Rainfed wheat yields under climate change in northeastern Iran

Authors


Abstract

Projecting agricultural crop, such as bread wheat, yield under future climate plays a vital role in planning for supply and demand, especially in developing countries. The objective of this study was to investigate impacts of climate change on grain yield of rainfed wheat in the Kashafrood basin situated in northeastern Iran. Climate projections of two General Circulation Models (HadCM3 and CGCM2) under two climate change scenarios (A2 and B2) were employed in this study. A statistical downscaling method was employed for developing the quantitative relationship between large-scale atmospheric variables (predictors) and local variables (observations), also daily climatological parameters performed by the LARS-WG5 stochastic weather generator, the Decision Support System for Agrotechnology Transfer (DSSAT) Version 4.5 adjusted to evaluate rainfed wheat performance. Results of this study represented substantial difference between GCM models and scenarios for wheat yield in the study area. The A2 scenario indicated more negative impact on wheat yield than the B2 using both GCM models. Highest increase and decrease of grain yield in comparison with the baseline belongs to the 2010–2039 period (+15%) using the HadCM3 model under the B2 scenario and the HadCM3 model under A2 in the 2040–2069 period (−50%), respectively. In general, growth period length indicated a markedly decreasing trend in both periods. There were strong relations between wheat yield and precipitation rate and growth period length in this study. Copyright © 2011 Royal Meteorological Society

1. Introduction

Crop yields are affected by variations in climatic factors such as air temperature and precipitation, and the frequency and severity of extreme events such as droughts, floods, hurricanes, windstorms and hail (Alexandrov and Hogenboom, 2000; Bannayan et al., 2010). Climate change processes including increasing concentration of atmospheric carbon dioxide (CO2), temperature and variability of precipitation directly impact agricultural products (Bannayan et al., 2005; Morison and Morecroft, 2006, Bannayan et al., 2011). IPCC reports prominently mention that the global warming process is occurring rapidly, and that roughly 1 °C of warming, relative to the late twentieth century, is expected globally by 2030 regardless of what happens to the intensity of greenhouse gases (IPCC, 2007).

Atmospheric general circulation models (GCMs) are in use for seasonal forecasting by projecting a full set of meteorological variables at a sub-daily time step (Ines and Hansen, 2006). GCMs contain significant uncertainties and there are different models and scenarios, such as A2 and B2, for predicting future climate in different situations (Wilby et al., 1999). The A2 scenario is one of the most extreme scenarios, with carbon emissions rising from about 10 Gt currently to over 25 Gt in 2100 (medium to high carbon emission) (Prudhomme et al., 2010). This scenario describes the maximum potential impact of future climate on specific dynamics. The B2 scenario is more optimistic (low to medium carbon emission) (Hewitson and Crane, 1996). GCM projections can be used to provide weather data for future-oriented crop simulations (Reddy and Pachepsky, 2000; Mall et al., 2004). How crop yield responds to climate change will affect the food security of a nation (Liu et al., 2010). Laux et al. (2010) predicted higher temperature under the A2 scenario during 2020–2080 in dry regions of Cameroon would decrease the economic yield of both maize and groundnut: however, increasing CO2 concentration may mitigate this decrease. Guo et al. (2010) reported average wheat yield and maize yield under A2A and B2A scenarios using the HadCM3 global climate model will respectively increase by 9.8 and 3.2% without CO2 fertilization in the North China Plain. Furthermore, rice yield will increase between 1 and 16.8% in 2070 which is more than 30% in optimistic considerations (Aggarwal and Mall, 2002). However, Liu et al. (2010) reported a 2–5 °C increase in temperature under climate change in China could have more impacts than CO2 concentration than on wheat yields, though it had a mainly negative impact.

Wheat is one of the most widely grown crops in the world and approximately one sixth of the total arable land in the world is under wheat cultivation (Satorre and Slafer, 1999). According to FAO Statistics (2008) there were 4.7 billion ha under wheat cultivation in Iran. In addition, the average yield of wheat ranged between 4.2 to 1.1 t per ha for irrigated and 1.1 t per ha for rainfed conditions in this country. Khorasan province, which is situated in the northeast of Iran, contains the most cultivated lands (72 000 ha), and its average rainfed wheat yield is about 1 t ha−1 (Khorasan Ministry of Agricultural Statistics, 2008).

