1 Introduction

Global average temperature has risen by roughly 0.13 °C per decade since 1950 (IPCC 2007). An even faster pace of roughly 0.2 °C per decade of global warming is expected over the next two to three decades, with substantially larger trends likely for cultivated land areas (IPCC 2007 ). In China, the temperature has increased by 1.2 °C since 1960, and winter warming (0.04 °C per year) is about four times the rate of summer warming (0.01 °C per year) (Piao et al. 2010). Although no significant long-term trend is observed in country-average precipitation since 1960, there are significant regional precipitation trends. The drier regions of North China and Northeast China are receiving less and less precipitation in summer and autumn (a 12 % decline since 1960). By contrast, the wetter region of southern China is experiencing more rainfall during both summer and winter.

The impact caused by the climate change on agriculture is not well understood so far (Lobell et al. 2011). Our understanding on the impacts of climate change on agricultural production in future, as well as the potential adaption options, can be further improved by shedding insights on the historical experiences in the past few decades (Olesen et al. 2011; Waha et al. 2013; Gourdji et al. 2013). Along this line, there are increasing efforts worldwide to investigate the observed climate change and its impacts to date at various spatial scales ranging from site to globe (e.g., Nicholls 1997; Lobell and Asner 2003; Peng et al. 2004; Tao et al. 2006, 2008; Sheehy et al. 2006; Lobell and Field 2007; You et al. 2009; Zhang et al. 2010; Li et al. 2010; Kristensen et al. 2011; Lobell et al. 2011). These studies generally suggest that improved datasets on climate and crop production, as well as novel analysis approaches, are crucial to accelerate understanding of climate impact on crop yields. Also climate–yield relationships are scale-dependent, and the results on county scale could be different from those on national or provincial scale.

For China, most of previous investigations are at provincial scale based on the agricultural census data at provincial level (e.g., Tao et al. 2008; You et al. 2009; Li et al. 2010). For example, Tao et al. (2008) investigated the climate–crop relationships, recent trends in seasonal climate, and their impacts on yields of major crops throughout China over the last few decades, based on the census data at provincial scale and the monthly dataset from the Climatic Research Unit (CRU, University of East Anglia, UK) (Mitchell and Jones 2005). Recently, some improved datasets on climate and agricultural production become available, including the agricultural census data at county scales, the detailed crop phenology records at hundreds of stations across China, and particularly an updated homogenized temperature dataset during 1960–2008 for mainland China (Li and Yan 2009). In the updated homogenized temperature dataset, all major breakpoints in temperature series caused by non-natural changes were detected and corrected, which is more reliable than the original in estimating local climate trends (Li and Yan 2009). Here, we used the improved datasets and analysis methods to investigate the climate–crop relationships, recent trends in seasonal climate, and their impacts on the yields of major crops (i.e., rice, wheat, maize, and soybean) by county throughout China during the period of 1980–2008. Our purposes are (1) to improve understanding the mechanisms, extent, and degree of climate impacts on major crop yields in China in the past few decades and (2) to identify the particular crops and regions that are vulnerable to climate change, as well as the potential adaption strategies.

2 Methods and data

2.1 Data

Time series of yields for rice, wheat, maize, and soybean by county from 1980 to 2008 were obtained from the Agricultural Yearbook of each province (autonomous region or municipality) in China (China Agriculture Press, Beijing 1980–2009 [annual]) and some unpublished data from county-level census bureau.

The phenology records of rice, wheat, maize, and soybean at hundreds of agro-meteorological experimental stations across China were obtained from China meteorological administration (CMA). The records include the dates of major phenological events such as planting, transplanting, flowering, and maturity since 1980. For each phenological event, mean date during the period of 1980–2008 was firstly computed at each station, and then, the mean date of the phenological event for each county was computed from the nearest station. Such detailed phenology records provide a unique chance to determine exactly crop growing period and therefore seasonal climate over the growing period at each county.

