This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

Transcription

1 This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier s archiving and manuscript policies are encouraged to visit:

2 Agricultural and Forest Meteorology 150 (2010) Contents lists available at ScienceDirect Agricultural and Forest Meteorology journal homepage: Responses of rice yields to recent climate change in China: An empirical assessment based on long-term observations at different spatial scales ( ) Tianyi Zhang a,b,, Jiang Zhu a, Reiner Wassmann b,c a LAPC, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, , China b International Rice Research Institute, DAPO Box 7777, Metro Manila, Philippines c Karlsruhe Research Center, Institute for Meteorology and Climate Research, Germany article info abstract Article history: Received 19 August 2009 Received in revised form 12 April 2010 Accepted 21 April 2010 Keywords: Climatic variability Radiation Drought Temperature Irrigation water availability This empirical study (i) assessed rice yield responses to recent climate change at experiment stations, in counties and in provinces of China for the period of and (ii) identified the climatic drivers determining the trend of yields at each spatial scale. Our empirical results, based on 20 experiment stations during study periods of years, indicate that rice yields were positively correlated to solar radiation, which primarily drives yield variation. At most stations, yields were positively correlated to temperature and there was no significant negative correlation between them. Therefore, our empirical results argue against the often-cited hypothesis of lower yields with higher temperature. We explain this by the positive correlation between temperature and radiation at our stations. Empirical analysis to yield at a regional scale (20 counties and 22 provinces) indicates a varying climate to yield relationships. In some places, yields were positively regressed with temperature when they were also positively regressed with radiation, showing the similar pattern at above experiment stations. But, in others, lower yield with higher temperature was accompanied by positive correlation between yield and rainfall, which was not happened at stations. We explain this by irrigation water availability, which played a crucial role in determining climatic effects (radiation or rainfall) on yield variability at a regional scale in China. However, temperature s negative effect is still weak at any scale. This study showed how rice yields respond to recent climate change from 1981 to 2005 at station and regional scales in China and identifies the major climatic driver for yield variation. The empirical findings presented here provide a foundation for anticipating climate change impacts on rice production in China. Crown Copyright 2010 Published by Elsevier B.V. All rights reserved. 1. Introduction There is a broad agreement that global food production is and will be affected by climate change in a significant manner (Parry et al., 2004; Schmidhuber and Tubiello, 2007). Often, the term response is used to describe the sensitivity of yield to a change in a given climatic parameter, such as temperature, in an attempt to quantify climatic effects on crop yields (Landau et al., 1998; Lobell and Asner, 2003; Peng et al., 2004; Sheehy et al., 2006a). The response can be derived from a specific yield change with respective change in climatic parameter, for instance, in tons per hectare per degree centigrade. Corresponding author at: LAPC, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, , China. Tel.: addresses: zhangty@mail.iap.ac.cn, ian zhangty@163.com (T. Zhang). An often-cited value of rice yield response to climate change was measured by Peng et al. (2004), who found rice yield correlation was more significant with minimum temperature than with other climatic parameters based on records from continuous field experiments during at the International Rice Research Institute (IRRI). Rice yields declined by 10% for a one degree centigrade increase in minimum temperature. In addition, other investigators have also empirically investigated climatic effects on rice alongside of other cereals (Lobell et al., 2005; Lobell and Ortiz- Monasterio, 2007; Sheehy et al., 2006b) using regional crop yield data on a subnational scale (Lobell et al., 2007; Tao et al., 2008), national scale (Lobell, 2007), and global scale (Lobell et al., 2008). In spite of these efforts, above studies did not reach a consensus and led to new arguments. For example, Lobell (2007) found that rising of maximum temperature is more harmful to rice yields than minimum temperature in most countries, which contradicts the major negative effect of minimum temperature inferred by Peng et al. (2004). More interesting, Sheehy et al. (2006b) reinvestigated /$ see front matter. Crown Copyright 2010 Published by Elsevier B.V. All rights reserved. doi: /j.agrformet

3 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) Table 1 Data sources and use in the study. Scale Data Source Use Experiment station Yield CMA archives (2009) To calculate rice yield response to climate change at experiment stations (Fig. 1) Phenology CMA archives (2009) To determine the time windows for calculating climate variables at stations Climate CMDSSS (2009) To annually calculate average climate variables based on phenology County Yield statistics CMA archives (2009) To calculate rice yield response to climate change at county scale Province Yield statistics IRRI database (2009) To calculate rice yield response to climate change at province scale Crop calendar Maclean et al. (2002) To determine the time windows for calculating climate variables at province scale Climate CMDSSS (2009) To annually calculate average climate variables based on crop calendar for each province (Fig. 2) Irrigation water quota DSIES (2009) To quantify irrigation water availability in each province the IRRI dataset Peng et al. (2004) used and argued against Peng s interpretation of the yield decline and pointed toward covariation of rising minimum temperature and reduced solar radiation; thus, a decrease in solar radiation could also explain the reduction in yield. These arguments may indicate that current understanding of the impacts of temperature on crop yields is still inadequate. The effects of climate change on rice production are particularly of concern in China since it provides staple food for Chinese people. The responses of Chinese rice yields to climate were routinely assessed by modeling studies (e.g., Matthews et al., 1995; Erda et al., 2005; Xiong et al., 2007; Yao et al., 2007), which usually show that increase in temperature have a significantly negative effect on rice yields. However, empirical study that can be used to test validity of the modeling project is still few in China and was not addressed in a systematic manner in the early works (Tao et al., 2006, 2008) since the large territory of China and diversity of growing conditions. This lack of empirical testing is a restriction for mechanistic models of rice yield responses to climate change in China. Therefore, in this study, we showed the results of how rice yields responded to recent interannual change in climate based on variety of spatial scales in China. Extensive rice yield and climatic data were collected at the scales of experiment stations, counties and provinces. The data cover most rice production area and are of representative Chinese rice production system at different levels. The objectives were: (i) To assess the responses of rice yields to climatic parameters at different spatial scales; (ii) To identify the major climatic drivers for yield variations. Table 1 summarizes the source and use for the data in the study. Crops at experiment stations were grown under optimum growing conditions, whereas crop yields in counties and provinces were obtained in the environment of local farmers fields. Data of experiment stations were obtained from 20 agricultural experiment stations operated by the Chinese Meteorological Administration (CMA archives, 2009), with continuous monitoring of rice phenology and yields. Optimum growing conditions were generally maintained for crop development at these experiments. The stations cover the main rice production area in China, with observation periods of years (Fig. 1 and Table 2). Eleven stations grow one-harvest rice whereas the other nine are doubleharvest rice stations. The crop data include rice phenological dates (transplanting and maturity) and field yields. Minimum temperature (T min ), maximum temperature (T max ), mean temperature (T mean ), sunshine hour, and rainfall (Rain) for the study period and stations were downloaded from the China Meteorological Data Sharing Service System (CMDSSS, 2009). Sunshine hour values were converted to radiation (Rad) values using the Ångström formula (Ångström, 1924). Yield statistics in the respective counties where the above experiment stations are located (Fig. 1) were also obtained from the CMA (CMA archives, 2009) in the same study period. The original source of these county-level yields is the local county bureau of statistics, which made investigation on rice production and rice sown area for each village and aggregated to county-average yield for each season. County-level yields are not the same with those at experiment stations since the yields were produced by farmers and represent the yields on farmer s fields in counties. Yield statistics in the 22 Chinese provinces highlighted in Fig. 2 were downloaded from the IRRI database (IRRI database, 2009), sources of which are the China Agricultural Yearbook published by National Bureau of Statistics of China. Being consistent with county level, provincial-level yields also represent yields obtained by farmers. The study period is from 1981 to For northeast, north, and northwest China (NEC, NC, and NWC), in the subhumid and semiarid region, one-harvest rice is mainly grown from April to September. The farmers in southern China (SC), the humid region, plant two-harvest rice. The main crop calendar is from February to June for the early rice season and from June to November for the 2. Materials and methods 2.1. Data at experiment stations, in counties, and in provinces Fig. 1. The locations of agricultural experiment stations and distribution of rice production in China. indicates single-harvest rice; indicates double-harvest rice. See Table 2 for details. Rice production is the average value during in the provinces (IRRI database, 2009).

4 1130 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) Table 2 The geographic backgrounds and mean growing-season climatic parameters (minimum and maximum values are shown) of agricultural experiment stations. ID a Station and county Long. ( E) Lat. ( N) Time period Season T min ( C) T mean ( C) T max ( C) Rad (MJ m 2 d 1 ) Rain (mm d 1 ) 1 Zunhua Single rice Gushi Single rice Xinyang Single rice Xuzhou Single rice Hefei Single rice Liuan Single rice Fangxian Single rice Zhongxiang Single rice Nanchong Single rice Leshan Single rice Yibin Single rice Lishui Early rice Late rice Longquan Early rice Late rice Changde Early rice Late rice Hengyang Early rice Late rice Meixian Early rice Late rice Heyuang Early rice Late rice Yangjiang Early rice Late rice Nanning Early rice Late rice Hechi Early rice Late rice a The ID number corresponds to those in Fig. 1. late rice season (Maclean et al., 2002). T min, T max, T mean, Rad (converted from sunshine hours), and Rain from 397 weather stations (Fig. 2) were downloaded from the CMDSSS (CMDSSS, 2009) Quantification of irrigation water availability in provinces Irrigation could also be a factor in determining the sensitivity of rice