Spatial sensitivity of grassland yields to weather variations in Austria and its implications for the future.

Agricultural production fulfills economic, ecological and structural functions. Despite technological advances, agricultural production remains sensitive to climate variations. In central Europe, climate change is predicted to bring more rainfall in winter, less rainfall in summer, and increased drought risk among other effects. Grassland agriculture, which is the dominant land use in Alpine regions, may be significantly affected by these climatic changes in the future. Motivated by this issue, the susceptibility of grassland yields to weather variations in Austria is empirical evaluated as a case study. The major objective of this study is to derive spatially distributed indications for climate change exposure by assessing the impacts of weather variations on past yield. It is assumed that reduced water supply during summer constitutes a threat to grassland productivity in regions that are warmer and drier already today. On the contrary, increased spring temperatures may improve grassland productivity in cooler regions like Alpine valleys, since the earlier snow melt leads to an extension of the growth period. Regression analyses are used for evaluating the relation between yearly yields and spring temperatures or water supply in summer, respectively. Water supply is thereby expressed by aggregated precipitation sums and the Climatic Water Balance (CWB). Input data are a meteorological time series as well as yearly yields available for 25 years between 1970 and 2010 and 99 districts in Austria. Yearly yields show a significant (P < 0.05) and positive dependency on water supply in summer for the eastern Austrian lowlands. The combination of temperature in spring and CWB in summer is only significant for six districts in the east of Austria. The positive impact of higher spring temperatures could not be verified. Generally, the regression coefficients are not very high, which indicates that temperature and water supply do not fully describe grassland productivity. Projected climate change may increasingly constitute a risk to yield reliability in the east of the country. That in turn, requires consideration in agricultural development plans and a quantification of these impacts from a social-economic perspective.

Introduction

Agricultural production contributes to the economic, ecologic and spatial structure of regions. Food security, biodiversity, soil loss protection, and diversity of agriculture are among the objectives of the rural development plans of, e.g., the European Commission. Despite technological advances and socio-economic measures, agricultural production is sensitive to climate variations. Climate change intensifies variations in weather conditions. Agricultural development plans need to consider potential impacts of climate change and develop mitigation strategies. The adaptation to changed conditions requires technological as well as institutional innovation as discussed in Rodima-Taylor, Olwig, and Chhetri (2012).

Predictions regarding the effects of climate change vary and the impacts cannot be precisely foreseen. Consequently, the perceptions of risks related to climate change differ. A study by Barnes, Islam, and Toma (2013) analyzed differences in the perception of climate change risk for the specific group of dairy farmers. They conclude that climate change risks need to be communicated more clearly.

The communication of climate change risk requires an understanding of potential impacts. Based on this motivation, this paper evaluates general propositions of climate change on grassland farming in Austria. Grassland farming constitutes a substantial element of the agrarian economy in Austria. Besides its general importance for agrarian economy, grassland exhibits further considerable ecological functions (Nitsch, Osterburg, Laggner, & Roggendorf, 2010). The permanent vegetation cover stores water, protects soils from erosion or nutrient loss and serves as a substantial carbon dioxide sink (Pötsch, 2009).

Given these basic functions of grassland farming, outside disturbances to the system may have far-reaching consequences. Among the parameters that may influence agricultural productivity, weather and climate are the most important factors (Eitzinger et al., 2007, 30 pp.). In addition to normal weather variations, climate change may bring about a substantial shift in weather conditions, which also affects yield reliability. Besides other effects, climate change is predicted to bring more rainfall in winter, less rainfall in summer, and an increased drought risk in central Europe (European Commission, 2013). Since grassland productivity is related to animal outputs such as milk and meat, climate and weather also affects the overall economic success and revenues gained by land owners. This especially applies to the production of milk, which exhibits a significant dependency on grassland production (Smit, Metzger, & Ewert, 2008).

In order to identify regions in Austria that might be susceptible to future climatic changes, grass yield sensitivity to weather variations is examined. For this purpose, a spatially explicit, empirical approach is used for the assessment of yearly weather influences on yearly grassland yields. Thereby, meteorological time series as well as yearly yields available for 25 years between 1970 and 2010 and 99 districts in Austria, are analyzed by means of statistical regressions. The major aim is to derive indications from the time series for location dependent yield sensitivity rather than making explicit yield forecasts. Based on the spatial intersection of yield sensitivity and grassland farming regions, possible consequences for the development of the sector are discussed.