Daily weather data including mean temperature (Tmean) and precipitation (P) were from the 756 national standard stations (NSSs) of CMA. Inhomogeneity in the record data of daily temperature series is almost unavoidable due to various non-natural changes such as those in the observing location, the environs, and many other factors (Manton et al. 2001; Li and Yan 2009). In the present study, for daily Tmean, we used the China Homogenized Historical Temperature Dataset (CHHT, Li and Dong 2009), in which inhomogeneities in the daily Tmean series from 1960 to 2008 at 549 NSSs were analyzed and corrected using the Multiple Analysis of Series for Homogenization (MASH) software package (Li and Yan 2009). At county level, daily Tmean in a county was computed from its neatest three NSSs using inverse distance weighted interpolation method. Daily precipitation in a county adopted the observation from its neatest NSS.

2.2 Methods

Datasets on crop locations, crop phenology, crop yields, climate, etc. were combined in a panel analysis of four major crops (i.e., rice, wheat, maize, and soybean) for all the counties in the cultivation areas. The trends in climate variables and overall yield were investigated by linear regression and Kendall’s tau statistic for 1980–2008 as the following:

$$ {Z}_{i,t}={b}_i+{a}_i\ast \mathrm{year} $$
(1)

where Z i,t is the climate variables or reported yield at county i in year t, b i is a county fixed effect, a i is the county-specific linear time trend.

To investigate climate–yield relationship by county, we computed the first-difference time series (i.e., the difference in values from 1 year to the next) for yield (ΔYield) and climate during crop growing period (i.e., from planting to maturity) (Lobell and Field 2007); the first-difference time series for mean temperature (ΔTmean) and precipitation (ΔP) was computed respectively. This approach avoids the confounding influence of long-term variations such as changes in crop management (Lobell and Field 2007; Tao et al. 2008). The relationships between ΔYield and ΔTmean and ΔP were evaluated using Pearson correlation analyses, respectively.

To quantify the impact of climate trends in the past few decades on yield, the methods developed by Lobell et al. (2011) were applied here. The model is of the form

$$ \mathrm{Log}\left({Y}_{i,t}\right)={c}_i+{d}_{1i}\ast \mathrm{year}+{d}_{2i}\ast {\mathrm{year}}^2+\beta {X}_{i,t}+{\varepsilon}_{i,t} $$
(2)

where Y i,t is the reported yield at county i in year t, c i is a county fixed effect, d 1i is the county-specific linear time trend, d 2i is the county-specific quadratic time trend, β is a vector of coefficients and X is a vector of variables (Tmean, Tmean2, P, P 2), and ε i,t is an error term. The statistical model uses quadratic terms for temperature and precipitation to account for the fact that crops perform best at moderate temperature and precipitation and are harmed by extreme weather (Lobell et al. 2011).

The impact of climate trends was calculated in a four-step process. First, let Tmean*i,t = predicted temperature for county i at year t, based on linear fit for 1980–2008

$$ T{\mathrm{mean}\mathrm{d}}_{i,t}=\mathrm{detrended}\ \mathrm{temperature}=T{\mathrm{mean}}_{i,t}-T\mathrm{mean}{*}_{i,t}+T\mathrm{mean}{*}_{i,1980} $$
(3)

With detrended precipitation similarly defined, we then use the regression model F(Tmean, P) (i.e., the regression model in Eq. (2)) to compute

  1. (i)

    F(Tmean, P) = predicted yields with observed weather

  2. (ii)

    F(Tmeand, P) = predicted yields with detrended temperature and observed precipitation

  3. (iii)

    F(Tmean, Pd) = predicted yields with observed temperature and detrended precipitation

  4. (iv)

    F(Tmeand, Pd) = predicted yields with detrended temperature and detrended precipitation

We compute the trends of the differences (i)–(ii), (i)–(iii), and (i)–(iv) to quantify the yield effect of trends in Tmean, P, and both Tmean and P, respectively. Estimates of the fifth to 95th percentile confidence interval were obtained in all cases by bootstrap resampling, where the historical data were resampled and a new regression model was fit to the data. In each case, 1000 bootstrap samples were used. Median estimates of the impacts of temperature and precipitation trends were used for further analyses.