Information needs for agricultural decision making at all levels are increasing rapidly due to increased demands for agricultural products and increased pressure on land, water and other natural resources. Traditional agronomic experiments are conducted at particular points in time and space, making results site- and season-specific, time consuming and expensive (Bannayan et al., 2007). Dynamic crop growth models have dramatically improved analytical solution of problems in crop science (Chipansky et al., 1997). Studies to assess climate change impacts on agriculture based on crop simulation models have used a variety of climate and agricultural models. However, large variation in climatic conditions and agricultural crops around the world makes it difficult to apply this information directly to a particular region and requires local assessment. In this study the Decision Support System for Agrotechnology Transfer (DSSAT) (Jones et al., 2003) comprising six models for simulating the growth of 16 crops has been used. The model has demonstrated high reliability under different climates, soil and management conditions (Bannayan et al., 2003). The CERES-Wheat model is one of the most popular and high visibility wheat models (Pecetti and Hollington, 1997). Furthermore, it has been tested in many sites across the world and the results indicated its capability to simulate grain yields under dry conditions (Rinaldi, 2004).

The objectives of the present study were to assess the potential impact of climate change on yield, growth period length and precipitation during growth period of rainfed wheat in the Kashafrood basin which situated in northeast of Iran.

2. Methods and materials

2.1. Study area

The Kashafrood basin, which covers 16 500 km2, is located in the northeast of Iran (Khorasan region) (Figure 1), between 35°40′ and 36°3′N and 58°2′ and 60°8′E. The climate pattern of the basin is cold and arid. The mean annual temperature is 13.6 °C and the average annual precipitation is 220 mm (IMO, 2009). In the mountainous part of Khorasan province, the annual precipitation exceeds 230 mm with the record of 324 mm at the highest location. In the Kashafrood basin, more than 86% of the total water consumption is in the agricultural sector. Water is mainly extracted from ground water, but 14% comes from surface water (IMO, 2009).

Figure 1.

Study area location in northeastern Iran

2.2. Data

Daily climate data, including maximum and minimum temperature ( °C) and precipitation (mm) were obtained as baseline data for the period of 1961–1990 from the Mashhad climatological station, which is situated in the Kashafrood basin. The climate projections for two future scenarios were based on two GCMs: the Canadian Climate Center (CGCM2) (Falto and Boer, 2001) and United Kingdom Met Office Hadley Centre (HadCM3) (Mitchell et al., 1995) and scenarios were A2 and B2. HadCM3 is a coupled atmosphere-ocean GCM that is described by Gordon et al. (2000) and resolved at a spatial resolution of 2.5° latitude by 3.75° longitude. CGCM2 was resolved at a spatial resolution of 3.7° latitude by 3.7° longitude (Salathe, 2003). A statistical downscaling method was used for developing quantitative relationships between large-scale atmospheric variables (predictors) and local variables (observations). An automated regression-based Statistical Downscaling model (ASD) (Hessami et al., 2007), was used in this study for downscaling. ASD was inspired by an existing statistical downscaling model (SDSM) (Wilby et al., 2002), which was developed using Matlab software. Selected future climate projection scenarios were chosen to evaluate climate impacts for the periods 2010–2039, 2040–2069 and 2070–2099. Monthly means of maximum and minimum temperature (Table I), precipitation, percentage of wet days and maximum number of Consecutive Dry Days (CDD) (Table II) for each month under future climate change of the Kashafrood basin in the three periods were calculated. Daily climatic data (one stochastic growing season for each study period) which are required for the crop simulation model, including maximum and minimum temperature, precipitation and solar radiation for each period of future climate, produced by the LARS-WG5 stochastic weather generator (Semenov and Brooks, 1999) by using absolute and relative change of temperature and precipitation in comparison with the baseline obtained by the downscaling model (ASD) from the GCM models. The model validation process was performed by a field experiment which was conducted in the Shirvan Dryland Agricultural Research Institute which is located 50 km from the Mashhad site. This experiment investigated the impacts of different planting densities on yield and yield components of rainfed wheat (Sharifi and Rahimian Mashhadi, 2001). Five levels of planting densities (70, 100, 130, 160 and 190 kg seed per ha) and two levels of irrigation (rainfed and supplement irrigation) were conducted with a split plot factorial experiment based on a randomized complete block design with three replications.

