Sunteți pe pagina 1din 12

Climate change scenarios with high spatial and temporal resolution for agricultural applications

E. M. BARROW1 AND M. A. SEMENOV2


1 2

Climatic Research Unit, University of East Angha, Norwich, NR4 7TJ, England Long Ashton Research Station, IACR, University of Bristol, Long Ashton, Bristol, BS18 9AF, England

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

Summary Scenarios of climate change with high spatial and temporal resolutions are required for the assessment of the impact of such change on agriculture. A method of producing high resolution scenarios based on regression downscaling techniques linked with a stochastic weather generator is described. Regression relationships were initially determined between observed large-scale and site-specific climate. By assuming that these relationships would be valid in a future climate, they were subsequently used to downscale general circulation model (GCM) data. The UK Meteorological Office high resolution GCM transient experiment (UKTR) was used to construct the climate change scenarios. Site-specific, UKTR-derived changes in a number of weather statistics were used to perturb the parameters of the local stochastic weather generator (LARS-WG), which had initially been calibrated using observed daily climate data. LARS-WG was used to simulate the site-specific daily weather data required by crop growth simulation models. This method permits changes to a wider set of climate parameters in the scenario, including variability. Simulated wheat yields were shown to be more sensitive to changes in climate variability than to changes in the mean. Results are presented for two European sites.

Introduction To construct scenarios for the assessment of the impact of climate change on crop production, it is necessary to analyse the sensitivity of a crop system to climate variables. The changes in weather parameters that may result in noticeable changes in yield potential or associated agricultural risk should be incorporated in the climate change scenarios. To date, in most climate change studies, mean changes in climate variables have been applied to historical weather or climate data to construct climate

, ,/- c change scenarios tor impact assessment (Giorgi and Mearns, 1991; Kenny et al., 1993; Rosenzweig et al., 1993), but changes in variability of climate can have a significant effect on crop growth and development. For example, changes in the variability of temperature can greatly influence dry matter production since both high and low temperatures decrease the rate of dry matter production and, at the extreme, can cause production to cease (Grace, 1988). Severe water deficits immediately before flowering can lead to pollen sterility and a decrease in grain
f, Vol. 68, No 4, 1995

350

FORESTRY

set (Saini and Aspinall, 1982). Furthermore, there is a non-linear relationship between precipitation amount and the water-use efficiency of plants. Average amounts of precipitation are used relatively more efficiently by plants than larger amounts because of the rapid movement of excess water to deep and unavailable layers in the soil (Keulen and Wolf, 1986). In turn, this can lead to excess nitrate leaching to ground water, especially if the rate of nitrogen mineralization by soil microbes is stimulated by warmer temperatures. Extreme weather events, such as drought or hot or cold spells, can have severe consequences for crops and their frequency of occurrence is better correlated with changes in the variability of climate variables, as opposed to changes in the mean values (Katz and Brown, 1992). Crop simulation models reflect the mixture of linear and non-linear responses and, in the broadest sense, transform a distribution of weather sequences into a distribution of total dry matter and, in the case of crop plants, harvestable yield. Assessments of the effects of climate on agricultural production, and the appraisal of associated risks to the food supply, need to bear the above in mind. General circulation models (GCMs) are the tools that are now most widely used to generate scenarios of climate change for impacts assessment (Giorgi and Mearns, 1991; Viner and Hulme, 1994). There are a number of factors, however, which limit direct use of their results in scenario construction. These include: (1) the ability of the control experiment to adequately simulate the large-scale features of present-day climate. This is one of the reasons that the difference between the control and perturbed integrations is used, rather than the raw data from the integrations themselves; (2) the coarse spatial gridoutput is on the scale of hundreds of kilometres rather than the tens of kilometres needed for impacts assessment (e.g., 2.9 latitudelongitude grid box resolution is approximately equivalent to 300 X 300 km). This coarse resolution also means that sub-grid scale processes, such as precipitation, are not adequately represented and important regional topographic features are omitted. Hence, although GCMs may be able to simulate largescale features of climate well, their simulation of regional climate is considerably poorer.