The focus on weather related impacts on productivity has previously been proven useful in a study on grassland vegetation in Inner Mongolia (Li et al., 2013). Li et al. (2013) linked vegetation growth values as derived from remote sensing data with meteorological parameters like temperature, humidity and precipitation.

In the following section theoretical responses of yearly grassland yields to weather and possible regional climate changes are discussed. This is followed by a chapter that describes the study area and data used in the regression analysis. Further, the derivation of independent regression variables and the calculation of regressions are described. Subsequently, results are shown and discussed, including regional implications for grassland agriculture. The paper ends with conclusions and an outlook to further work.

Assumptions and hypothesis

In Europe, regions with higher annual precipitation also have higher grassland yields (Smit et al., 2008). Buchgraber and Gindl (2004) emphasize the importance of a sufficient water supply required for ideal growth characteristics of grassland. As a rule of thumb a yearly precipitation of 800 mm is stated as a minimum requirement for ideal growth characteristics. However, apart from the yearly precipitation sum, the seasonal distribution of the precipitation supply crucially affects growth rates. Bohner (2003) refers to a general growth depression during the summer season, which is caused by insufficient water supply. Significant drops in grassland productivity can usually be attributed to drier summers such as those in 2003 and 1976 (Smit et al., 2008).

Due to its warmer and drier climate, the eastern part of the country is more susceptible to changing water supply rates during summer. According to Eitzinger (2007) the transition zone from grassland to arable farming along the eastern edge of the Alps is more sensitive to a future climatic reduction of water supply. Regional climate projections show decreasing precipitation rates for the summer season in conjunction with higher temperatures (e.g., European Commission, 2009; Schaumberger, 2011; Suklitsch, Gobiet, Truhetz, Leuprecht, & Themeßl, 2007). This trend is considered to be a major threat to grassland productivity in Austria (Birnhuber, Hiess, Jiricka, Kleinbauer, & Pröbstl, 2011).

In addition, even though temperatures generally affect grassland yields to a lesser extent (e.g., Buchgraber & Gindl, 2004; Smit et al., 2008), low temperatures in spring restrict the yields in cool upland regions (cf. Bohner, 2003). Low temperatures during the initial phase of plant growth may shift and therefore reduce the growth season. Thus, in contrast to water supply, the climate change trend of increasing temperatures may have a positive effect on grassland yields (Balas, 2009). This tendency is especially apparent in Alpine regions, which already exhibited an extension of the growing season in the last decades. Field experiments in the Alpine Enns Valley showed a 3 week extension of growth periods between 1987 and 1994 (BMWFJ, 2009), whereby both an earlier growth initiation and a longer growth period was observed. This also led to intensification of the harvesting frequency, from three to four cuts in valleys and two to three cuts in upland regions. High spring temperatures in conjunction with an earlier snow melt are considered as the main reason.

Accordingly, temperatures in spring and water supply during the summer season are considered as the major driving factors of yearly yield variability in this study. The selection of the methodology for the evaluation of the spatial sensitivity of grassland yields to weather variations is based on these assumptions.

The hypothesis of this study is that there are areas in Austria where yields are significantly linked to precipitation and temperature. Areas showing a linkage between yields and meteorological conditions are assumed to be more sensitive to changes in these conditions than other regions. It is expected that the eastern region of Austria will be identified as the area most sensitive to variations in summer water supply, whereas the western region may exhibit a higher dependency on spring temperatures due to the cooler Alpine climate.

Study area and data

Austria is located in central Europe; the western part of the country is covered by the Alps and in the north and east of the country are lowlands. The variations in topography have effects on the suitability of areas for agricultural production. Grassland production is predominant in Alpine regions and in areas bordering with the Alps. Grassland production plays an important role in the agrarian economy of Austria. About 7 million tons of dry matter are harvested each year, which serves as a basic food resource for 2.5 million animals (Buchgraber & Schaumberger, 2006). Permanent grassland occupies an overall area of 1.73 million hectares (Seidelberger et al., 2009), which accounts for about 50% of the agricultural land of the study area (Schaumberger, 2005).

The assessment of impacts on grassland productivity is based on a spatially interpolated meteorological time series. The European Climate Assessment & Dataset Project provides freely available gridded observational data (termed E-OBS) for daily minimum temperature, daily maximum temperature, daily mean temperature and daily precipitation sum in NETCDF format for the time period 1950–2012. Currently, the newest release (Version 8.0) achieves a spatial resolution of 0.25° (regular grid) interpolated from 2316 weather stations, which are distributed over an area of about 106 km2 (Haylock et al., 2008).