yields to climate change (Simelton et al., 2009; Zhang et al., 2008a). We used the data of irrigation water quota to quantify the spatial difference in irrigation water availability for each province. China s Ministry of Water Resources allocates for each village an annual water quota for agricultural irrigation, which is referred to as irrigation water quota (m 3 ha 1 year 1 ). This indicator is determined according to factors such as local irrigation area, major crops, soil types, water resource conditions, population, and local irrigation facilities. Irrigation water quotas were adjusted on a year-to-year basis for consideration of local agricultural policy and new availability of agricultural resources. The irrigation water quotas in our study were obtained from the Data Sharing Infrastructure of Earth System Science (DSIES, 2009). The database consisted of the indicator from China s Ministry of Water Resources for (except 1998, which has no records), and aggregated it to provincial data. We downloaded the provincial results and calculated the average value of irrigation water quota in each province to quantify irrigation water availability at a provincial scale. But the data in counties are unavailable in the database; thus, an analysis on a county scale was omitted Methods Fig. 2. Weather stations for calculating province-average climatic parameters in each year; the regions of NEC, NC, and NWC are single-harvest areas while the SC region is a double-harvest area. Responses of rice yields to climatic parameters were derived as the steps illustrated in Fig. 3. Step 1: For each station and year, we calculated the daily mean climatic parameters for the period from transplanting to maturity defined as growing season. Daily T min, T max, T mean, Rad, and Rain for each year and each season were then obtained. Climatic parameters in a county were assumed to be identical to those calculated for the respective experiment station in the respective year due to no data of crop duration data in those counties. For provincial-scale climatic parameters, the time windows for calculating mean growing-season daily climatic parameters are roughly set to 1 April to 30 September for one-harvest area (the NEC, NC, and NWC) and 1 February to 30 November (including both early and late rice) for two-harvest area (the SC) based on the crop calendar in China (Maclean et al., 2002). We calculated

5 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) Fig. 3. A schematic of the steps to estimate yield to climate correlations at stations (STA), in counties (COT), and in provinces (PRO). daily climatic parameters summed over growing season for each province and each year. Then, provincial-scale climatic parameters were derived by regional and seasonal averaging of climate data at those weather stations. Step 2: All time series are converted to values of first differences of yields and climatic parameters for de-trending (i.e., YLD STA, YLD COT, T min, T max, T mean, Rad, Rain, YLD PRO, T mean PRO, Rad PRO, Rain PRO). This step is for minimizing the possible confounding non-climatic effects (Nicholls, 1997; Lobell et al., 2008). The following equation was used in the conversion: X diff (t) = X (t) X (t 1) (1) where X diff(t) denotes the first difference of X at time t and X (t) denotes the value of the time series X at time t. The first year in each time series was omitted because of calculating the first-difference values in the step. Step 3: These first-difference values were used to regress in a linear model for each pair of climate and yield data to derive the responses of yield to T min, T max, T mean, Rad, and Rain for each station, county, and province. 3. Results 3.1. Temperature regimes, weather correlation, and yield responses to climate change at experiment stations Our experiment stations span a wide range of temperature regimes (Table 2). The range of C is for T min, C is for T max, and C is for T mean, respectively. The temperature ranges cover the temperate regimes at IRRI s experiment stations Peng et al. (2004) used ( C for T min, C for T max, and C for T mean ). The correlations of Rad with T min, T max, T mean, and Rain at each station/season are given in Table 3. Rad was positively regressed with temperature (including T min, T max, and T mean ) and the majority of the relationships reached a statistically significant level. These positive correlations are contrary to those at IRRI (Peng et al., 2004; Sheehy et al., 2006b). But, there is a negative correlation between Rad and Rain, with 20 stations/seasons showing a significant negative correlation, which is consistent with those at IRRI (Peng et al., 2004; Sheehy et al., 2006b). Correlation coefficients between yields and climatic parameters at each station are listed in Table 4. Rice yields were positively correlated to T min, T max, and T mean at the majority of stations. None of the stations had a significant negative correlation to temperature (T min, T max, and T mean ), but significant positive correlations were observed at four stations for T max and T mean. Above results suggest that increase in rice yields were accompanied by increase in temperature at most stations. Rad was found to be positively regressed with yields, five stations reaching a statistically significant level. Estimations of significant responses of yield to a change in Rain are characterized by a negative sign in the correlation coefficient, for which six significant correlations can be found Yield responses to climate change in counties vs. at stations Table 4 also presents the correlation coefficients between yield and climatic parameters in respective counties. There is a varying temperature to yield relationship in counties. Not only a significant positive yield to temperature correlation was observed but also in several counties the correlations were significant negative.