3 Results

3.1 Climate trends over crop growing period

We presented the spatial explicitly patterns of significant climate trend (p < 0.05) during crop growing period over the period of 1980–2008 (Fig. 1). During rice growing period, Tmean increased generally across the cultivation areas (Fig. 1a). Based on the highest value of median estimates in each province, particularly, it increased by more than 0.60 °C/decade at some portions of northeastern China (0.59 °C/decade in IM) and the middle and low reaches of Yangtze River (0.66 °C/decade in SH) (see supplemental materials Fig. S1 for provinces or the river mentioned). By contrast, the number of counties with significant precipitation trends was relatively small. Precipitation decreased sporadically in some major rice production areas including northeastern China, some portions of the middle and low reaches of Yangtze River, and southwestern China (Fig. 1b). During wheat growing period, Tmean increased more dramatically, particularly in northern (0.65 °C/decade in IM) and eastern China (0.71, 0.73 °C/decade in JS and SH) (Fig. 2c); precipitation increased in the southwestern China (28.49 mm/decade in SC) (Fig. 1d). During maize growing period, Tmean also increased generally, with the most increase located in the middle and low reaches of Yangtze River (0.76, 0.78 °C/decade in HN and HB) (Fig. 1e); precipitation decreased particularly in some portions of northwestern China (−78.36 mm/decade in SN) (Fig. 1f). During soybean growing period, Tmean increased generally across China, (0.48 and 0.55 °C/decade in IM and ZJ) (Fig. 1g); precipitation decreased at a large area (−111.74 mm/decade in SC), however increased in some portions of eastern China (23.10 mm/decade in AH) (Fig. 1h).

Fig. 1
figure 1figure 1

Trends in mean temperature (a, c, e, g) and total precipitation (b, d, f, h) over growing period of rice (a, b), wheat (c, d), maize (e, f), and soybean (g, h) by county, respectively, during the period of 1980–2008

Full size image
Fig. 2
figure 2figure 2

Correlations between mean temperature (a, c, e, g); total precipitation (b, d, f, h); and yields of rice (a, b), wheat (c, d), maize (e, f), and soybean (g, h) by county, respectively, during the period of 1980–2008

Full size image

3.2 Climate–crop yield relationships by county

Notable correlations (p < 0.1) between climate anomalies and crop yields could be observed directly at some regions (Fig. 2). Rice yields were generally favored by cooler and wetter years in southwestern China, in contrast favored by hotter and drier years in southeastern and northeastern China (Fig. 2a, b). Wheat yields were generally favored by cooler and wetter years in northwestern, northeastern, and southwestern China, in contrast favored by drier years in southeastern China (Fig. 2c, d). Maize and soybean yields were favored by cooler and wetter years in northern and southwestern China, but favored by drier years in southeastern China (Fig. 2e–h).

3.3 Sensitivity of crop yields to climate trends

We identified the areas sensitive to climate change for crop production by investigating crop yield change for each 1 °C increase in Tmean and for 10 % decrease in precipitation during crop growing period, respectively. Based on the highest value of median estimates in each province, for each 1 °C increase in Tmean during crop growing period, rice yield could decline by 0–5 % or more in southwestern China (−3.27 % in YN) and increase by 0–5 % or more in southern (3.78 % in HN) and northeastern China (2.90 % in HLJ) (Fig. 3a). Wheat and particularly maize yield could be more negatively sensitive and could decline at quite a large area (Fig. 3c, e). Wheat yield in northeastern China (−3.56 % in HLJ) and maize yield in northern and northeastern China (−2.33 % in LN) were especially sensitive to climate warming, which could decline by up to 5–10 % or more. Soybean yield could decline by up to 5–10 % or more in portions of northern and northwestern China (−3.49 % in SN); in contrast, it could increase by 5–10 % or more in portions of central and southwestern China (0.91 % in YN) (Fig. 3g).

Fig. 3
figure 3figure 3

Yields change for 1 °C increase in mean temperature (a, c, e, g) or 10 % decrease in total precipitation (b, d, f, h) over growing period for rice (a, b), wheat (c, d), maize (e, f), and soybean (g, h) by county, respectively. Values are expressed as percent of average yield

Full size image

Similarly, with precipitation decreasing by 10 %, rice (0.43 % in ZJ), wheat (0.62 % in AH), and maize production (0.39 % in ZJ) could be generally benefited in southeastern China (Fig. 3b, d, f). However, wheat, maize, and soybean yield could decline by up to 1.0–2.0 % or more particularly in northern and northeastern China, especially in IM (Fig. 3d, f, h).