In contrast to most standard grid and image formats, the NETCDF data format is especially suitable for the storage of multidimensional data. The datasets used in the presented analysis comprise the variables longitude, latitude, time and an additional variable, which stores the meteorological value. This value is described by a multidimensional index based on longitude, latitude and time. Thus, the data structure allows for retrieving single elements as well as time slices within the three dimensional vector. In order to speed up processing times, the original size of the datasets was reduced to a bounding box encompassing the Austrian border. Moreover, the temporal dimension was adapted to the yearly grassland yield statistics, which was requested from Statistics Austria for the time period 1970–2010. For the slicing of the NETCDF files, a freely available collection of command-line programs called NETCDF Operators was used. Those smaller data chunks derived from the original datasets eventually exhibited an extent of 13 latitude raster cells by 34 longitude raster cells by 14,975 days, respectively.

The yield statistics refer to the absolute yearly yields in quintals (100 kg) of dry matter as well as area-specific yields in quintals per hectare for each Austrian district. Due to the lack of continuous yield records in some of the districts, the original number of 121 districts was reduced to 99 districts. The statistics obtained distinguishes between meadows harvested once and meadows harvested at least twice a year. Since some environmental programs exist which encourage extensive utilization of meadows with subsidies (Isselstein, Jeangros, & Pavlu, 2005), grassland productivity may be below maximum yield limits on some extensively used meadows. Due to the intention of evaluating these limits and to minimize the biasing effect, the extensive category (harvested once) as well as special types such as meadows with scattered fruit trees or pasture was excluded from further analyses.

On the contrary to arable crop yields, grassland yields are not directly measured. Grassland yields are estimated by about 2000 volunteers (mostly farmers) in Austria. The estimation of grassland dry weight is based on measured round bale diameters which are converted to quintals per hectare using standardized conversion tables. Due to the high number of volunteers, about 85% of the permanent grassland areas can be covered. Subsequently, these reference areas are extrapolated to the entire municipality by weighting the samples as a function of estimated sample area. In order to guaranty data accuracy, the yield estimations are also validated by the comparison with further samples taken by experts (Bader, Personal communication).

Methodology

The dependency of yearly grassland yields on yearly weather variations is evaluated using statistical regression analyses. The water supply in the summer months June, July and August (JJA) was selected as an independent variable. Water supply was considered as the Climatic Water Balance (CWB), which is defined by the difference between precipitation sums and potential evapotranspiration (Häckel, 2005). Moreover the yearly mean temperatures in the spring months March April and May (MAM) and precipitation sums in JJA were utilized as further independent regression variables.

Tools

For the calculation of the independent regression variables from the multidimensional NETCDF datasets, the programming language Python was used. The Python module NETCDF4 enabled us to import and read the files in the script. However, in order to perform algebraic operations with reasonable processing times, a conversion to so called Numpy Arrays was required. Numpy is an open source library for scientific computing, which contains a long list of useful mathematical functions. In order to assign the derived raster values to the respective districts (polygons), the ArcGIS library for Python (ArcPy) was imported to the script. ArcPy enables an access to a multitude of operations for the analysis and manipulation of spatial data. Moreover, the software SPSS was used for the calculation of statistical regressions.

Calculation of independent regression variables

One essential component of the CWB is the potential evapotranspiration. In the presented model, potential evapotranspiration was expressed by the so called reference crop evapotranspiration (ETo). The calculation of ETo was based on the Food and Agriculture Organization of the United Nations (FAO) guidelines introduced by Allen, Pereira, Raes, and Smith (1998). The guidelines refer to a reference crop surface, which is not short of water. Thereby, the reference surface is defined as grassland with hypothetically assumed characteristics such as a crop height of 0.12 m, a fixed surface resistance of 70 s m−1 and an albedo of 0.23. The surface resistance implies a moderately dry soil surface resulting from a weekly irrigation frequency.

