Harou et al. construct a synthetic drought in 2020 based on the paleo-record, rather than GCM projections. Their results regarding agricultural water scarcity and environmental flows are consistent with other CALVIN-SWAP studies. Environmental flows are also extremely restricted. Marginal opportunity costs of environmental flows rise by one or more orders of magnitude with extreme drought as compared to the historic baseline, with the Trinity, Clear Creek, and Sacramento Rivers experiencing the highest increase. Average agricultural water scarcity increases 3900% across the entire state under extreme drought even under well-functioning water markets, which seems somewhat implausible and may result from an overly restrictive model. Although Dale et al. do not calculate scarcity costs, they find that a 60-year drought with 70% reduction in surface flows only moderately impacts the total amount of irrigated acreage in the Central Valley, which declines from 2.4 million hectares to 2.1 million. This suggests that Central Valley farmers tend to have a relatively inelastic groundwater demand, compensating for the loss in surface water with groundwater rather than fallowing. Within the Valley, they find that Tulare Basin has a greater increase in fallowing than the San Joaquin Basin since the former is historically more dependent on groundwater. Dale et al. are also able to capture the increase in aquifer subsidence due to increased withdrawals during the prolonged drought,nursery pots suggesting that the quality of the aquifer will decline through time with excessive pumping.
Joyce et al. use WEAP-CVPM to model climatic changes with 6 GCMs under B1 and A2 scenarios for 2006–2099. Unlike the CALVIN-SWAP studies, they model irrigation efficiency by assuming that vegetable and fruit and nut crops in the Central Valley will be entirely converted to drip irrigation, and half of field crops will be converted by mid-century. They find that these adaptations tend to offset increasing water demands caused by increasing temperatures and periods of drought. The model even projects a reduction in annual groundwater pumping compared to the historical period until mid-century under some scenarios. Unfortunately, this positive effect of switching to drip irrigation is lost toward the end of the century as higher temperatures drive up crop water demands.One of the earliest econometric evaluations on the impacts of human-induced climate change to US agriculture developed out of a reaction to the limited substitution options in earlier production function models . Given long-run production decisions, one could use a cross-section of county farm data to assess how climate could impact agricultural rents. The underlying assumptionis that farmers have fully adapted to their environmental circumstances . While one cannot explicitly identify these decisions, these models are an improvement over older models because one could at least measure the consequences of adaptive decision-making . Other econometric models are able to capture annual decision-making using panel data. However, these models are unable to capture the idea of adaptation because farmers are very likely unresponsive to changes in weather from one year to the next. In their classic paper, Mendelsohn et al. evaluate the impact of climate variables on the expected present value of future rents in US agriculture, which they assume is proportional to farm land value under a few simplifying assumptions.
They regress farm land values on climate, soil, and socioeconomic variables using cross-sectional data for 2933 US counties. Their results reveal the seasonality and non-linearity in the relationship between climate variables and farm land value. Beyond this, both the direction and magnitude of their estimated climate parameters have been criticized and subsequently revised . One criticism with the original model, now obvious with hindsight, is the omission of ground and surface irrigation variables . Indeed, the negative effect of summer precipitation on land value in Mendelsohn et al. is potential evidence of omitted variable bias and, as Schlenker et al. suggest, misspecification. In response, Mendelsohn and Dinar include a surface water variable interacting it with annual temperature and precipitation. They find that the former is positive while the latter is negative, suggesting that counties with more surface water can tolerate higher annual temperatures and lower annual precipitation. Schlenker et al. indicate that dryland and irrigated counties require two separate estimation equations, unlike the single estimation equation in Mendelsohn and Dinar . A single equation erroneously implies that dryland farms requiring irrigation in the future will have access to analogous large-scale water projects peculiar to the western US at a given point in history. In testing the null hypothesis that each of the 16 climate variables in the original analysis are individually the same in the dryland and irrigated farm sub-groups, they find that between 4 and 6 coefficients are significantly different from 0, depending on the weighting method.
