Mean summer (June–August) climate for the conterminous United States. Summer, rather than annual, averages of temperature and humidity are shown because the bulk of dairy production losses occur during this season (see Figure 3 and associated discussion for details). (Color figure available online.) 

Contexts in source publication

Context 1

... of the mean summer (June-August) climate are shown in Figure 4. Specifically, the two maps show average maximum temperature and afternoon humid- ity for the period from 1950 through 1999. ...

Context 2

... mod- erate RH. An interesting case is Tillamook, Oregon, which is humid but generally quite cool; it nevertheless shows production losses resulting from days that are dry but warm, where the impact of increasing temperatures outpaces the influence of decreasing humidity on heat stress. This strong geographic variability in the heat sensitivity of milk production follows directly from the geographic variability in temperature, diurnal temperature range, and humidity. Previous approaches that model milk production loss as a simple linear relationship with temperature give the same sensitivity to warming regardless of whether an area is climatologically humid or dry. Figure 2 includes the observed milk production results from Ahmed and El Amin (1997; downward-pointing ◦ triangles), which are the basis for the 1.15 kg/ C sensitivity used in Hayhoe et al. (2004) and other studies. The reported values are weekly averages of daily milk yield, which we have converted to production loss by subtracting from our baseline production estimate of 30 kg/day. As a direct test of the THI equation, the upward-pointing triangles in Figure 2 show the results for milk loss computed by applying Equation 2 to the observed temperature and humidity reported in Ahmed and El Amin (1997). 4 As can be seen from Figure 2, the milk loss computed from Equation 2 (upward-pointing triangles) is in good agreement with the observed loss (downward-pointing triangles). Based on the results of Ahmed and El Amin (1997), Hayhoe et al. (2004) used two thresholds for heat stress ◦ ◦ (25 C and 32 C) beyond which they modeled milk loss ◦ using a linear 1.15 kg/ C sensitivity. These linear relationships are also plotted for comparison. The linear ◦ relationship with the 32 C threshold clearly does not ◦ match the observations. In fact, 32 C is the coolest weekly average temperature observed by Ahmed and El Amin (1997), so these data cannot be used to infer the production level of unstressed cows without significant extrapolation to lower temperatures. Comparing the loss estimates of the THI method (Equation 2) to those of the linear Ahmed and El Amin (1997) relationship, it is clear that the latter is only consistent with the driest climates of the United States. Indeed, in Sudan the weekly average RH observed by Ahmed and El Amin ranged from 11 percent to 18 percent. Because humidity is an important determinant of heat stress and varies significantly across the conterminous United States, a model based solely on temperature is not sufficient to capture observed variations in U.S. dairy production. Because Equation 2 is nonlinearly dependent on daily temperature and humidity, it follows that there should be a strong seasonal cycle in production loss. Figure 3 shows the mean annual cycle of lost milk production for ten U.S. counties (listed in Table 1), selected to represent a cross section of the primary milk-producing regions of the country. For each location, Figure 3 shows the mean daily loss for each day of the year. For clarity, the data were smoothed using a seven-point moving average. Each plot shows the daily milk production loss due to heat stress predicted from Equation 2, for both the historical period and the 2050s. For the historical period only, each plot also includes the linear loss estimates used by Hayhoe et al. (2004; ◦ ◦ ◦ loss of 1.15 kg/ C for 25 C and 32 C thresholds; see Figure 2). Figure 3 highlights a number of interesting implications regarding climate impacts on milk production. First, there is a striking contrast between production losses in different regions, ranging from Tillamook, Oregon, where losses are almost negligible, to Maricopa, Arizona, where even at present the mean daily losses in summer approach 50 percent of total production. Large contrasts are seen across all ten locations shown, some stemming from seemingly small differences in climate. Second, whereas production losses under historic conditions are mainly confined to summer, climate change yields losses from spring to au- tumn in many locations. Third, the difference between loss computed from Equation 2 and the linear temperature relationship from Hayhoe et al. (2004) is strongly dependent on the climate of each location. In the dry climate of Maricopa, Arizona, for instance, the two are in fairly close agreement. In contrast, in the humid climate of western New York (Wyoming County), Equation 2 predicts substantial losses, whereas the temperature-based method does not predict any loss. Finally, and perhaps most significantly, those regions that are currently experiencing the greatest losses are also the most susceptible to additional losses: They are projected to be affected