This approach obviates the need to define a separate evapotranspiration level for each stage of growth. The only factors affecting ETo are climatic parameters. In the FAO guidelines the widely used and calibrated Penman–Monteith method is considered as the most suitable approach for calculating ETo. However, the comprehensive set of required spatiotemporal data on air humidity, wind speed or barometric pressure could not be acquired for the study area. Thus, the alternatively suggested Hargreaves equation was chosen (see equation (1)). Schaumberger, Eitzinger, and Formayer (2008) conducted a regression analysis to compare these approaches. The results showed an R2 of 0.79 and a slope of 0.95. Given these key figures, the approach was considered to be appropriate for the purpose of this study.

The variables Tmean, Tmax and Tmin in equation (1) refer to the daily mean, maximum and minimum temperatures in °C. The variable Ra is the daily extraterrestrial radiation [mm day−1] expressed by equation (2):where Gsc is the solar constant (0.082 MJ m−2 min−1), dr is the inverse relative distance Earth–Sun, ωs is the sunset hour angle [rad], φ is the latitude [rad] and δ is the solar declination [rad].

The inverse relative distance Earth–Sun (dr) and the solar declination are given by equations (3) and (4):where J is the number of the day in the year between 1 and 365 or 366, respectively. In addition, the sunset hour angle is given by equation (5):

The result of Ra is in MJ m−2 min−1 and had to be converted to the corresponding equivalent evaporation by multiplying Ra by 0.408 before it was inserted in equation (1). Subsequently ETo was subtracted from the daily precipitation sum to get the daily CWB [mm day−1]. These daily water balances were summed up for the summer season of each year between 1970 and 2010. The same simple arithmetic operations were applied for the calculation of seasonal mean temperatures in spring and precipitation sums in summer.

Since yield statistics have been obtained for each district, the aggregated raster values had to be adjusted to these administrative units. In order to assign the derived values to districts, an area weighted mean based on the spatial overlap between district polygon and raster cells was calculated. Thereby, the influence of a certain raster value on the calculated district mean value of the respective year is proportional to the overlap area.

Regression analysis

Regression analyses are intended to estimate a dependent variable by one or more explanatory (independent) variables. In this study the yearly aggregates such as the CWB in JJA, precipitation sums in JJA or mean temperatures in MAM are utilized as independent variables, in order to assess their contribution to the variance of the dependent variable of yearly grassland yields. Thereby, the statistical analysis encompasses three different regression models, which were applied to each of the 99 districts considered. This comprises linear regressions of yearly yields on the yearly precipitation sum in JJA, yearly yields on the yearly CWB in JJA and a multiple linear regression of yearly yields on the yearly CWB in JJA and mean temperature in MAM.

The significance check in the linear regression analysis was based on an F-test. In case of the multiple linear regression, the significance of single independent variables was additionally assessed by a t-test. For the selection of variables in the multiple regression a stepwise approach was applied. Thereby, each variable is entered in sequence and its value assessed. If the respective variable contributes to the model, it is retained. Other variables are simultaneously re-tested if they still contribute to the success of the model. If they no longer significantly contribute they are removed (Brace, Kemp, & Snelgar, 2006). The significance is evaluated at the 5% level.

Results

The performed linear regression analyses showed significant dependencies of grassland yields on precipitation in summer and CWB in the east of Austria. The combination of spring temperature and CWB in summer in a multiple regression resulted in higher explanatory power of regressions. However, only in a few districts exhibited significant dependencies. Regressions showing significant dependencies only on spring temperatures are scattered over the entire study area without any spatial clustering. While variables in most regressions were positively related to yearly yields, regressions based solely on spring temperature showed negative dependencies. A summary of all significant regression results can be found in the Appendix A.

In the following sections, the obtained results are described in detail. In order to show both, the spatial distribution of correlations and the value distribution, maps as well as diagrams were generated for each regression model.

Precipitation in summer and yearly yields

The regression analyses of yearly yields on the yearly precipitation sum in JJA resulted in a clear spatial clustering of significant districts along the east and south Alpine foothills (see Fig. 2). Only 26 of the 99 analyzed districts exhibited a significant dependency at the 5% level. Further 6 districts had to be excluded from this analysis due to a lack of complete precipitation records for the selected time period.

Fig. 2

Coefficient of determination (R2) of yearly yields on the yearly precipitation sums in JJA per district (labeled with official district abbreviations).

The distribution of yearly yield and precipitation values considered in this regression is exemplarily represented for the district HB (see Fig. 1), which has the highest correlation on precipitation in summer of all districts considered. The same dependency was analyzed separately for each district. The graph in Fig. 3 summarizes the regression analysis results for all districts with significant correlation coefficients. The graph shows the values of the correlation coefficients as well as the positive or negative relation between yields in the districts and precipitation totals in JJA. Districts with significant dependencies exhibited R2 values ranging from 0.16 to 0.46. Slopes of the linear regression functions vary between −0.07 and 0.1, whereby only district HL shows a negative correlation.