Even though Schlenker et al. did not have access to water data on irrigated counties at the time, their F-test was still able to provide sufficient proof of the bias in pooling dryland and irrigated counties into one model. After studying climate impacts to dryland agriculture in the US , Schlenker et al. study the impact of water availability and degree days on California farmland values. Their cross-sectional dataset represents individual farms, rather than county aggregates.5 Including groundwater and surface water supply corrects for the omitted irrigation variable bias in Mendelsohn et al. . Importantly, Schlenker et al. include a nonlinear measure of temperature effects on crop growth known as degree days.6 Their results suggest a positive relationship between the long-run annual availability of surface water and farmland value . They find that the coefficient on surface water is sensitive to water price: as water price per acre-foot increases, this coefficient decreases. They also find that the coefficient on degree days is positive and statistically significant, while degree days squared is negative and statistically significant. They do not use these relationships to estimate impact to farmland value under future climate scenarios. A criticism of degree days, as used in Schlenker et al. , is that it is a measure of weather not climate . In contrast to cross-sectional analysis, Deschenes and Greenstone estimate the impact of yearly fluctuations in weather on annual farm profits using US county level panel data , under 3 climate scenarios: uniform, Hadley II , and Hadley II . When they account for county and year effects, the results from all three models show a negative impact on profits. With the addition of state-by year fixed effects, all three models show a positive impact on annual profits. Fisher et al. find data and coding errors in the Deschenes and Greenstone model, biasing the original results in the positive direction. Specifically,large pots plastic the climate variable on the average number of degree days has a zero value for several counties, and climate projections varied by state while their historic climate data varied by county. Both of these errors tend to result in a regression toward the mean effect, with warm counties projected to get cooler, and vice versa. In response, Deschenes and Greenstone acknowledge the data and coding errors, and find that the $1.3 billion benefit in annual profits under Hadley II is actually a $4.5 billion loss.7 However, Deschenes and Greenstone disagree that state-by year fixed effects are misspecified. Like Fisher et al. , they find that state-by-year fixed effects tend to absorb most of the weather variability, resulting in a positive impact on profits. Their purpose in including state-by-year fixed effects is to control for state-level shocks in prices and productivity. To test for this, Deschenes and Greenstone include two additional specifications of year fixed effects: varying according to 9 USDA Farm Resource Regions, and varying according to 9 US Census Divisions.
The results from an F-test reject the null hypothesis of zero local shocks. Just as excluding all year effects, as in Fisher et al. , may bias the results downward, including fixed effects at the state level may be too strict, biasing the results upwards. The two intermediate cases of region-by-year fixed effects may present a “happy medium” to this problem. Schlenker and Roberts use panel data to study yield impacts to cotton in the western US.8 Constructing a dataset that incorporates the entire distribution of temperatures within a day, and across all days of the growing season, they find that the level of yield decline is greater under nonlinear temperature effects than linear ones. Even under a moderate emissions scenario , cotton yields decline across the western US by approximately 30%. Their approach is analogous to statistical crop studies discussed in the Impacts of Climate Change to California section . Massetti and Mendelsohn test the use of panel data on a Ricardian model using the same Agricultural Census data as Deschenes and Greenstone . They test the stability of climate variables using two panel data approaches against a repeated cross-section Ricardian model. Both panel models have relatively stable climate variables across the six Census years tested. There are $15 billion in welfare gains for a uniform 2.7 C warming and 8% precipitation increase for both panel models, although this ignores distributional welfare impacts. In contrast, the climate variables of a repeated cross-section Ricardian model vary through time. As a result, the welfare calculations also vary through time. Deschenes and Kolstad use panel data on aggregate county-level farm profits to study the differential effects of climate and yearly fluctuations in weather . Their climate variables include a 5-year moving-average of the annual degree days and precipitation, while weather variables are represented by annual degree days and precipitation. While none of the coefficients on annual degrees days are statistically significant with either the historical or CCSM models, their study is instructive in finding that the climate variable has a greater magnitude than the weather variable both in the baseline and climate change scenarios. This corroborates the theory that long-term changes in weather are more costly for some farmers than short-run fluctuations. They tease out which farmers may be most affected by changes in climate by analyzing 15 of the largest crops . They find that certain crop revenues respond positively to degree days , while others respond negatively . A few econometric approaches study specific adaptations. Mukherjee and Schwabe evaluate the benefits of access to multiple water sources for irrigated agriculture in California. Using a hedonic property value approach, they find that the marginal value of average water supplies from the Central Valley Project or State Water Project decreases as access to other sources increases. Lobell and Field study the use of federal crop insurance and emergency payments/loans in California from 1993–2007. They find that the most common cause of insurance and disaster payments during this period is excess moisture. Cold spells and heat waves are also important causes.We return to the question posed in the title. What have we been assessing with respect to the human and institutional responsiveness known as adaptation to climatic change in more recent studies on the topic? Several sub-questions are subsequently discussed. To what extent have study results identified economically efficient adaptations? To what extent have economically efficient adaptations reduced vulnerability to climatic changes and/or welfare losses? Have these studies identified limits to adaptive capacity in the agricultural sector, tempering the optimism of earlier studies? Normative approaches have provided insight into which adaptations may be economically efficient equating this with the optimal solution to the farmer’s objective function. There are two such adaptations explicitly represented in the CALVIN/SWAP models: changes to crop mix and water transfers/markets.9 As water resources decline, the resulting crop mix will reflect a decline in field crop acreage, with relatively less change for specialty crops . CALVIN includes water markets as an institutional adaptation. Under climate-induced water reduction scenarios, water is transferred from low-value to high-value use. Implicit in this is the transfer of land from agricultural to urban uses, though this is not directly modeled in these studies. In the WEAP-CVPM framework, Joyce et al. implicitly model the potential for converting to drip irrigation, particularly for thirsty field crops. By contrast, economically efficient adaptation is assumed, rather than modeled, in positive approaches, such as Ricardian models. Ricardian approaches have thus studied how climatic change will impact agriculture in the presence of long-run economically efficient adaptations.