Consequently, it is necessary to 'downscale' the coarse-resolution GCM output to the scales required in regional impacts assessment. Downscaling should be tailored to the impacts application for which the climate change scenarios are being constructed. For example, in the case of agriculture, information is usually needed for a suite of climate variables at a high spatial and temporal resolution. Information about changes in climate variability is also important for the reasons mentioned above. There are currently two main generic methods in use to downscale large-scale climate information to the site level. These are: (1) regression methods (e.g., Kim et al., 1984; Wigleycta/., 1990; Karl et al, 1990) and (2) circulation patterns (e.g. Bardossy and Caspary, 1991; Matyasovszky et al., 1993). Both methods make the basic assumption that the present empirical relationships between large-scale and local climate and the observed relationships between variables will continue to be valid in any future climate. Each method uses different sets of GCM-derived variables and hence, to some extent, makes assumptions about the reliability of such variables. For example, regression techniques may use any number of variables including mean sea level pressure (MSLP), temperature, precipitation etc., while the circulation pattern (CP) method restricts the number of variables to MSLP and possibly the height of another pressure surface (e.g., 500 mbar). Both methods use existing instrumental databases to determine the relationships between large-scale and local climate. Regression techniques develop statistical relationships between local station data and grid-point scale, area average values of, say, temperature and precipitation and other meteorological variables, whereas in the CP approach atmospheric circulation is classified according to type and links are then determined between the circulation pattern and the relevant climate variable (e.g., precipitation). In this paper, the downscaling process and its application will be described, from the construction of a high resolution climate change scenario using regression techniques, to its use in a crop-growth simulation model. In order to produce climate data at the correct temporal resolution for such a simulation model, the

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

CLIMATE CHANGE SCENARIOS

351

downscaling technique was used in conjunction with a stochastic weather generator (Racsko et al., 1991; Semenov and Porter, 1994). This is a methodologically consistent approach to incorporate changes in climate variability into climate change scenarios instead of using perturbed historical data as input to a crop simulation model (as was done, for example, in Semenov etal., 1993). A stochastic weather generator allows temporal extrapolation of observed weather data for agricultural risk assessment as well as providing an expanded spatial source of weather data by interpolation between the point-based parameters used to define the weather generator (Hutchinson, 1991). For analysis of the effects of climate change, a stochastic weather generator can play an important role by providing flexibility in the construction of weather scenarios and by linking information about possible climate changes, derived from GCMs using downscaling methods, to local weather characteristics (Wilks, 1992). The procedure is analogous to the conventional practice of applying changes in means to observed data. The essential difference is that changes in the variability of climate parameters are permitted.

It is difficult to relate the control and perturbed integrations of the UKTR experiment directly to calendar years for a number of reasons, including the 'cold start' problem (Viner and Hulme, 1993; Murphy, 1995). However, by combining the UKTR global-mean warming for this decade (1.76C) with the results from a simple climate model, e.g., MAGICC (Model for the Assessment of Greenhouse gas Induced Climate Change; Wigley and Raper, 1992; Wigley, 1994; Hulme et al., 1955), it is possible to calculate a range of future dates when this warming may occur. If the negative effects of sulphate aerosols on global warming are omitted (the UKTR experiment did not include the negative forcing of aerosols on climate), then the UKTR global warming for decade 66-75 may be reached as early as 2038 if climate sensitivity is high, or after 2100 if the climate sensitivity is low. The best estimate data for a global-mean warming of 1.76C to be reached is 2065. Downscaling: calibration of the regression equations Two sites were selected in Europe (Rothamsted, UK, and Sevilla, Spain) for which downscaling was to be undertaken. Unfortunately, insufficient Spanish data were available to construct regional area averages of mean temperature and precipitation, and hence the regression relationships, in time for inclusion in this paper. Downscaling was therefore undertaken only at Rothamsted and the direct UKTR changes for the relevant grid box were used for Sevilla. The first step in this process was the calculation of regression relationships between the observed large-scale and local climate for each month. These relationships were formulated in terms of anomalies from the long-term 1961-90 mean for the variable under consideration. Deriving relationships in this form simplifies the process of applying the GCM changes to these equations. The observed large-scale climate was determined by averaging data from a number of sites located within the appropriate grid box. Figure 1 shows the locations of the sites chosen in relation to the UKTR grid boxes for the Rothamsted site. For this site, anomalies were simply averaged to produce regionally-averaged values. No weighting of sites was necessary

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

Development of high resolution climate change scenarios


GCM scenarios The climate change scenarios used here have been constructed using the UK Meteorological Office high resolution (2.5 latitude by 3.75 longitude) GCM transient experiment (UKTR; Viner and Hulme, 1993; Murphy, 1995; Murphy and Mitchell, 1995). Mean monthly changes for a number of climate variables (mean temperature, precipitation, MSLP and northsouth and east-west pressure gradients) were calculated for each grid box in the European area for the last decade (years 6675) of the model experiment using the equivalent years of the control and perturbed integrations. The change in temperature variability was also determined by analysing the daily mean temperature variance of the control and perturbed integrations for this decade.