Table I. Baseline (current) and projected climate characteristics for monthly means of maximum and minimum temperature with HadCM3 and CGCM2 under A2 and B2 scenarios in all studied years (2010–2039, 2040–2069 and 2070–2099)
   Monthly minimum temperature
GCM modelScenarioPeriodsJanuaryFebruaryMarchAprilMayJuneJulyAugustSeptemberOctoberNovemberDecember
HadCM3A21961–1990− 5.2− 2.52.58.212.215.817.915.010.15.20.9− 3.0
  2010–2039− 5.5− 2.53.18.012.215.918.716.211.06.11.2− 2.6
  2040–2069− 4.5− 1.73.48.613.016.619.517.412.07.01.4− 2.1
  2070–2099− 4.1− 1.44.59.413.917.620.719.113.47.72.7− 1.2
CGCM2B22010–2039− 5.3− 3.33.28.212.215.718.816.610.85.30.8− 3.1
  2040–206− 5.0− 2.53.88.513.016.919.217.811.26.11.1− 2.5
  2070–2099− 4.2− 2.34.48.613.617.320.218.712.46.41.4− 2.7
 A22010–2039− 6.0− 2.72.57.512.215.717.915.710.75.30.9− 3.4
  2040–2069− 6.1− 2.62.47.612.416.118.416.411.05.71.0− 3.4
  2070–2099− 6.2− 2.22.47.612.816.618.817.211.36.11.2− 3.4
 B22010–2039− 6.0− 2.72.57.512.115.718.015.710.75.30.9− 3.5
  2040–2069− 6.1− 2.62.47.612.415.918.316.110.95.61.0− 3.4
  2070–2099− 6.1− 2.42.57.512.516.118.516.511.05.71.0− 3.4
 Monthly maximum temperature
HadCM3A21961–19906.69.014.421.427.332.734.432.628.321.615.29.3
  2010–20397.310.016.021.328.033.435.433.829.023.517.711.7
  2040–20699.311.017.122.530.235.336.835.330.725.118.712.9
  2070–209910.813.019.925.333.137.838.937.632.726.621.014.9
CGCM2B22010–20397.99.016.322.028.233.937.034.529.223.317.411.2
  2040–20698.810.917.622.830.436.538.236.129.925.119.112.8
  2070–209910.411.618.723.831.937.540.937.532.426.120.113.3
 A22010–20396.28.814.319.727.832.833.832.928.822.316.49.5
  2040–20696.18.814.019.728.433.233.733.029.222.716.69.4
  2070–20995.99.013.919.529.434.033.633.129.823.117.09.4
 B22010–20396.28.814.219.827.732.833.832.928.722.316.49.4
  2040–20696.18.914.219.728.433.133.733.029.122.616.69.5
  2070–20996.18.914.019.728.733.433.733.029.322.716.69.3
Table II. Baseline (current) and projected climate characteristics for monthly average precipitation, percentage of wet days (wet-day) and Maximum Number of Consecutive Dry Days (CDD) with HadCM3 and CGCM2 under A2 and B2 scenarios in all studied years (2010–2039, 2040–2069 and 2070–2099)
   Monthly mean precipitation
GCM modelScenarioPeriodsJanuaryFebruaryMarchAprilMayJuneJulyAugustSeptemberOctoberNovemberDecember
HadCM3A21961–199033.136.052.048.825.53.00.90.71.511.215.726.9
  2010–203922.025.944.144.045.96.80.00.08.515.630.025.7
  2040–206921.031.252.944.840.42.90.00.07.314.926.029.1
  2070–209920.231.754.936.242.51.60.00.06.212.027.030.1
 B22010–203920.829.844.944.643.04.80.00.011.217.833.628.1
  2040–206920.030.248.643.444.95.40.00.011.214.429.827.0
  2070–209919.135.447.742.945.73.40.00.08.317.426.028.2
CGCM2A22010–203923.825.945.744.425.11.61.70.61.26.715.523.2
  2040–206924.226.645.044.524.61.21.50.81.36.616.322.6
  2070–209924.326.745.445.324.51.41.40.81.36.515.623.7
 B22010–203923.126.744.644.828.42.91.90.01.37.415.824.6
  2040–206923.726.244.745.526.72.42.00.01.17.315.523.6
  2070–209923.526.045.446.527.41.22.10.01.06.815.622.0
Percentage of wet days (wet-day) (monthly)
HadCM3A21961–199019.022.427.025.214.22.61.00.81.66.59.014.6
  2010–203912.714.822.023.120.03.31.00.82.77.213.312.1
  2040–206912.316.625.724.517.61.80.80.83.06.911.812.9
  2070–209912.016.825.222.417.31.00.40.83.46.211.913.3
 B22010–203912.116.322.223.918.82.71.10.83.27.614.813.3
  2040–206911.916.623.723.619.52.70.90.83.56.612.912.6
  2070–209911.718.722.624.319.31.90.60.83.38.311.612.5
CGCM2A22010–203917.020.426.126.114.32.31.40.81.96.77.913.3
  2040–206917.320.825.926.314.22.41.40.81.96.78.312.3
  2070–209917.320.726.026.414.32.31.40.81.96.98.312.5
 B22010–203916.620.825.926.914.92.41.60.92.17.38.312.6
  2040–206916.820.426.227.014.42.41.51.01.96.88.112.3
  2070–209916.820.326.127.414.62.31.60.91.87.18.012.0
HadCM3A21961–199013119111627293027221714
  2010–2039151310111125282825201515
  2040–206915129101227292825211615
  2070–209915129101228292824211614
 B22010–2039151210101125282824201414
  2040–2069151210101125282823211515
  2070–2099161110101127292824191615
CGCM2A22010–20391210981325282926191814
  2040–20691110881325282926191714
  2070–20991210881325282926191714
 B22010–20391210981325272926191714
  2040–20691210881325282926191715
  2070–20991210881325272926191815

