21-year dataset of individual cropping decisions, we provide evidence that access to a large groundwater banking project, the Kern Water Bank, increased the rate at which farmers switched from lower-value annual crops such as wheat and alfalfa, into high-value perennial nut crops, primarily almonds and pistachio.