全球变化知识资源中心

a, Trends in total cropland area for 1910–2014. b, Weather stations, and their corresponding trends in JJA 95th percentile daily maximum temperatures (Tx95), are grouped according to trends in total cropland area. c–f, Similar to a,b, but for irrigated area (c,d), and area-normalized crop net primary productivity (NPPan) (e,f) calculated from USDA survey data on areas and yields of 12 major summer crop types. Median temperature trends for each subset of data are indicated by horizontal red lines. Dashed vertical whiskers show the range of temperature trends, with any values exceeding 1.5× the interquartile range indicated by cross marks. Asterisks indicate that mean cooling across stations is significant for a given subset at P < 0.05 (single) or P < 0.01 (double) using a double-sided test. Trends in 95th percentile maximum temperatures are calculated as in Fig. 2, excluding the Dust Bowl and the period of maximum aerosol-induced cooling. Boxes are area-weighted (Supplementary Information), and bins overlapping with zero have been collapsed to 1/5 of the original width. The lightest grey counties for c and e indicate insufficient data.

Observational and modelling studies have demonstrated the ability of irrigation to cool surface temperatures through greater soil moisture and evapotranspiration^{12, 13, 14, 15, 16}. Our analysis shows a significant cooling effect associated with increases in irrigated area calculated from agricultural census data (Fig. 3c, d). Weather stations in counties with increases in irrigated area of greater than 10% of county area per decade show significant cooling. Where irrigation increases have been largest, such as in eastern Nebraska with trends greater than 7% of county area per decade, 95th percentile temperatures have cooled at a rate of 0.30 °C per decade (P < 0.01). Significant cooling associated with increased irrigation is, however, generally found only in Nebraska, Arkansas and the western US; and amounts only to around 11% of the 134 million hectares cooling at rates of at least 0.2 °C per decade (area calculations are performed using Voronoi polygons associated with each weather station). Comparison of local cooling to local irrigation trends seems appropriate, as high-resolution model results indicate that the cooling effects associated with irrigation are localized¹³.

Another major change over the past century has been the marked intensification of crop management and productivity. Several lines of evidence suggest that changes in management practices, cultivar properties and crop choice associated with more intensive land use would lead to elevated evapotranspiration rates, even for rainfed croplands. Widespread increases in fertilization have largely alleviated nitrogen stress, which can otherwise reduce photosynthetic rates, stomatal conductance, leaf area index and root development^{32, 33}, resulting in decreased magnitude^{32, 34} and duration³⁴ of peak evapotranspiration in the field. Evapotranspiration can also be affected by the frequency of fallow¹⁷, planting density³⁵ and shifts in crop types³⁶. In particular, there has been a transition towards more maize and soybean acreage at the expense of hay and shorter-season³⁷ oats (Supplementary Fig. 2). Increased adoption of no-till systems can prevent early-season soil evaporation and conserve water for transpiration³⁸. Evidence from the crop breeding literature also suggests that rooting and transpiration characteristics of cultivars have changed over time in ways that allow for greater evapotranspiration potential (Supplementary Information).

The inference of trends towards greater potential for evapotranspiration is supported by observations of increased specific and relative humidity across the Midwest during summer^{39, 40}, a positive evapotranspiration trend in the region inferred from a Mississippi basin water balance⁴¹, and the presence of a smaller diurnal temperature range over US croplands as compared with forested landscapes⁴². The cooling influence of elevated evapotranspiration on temperatures is expected to be most pronounced for high-temperature days when evaporative demand is the greatest⁴³, consistent with observed temperature trends.

To quantitatively evaluate whether changes in agricultural intensity correspond with the observed spatial pattern and magnitude of cooling across the US, we calculate an index of agricultural intensity since 1910 using county-level USDA (US Department of Agriculture) survey records of crop harvested area and yield. We use 12 major summer crop types and calculate crop carbon fixation by county using conversion factors relating crop yield to whole-plant carbon content. Adjusting for county area provides estimates of area-normalized net primary productivity (NPPan) in grams of carbon fixed per year and per unit area (g C m⁻² yr⁻¹). This metric accounts for land use influences on evapotranspiration through integrating the effects of crop type, harvested area, and productivity. Examples of county-specific NPPan data are shown in Supplementary Fig. 3.

Changes in agricultural intensity closely correspond to the pattern of Midwest cooling (Fig. 3e, f and Supplementary Tables 1 and 2). Counties with local trends in NPPan greater than 4.5 g C m⁻² yr⁻¹ show a highly significant cooling in 95th percentile temperatures of 0.22 °C per decade (P < 0.01), and counties with sequentially lower trends in NPPan show correspondingly smaller rates of cooling. These findings are robust to the exclusion of locations with >10% irrigated area, and support our hypothesis of increased evapotranspiration potential in high-productivity croplands cooling high temperatures. Increasing NPPan is associated with seven times more land area undergoing cooling of at least 0.2 °C per decade than irrigation. It is remarkable that cooling has occurred despite increased atmospheric carbon dioxide, which is associated with lower rates of transpiration, increased water use efficiency⁴⁴, and—along with other greenhouse gases—general surface warming.

Increased precipitation also favours greater evapotranspiration and may itself be influenced by evapotranspiration. Recent work in the Canadian Prairies has shown that increased crop evapotranspiration is correlated with greater specific humidity and precipitation¹⁷; such changes to the surface energy balance can favour deep convection and increase the frequency and severity of precipitation events^{17, 45}. Evapotranspiration of irrigated water would also lead to increased precipitation and moisture recycling, and it has previously been suggested that Midwest growing-season precipitation increases may partly be attributed to irrigation^{28, URL:}