6 1132 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) Table 3 Correlation coefficients between Rad and T min, T mean, T max, and Rain (first-difference value) at experiment stations. ID a Station name Sample size (years) Season Rad vs. T min T mean T max Rain 1 Zunhua 13 Single rice * 0.66 * 2 Gushi 24 Single rice 0.64 ** 0.84 ** 0.92 ** 0.46 * 3 Xinyang 24 Single rice 0.55 ** 0.84 ** 0.94 ** 0.63 ** 4 Xuzhou 23 Single rice 0.70 ** 0.79 ** 0.83 ** 0.51 * 5 Hefei 24 Single rice 0.53 ** 0.69 ** 0.78 ** 0.63 ** 6 Liuan 17 Single rice 0.68 ** 0.87 ** 0.94 ** 0.77 ** 7 Fangxian 22 Single rice ** 0.87 ** 0.63 ** 8 Zhongxiang 24 Single rice 0.70 ** 0.85 ** 0.91 ** 0.48 * 9 Nanchang 24 Single rice ** 0.80 ** Leshan 18 Single rice * 0.68 ** 0.60 ** 11 Yibin 24 Single rice ** Lishui 22 Early rice ** 0.79 ** 0.69 ** 22 Late rice 0.54 ** 0.75 ** 0.86 ** Longquan 24 Early rice ** 0.83 ** 0.43 * 24 Late rice * 0.75 ** 0.68 ** 14 Changde 24 Early rice ** 0.75 ** 0.59 ** 24 Late rice 0.56 ** 0.69 ** 0.74 ** 0.42 * 15 Hengyang 24 Early rice * Late rice ** 0.78 ** 0.78 ** 16 Meixian 23 Early rice ** 0.85 ** 0.75 ** 23 Late rice ** 0.89 ** 0.50 ** 17 Heyuang 22 Early rice 0.62 ** 0.81 ** 0.88 ** 0.58 ** 22 Late rice * 0.71 ** Yangjiang 23 Early rice * 0.53 ** 0.41 * 23 Late rice * 0.63 ** 19 Nanning 24 Early rice 0.47 * 0.63 ** 0.71 ** 0.61 ** 24 Late rice ** 0.78 ** Hechi 18 Early rice * Late rice ** 0.77 ** 0.36 a The ID number corresponds to those in Fig. 1. * Significant at P < 0.05; ** Significant at P < Table 4 Correlation coefficients between yield and climatic parameters (first-difference value) at stations and in counties. ID a Name Sample size(years) Season YLD STA vs. YLD COT vs. T min T mean T max Rad Rain T min T mean T max Rad Rain 1 Zunhua 13 Single rice * Gushi 24 Single rice Xinyang 24 Single rice Xuzhou 23 Single rice Hefei 24 Single rice ** 6 Liuan 17 Single rice Fangxian 22 Single rice ** 0.59 ** 0.55 ** * 8 Zhongxiang 24 Single rice Nanchang 24 Single rice Leshan 18 Single rice Yibin 24 Single rice * 0.55 ** 0.51 * Lishui 22 Early rice * 0.57 ** 0.64 ** 0.46 * Late rice Longquan 24 Early rice * Late rice * 0.41 * Changde 24 Early rice * 0.62 ** 0.65 ** 0.70 ** * 0.56 ** 0.45 * 0.57 ** 24 Late rice Hengyang 24 Early rice Late rice Meixian 23 Early rice Late rice * Heyuang 22 Early rice * Late rice * 0.42 * * 0.46 * Yangjiang 23 Early rice * 0.55 ** 0.42 * 0.48 * 0.49 * ** 23 Late rice * 0.55 ** 0.57 ** 0.51 * 0.54 ** 19 Nanning 24 Early rice ** 24 Late rice Hechi 18 Early rice * Late rice a The ID number corresponds to those in Fig. 1. * Significant at P < 0.05; ** Significant at P < 0.01.

7 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) negative, yields were also negatively correlated to Rad (quadrant III on Fig. 4c) and positively correlated to Rain (quadrant II on Fig. 4d) Yield responses to climate change in provinces Fig. 4. Pattern of yield responses to climatic parameters at experiment stations (a, b) and in counties (c, d). We plotted the correlation coefficient between yield and climatic parameters at experiment stations and counties from Table 4 in one graph (Fig. 4). For the results at experiment stations, the majority of data points fell into quadrant I on Fig. 4a, reflecting positive correlations of yields to T mean and coinciding with positive correlations to Rad. Fig. 4b shows that data points fell into quadrant IV, which indicates that positive yield correlations to T mean were accompanied by negative correlations to Rain. However, in their respective counties, the distribution pattern was more complex. In addition to above features in Fig. 4a and b, it is also clear that, when the correlation between yields and T mean was Spatial distributions of yield responses to T min, T max, T mean, Rad, and Rain in provinces are given in Fig. 5, and the correlation coefficients in each province are listed in Table 5. There is a varying correlation between yields and temperature (T min, T max, and T mean ). Negative responses to temperature were concentrated in the provinces of NC and NWC (excluding the Hebei and Ningxia Province), while a positive response to temperature could be found in NEC and SC (excluding the Sichuan, Guizhou and Jiangxi Provinces). A broad similar spatial distribution was also exhibited for yield correlations to Rad. More importantly, there is a clear spatial distribution for yield correlation with Rain. Rice yield variability was positively correlated to change in Rain in NC and NWC, except for Ningxia, whereas, in other regions, a negative correlation could be observed. Correlations between climatic parameters in these provinces (Table 5) are similar to observations at experiment stations (Table 3). Fig. 6 displays pattern of climate to yield relationships using provincial-scale climate and yield data in Table 5. The size of the dots donates irrigation water quota. Similar results at county level (Fig. 4c and d), dots are predominantly fallen in quadrants I and III on Fig. 6a and in quadrants II and IV on Fig. 6b. In addition, another noteworthy feature is that the data points with positive correlation between T mean and yield (in quadrant I on Fig. 6a and quadrant IV on Fig. 6b) have a higher irrigation water quota; whereas the irrigation water quota is relatively low for the data points with negative correlation (in quadrant III on Fig. 6a and quadrant II on Fig. 6b). Fig. 5. Correlation coefficients of rice yield responses to T min (a), T mean (b), T max (c), Rad (d), and Rain (e) summed over crop duration in China. See Table 5 for correlation coefficients.