Fig. 1

Regression of yearly yields on precipitation sums in JJA for district HB.

Fig. 3

Coefficient of determination (R2) and slope of the linear regression of yearly yields on the yearly precipitation sums in JJA per district (labeled with official district abbreviations).

The spatial visualization of R2 values clarifies several noticeable coherences (see Fig. 2). The district exhibiting the highest R2 (HB) is located in the center of significant districts, whereas the lowest values such as VL, ZT, KL or HL are located at the edge of this spatial cluster. In addition, the only district exhibiting a negative dependency (HL) is located in the outer north of the country. Thus, the cluster of significant districts depicts a peak in the center and a tendency of decreasing values towards the edge.

Climatic Water Balance in summer and yearly yields

The regression of yearly yields on the CWB in JJA showed a similar spatial pattern of significant dependencies as the regression on precipitation sums (see Fig. 4). However, the CWB in JJA allowed for the derivation of additional significant regressions. The regression model resulted in 35 significant districts, whereas 10 districts had to be excluded due to missing meteorological records.

Fig. 4

Coefficient of determination of yearly yields on the Climatic Water Balance in JJA per district (labeled with official district abbreviations).

The cluster extended to the north towards district ZT. Moreover, the spatially outlying districts LZ and RI exhibited significant dependencies. Even though the overall R2 values increased compared to the precipitation regression, RI and LZ, as well as districts located at the edge of the main cluster, again showed the lowest values.

In contrast to the precipitation regression model, all significant district regressions of yearly yields on the CWB indicated positive dependencies (see Fig. 5). District HB is again an outlier with an R2 value of 0.58. The lowest R2 of 0.16 was calculated for district KF. Slopes are ranging from 0.03 to 0.1, whereby higher coefficients are generally associated with steeper regression slopes.

Fig. 5

Coefficient of determination and slope of the linear regression of yearly yields on the Climatic Water Balance in JJA per district (labeled with official district abbreviations).

Spring temperature, Climatic Water Balance and yearly yields

For the multiple linear regressions, CWB in JJA was supplemented with mean temperatures in MAM as an additional independent variable. Only 6 districts exhibited significant multiple regressions based on both variables. In further 8 districts the null hypothesis was rejected for mean temperatures in MAM, whereas yields in 29 districts exclusively depend on CWB in JJA. Moreover, 10 districts were excluded due to missing meteorological records (see Fig. 6).

Fig. 6

Coefficient of determination of yearly yields on the Climatic Water Balance in JJA and mean temperature in MAM per district (labeled with official district abbreviations).

The incorporation of the temperature variable resulted in a second peak of R2 located close to the Slovenian border. Regressions based on both variables showed higher values of R2 ranging from 0.38 (RA) to 0.63 (VK), whereby the CWB as well as mean temperatures are positively related to yearly yields.

Contrarily to the assumption discussed in the hypothesis chapter, all significant statistical regressions, only associated to mean temperatures in MAM, were negatively related to yearly yields. This means low mean temperatures were related to high yields and vice versa, even though relatively low values of R2 ranging from 0.17 to 0.28 suggested weak dependencies (see Fig. 7). The regression slopes, which vary between −2.4 and −8.5, indicate a pronounced indirect dependency. For instance, the regression derived for district BL (district with highest R2 value) estimates a reduction of yearly grassland yields of 8.5 quintals per hectare for a temperature increase of 1 °C in MAM. The variation of considered time periods such as mean temperatures in March or March and April did not crucially change this rather counter intuitive outcome.

Fig. 7

Coefficient of determination and slope of the linear regression of yearly yields on the mean temperature in MAM (labeled with official district abbreviations).