2.3. Model calibration

The Sardari cultivar is one of the common used varieties of rainfed wheat in Iran which has been cultivated for about 20 years in the Khorasan region (Khorasan Ministry of Agricultural Statistics, 2008). The crop model was calibrated based on measured data of two experiments which were performed at the experimental station of the College of Agriculture, Ferdowsi University of Mashhad (36°15′N, 59°28′E: 928 m asl) in the central part of Khorasan province in the 1997–1998 and 1999–2000 growing seasons. The effects of planting densities and different cultivars of rainfed wheat in the Khorasan province situation (Sharifi and Rahimian Mashhadi, 2001) investigated in the first and second experiments was about the impact of different levels of phosphorus and seed rate on yield of rainfed wheat (Koocheki and Azemzade, 1999). Common planting date of rainfed wheat in the study area is early November. The model was calibrated using measured weather, soil and the treatments which were employed on field experiments. Measurements including grain yield, shoot biomass, leaf area index in different growth stages, plant height, tiller numbers, 1000 seed weight, days to heading and days to maturity, harvest index and spikelet in square metres, were provided for the model as observed data using ATCreate 4.5. ATCreate is a sub-program of DSSAT which was used for declaration of the observed data. A and T files produced by ATCreate were linked to a Genetic Coefficient Estimator (Gencalc) for estimated genetical coefficients of the Sardari cultivar (Hunt et al., 1993). Genetic coefficients of the Sardari cultivar are represented in Table III.

Table III. Genetical coefficients of Sardari cultivar
P1VP1DP5G1G2G3PHINT
  1. P1V, Days at optimum vernalizing temperature required to complete vernalization; P1D, Percentage reduction in development rate in a photoperiod 10 h shorter than the threshold relative to that at the threshold. P5, Grain filling (excluding lag) phase duration ( °C.d), G1, Kernel number per unit canopy weight at anthesis; G2, Standard kernel size under optimum conditions (mg); G3, Standard, non-stressed dry weight (total, including grain) of a single tiller at maturity (g); PHINT, Interval between successive leaf tip appearances ( °C.d).