the most by climate change. The preceding differences stem not only from the nonlinearity of the response curve to temperature, all else being equal, but also from the sensitivity to humidity and the diurnal range in THI. Maps of the mean summer (June–August) climate are shown in Figure 4. Specifically, the two maps show average maximum temperature and afternoon humidity for the period from 1950 through 1999. Note that we have chosen to focus this discussion on afternoon conditions in summer, following Figure 3 and the previous discussion. These maps are shown to provide context to subsequent figures showing the nationwide pattern of production losses. The pattern in maximum temperature is primarily modulated by latitude, elevation, and proximity to coasts, with a notable maximum in the U.S. Southwest. Not surprisingly, the spatial pattern in humidity is different, determined by the influence of coastal air (Pacific, Atlantic, Gulf of Mexico), as moderated by topography (e.g., rain shadows) and regional variations in weather patterns. Note that the influence of moist Pacific air is largely confined by coastal mountains, whereas humid air from the Gulf of Mexico extends far into the continent. The complex patterns of humidity and temperature are reflected in the estimates of lost milk production per cow for the historical period (Figure 5). We see, for example, the dominance of high temperatures in Arizona as well as the increased sensitivities resulting from the high humidity found in the Southeast. Also apparent is that many of the regions that are currently strong in dairy production (see Figure 1) are located in areas with relatively mild summer climates, where the impacts of heat stress are small relative to nearby regions. This is not universally true, however, as exem- plified by the milk-producing regions in central California, Arizona, and Florida, where the impacts of heat stress can be quite severe. The top panel of Figure 6 shows the projected changes in summertime temperature for the 2050s relative to the historical period. Note that the projected temperature changes vary smoothly across the domain, a result of the low resolution of the global models, with substantially higher warming throughout the interior than along the coasts. As described earlier, these changes in temperature were applied to the higher resolution historical record to estimate future losses in milk production. The bottom panel of Figure 6 shows the difference in mean daily milk loss per cow for the 2050s relative to historical data. The pattern of changes in lost production is quite different from the pattern of changes in temperature, a result of the nonlinearity of the dependence on temperature and humidity. For example, despite relatively slight warming over Florida (Figure 6A), substantial production losses are projected (Figure 6B) due to the high temperature sensitivity of this humid region (Figure 2). Although not shown, results for the 2080s are qualitatively similar to those shown for the 2050s and are summarized in Tables 1 and 2 and in the supple- mentary material. The data show that there are few parts of the country that are both amenable to dairy farming (e.g., adequate supply of water, topographically suitable) and unaf- fected by climate change. Furthermore, future changes in climate are projected to accentuate regional disparities, with the areas that are already experiencing greater losses per cow—Arizona, the San Joaquin Val- ley in California, much of the Southeast, and the south- ern Midwest states—facing the most severe additional losses. Table 1 shows our estimates of milk production losses for ten selected counties; similar data for 2,801 counties in the conterminous United States are included in the online supplement, plus totals for the conterminous United States as a whole and for each of the lower forty-eight states. 5 A weighted average for the entire United States (shown in Table 2, along with results for selected states) yields a per cow loss of 0.57 kg/day for the historical period, 1.4 kg/day for the 2050s, and 1.9 kg/day for the 2080s. Using 30 kg/day as a baseline for production without any heat stress, loss per cow rises from 1.9 percent of the baseline for the historical period to 4.7 percent for the 2050s and 6.3 percent for the 2080s. This corresponds to economic losses for the country as a whole of $ 670 million per year in the historical period, $ 1.7 billion per year in the 2050s, and $ 2.2 billion per year in the 2080s. St- Pierre, Cobanov, and Schnitkey (2003) estimated that optimal heat abatement can reduce losses by 29 percent. Applying this reduction to our figures yields ball- park estimates of losses with optimal heat abatement of $ 470 million per year for the historical period, rising to $ 1.2 billion per year for the 2050s and $ 1.6 billion for the 2080s. When interpreting these numbers, it is important to note that they are based on the annual average in milk production loss, whereas the vast majority of the impacts are felt during the summer season (see Figure 2). As a consequence, daily production losses in June through August are approximately triple those quoted earlier (i.e., average historical production losses for June–August are ...