8 1134 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) Table 5 Correlation coefficients between yield and climatic parameters (first-difference value) and the correlation between climatic parameters in provinces. ID a Provinces Yield PRO vs. Rad PRO vs. T min PRO T mean PRO T max PRO Rad PRO Rain PRO T min PRO T mean PRO T max PRO Rain PRO NEC 1 Heilongjiang * 0.74 ** 2 Jilin * 0.70 ** 3 Liaoning ** 0.73 ** NC 4 Hebei * 0.67 ** 0.65 ** 5 Tianjin ** 0.54 ** 6 Shandong * ** 0.78 ** 0.62 ** 7 Henan ** 0.91 ** 0.61 ** NWC 8 Shanxi ** 0.81 ** 9 Shaanxi 0.49 * 0.54 ** 0.43 * ** 0.88 ** 0.68 ** 10 Ningxia * 0.45 * ** 0.48 * SC 11 Jiangsu ** 0.78 ** 0.78 ** 12 Anhui * 0.41 * ** 0.88 ** 0.84 ** 13 Hubei ** 0.86 ** 0.92 ** 0.62 ** 14 Sichuan Zhejiang * * 0.75 ** 16 Hunan ** 0.79 ** 0.42 * 17 Jiangxi * ** 0.51 ** 18 Guizhou * ** Fujian ** 0.63 ** 20 Yunnan 0.48 * 0.48 * 0.41 * ** 0.56 ** 21 Guangxi 0.40 * 0.40 * * Guangdong * 0.60 ** a The ID number corresponds to those in Fig. 5. * Significant at P < 0.05; ** Significant at P < Fig. 6. Pattern of yield responses to climatic parameters in provinces; the size of data points denotes irrigation water quota. 4. Discussion 4.1. Uncertainties Uncertainties of above results may come from several sources. Firstly, we cannot calculate county-average climatic parameters since there is only one weather station set by CMA for a given county, which may introduce some uncertainties in calculating yield response to climate at county scale. Secondly, at provincial scale, a lack of rice cropping area data within one province may also lead to some biases in calculating provincial climatic parameters. Thirdly, as mentioned in previous empirical studies (Lobell et al., 2007), de-trending technique may not completely remove non-climatic effect. It is possible that management changes from year by year, as a consequence of climate or agricultural policy, also affected interannual change in yield. But such effect is fairly difficult to remove without detailed management data of farmers in each year. The empirical study is based on the common assumption that interannual change in yields is mainly dependent on climate. Above uncertainties also exist in other empirical studies and inevitable unless there is a substantial improvement on data quality. The data used and data aggregation method we are based on in the study is similar to previous case studies focusing on China (Tao et al., 2006, 2008) Responses of rice yields to climate change at experiment stations According to the results of experiment stations (Table 4), there is a clear pattern that yields were positively correlated with Rad at most stations. The results show that Rad benefits rice yields, which can be supported by a number of experimental studies; an increase in Rad promotes photosynthesis, thus increasing yield (Evans, 1993; Yang et al., 2008). Moreover, yield responses to temperature were broadly positive, which means that yields were not limited by an increase in T min, T max,ort mean in spite of the similarity of temperature regimes in IRRI s experiments (Peng et al., 2004). Results for temperature to yield correlations at our stations are the opposite from IRRI s experiments, in which negative T min to yield correlations were observed (Sheehy et al., 2006b). Our hypothesis is that Rad is the major climatic driver for yield fluctuations at these Chinese experiment stations, and the positive yield correlation to temperature can be explained by the correlations between Rad and temperature, which were positive at most stations. Thus, we observed that positive effect of Rad overwhelmed temperature s effect on rice yield variation. Many previous empirical studies based on IRRI have attributed yield variation to Rad in experiments, being in favor of above conclusion. For example, Dobermann et al. (2000) reported that 54% of

9 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) change in rice yields was due to increase in Rad in a long-term continuous rice cropping system of IRRI. Yang et al. (2008) explained the yield gap between dry season and wet season by higher Rad in dry season. Evans and De Datta (1979) investigated rice yield trend for a 10-year period at IRRI, concluding that high Rad at any stage after PI would promote rice yield. Sheehy et al. (2006b) compared empirical and modeling evidence, suggesting that change in yield was more responsible for Rad variation at IRRI while temperature effect is actually lower than expected Responses of rice yields to climate change at a regional scale To the yield farmers produced at regional scale (county and province), it was observed that positive yield correlation with temperature was accompanied by positive correlation with Rad while yields were also observed negatively correlated to temperature in several places when there is a positive correlation between yield and Rain accompanied (Figs. 4c, d and 6). The former pattern is similar to results at experiment stations but the latter was inversed and did not occur in experiment stations. Our results are consistent with Tao et al. (2008), who also observed the varying correlation between rice yield and