14045013411.560

2.4. Crop model validation

Validation involves comparing model outputs to experimental results and observations or measurements of the real system (Beheydt et al., 2007). Several criteria (Smith et al., 1997; Mall et al., 2004) were calculated to quantify the difference between simulated and observed data. The root mean-squared error (RMSE) was computed to measure the coincidence between measured and simulated values, while mean deviation (RMD) was calculated to evaluate systematic bias of the model. Model efficiency (EF) is calculated to estimate model performance in relation to the observed mean (Nash and Sutcliffe, 1970). Moreover, linear regression detected between simulations and observations was used to evaluate model performance and correlation coefficient (R2) determined for each simulation:

equation image(1)
equation image(2)
equation image(3)

where P and O are simulated and observed data, respectively, Ō is the mean of observed data and n is the number of observations. The RMSE illustrated the model prediction error by heavily weighting high errors, whilst the RMD weights all errors the same, which tends to smooth out discrepancies between simulated and observed data. EF indicated the efficiency of the model and can be positive or negative (Brisson et al., 2002; Huang et al., 2009). Simulated and observed data, including grain and biomass and maximum leaf area index were compared in this study.

3. Results and discussion

3.1. Crop and climate model evaluation

The correct estimation of crop yield and leaf area index plays a vital role for the accurate evaluation of crop growth simulation models at a specific location. Model validation was conducted by three factors, including grain, biomass and maximum leaf area index using four error criteria [root mean-squared error (RMSE), root mean deviation (RMD), model efficiency (EF) and R2]. Evaluation results showed small values for grain and biomass (Table IV) and there was significant correlation between observed and simulated grain and biomass (Table IV). As indicated in Table IV, the model-predicted maximum leaf are index is ± 8% of the measured yields (RMSE = 8.1), but there is high correlation (0.94) between observed and simulated values.

Table IV. Comparison of simulated and observed grain and biological yield and maximum leaf area index by root mean-squared error (RMSE), root mean deviation (RMD), Model efficiency (EF) and R2 methods
ParametersRMSERMDEFR2
Grain yield5.314.120.100.88
Crop biomass4.814.5− 4.70.51
Maximum leaf area index8.107.140.190.94

The leaf area index parameter showed close relation with environmental factors owing to the fact that predicted values for this parameter were relatively higher than the observed data especially in rainfed situations in dry regions in which the crop may exposed to drought stress nearly across all growth stages (Teruel et al., 1997; Xue et al., 2003).

The precise estimate of weather variables such as maximum and minimum temperature and precipitation in the baseline period may show the accuracy of downscaling models in climate change studies (Viglizzo et al., 1997). Downscaling models represented high accuracy projection of maximum (RMSE = 1.7) and minimum (RMSE = 19) temperature and precipitation (RMSE = 11) in the study area (Figure 2(a–c)).

Figure 2.

Comparison of simulated (white) and observed (black) climatic parameters including (a) minimum, (b) maximum and (c) precipitation

3.2. Grain yield

The results of crop growth simulation performed with the CGCM2 model under the A2 scenario showed a marginal drop for rainfed wheat grain yield in 2010–2039, however the grain yield trend indicated slight growth in 2040–2069 compared to the baseline yield (Figure 3(a)). Yield trend in 2070–2099 gradually decreased (−18%) in comparison with baseline yield (Figure 3(a)). There was no difference between baseline yield and projected yield in 2010–2039 obtained under the B2 scenario, although yield of rainfed wheat showed a gentle increase in 2040–2069 compared to the baseline (Figure 3(b)). Grain yield indicated a moderate slump (−13%) in 2070–2099 (Figure 3(b)); this decrease intensified under the A2 scenario.

Figure 3.

Grain yield of wheat and percentage of change in compare with baseline under climate change process performed with CGCM2 model under (a) A2 and (b) B2 scenarios in different periods

In addition, there was a negligible increase in grain yield under future climate derived by the HadCM3 model under the A2 scenario in 2010–2039, but there was a noticeable decrease (−50%) in 2040–2069 compared with the baseline (Figure 4(a)). However, crop-simulated data indicated relative pick up by + 7% of wheat grain yield during 2070–2099 compared to the baseline yield (Figure 4(a)). Fluctuations and trend of rainfed wheat grain yield under the B2 scenario using the HadCM3 model was similar to the A2 scenario (Figure 4(b)). However, increasing grain yield in 2010–2039 compared to the baseline under B2 (+15%) was more than that for A2 (+5%).

Figure 4.

