One of the most vulnerable sectors to a change in climatic conditions is agriculture, where climatic variables such as temperature and precipitation are direct inputs into the production function. Previous studies of the potential impact of global warming on U.S. agriculture have yielded widely varying results, with some predicting large damages and others suggesting that U.S. agriculture may even benefit, in at least one likely scenario associated with a doubling of greenhouse gas concentrations. In this paper we show that much, if not all, of the difference in estimates can be explained by the failure to adequately allow for differences between rain-fed and irrigated agriculture, and proximity to urban areas, in the estimation of the relationship between farmland values and climatic and other variables.

A cross-sectional data set of counties in the continental U.S. is employed to estimate a hedonic value function. We first model the error term structure and show that our derived set of weights is best at explaining the heteroscedasticity and spatial correlation of the error terms. Second, we use bootstrap simulations to assess the variability of the damage estimator. Third, Chow tests and a Bayesian outlier analysis show that irrigated and urban counties severely bias the damage estimator. When we limit the analysis to dryland and non-urban counties, the different damage estimators from previous studies overlap and the confidence intervals are cut by up to half. Dryland U.S. agriculture is unambiguously damaged under the CO2 doubling scenario, and the damages are quite large relative to recent estimates in the literature.