Discussion

Sensitivity of yields to weather variations

The regression results based on hydro-climatic variables revealed a distinct clustering of significant districts in the eastern part of the country. This is probably due to the general tendency of decreasing precipitation rates towards the east. Whereas summer seasons with lower precipitation sums in western regions do not fall below the required water supply thresholds, varying summer precipitation significantly affects yearly yields in eastern districts. In system theory, this behavior is explained by homeostasis. The system is able to withstand outside disturbances up to a certain magnitude. As long as this outside disturbance remains within a certain range, the homeostatic processes maintain control (Ford, 2010). The climate of the considered time period (1970–2010) is already close to the upper boundary of homeostatic influence. In addition, the incorporation of modeled reference crop evapotranspiration in the CWB substantially increased the fit of the model. This implicitly includes the impact of summer temperatures and also points out that higher temperatures lead to lower yields. Thus, further warming of the summer season in conjunction with lower precipitation rates may increasingly disturb yield reliability in the future in eastern districts. Therefore, the assumption that this region is more sensitive to a reduction of summer water supply was confirmed by the results obtained (see chapter 2).

Contrarily to the hydro-climatic variables, temperature variations in spring do not allow for an unambiguous conclusion. Whereas temperature variables in multiple regressions are positively related to yearly yields, regressions exclusively based on temperature showed an unexpected negative relation. Beyond that, only a few districts, which exhibit a rather arbitrary spatial distribution, show a significant temperature dependency. Thus, the assumption of yield sensitivity to spring temperatures could not be clearly evaluated by the results obtained (see chapter 2). This may also be caused by the height distribution of the grassland type considered in this study (harvested at least twice). While intensive grassland farming is usually restricted to valleys and lowlands, cool temperatures in spring are assumed to influence yearly yields especially in extensively used upland regions (Schaumberger, 2005). Consequently, an analysis of grassland areas harvested once would probably display stronger dependencies on spring temperatures than the land use type being considered. The negative correlation of intensive grassland might be explained by the reduction of evapotranspiration associated with lower temperatures, which leads to less drought stress and thus improves grassland productivity.

Implications of changing yield reliability

Possible effects of changes in grassland yield reliability strongly depend on regional agricultural specializations. In Austria grassland agriculture is predominant in Alpine valleys and along the foothills north and northeast of the Alps where climatic conditions aren't conducive to crop farming. The formation of grassland monoculture in these regions exacerbates the susceptibility to climate variations, since agrarian economy heavily relies on grassland biomass as a forage resource. An adaptation to changing environmental conditions by diversifying the agricultural specialization is in many places limited by the topography. Crop farming on steep slopes would lead to soil nutrient loss and high erosion rates. Thus, grassland regions sensitive to a reduced water supply such as those northeast of the Alps (e.g. districts AM, SB, PL) may be increasingly exposed to these problems (see Fig. 8).

Fig. 8

Grassland distribution and regions with significant yield dependency on CWB in summer.

Possible effects are the agricultural marginalization of these regions associated with land abandonment and afforestation. These processes are most likely to be found in extensive farming regions and those where small-scale farming is prevalent (Brouwer, Baldock, Godeschalk, & Beaufoy, 1997). Although farm productivity and average farm size significantly increased over the last decades (Weiss, 1998), grassland farms are still small size compared to farms in neighboring countries such as Switzerland and Germany which exhibit about twice the size of Austrian farms (Ortner, Hambrusch, & Kirner, 2006). The still ongoing process of structural change may even compensate the decreasing grassland productivity by increasing production areas in grassland agriculture (Eitzinger, 2007).

Lower harvesting yields due to possible increase in water shortage may also lead to a detensification of agriculture as discussed in Olesen and Bindi (2002). One the one hand side this detensification is driven by economic considerations. One the other hand, however, also the reduced livestock carrying capacity of affected areas acts as a crucial ecological constraint in this system (cf. McKeon et al., 2009). As a consequence of possible detensification or marginalization, land use alternatives become more and more important. Birnhuber et al. (2011) pointed out the usage of unprofitable grassland areas for the production of renewable energy by means of photovoltaic plants.

In order to sustain current land use in regions affected by a drier climate, considerable adaptation efforts would be required. Subsidies, which are already today needed to ensure farming in less favored areas in Austria (Schmitzberger et al., 2005), have to be adjusted to the changing environmental conditions. However, current trends of agricultural policy changes on a national and European level are rather aiming at fostering competition among farms. Although direct payments in Austria for small-scale farms are slightly higher and compensate for natural disadvantages, the decrease in intervention prices for dairy products and the decoupling of subsidies from milk production leads to an increasing pressure on grassland farms to become more efficient (Ortner et al., 2006). However, the options to improve the economic efficiency might be more and more restricted by natural limitations.