Grain yield of wheat and percentage of change in compare with baseline under climate change process performed with HadCM3 model under (a) A2 and (b) B2 scenarios in different periods

The A2 scenario is one of the most intense scenarios, with carbon emission rising from about 10 Gt currently to over 25 Gt in 2100 (medium to high carbon emission): economical development is slow and population growth shows a sharp increase (IPCC, 2001). This scenario indicated maximum potential impact of future climate on specific dynamics. The B2 scenario is more optimistic (low to medium carbon emission), with economical pick up rate is high with environmental friendly technologies and low population growth rate (Hewitson and Crane, 1996). The A2 scenario showed more decrease and less increase of grain yield using both GCM models (Figures 3 and 4).

Changes of precipitation amount and distribution under climate change have major influence on the yield of rainfed crops (Rockstrom, 2003). It seems that lower precipitation in 2070–2099 was the main reason of grain yield drop using CGCM2 model under both scenarios (Figure 5). Furthermore, impacts of lower precipitation on the growth period of wheat were the main reason for dramatic decrease of yield using the HadCM3 model under both scenarios (Figure 5).

Figure 5.

Precipitation rate in growth period of wheat under CGCM2 and HadCM3 models under A2 and B2 scenarios in different time periods. (CGCM2 (A2): large checker board, CGCM2 (B2): large confetti, HadCM3 (A2): dark horizontal and HadCM3 (B2): trellis)

3.3. Growth period length

Growth period length is effective on dry matter accumulation and respiration rate and its duration directly impacts yield and productivity of crops (Wiedenfeld, 2000). In general, there was a gradual decrease in growth period length in comparison with the baseline (Figure 6(a) and (b)) using the CGCM2 model under both A2 and B2 scenarios in 2070–2099 (Figure 6(a)). In addition, there was a significant fall in growth period length in 2040–2069 obtained using the HadCM3 model under both scenarios (Figure 6(b)).

Figure 6.

Growth period length of wheat under climate change process performed with (a) CGCM2 and (b) HadCM3 models under A2 and B2 scenarios in different time periods. (A2: large checker board and B2: dark horizontal)

It seems that a steady increase in temperature variables, especially in maximum temperature in the early growth period (Table II) disturbed the vernalization process in rainfed wheat. Direct HadCM3 climate model output points to an annual temperature increase of between 2 and 4 °C for most of the Mediterranean area in the period 2070–2099 (Hertig and Jacobeit, 2008). This increase in minimum temperature could have delayed the vernalization process and exposed the sensitive wheat growth stage to extreme environmental conditions (Rawson et al., 1998; Yan and Hunt, 1999).

3.4. Trend detection

Simulated wheat grain yield indicated strongly significant positive correlation with precipitation (Figure 7(a)). Furthermore, this parameter showed significant positive trend along with growth period length (Figure 7(b)). There was also a significant positive correlation between precipitation rate and growth period length (Figure 7(c)). A strong relationship between precipitation and crop productivity in rainfed conditions is shown for wheat (Anderson, 1992; Lazar et al., 1995) and bean (White et al., 1994). In addition, growth period length was highly correlated with grain yield, the time course of the partition coefficient of photosynthetic products between vegetative and reproductive growth which maximizes final seed yield (Cohen, 1971).

Figure 7.

(a) Relationship between simulated yield and precipitation, (b) growth period length and (c) between precipitation and growth period length

4. Conclusion

The main conclusions that can be drawn from this work are that, as a result of climate change, there is likely to be a sharp decrease in rainfed wheat yield in the study area, especially in the next few decades. It seems climate change processes may have a severe impact on agricultural production in dry regions such as northeast Iran in comparison with other parts of world because of a greater increase in temperature (Morison and Morecroft, 2006). Noticeable fluctuations of wheat grain yield under rainfed condition using two different general circulation models and two scenarios in various future periods (2010–2039, 2040–2069 and 2070–2099) were detected. The temperature increase and precipitation in growth period of rainfed wheat under the A2 scenario is conspicuously larger than those of the B2 scenario. CGCM2 model outputs in both scenarios showed a slower decrease than the baseline yield but this trend was reversed under the HadCM3 model. There was a steep decline of grain yield in both scenarios of this GCM. This decreasing trend with less intensity also was indicated using the CGCM2 model for 2070–2099. Growth period length showed a decreasing trend in comparison with the baseline. There were significant correlations with positive trend between simulated grain yields and precipitation and growth period length, in addition similar trend detected between precipitation and growth period length.

Ancillary