352

FORESTRY

62.5

60.0

57.5

55.0

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

1 52.5
50.0-

47.5 -11.25

3.75

Figure 1. Location of the sites used to calculate area-average mean temperature and precipitation for the grid boxes relating to IACR-Rothamsted, UK. The shaded area illustrates UKTR land grid boxes. Unshaded cells represent ocean area.

because of their approximate even distribution throughout the grid and the likely homogeneity of temperature and precipitation anomalies in this particular area (M. Hulme, personal communication). Anomalies for three pressure variables were also calculated for this grid box: MSLP and the north-south and eastwest pressure gradients. Once the anomalies had been calculated, the data were divided into two time periods, 1961-83 and 1984-88. (The gridded MSLP record initially used was complete up to 1988, although it was extended to 1990 in the latter stages of this work.) The first period was used to calibrate the regression equations, while the latter period was used to verify the performance of the regression models. Regression analyses were then undertaken for each month using sta-

tion mean temperature and precipitation anomalies as predictands and the regionallyaveraged anomalies of temperature, precipitation, MSLP and northsouth and east-west pressure gradients as predictors. Table 1 illustrates the performance of the regression models for each month for the Rothamsted site. To be confident that the regression model works well, it must explain a high proportion of the variance in the data and for it to be applicable for any future climate, the scenario changes in temperature and precipitation should be within the observed anomaly range. Table l(a) indicates the variance explained by the regression model for each month, while Table l(b) shows the verification results. As would be expected, the model generally performs better for temperature than for precipitation, especially in spring and

CLIMATE CHANGE SCENARIOS Table 1: Performance of the regression models (a) Calibration of regression model: 1961-83; Variance explained (%)

353

Jan
TMP PPT 99.2 85.9

Feb 99.1 90.3

Mar 98.6 88.2

Apr 98.0 73.9

May 97.3 76.4

Jun
98.3 81.9

Jul
98.5 79.7

Aug 98.5 80.9

Sep 96.8 90.3

Oct 99.1 90.1

Nov 97.9 93.5

Dec 99.0 93.7

(b) Verification of regression model: correlations between observed and predicted; 1984-88 Jan TMP PPT 0.999 0.960 Feb 0.999 0.961 Mar 0.994 0.896 Apr 0.991 0.989 May 0.995 0.859

Jun
0.988 0.953

Jul
0.989 0.918

Aug 0.980 0.956

Sep 0.978 0.977

Oct 0.935 0.996

Nov 0.993 0.990

Dec 0.986 0.988

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

TMP = temperature; PPT = precipitation.

summer. The correlations shown in Table l(b) may be misleadingly high because of the small number of data points used. Table 2 shows the mean square correlations between predictands (expressed as a percentage) for mean temperature (Table 2(a)) and precipitation (Table 2(b)). Table 2(a) illustrates that the areal-mean temperature anomaly is the most important variable in determining Rothamsted mean temperature. While the areal-mean precipitation anomaly is also the most important variable for Rothamsted precipitation, MSLP also has an important role in determining the site precipitation anomalies (Table 2(b)). Downscaling: application ofGCM changes Once the regression equations had been determined using observed data, the next step was to calculate the site changes in mean temperature and precipitation using the changes derived from UKTR for the predictor variables. The UKTR grid box values of the predictor variables are assumed to be equivalent to the regionally-averaged values derived from the observational data. The appropriate UKTR changes were substituted into the regression equations to obtain predictions for the Rothamsted site. Table 3 illustrates the changes in mean temperature (Table 3(a)) and precipitation (Table 3(b)) for the appropriate grid box and