3.4 Crop yields change due to climate trend over the period of 1980–2008

During the period of 1980–2008, crop yield changes due to Tmean trend were generally from −2.5 to 2.5 % (Fig. 4). Based on the highest value of median estimates in each province, Tmean trend increased rice yields by up to 5.0–6.3 % in northeastern China (2.12 % in HLJ) (Fig. 4a), however reduced rice yields by 2.5–5.0 % or more in northern and northeastern China (−0.52 % in IM). Tmean trend reduced yields of wheat, maize, and soybean by up to 2.5 % at quite large areas (Fig. 4c, e, g). Particularly, Tmean trends reduced wheat yields by 2.5–5.0 % or more in portions of northern and northeastern China (−2.03 % in IM), reduced maize yields by 2.5–5.0 % or more in portions of northwestern and northeastern China (−1.50 % in SN) (Fig. 4e), and reduced soybean yields by 2.5–5.0 % or more in portions of northern China (−0.44 % in SX) (Fig. 4g).

Fig. 4
figure 4figure 4

Estimated net impact of temperature (a, c, e, g) and precipitation (b, d, f, h) trends on yields of rice (a, b), wheat (c, d), maize (e, f) and soybean (g, h) by county, respectively, during the period of 1980–2008. Values are expressed as percent of average yield

Full size image

In comparison with trend in Tmean, trend in precipitation had smaller impacts on crop yields, generally from −1.0 to 1.0 % (Fig. 4). Particularly, precipitation trends reduced maize and soybean (−0.37 and −1.02 % in IM, respectively) yields by 1.0–2.0 % or more in portions of northern and northeastern China (Fig. 4f, h).

3.5 Relative importance of climate relative to all other factors

We divided the climate-induced yield trend by the overall yield trend for 1980–2008 in each county. The ratio provides a useful measure of the relative importance of climate relative to all other factors including technology and CO2 fertilization (Lobell et al. 2011), with values of –0.1 indicating that 10 years of climate trend is equivalent to a setback of roughly 1 year of technology gains. For rice, cases where negative climate impacts represented a large fraction of overall yield gains were in the North China Plain and portions of southwestern China (Fig. 5a). In contrast, in northeastern China, positive climate impacts represented a large fraction of overall yield gains. For wheat, maize, and soybean, cases where negative climate impacts represent a large fraction of overall yield gains were more extensive across the cultivation regions, although cases where positive climate impacts represented a large fraction of overall yield gains were also found in many regions (Fig. 5b–d). The impact of climate trends on crop yield was large enough to offset a notable portion of the increases in average yields that arose from technology and other factors in central China for wheat, in portions of northern and northeastern China for maize and soybean. Trends in temperature played a dominant role in comparison with trends in precipitation (Fig. S2).

Fig. 5
figure 5

Estimated net impact of climate trends for 1980–2008 on yields of rice (a), wheat (b), maize (c), and soybean (d) by county, divided by the overall yield trend per year for 1980–2008. Values represent the climate effect in the equivalent number of years of overall yield gains. For example, value of –0.1 indicates that 10 years of climate trend is equivalent to a setback of roughly 1 year of technology gains

