Cropping frequency and area response to climate variability can exceed yield response

The sensitivity of agricultural output to climate change has often been estimated by modelling crop yields under climate change scenarios or with statistical analysis of the impacts of year-to-year climatic variability on crop yields1,2. However, the areaof croplandandthenumberof cropsharvestedpergrowing season (cropping frequency) both also a ect agricultural output and both also show sensitivity to climate variability and change3–9. We model the change in agricultural output associated with the response of crop yield, crop frequency and crop area to year-to-year climate variability in Mato Grosso (MT), Brazil, a key agricultural region. Roughly 70% of the change in agricultural output caused by climate was determined by changes in frequency and/or changes in area. Hot andwet conditions were associatedwith the largest losses andcoolanddryconditionswith the largestgains.All frequency and area e ects had the same sign as total e ects, but this was not always the case for yield e ects. A focus on yields alone may therefore bias assessments of the vulnerability of agriculture to climate change. E orts to reduceclimate impacts to agriculture should seek to limit production losses not only from crop yield, but also from changes in cropland area and cropping frequency. Yearly agricultural crop production in a given region is equal to the sum over each of the region’s harvested crops of that crop’s yield multiplied by its area10. Although in theory each component of this production equation—crop yield, cropping area and crop frequency—could be sensitive to climate change and/or climate variability, this is the first empirical study, to our knowledge, to estimate the response of agricultural output to climate shocks as a function of each of these three components. Research on impacts of climate on agriculture has focused on the impacts of climate change, decadal climate variability, and interannual climate variability on crop yields11,12. Hotter (and sometimes drier) conditions can cause abrupt and/or persistent declines in agricultural yields2; wet conditions can also reduce yields when they interrupt sowing, harvesting, or both13. Interannual climate variability is associated with a substantial share of yield variation, but the most damaging temperature and precipitation anomalies vary greatly across crops and regions14,15. Crop sensitivity to interannual climate variability depends strongly on the portion of the growing season during which the anomaly occurs. Anomalies occurring at different stages of crop development have varied impacts on agricultural output16. Research on the response of cropland area to climate has estimated the association of cross-sectional climatic variation with cropping area17 or has identified the relationship between cropping area under region-wide climate shocks, such as El Niño, as compared with more typical production years7,8. The subset of this research that investigates the impacts from spatio-temporally variable climate shocks generally uses area data that are aggregate measures. These do not allow for the disentangling of year-to-year fluctuations in cropland utilization versus more persistent changes in agricultural land use associated with agricultural expansion or abandonment. One recent study proposed a useful metric for assessing yield potential—the utilization fraction18. However, this conflates two sources of utilization change that are likely to have different drivers and should ideally be studied separately— ephemeral changes versus persistent changes. Thanks to a spatially explicit, sub-annual, satellite remote-sensing-derived agricultural land-use data set, we can explicitly distinguish between frequency of cropland utilization (cropping frequency) and changes in cropland area that persist for two or more years (cropland area). In production systems with more than one crop per growing season, one recent study showed that the first and second crops exhibit some common and some differential responses to interannual climate anomalies4. Another study used cropmodelling to show that second-crop sensitivity to climate can be substantial relative to first-crop effects and can partially ‘offset’ modelled losses from first crops under warming5. However, neither study allows for analysis of the relative importance for agricultural output of changes in cropping yield versus cropping frequency versus cropland area associated with a given set of climate anomalies. We focused our analysis on an emerging tropical agricultural production centre, the Brazilian State of Mato Grosso (MT). MT is a 90-million-hectare state where agricultural output increased threefold from 2000 to 201019. In 2013, agriculture comprised roughly 40% of the state’s land cover and 72% of the state’s GDP (gross domestic product). In 2013, MT produced 10% of global soybeans on 10 million hectares of cropland. Ranging from 7.23–17.87 S to 50.57–61.52 W and with mean annual precipitation ranging from 1,000mm in the southeast to over 2,500mm in the northwest, in typical years, much of MT is suitable for the production of two rainfed crops per growing season (Supplementary Fig. 5). In 2010, roughly half of MT’s cropland produced two commercial crops per growing season, usually a soy harvest followed by a corn harvest20. In net, the area of cropland and the frequency of cropping grew steadily from 2000–2010 but thismasks substantial instability—over the period 3million hectares of agricultural abandonment occurred and 3 million additional hectares of double-cropping abandonment occurred20.