for Rothamsted. It is apparent that downscaling the grid box changes to the Rothamsted site makes little difference to temperature. The observed climate within the grid box containing Rothamsted is actually quite homogeneous, and so it is to be expected that there is little difference between site and areal climate and hence, little difference is obtained by downscaling GCM data. Larger differences between grid box and site values may have been obtained by using a site at an altitude which was considerably different from that of the grid box itself. However, downscaling has a larger effect on precipitation. In some cases, the sign of the precipitation change is reversed, e.g., in May the grid box change indicates an increase of 0.09 mm day"1, whereas at the site a decrease of 0.07 mm day"1 is predicted. Table 3(c) shows the observed 1961-90 anomaly range for the predictor variables. In the case of temperature (Table 3(a)), it is apparent that the grid box changes are outside the anomaly ranges shown in Table 3(c) on only a few occasions. For precipitation the changes are inside the anomaly ranges on all occasions. As well as changes in mean values being extracted from UKTR, daily data were also analysed from the last decade of the model experiment (years 66-75) in order to derive changes in temperature variability and length of wet and dry spells.

354

FORESTRY

Table 2: Mean square correlations between predictands, expressed as a percentage (1961-83). (a) Rothamsted mean temperature Tra Pua 11.70 15-52 1.32 0.16 0.52 3.28 26.52 37.58 6.76 4.49 6.40 0.25 MSLP 0.07 0.23 2.19 2.37 1.72 0.006 4.49 32.04 0.66 4.88 1.23 5.66 Ap ra 72.59 44.49 43.30 1.23 4.37 0.74 15.21 21.25 12.11 0.85 16.81 35.64 Apew 3.53 9.42 1.61 3.50 16.40 30.69 0.05 16.89 8.35 28.52 14.98 0.0001

Jan

Feb Mar Apr May

Jun Jul
Aug Sep Oct Nov Dec

99.20 99.00 98.01 96.24 95.26 97.61 98.41 97.81 96.04 98.21 97.61 98.60

(b) Rothamsted precipitation T, n MSLP 80.82 90.06 86.86 70.06 69.22 80.28 67.08 77.79 90.25 89.87 92.93 92.54 23.33 46.92 60.99 44.62 30.36 34.46 7.95 42.14 63.52 44.49 55.35 43.43 Apn 27.67 18.58 6.15 15.68 6.45 0.08 1.30 17.14 13.84 26.73 18.75 9.18 21.90 22.85 22.37 17.56 11.16 8.18 0.14 23.81 25.30 15.76 26.11 34.22

Jan

Feb Mar Apr May

Jun Jul
Aug Sep Oct Nov Dec

9.36 12.32 1.32 3.42 12.32 1.30 27.35 40.45 5.06 3.13 6.50 0.46

long dry series, a drought, affects crop growth and development and can dramatically decrease the yield. Hence, in order to accurately assess agricultural risk, it is important that such events are modelled well. The distribution of other weather variables, such as temperature and solar radiation, is based on the current status of the wet or dry series. Mixed exponential distributions were used to model dry and wet series so that the model would be applicable to a wide range of locations in Europe. Daily minimum and maximum temperature and radiation were considered as stochastic processes with daily means and standard deviations conditioned on the wet and dry series. The techniques used to analyse the process are very similar to those presented in Richardson (1981). The seasonal cycle of means and standard deviations was removed from the observed record and the residuals approximated by a normal distribution. These residuals were used to analyse a time correlation within each variable. Fourier series were used to interpolate seasonal means and standard deviations. Where radiation data were unavailable, sunshine hours were used in the simulation. Sunshine hours can be converted to radiation by means of the regression relationships between these two variables (Rietveld, 1978). Adjustment of LARS-WG parameters for climate change scenarios The regression downscaling procedure described above generated site specific (for Rothamsted) changes in mean monthly temperature and monthly relative changes in precipitation amount. The parameters of LARS-WG were adjusted accordingly to allow the generation of daily weather sequences for the climate change scenarios. However, information about changes in temperature and precipitation means is not enough to define precisely the changes in all of the LARS-WG parameters, especially changes in daily temperature variability and the duration of dry and wet spells. Additional information is required. It seems unlikely that the regression technique used above for downscaling mean values will be applicable to analyses of climate variability, although more research is necessary in this area. In the current study,

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

= areal mean temperature; P , r a = areal mean precipitation; MSLP - mean sea level pressure; Ap r a = northsouth pressure gradient; Ap,-. = eastwest pressure gradient.