In order to stay competitive, the adoption of new farming techniques may gain increasing importance. Limited water supply can, for instance, be compensated by irrigation. The irrigation of grassland is relatively common in several regions in the Netherlands, Italy and UK (Wriedt, Van der Velde, Aloe, & Bouraoui, 2008). However, the willingness of grassland farmers to adopt grassland irrigation techniques is eventually a matter of local irrigation costs (Eitzinger, 2007).

Methodology transfer and data limitations

Permanent grassland occupies a large proportion of the European agricultural area including about 65% of the British, 67% of the Alpine and about 37% of Mediterranean and East European agricultural areas (Olesen & Bindi, 2002). Consequently, changing climatic conditions and the understanding of potential impacts on this land use is also relevant on a larger European scale. Thus, the presented contribution can be seen as case study for regions facing similar challenges. Apart from the yield statistics, which are ascertained separately for each country, all data being used covers the entire European Union. Thus, the methodology presented in this case study is widely transferable to other European regions. A comprehensive discussion of data available for studying agricultural systems across Europe is given by Janssen, Andersen, Athanasiadis, and van Ittersum (2009). Potential problems and limitations associated with the transnational data being used in this study can be concluded as follows:

The coarse spatial resolution of the meteorological data might be a source of error. The resolution of 0.25° does not perfectly represent weather variations in regions with complex alpine terrain. In addition, the spatial interpolation is only based on 2316 stations. For the given spatial resolution and area covered by the dataset, about 16,000 stations would be required for an accurate ascertainment of meteorological conditions (Haylock et al., 2008). Despite of these inaccuracies, the interpolation of the E-OBS data is based on the largest available pan-European dataset (Hofstra, Haylock, New, & Jones, 2009). Thus, no data with higher network density is currently available on a European scale.

Furthermore, the yield estimates may differ from the actual yields. Actual grassland yields can only be ascertained by weighing the harvested biomass. Due to the high efforts associated with this procedure, sensor based methods are currently discussed. Fricke, Richter, and Wachendorf (2011), for instance, validated the method of grassland biomass assessment by ultrasonic devices. Unfortunately, reliable sensor based field-measurements are still restricted to test areas and not available for extensive areas in Austria. Advances in remote sensing technology might help to overcome this problem in the future. For now, yield estimates, as described in chapter 3, are the only data source available.

In addition to errors caused by data inaccuracy, several factors of influence were neglected in this study. Soil types and textures, for instance, significantly influence water storage capacities and therefore determine the ability to compensate dry periods. Consequently, the spatial pattern of yield sensitivity is not only a matter of climate, but is overlaid by the hydrological properties of soil. In contrast to this temporally static factor, different parameters of agricultural management such as the utilization of fertilizers may have changed throughout the considered time period. This factor was neglected due to a lack of reliable time series. Moreover, increasing atmospheric CO2 concentrations directly enhances plant productivity and also increases resource efficiencies by a reduction of stomatal conductance and thus transpiration (Olesen & Bindi, 2002). This indirect effect of anthropogenic climate change was neglected due to uncertainties still given for the non-linear response of grass species to changes in CO2 (cf. Tubiello, Soussana, & Howden, 2007).

The chosen data and methodology is considered to be appropriate for the derivation of regional sensitivity patterns, however, doesn't enable the forecast of future grassland yields. Nevertheless, the results of this study improve the understanding of spatial differences in the susceptibility of grassland agriculture to possible climate variations.

Conclusion and outlook

The hypothesis of susceptibility to the water supply in summer of eastern districts could be confirmed, whereas anticipated dependencies on spring temperatures were not apparent from the results obtained. The clustering of significant effects of variations in water supply in the climatically drier and warmer eastern region suggests a dominance of the climatic regime compared to factors such as the hydrologic properties of soil, or grassland management.

Short term weather variations affect yearly yields, but the climate determines the disposition of a region. Thus, the climate change prediction of decreasing precipitation sums in summer is on the whole likely to exacerbate the situation in areas which are already susceptible to drier summer seasons. In addition, the higher explanatory power of the CWB compared to regressions based on precipitation sums underlines risks associated with increasing evapotranspiration rates due to higher summer temperatures.

Districts in the northeast of the Alps are exposed to the highest risk, since results suggest sensitivity to a reduction of water supply in summer for this region with predominant grassland agriculture. Problems associated with a possible reduction of competitiveness might even be exacerbated by current agricultural policies, which increase the pressure on grassland farms to become more efficient. Possible options of adaptation are the change to an alternative non-agricultural land use or the adoption of new farming techniques such as the irrigation of grassland.