Full size image

4 Discussion

In previous studies, Peng et al. (2004) suggested rice grain yield declined by 10 % for each 1 °C increase in growing-season Tmin, whereas the effect of Tmax on crop yield was insignificant, based on an analysis of yield and weather data for the years 1992–2003 at the research farm of International Rice Research Institute, Philippines. However, Sheehy et al. (2006) found there was a significant negative correlation between solar radiation and Tmin in the datasets used by Peng et al. (2004) and suggested that the effect of Tmin on rice yield was overestimated. Based on the analyses at provincial scales throughout China, Tao et al. (2008) suggested that the warming trend in the last few decades increased rice yield in northeastern China and soybean yield in northern and northeastern China; however, it decreased maize yield in seven provinces (autonomous region or municipality) and wheat yield in three provinces. Zhang et al. (2010) suggested that rice yields were positively correlated to solar radiation, which primarily drives yield variation. In this study, asymmetric changes in Tmax and Tmin, as well as the correlations between crop yields and Tmin, Tmax, and solar radiation were also investigated (Fig. S3). We found that the correlation patterns between crop yields and Tmax, Tmean, and solar radiation were quite similar. Notable negative correlations between crop yields and Tmax, Tmean, and solar radiation were observed in southwestern China for rice, in northeastern and northwestern China for wheat, in northern and southwestern China for maize, and in northern and northeastern China for soybean. The regions where crop yields were negatively correlated with Tmax or Tmean were usually the regions where crop yields were positively correlated with precipitation, particularly for wheat, maize, and soybean, suggesting temperature should indirectly affect crop yield through increasing evapotranspiration and droughts, which is consistent with our previous results (Tao et al. 2003). We showed that crop production in arid or semi-arid areas in mid- to high-latitude regions was more vulnerable to climate warming due to warming-induced droughts in the past few decades, although generally, it was believed that moderate warming benefits crop yields in mid- to high-latitude regions (IPCC 2007). In portions of southeastern China, where continuous rain and low solar radiation are the primary limiting factors for crop productivity, crop yields were negatively correlated with precipitation and Tmin and positively correlated with Tmax and solar radiation (Fig. S3). The relative roles of Tmin and Tmax in affecting crop yields were dependent on crops and regions.

We found each 1 °C increase in Tmean can reduce crop yield by up to 5–10 % or more at some regions, and trends in Tmean during the period of 1980–2008 reduced crop yields by up to 2.5–5.0 % or more at some regions, particularly for maize and wheat. As a whole, based on the crop planting areas and total production at county level in 2008, planting area-weighted average of yield change in the whole country due to trends in Tmean was 0.76, −0.27, −0.37, and 0.24 %, respectively, for rice, wheat, maize, and soybean (Fig. 6). The corresponding value due to trends in Tmean and precipitation together was 1.16, −0.31, −0.40, and 0.11 %, respectively, which increased rice production by about 4.6 × 108 t and soybean production by 3.9 × 106 t, however reduced maize production by about 1.5 × 108 t and wheat production by about 2.5 × 107 t (Fig. 6). The results are much less than previous estimates at provincial scale (e.g., Tao et al. 2008; You et al. 2009) and national scale (Lobell et al. 2011) that used different datasets on climate and crop yields and analysis methods.

Fig. 6
figure 6

Estimated net impact of climate trends for 1980–2008 on crop yields and production in China. Values are planting area-weighted average across China

Full size image

The current study represents several important improvements in terms of datasets on climate, crop phenology, and crop yield, as well as the statistical models. The climate-induced changes in cropland area and cropping system (Dong et al. 2009; Olesen et al. 2011; Waha et al. 2013) were however not taken into accounted here. In addition, the study did not fully incorporate long-term adaptation possibilities, nor did the impacts of extreme temperature or precipitation events. The direct effects of elevated atmospheric CO2 concentration on crop yield were also not accounted, which could increase C3 crop (i.e., rice, wheat, soybean) yield approximately by 3 % when CO2 concentration increased from 339 ppm in 1980 to 386 ppm in 2008 (Lobell et al. 2011).

5 Conclusion

We used the improved datasets on climate, crop phenology, and yield, as well as statistical methods, to investigate climate trends and the implications for crop production during the period of 1981–2008. The temporal and spatial patterns of climate trends and the impacts on major crop yields were presented. We found that past climate trends had caused measureable impacts on major crop production, which can offset a notable portion of the increase in average yields that arose from technology and other factors. Since climate warming rate is projected to be more progressive in future, adaptation efforts have to be taken. The particular crops and regions that have been most affected and should be the priorities to adapt are maize and wheat in the arid and semi-arid areas of northern and northeastern China. Climate warming-induced droughts were one of major impact mechanisms. Among other things, advanced water-saving technologies in irrigated agriculture as well as soil and water conservation technologies in rainfed agriculture should be largely popularized. The study focused on climate trend and did not take into account the impacts of extreme temperature or precipitation events. Due to the increasing climate extreme events in recent decades under ongoing climate change, the risk of crop production to climate variability or extreme weather events should be stressed in further studies.