LARS-WG: stochastic weather generator LARS-WG is a development of the stochastic weather generator described in Rascko et al. (1991), to which modifications were made in order to match the output to the meteorological input data required by crop simulation models. The weather generator is based on distributions of the length of continuous sequences, or series, of dry or wet days. Long dry series are simulated better using this approach compared to the Markov chain method (e.g., Richardson, 1981) of simulating precipitation occurrence. A

Table 3: Grid box and site changes derived from UKTR, years 66-75 (a) Mean temperature (C) change Rothamsted UKTR grid box Site 3.49 1.03 2.53 0.54 1.93 1.65 2.30 3.72 3.83 1.50 2.00 2.33

Jan
Feb Mar Apr May

Jun Jul
Aug Sep Oct Nov Dec

3.43 0.98 2.35 0.54 1.61 1.42 2.15 3.55 3.72 1.48 2.04 2.18

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

(b) Precipitation change (mm day"1) Rothamsted UKTR grid box Jan Feb Mar Apr May 0.13 0.30 0.03 -0.14 0.09 0.04 -0.11 -0.33 -0.25 0.12 0.35 0.12 Site -0.07 0.25 0.08 -0.12 -0.07 -0.02 -0.10 -0.59 -0.31 0.06 0.45 0.13

Jun Jul
Aug Sep Oct Nov Dec (c) 1961-90 anomaly ranges MSLP (mbar) Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec -13.8,+15.0 -17.9,+17.4 -12.3,+12.1 -11.6,+9.6 -9.3,+7.8 -6.6,+5.0 -6.9,+4.8 -5.3,+5.7 -7.1,+8.1 -12.8,+8.6 -15.2,+9.6 -19.4,+14.8

Apm (mbar) -2.2,+4.4 -4.0,+3.8 -2.0,+2.0 -1.5,+1.5 -1.9,+1.6 -1.0,+1.4 -1.7,+1.3 -1.7,+1.7 -1.4,+2.1 -2.3,+1.9 -2.0,+2.0 -2.5,+2.1

Apew (mbar) -2.2,+1.9 -1.8,+2.4 -1.3,+1.5 -1.2,+1.8 -0.8,+1.0 -0.8,+1.1 -0.8,+0.7 -0.6,+0.6 -0.7,+0.6 -1.4,+0.8 -1.1,+1.3 -1.6,+1.8

Tarea (C) -6.8,+3.0 -5.2,+4.0 -3.0,+2.5 -2.0,+2.2 -1.3,+23 -2.4,+3.0 -2.0,+3.6 -1.9,+2.6 -2.4,+1.6 -3.1,+2.2 -2.3,+1.8 -3.9,+3.2

Parc (mm day"') -0.8,+0.8 -0.8,+1.7 -0.9,+1.1 -0.9,+0.9 -0.7,+1.6 -0.8,+1.3 -0.8,+1.5 -0.7,+0.8 -0.7,+1.3 -0.9,+1.6 -0.6,+1.5 -0.8,+1.0

356

FORESTRY

possible changes in temperature variability and in the duration of dry and wet spells were derived from the analysis of daily grid-box UKTR data from the last decade (years 66-75) of the transient experiment.

Sensitivity of crop systems to climate variability


The importance of considering effects of climate variability on crop growth and development arose from climate change studies (Mearns et al., 1992; Semenov et al., 1993, Semenov and Porter, 1994). Mearns et al. (1992) investigated how changes in climate variability could affect wheat production and performed sensitivity analyses using the CERES-Wheat crop simulation model and perturbed historical climate data in order to increase the inter-annual variance of the climate variables. Semenov and Porter (1994) used a stochastic weather generator to simulate and alter characteristics of weather sequences instead of using historical weather data alone as input to the AFRC

WHEAT2 model (Porter, 1993). This approach provides a more consistent way of changing the weather parameters, including their variance or the distribution itself. A sensitivity analysis of the AFRCWHEAT2 model to changes in temperature was performed. The effects of changes in climate variability and changes in mean climate on wheat growth and development were compared. The analysis was undertaken for IACR-Rothamsted, UK, using winter wheat cv. Avalon (Figure 2). Changes in the variability of temperature had a similar effect on potential grain yield as changes in the mean values, but a larger effect on the coefficient of variation (CV) than an increase in the temperature mean value. It was concluded that, potentially, increases in temperature variability can decrease potential crop yield and increase agricultural risk more than changes in mean temperature. In this study, the SIRIUS-Wheat growth simulation model (Jamieson, 1989; Jamieson, 1993) was used to illustrate the importance of incorporating downscaled changes in mean climate and also changes in climate variability in climate change scenarios at two sites: IACR-

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

-100

Figure 2. Temperature sensitivity analyses for the AFRCWHEAT2 model. Relative changes in average grain yield and coefficient of variation (CV) for cv. Avalon at IACR-Rothamsted compared with the baseline climate for different sensitivity scenarios. T+2 and T+4, increase in mean daily temperature by 2C and 4C respectively; sd*2, doubling of temperature variability; (T+4)sd*2, a scenario combining an increase in mean temperature with an increase in variability.