In order to assess the impacts of climate change on the grassland farming sector, a more comprehensive system evaluation is required. Therefore, an integrated impact assessment is planned that studies the influence of climate and the response of humans with the objective of adapting to a changing environment. Integrated approaches have previously been suggested for example in the context of reducing uncertainties in evaluating climate impacts on agriculture and water (Falloon & Betts, 2010).

Bringing together social and physical factors in a model is a challenge. According to Câmara, Vinhas, Davis, Fonseca, and Carneiro (2009 p. 213, 16 pp.) “[t]here is no proven scientific paradigm for human-environmental modeling”. The anticipated model will be based on spatial system dynamics (SSD), which is considered as an appropriate approach for combining the physical and social worlds. The focus of that dynamic model will be on simulating general mechanisms in agrarian grassland systems to assess the regional exposure to land abandonment and land use changes.

Table 1

Summary of significant regression results.

Districts	Prec. JJAa		CWB JJAb		CWB JJA/Temp. MAMc
	Coef. (R)	Prob.	Coef. (R)	Prob.	Coef. (R)	Prob.
AM	0.505	0.010	0.520	0.008	0.520e	0.008
BN	0.458	0.021	0.514	0.009	0.514e	0.009
BZ	–	–	–	–	−0.492f	0.013
BL	–	–	–	–	−0.525f	0.007
DL	0.548	0.005	0.584	0.002	0.584e	0.002
E	0.465	0.019	0.543	0.005	0.543e	0.005
EU	–	–	0.475	0.016	0.475e	0.016
FB	0.466	0.019	0.544	0.005	0.544e	0.005
FK	–	–	–	–	−0.446f	0.025
FE	–	–	0.564	0.010	0.564e	0.010
FF	0.560	0.004	0.663	0.000	0.663e	0.000
GU	0.413	0.040	0.480	0.015	0.480e	0.015
GS	0.584	0.002	0.670	0.000	0.670e	0.000
HB	0.675	0.000	0.761	0.000	0.761e	0.000
HL	−0.399	0.048	–	–	−0.431f	0.031
JE	0.583	0.002	0.631	0.001	0.631e	0.001
KL	0.422	0.035	0.475	0.016	0.654d	0.002
KF	–	–	0.398	0.049	0.398e	0.049
KR	–	–	0.422	0.036	0.422e	0.036
LB	0.481	0.015	0.538	0.006	0.649d	0.002
LE	–	–	–	–	−0.481f	0.015
LZ	–	–	0.414	0.039	0.414e	0.039
LF	0.468	0.018	–	–	–	–
L	–	–	–	–	−0.410f	0.042
MA	–	–	0.425	0.034	0.425e	0.034
ME	–	–	0.413	0.040	0.413e	0.040
MD	0.485	0.014	0.552	0.004	0.552e	0.004
NK	0.563	0.003	0.684	0.000	0.684e	0.000
OP	–	–	0.474	0.017	0.474e	0.017
OW	0.499	0.011	0.604	0.001	0.732d	0.000
RA	–	–	0.437	0.029	0.614d	0.005
RI	–	–	0.403	0.046	0.403e	0.046
PL	0.405	0.045	0.476	0.016	0.476e	0.016
SB	0.518	0.008	0.479	0.021	0.479e	0.021
SZ	–	–	–	–	−0.440f	0.028
VI	0.522	0.007	0.527	0.007	0.527e	0.007
VL	0.404	0.045	0.420	0.036	0.420e	0.036
VK	0.482	0.015	0.469	0.018	0.792d	0.000
WT	–	–	0.402	0.046	0.632d	0.004
WZ	0.476	0.016	0.538	0.005	0.538e	0.005
WL	–	–	–	–	−0.410f	0.042
WB	0.562	0.003	0.635	0.001	0.635e	0.001
WO	0.461	0.020	0.467	0.019	0.467e	0.019
ZT	0.396	0.050	0.501	0.011	0.501e	0.011

Linear regression of yearly yields on precipitation sums in June, July and August.

Linear regression of yearly yields on the Climatic Water Balance in June, July and August.

Multiple regression of yearly yields on mean temperature in March, April, May and climatic and water balance in June, July and August.

Significant dependency on Climatic Water Balance and temperature.

Significant dependency on Climatic Water Balance.

Significant dependency on temperature.