temperature depending on provinces in China and found it cannot be explained by current viewpoint of strong negative temperature effect. Here, we try to give an explanation for the varying yield correlation with temperature observed at region scale. At provincial scale (Fig. 5), there was a clear spatial distribution of yield responses to Rain; positive responses to Rain were shown in most area of NC and NWC while remaining regions had a negative correlation. This suggests that rice yields are more sensitive to drought in the NC and NWC compared with yields in other provinces of China. With the irrigation water quota dimension (Fig. 6), negative yield correlations with T mean tended to happen in the provinces with relative low irrigation water quota, which can be explained by the effects of drought. Water scarcity in NC and NWC is well known (Shu et al., 2001; Khan et al., 2009; Varis and Vakkilainen, 2001). The irrigation water shortage in the region is estimated to be 1.6 billion m 3 per year (Changming and Jingjie, 2001). Even though water requirement of rice in NC and NWC can be supplemented by irrigation (via river or groundwater), literatures have suggested drought indeed influenced Chinese rice production, especially in the north (Maclean et al., 2002; Simelton et al., 2009; Xiong et al., 1992). Under drought stresses, surface water and groundwater resources are becoming less, leading to lowering of groundwater tables, diminished river flows, and heavy pressures on irrigation supplies. This suggests drought will affect rice yield when its impacts exceed adaptive capacity of local irrigation. Compared with north, drought impacts were weakened in southern part of China because abundant water resources there (higher irrigation water quota and Rain) met the demands of crop water requirements in most year (Khan et al., 2009). This makes the field conditions of farmers in the region similar to those of experiment stations, yielding a similar yield response to climate that Rad drove yield variation (Fig. 6). Positive yield correlation with T mean can be seen a consequence of that Rad and T mean were positively correlated. Therefore, above results suggest that positive and negative yield correlation with temperature can be seen as a proxy of Rad or Rain effects via correlating with them, suggesting a relative weak temperature effect on change in rice yields. It is our view that irrigation water availability in different regions plays the role of a switch to determine which climatic drivers (Rad or Rain) are more dominant in yield variation in China. Recently, there are several studies emphasizing importance of considering human adaptation processes in climate to crop relationships (Challinor et al., 2009; Ewert et al., 2007; Simelton et al., 2009). In particular, Simelton et al. (2009) found the vulnerability of Chinese rice production to drought event is becoming lower on a time scale since the improvement of several socio-economic characteristics including betterment of irrigation, supporting our results. Our results reflected such interaction on a spatial scale. Here, it has to be stressed that one possible problem raised by the approach is that estimating irrigation water does not execute in an accurate way. This is because separating irrigation water for rice from other types of crops is quite difficult based on current data. Quantifying irrigation water availability of farmers is also inexact and with many uncertainties in other countries (Alcamo et al., 2008). The idea here is that drought impact on rice is more serious in north than other regions, which has been confirmed by literatures (Maclean et al., 2002; Xiong et al., 1992). Farmers could alleviate drought to rice more easily if there are more irrigation water resources. However, we did not try to quantify the interaction of the irrigation water quota on the climate to yield relationships since future work is still required on testing validity of irrigation water indicators Comparison with previous estimate Previous studies on temperature effects have elucidated four possible mechanisms for yield impacts. (1) Night temperature, i.e., T min, has previously been reported as being negatively correlated to rice yields at long-term experiment stations due to increased respiration losses (e.g., Peng et al., 2004). However, these results were not corroborated by our observations although the temperature regimes in the two cases are similar. In China, relationships between yields and T min are the most insignificant compared with those with other climatic parameters. This suggests that there is no universal linkage between higher T min and lower rice yields. In addition, a recent study based on experiments (Cheng et al., 2009) suggested magnitude that night respiration increased at high night temperature was much weaker than generally expected. (2) Temperature stress during the flowering period of rice has been experimentally linked to spikelet sterility (Horie et al., 1996; Nakagawa et al., 2003). Under humid conditions, spikelets cannot be fertilized if flowering plants are exposed to T max above approximately 35 C(Matsui and Horie, 