CLIMATE CHANGE SCENARIOS

357

Rothamsted, UK, and Sevilla, Spain. The sites represent different weather conditions. At IACR-Rothamsted changes in temperature have the largest effect on crops, whereas at Sevilla water availability is the most limiting factor and changes in temperature have little effect. SIRIUS-Wheat was calibrated for these sites using experimental data and then run using 30 years of generated data both for sensitivity analyses and the UKTR-derived climate change scenarios previously described. The direct effects of changes in CO2 concentration on the crop were not considered. The following sensitivity scenarios were selected in addition to using the observed climate as a baseline (BS): (1) an increase in mean temperature of 3C (BS+3); (2) a doubling of temperature variability without changes in mean values (TV); (3) a doubling of the length of dry series (PV). Two climate change scenarios based on the data derived from the UKTR GCM were used. For Rothamsted, the downscaled changes in mean climate were applied, whereas at Sevilla the appropriate grid box changes were used for the reasons mentioned earlier. In the first scenario (CC) a conventional approach was used where only changes in mean temperature and the amount of precipitation were applied. In the alternative scenario (CCV), temperature variability and the duration of the dry and wet series were perturbed in accordance with the analysis of daily UKTR data. The results are shown in Table 4. For Rothamsted there was very little difference between yields simulated using downscaled climate change data and those simulated using the appropriate UKTR grid box changes (not shown). This is because (1) there is very little

difference between the downscaled and grid box temperature changes, and, more importantly, (2) although there is a larger difference between grid box and site precipitation changes, precipitation is not limiting at this site. The difference is simulated grain yield for the CC and CCV scenarios is also not significant because: (1) the increase in temperature variability predicted by UKTR occurred during the time when it would have little effect on crop growth and development (late summer/early autumn), and (2) there are no significant changes in the duration of dry and wet spells at Rothamsted. There was a large difference in the grain yield predicted by the two scenarios CC and CCV at Sevilla, even without downscaling the UKTR grid box data. Under the first climate change scenario, CC, wheat production will benefit. Simulated grain yield is slightly increased with a decrease in the CV of 50 per cent. The second climate change scenario, CCV, which incorporates changes in climate variability indicates a decrease in mean grain yield of 37 per cent and an increase in the CV of 124 per cent. These results can be explained by the high sensitivity of grain yield to the duration of the dry spell at Sevilla. Analysis of daily UKTR data snowed that the length of the dry spell will increase during the growing season at this site under the climate change scenarios considered here. The cumulative probability functions of grain yield for the scenarios CC and CCV are presented in Figure 3. The lowest simulated yield for the CC scenario was 4 t ha"1, whereas in the CCV scenario the grain yield was less than 4 t ha"1 in more than 50 per cent of the simulations. This may result in wheat production becoming

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

Table 4: Average grain yield and coefficient of variation for winter wheat, as simulated by SIRIUS Wheat at IACR-Rothamsted, UK and Sevilla, Spain, using 30 years of data for sensitivity experiments (BS, BS+3, TV and PV) and UKTR climate change scenarios (CC and CCV) BS IACR-Rothamsted Yield (t ha"1) CV Sevilla Yield (t ha"1) CV 8.23 0.10 5.88 0.22 BSX3 7.06 0.15 5.60 0.12 TV PV 8.12 0.13 3.00 0.58 CC 7.85 0.07 5.95 0.11 CCV 7.98 0.09 3.67 0.48

8.48 0.15 5.77 0.26

358

FORESTRY

4 Grain yield, t ha" 1

Figure 3. SIRIUS What simulated grain yield cumulative probability distributions for winter wheat, cv. Alcala, at Sevilla, Spain, for UKTR climate change scenarios. CC incorporates only changes in mean temperature and precipitation amount; CCV incorporates changes in temperature variability and the length of dry and wet spells.

uneconomical in this region as a result of climate change. Thus, the incorporation of climate variability into climate change scenarios can change completely the conclusions concerning the future suitability of wheat production at Sevilla.

analyses and in the climate change scenario experiments for two locations in Europe. The results demonstrate clearly that changes in climate variability sometimes have a larger effect on grain yield and associated risk than changes in average conditions. Moreover, incorporation of climate variability in climate change scenarios can qualitatively change the prediction of the effect of climate change on wheat growth and development in a particular region. Changes in dry spells at Sevilla, derived from the analysis of UKTR data, resulted in a large decrease in potential yield and a large increase in risk which may make wheat production unsuitable in this region. Considerations of variability are important in the light of estimates of the effect of climate change on agriculture and world food supply (Adams et al., 1990; Rosenzweig et al., 1993). Such studies have not yet examined the possibility that a climate with different variability may have serious effects on food production and trade. If the variability of climate increases, then the frequency distribution of yields is likely to widen with sequences of years with low yields becoming more likely and consequent serious impacts on world food supply.

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

Discussion
GCMs are the most highly developed and internally-consistent tools currently available to model climate and the effects of anthropogenic climate change. The output variables from GCMs are at a coarse regional scale, often at around 100 000 km2. Agricultural models usually need daily weather data within a scale of 5 km 2 . Thus, it is necessary to downscale the information from GCMs to a finer spatial resolution. Here, regression downscaling was used in conjunction with a stochastic weather generator to produce climate change scenarios. The essential difference to the conventional approach is that the changes can be applied to the parameters of a stochastic weather generator which cover almost all the relevant statistics of local weather, including means and variances. The relative importance of changes in climate variability compared with changes in mean values has been assessed in the sensitivity

Acknowledgements
We would like to thank Dr P.D. Jamieson for providing the SIRIUS Wheat model for the simulation runs. We are also very grateful to Dr Mike Hulme for providing many constructive comments on this paper. The UKTR model data were provided by the Climate Impacts LINK Project (Department of the Environment Contract PECD 7/12/96) on behalf of the UK Meteorological Office. This work was funded by the Commission of the European Communities' Environment Programme (Contract EV5V-CT930294).

References
Adams, R.M., Rosenzweig, C , Peart, R.M., Ritchie, J.T., McCarl, B.A., Glyer, J.D., Curry, B., Jones, J.W., Boote, K.J. and Allen, L.H. Jr. 1990 Global climate change and US agriculture. Nature 345, 219-224. Bardossy, A. and Caspary, H.J. 1991 Conceptual model for the calculation of the regional hydrologic effects of climate change. In Hydrology for

CLIMATE CHANGE SCENARIOS the Water Management of Large River Basins. Proceedings of the Vienna Symposium, August 1991. IAHS Publ. No. 201. Giorgi, F. and Mearns, L.O. 1991 Approaches to the simulation of regional climate change: a review. Rev. Geophys. 29, 191-216. Grace, J. 1988 Temperature as a determinant of plant productivity. In Plants and Temperature. S.P. Long and F.I. Woodward (eds). The Company of Biologists Ltd, Cambridge, 91-107. Jamieson, P.D. 1989 Modelling the interaction of wheat production and the weather. In Integrated Systems Analysis and Climate Impacts. R.W.M. Johnson (ed.). Proceedings of a workshop on systems analysis, Wellington, 21-22 November 1989. Rural Policy Unit, MAF-Technology, Wellington, 133-140. Jamieson, P.D. 1993 SIRIUS Wheat. In Model and Experimental Meta Data. GCTE Focus 3 Associate Office, University of Oxford, Oxford. Hulme, M., Raper, S.C.B. and Wigley, T.M.L. 1995 An integrated framework to address climate change (ESCAPE) and further developments of the global and regional climate modules (MAGICC). Energy Policy (in press). Hutchinson, M.F. 1991 Climatic analyses in data sparse regions. In Climatic risk in crop production: Models and Management for the Semiarid Tropics and Subtropics. R.C. Muchow and J.A. Bellamy (eds). CAB International, Wallingford, 55-73. Katz, R.W. and Brown, B.C. 1992 Extreme events in a changing climate: variability is more important than averages. Climatic Change 21, 289302. Karl, T.R., Wang, W.-C, Schlesinger, M.E., Knight, R.W. and Portman, D. 1990 A method of relating general circulation model simulated climate to the observed local climate. Part I: Seasonal statistics. / . Clim. 3, 1063-1079. Kenny, G.J., Harrison, P.A. and Parry, M.L. (eds) 1993 The Effect of Climate Change on the Agricultural and Horticultural Potential in Europe. Environmental Change Unit Research Report No. 2, University of Oxford. Keulen, H. van and Wolf, J. 1986 Modelling of Agricultural Production: Weather, Soil and Crops. Pudoc, Wageningen. Kim, J.-W., Chang, J.-T., Baker, N.L., Wilks, D.S. and Gates, W.L. 1984 The statistical problem of climate inversion: determination of the relationship between local and large-scale climate. Mon. Weather Rev. 112, 2069-2077. Matyasovszky, I., Bogardi, I., Bardossy, A. and Duckstein, L. 1993 Space-time precipitation reflecting climate change. Hydrol. Sci. J. 38, 539-558. Mearns, L.O., Rosenzweig, C. and Goldberg, R. 1992