1992). However, in all years at the stations, T max was lower than this threshold over a range from 27.3 to 32.5 C; therefore, this mechanism can be excluded as a yield determinant. However, it is possible that this effect would become more important and even play a crucial role in determining rice yields under global warming (Matsui et al., 2001). Further studies using climate change scenarios are necessary to identify regions where temperatures will go beyond critical levels during the flowering period. (3) Lobell (2007) showed that negative correlations between rice yields and T mean are particularly obvious in the countries in which rainfed environments are prevailed. Therefore, he proposed an indirect effect of increased T mean on rice yields; greater water stress associated with warming might explain a reduction in yields with warming. Our results support this viewpoint since our analysis based on regional yields suggested that lower yield tended to occur with higher temperature in provinces having a low irrigation water quota, where variability of Rain explains a change in yield to some degrees. (4) Crop models simulated that higher temperature would seriously lower rice yields due to shorter crop duration (e.g., Tao et al., 2008; Xiong et al., 2007; Yao et al., 2007), which is inconsistent with our empirical findings. However, the results are

10 1136 T. Zhang et al. / Agricultural and Forest Meteorology 150 (2010) completely based on a crop model that includes a great variety of optimum and untested assumptions. Recent several empirical studies argue against such modeling project (Holmer, 2008; Zhang et al., 2008b). In particular, Zhang et al. (2008b) pointed out that constant thermal time accumulation, a parameter to calculate crop duration length in crop model, over environments assumed in modeling studies may bias yield simulation. Further investigation on the discrepancy between estimates of the climate impacts on crop yields based on observations and model simulation is necessary. 5. Conclusions This study examined the empirical relationships between rice yields and T min, T max, T mean, Rad, and Rain using extensive data at difficult spatial scales (experiment stations, counties, and provinces) in China. For most experiment stations, there is a positive correlation between yield and temperature that contradicts current viewpoint that warming declines rice yields. Instead, we saw a clearly positive correlation of yields to Rad for the majority of stations. This suggests that Rad is the major climatic driver in determining yield variability at experiment stations. At regional scale (county and province), besides Rad s effect happened at experiment stations, we also detected Rain s effect that affected rice yields in NC and NWC, where irrigation water availability is relative few. Therefore, the interaction between Rad, Rain and irrigation water availability determined yield variation at regional scale. However, temperature stresses on Chinese rice yields are observed relatively weak at either experiment stations or a regional scale. Yield correlation with temperature can be seen as a proxy of effects of Rad at stations and effects of Rad or Rain at regional scale in China. Evaluation on larger area is still required for a more general picture of how rice yields respond to climate change. Acknowledgments This research was jointly supported by Chinese Academy of Sciences (Grant Nos. KZCX2 YW 305, KZCX2-YW-202 and KZCX1-YW-12-03), China Meteorological Administration (Grant No. GYHY ) and National Basic Research Program of China (Grant No. 2006CB403600). The authors also thank the CMA, CMDSSS, and DSIES for providing data, anonymous reviewers for comments, and Dr. Kay Sumfleth for help on ArcGIS software. References Alcamo, J., Acosta-Michlik, L., Carius, A., Eierdanz, F., Klein, R., kromker, D., Tanzler, D., A new approach to quantifying and comparing vulnerability to drought. Reg. Environ. Change 8, Ångström, A.K., Solar and atmospheric radiation. Q. J. R. Met. Soc. 50, 121. Challinor, A.J., Ewert, F., Arnold, S., Simelton, E., Fraser, E., Crops and climate change: progress, trends, and challenges in simulating impacts and informing adaptation. J. Exp. Bot. 60, Changming, L., Jingjie, Y., Groundwater exploitation and its impact on the environment in the North China Plain. Water Int. 26, Cheng, W., Sakai, H., Yagi, K., Hasegawa, T., Interactions of elevated [CO 2] and night temperature on rice growth and yield. Agric. Forest Meteorol. 149, CMA archives, Chinese Meteorological Administration Archives. CMDSSS, China Meteorological Data Sharing Service System. Dobermann, A., Dawe, D., Roetter, R.P., Cassman, K.G., Reversal of rice yield decline in a long-term continuous cropping experiment. Agron. J. 92, DSIES, Data Sharing Infrastructure of Earth System Science. Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences. Erda, L., Wei, X., Hui, J., Xu, Y., Li, Y., Bai, L., Xie, L., Climate change impacts on crop yield and quality with CO 2 fertilization in China. Phil. Trans. R. Soc. B 360, Evans, L.T., Crop Evolution, Adaptation and Yield. Cambridge University Press, Cambridge, pp Evans, L.T., De Datta, S.K., The relation between irradiance and grain yield of irrigated rice in the tropics, as influenced by cultivar, nitrogen fertilizer application, and month of planting. Field Crops Res. 2, Ewert, F., Porter, J.R., Rounsevell, M.D.A., Crop models, CO 2 and climate change. Science 315, Holmer, B., Fluctuations of winter wheat yields in relation to length of winter in Sweden 1866 to Climate Res. 36, Horie, T., Matsui, T., Nakagawa, H., Omasa, K., Effect of elevated CO 2 and global climate change on rice yield in Japan. In: Omasa, K., Kai, K., Taoda, H., Uchijima, Z., Yishino, M. (Eds.), Climate Change and Plants in East Asia. Spring-Verlag, Tokyo, pp IRRI database, International Rice Research Institute online electronic database. content&task=view&id= 413&Itemid=352. Khan, S., Hanjra, M.A., Mu, J., Water management and crop production for food security in China: a review. Agric. Water Manage. 96, Landau, S., Mitchell, R.A.C., Barnett, V., Colls, J.J., Craigon, J., Moore, K.L., Payne, R.W., Testing winter wheat simulation models predictions against observed UK grain yields. Agric. Forest Meteorol. 89, Lobell, D.B., Asner, G.P., Climate and management contributions to recent trends in U.S. agricultural yields. Science 299, Lobell, D.B., Burke, M.B., Tebaldi, C., Mastrandrea, M.D., Falcon, W.P., Naylor, R.L., Prioritizing climate change adaptation needs for food security in Science 319, Lobell, D.B., Ortiz-Monasterio, J.I., Asner, G.P., Matson, P.A., Naylor, R.L., Falcon, W.P., Analysis of wheat yield and climatic trends in Mexico. Field Crops Res. 94, Lobell, D.B., Ortiz-Monasterio, J.V., Impacts of day versus night temperature on spring wheat yields: a comparison of empirical and CERES model prediction in three locations. Agron. J. 99, Lobell, D.B., Cahill, K.N., Field, C.B., Historical effects of temperature and precipitation on California crop yields. Climatic Change 81, Lobell, D.B., Changes in diurnal temperature range and national cereal yields. Agric. Forest Meteorol. 145, Maclean, J.L., Dawe, D.C., Hardy, B., Hettel, G.P. (Eds.), Rice Almanac. CABI Publishing, Wallingford, UK. Matsui, T., Horie, T., Effect of elevated CO 2 and high temperature on growth and yield of rice. 2. Sensitive period and pollen germination rate in high temperature sterility of rice spikelet at flowering. Jpn. J. Crop Sci. 61, Matsui, T., Omasa, K., Horie, T., The difference in sterility due to high temperature during the flowering period among Japonica-rice varieties. Plant Prod. Sci. 4, Matthews, R.B., Kropff, M.J., Bachelet, D., Van Laar, H.H., Modeling the Impact of Climate Change on Rice Production in Asia. CABI, Los Banos, Philippines. Nakagawa, H., Horie, T., Matsui, T., Effects of climate change on rice production and adaptive technologies. In: Mew, T.W., Brar, D.S., Peng, S., Dawe, D., Hardy, B. (Eds.), Rice Science: Innovations and Impact for Livelihood. International Rice Research Institute, pp Nicholls, N., Increased Australian wheat yield due to recent climate trends. Nature 387, Parry, M.L., Rosenzweig, C., Iglesias, A., Livermore, M., Fisher, G., Effects of climate change on global food production under SRES emissions and socioeconomic scenarios. Global Environ. Change Hum. Policy Dimension 14, Peng, S., Huang, J., Sheehy, J.E., Laza, R.C., Visperas, R.M., Zhong, X., Centeno, G.S., Khush, G.S., Cassman, K.G., Rice yields decline with higher night temperature from global warming. Proc. Natl. Acad. Sci. U.S.A. 101, Schmidhuber, J., Tubiello, F.N., Global food security under climate change. Proc. Natl. Acad. Sci. U.S.A. 104, Sheehy, J.E., Mitchell, P.L., Allen, L.H., Ferrer, A.B., 2006a. Mathematical consequences of using various empirical expressions of crop yield as a function of temperature. Field Crop Res. 98, Sheehy, J.E., Mitchell, P.L., Ferrer, A.B., 2006b. Decline in rice grain yields with temperature: models and correlations can give different estimates. Field Crops Res. 98, Shu, G., Yixing, Z., Minghua, Z., Smallwood, K.S., A sustainable agro-ecological solution to water shortage in the North China Plain (Huabei Plain). J. Environ. Plann. Manage. 44, Simelton, E., Fraser, E.D.G., Termansen, M., Forster, P.M., Dougill, A.J., Typologies of crop-drought vulnerability: an empirical analysis of the socioeconomic factors that influence the sensitivity and resilience to drought of three major food crops in China ( ). Environ. Sci. Policy, doi: /j.envsci Tao, F., Yokozawa, M., Liu, J., Zhang, Z., Climate-crop yield relationships at provincial scales in China and the impacts of recent climate trends. Climate Res. 38, Tao, F., Yokozawa, M., Xu, Y., Hayashi, Y., Zhang, Z., Climatic changes and trends in phenology and yields of field crops in China Agric. Forest Meteorol. 138, Varis, O., Vakkilainen, P., China s 8 challenges to water resources management in the first quarter of the 21st Century. Geomorphology 41, Xiong, Z., Cai, H., Min, S. (Eds.), Rice in China. Chinese Publishing Press of Agricultural Science and Technology, Beijing, China (in Chinese with English abstract). Xiong, W., Lin, E., Ju, H., Xu, Y., Climate change and critical thresholds in China s food security. Climatic Change 81,