359

Effects of changes in interannual climatic variability on CERES-Wheat yields: sensitivity and 2XCO2 general circulation model scenarios. Agric. For. Meteorol. 62, 159-189. Murphy, J.M. 1995 Transient response of the Hadley Centre coupled ocean-atmosphere model to increasing carbon dioxide. Part I: Control climate and flux correction. /. Climate 8, 3656. Murphy, J.M. and Mitchell, J.F.B. 1995 Transient response of the Hadley Centre coupled oceanatmosphere model to increasing carbon dioxide. Part II: Spatial and temporal structure of the response. J. Climate 8, 57-80. Porter, J.R. 1993 Af RCWHEAT2: A model of the growth and development of wheat incorporating responses to water and nitrogen. Eur. } . Agron. 2, 69-82. Racsko, P., Szeidl, L. and Semenov, M. 1991 A serial approach to local stochastic weather models. Ecol. Modell. 57, 27-41. Richardson, C.W. 1981 Stochastic simulation of daily precipitation, temperature, and solar radiation. Water Resour. Res. 17, 182-190. Rietveld, M.R. 1978 A new method for estimating the regression coefficients in the formula relating solar radiation to sunshine. Agric. For. Meteorol. 19, 243-252. Rosenzweig, C , Parry, M.L., Fischer, G. and Frohberg, K. 1993 Climate Change and World Food Supply. Environmental Change Unit Research Report No. 3, University of Oxford. Saini, H.S. and Aspinall, D. 1982 Effect of water deficit on sporogenesis in wheat {Triticum aestivum L.). Ann. Bot. 48, 623-633. Semenov, M.A. and Porter, J.R. 1994 The implications and importance of non-linear responses in modelling of growth and development of wheat. In Predictability and Non-linear Modelling in Natural Sciences and Economics. J. Grasman and G. van Straten (eds). Kluwer Academic Publishers, Wageningen, Netherlands. Semenov, M.A., Porter, J.R. and Delecolle, R. 1993 Climate change and the growth and development of wheat in the UK and France. Eur. J. Agron. 2, 293-304. Viner, D. and Hulme, M. 1993 The UK Met. Office High Resolution GCM Transient Experiment (UKTR). Technical Note 4 prepared for the UK Department of the Environment Climate Change Impacts/Predictive Modelling LINK, Contract Reference Number PECD 7/12/96. Climatic Research Unit, Norwich. Viner, D. and Hulme, M. 1994 The Climate Impacts LINK Project: providing climate change scenarios for impacts assessment in the UK. DoE/CRU Report, Norwich.

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

360

FORESTRY coarse-resolution general circulation model output. / . Ceophys. Res. 95(D2), 1943-1953. Willcs, D.S. 1992 Adapting stochastic weather generation algorithms for climate changes studies. Climatic Change 22, 67-84. Received 28 November 1994

Wigley, T.M.L. 1994 MAGICC Version 1.2: User's guide and scientific reference manual. NCAR, Boulder. Wigley, T.M.L. and Raper, S.C.B. 1992 Implications of revised IPCC emissions scenarios. Nature 357, 293-300. Wigley, T.M.L., Jones, P.D., Briffa, K.R. and Smith, G. 1990 Obtaining sub-grid-scale information from

Downloaded from http://forestry.oxfordjournals.org/ by guest on December 7, 2011

S-ar putea să vă placă și