International Journal of Biometeorology

, Volume 54, Issue 4, pp 423–431

Changing climate in Hungary and trends in the annual number of heat stress days

Authors

    • Adaptation to Climate Change Research GroupHungarian Academy of Sciences–Corvinus University Budapest (HAS-CUB)
  • Csaba Torma
    • Adaptation to Climate Change Research GroupHungarian Academy of Sciences–Corvinus University Budapest (HAS-CUB)
    • Department of MeteorologyEötvös Loránd University
  • Anikó Kern
    • Adaptation to Climate Change Research GroupHungarian Academy of Sciences–Corvinus University Budapest (HAS-CUB)
    • Department of MeteorologyEötvös Loránd University
  • Ákos Maróti-Agóts
    • Faculty of Veterinary ScienceSzent István University
  • Zoltán Barcza
    • Department of MeteorologyEötvös Loránd University
  • László Könyves
    • Faculty of Veterinary ScienceSzent István University
  • Olaf Berke
    • Department of Population MedicineUniversity of Guelph
  • Jenő Reiczigel
    • Adaptation to Climate Change Research GroupHungarian Academy of Sciences–Corvinus University Budapest (HAS-CUB)
    • Faculty of Veterinary ScienceSzent István University
Original Paper

DOI: 10.1007/s00484-009-0293-5

Cite this article as:
Solymosi, N., Torma, C., Kern, A. et al. Int J Biometeorol (2010) 54: 423. doi:10.1007/s00484-009-0293-5

Abstract

Global climate change can have serious direct effects on animal health and production through heat stress. In Hungary, the number of heat stress days per year (YNHD), i.e., days when the temperature humidity index (THI) is above a specific comfort threshold, has increased in recent years based on observed meteorological data. Between 1973 and 2008, the countrywide average increase in YNHD was 4.1% per year. Climate scenarios based on regional climate models (RCM) were used to predict possible changes in YNHD for the near future (2021–2050) relative to the reference period (1961–1990). This comparison shows that, in Hungary, the 30-year mean of YNHD is expected to increase by between 1 and 27 days, depending on the RCM used. Half of the scenarios investigated in this study predicted that, in large parts of Hungary, YNHD will increase by at least 1 week. However, the increase observed in the past, and that predicted for the near future, is spatially heterogeneous, and areas that currently have large cattle populations are expected to be affected more severely than other regions.

Keywords

CattleClimate changeHeat stressRegional climate modelSpatial heterogeneityTemperature humidity index

Introduction

In temperate regions, mean temperatures are expected to increase due to climate change. The increased temperature could have negative effects on agriculture (Ferris et al. 1998; White et al. 2006), biodiversity (Thuiller et al. 2005), energy sector (Smoyer-Tomic et al. 2003), hydrology (Tereshchenko et al. 2002), human (Poumadére et al. 2005; Diffenbaugh et al. 2007) and animal health.

Animal health-related phenomena attributed to climate change include the emergence of infectious diseases in new geographic areas and rising incidence in endemic regions (Wittmann and Baylis 2000; Purse et al. 2005; Casimiro et al. 2006; Gloster et al. 2007). Furthermore, climate change affects, both directly and indirectly, the health and productivity of food animals. Indirect effects are caused by changes in the nutritional environment, through the availability of livestock feeds and the quantity and quality of livestock pastures and forage crops. On the other hand, increasing temperatures due to climate change directly affect food animal production by increasing the frequency of heat stress days (Lee-Ann et al. 2008).

Lactating cows with high production levels may be very reactive to heat stress. Numerous studies have investigated the negative effects of heat stress on cattle physiology (Padilla et al. 2006). Heat stress can affect lactating cows in different ways. It may cause increased water intake and body temperature, reduced feed intake, and changes in blood hormone concentration (Armstrong 1994). These physiological changes may have effects on production (West 1994, 2003; Bouraoui et al. 2002; West et al. 2003) and reproduction (Morse et al. 1988; Ravangolo and Misztal 2002; Avendaño-Reyes et al. 2006; Morton et al. 2007) depending on the magnitude of the heat stress. By decreasing milk production, heat stress may reduce profitability in this sector. According to Welchext et al. (1965), the daily average milk production of cows shaded with slatted shades was 1.4 kg lower than cows with solid shades. Our results (Reiczigel et al. 2009) similarly showed that even a single heat stress day causes an average milk loss of 1.5–2 L per cow per day (5–10% of daily production).

To quantify the degree of heat stress, the temperature-humidity index (THI) is commonly used in human and veterinary epidemiology. The THI combines temperature and relative humidity into a single index, which is commonly used in association with heat stress in dairy cattle. Various authors have proposed different approaches to estimate THI (Thom 1959; Bianca 1962; National Research Council 1971; Yousef 1985). These approaches differ in the way temperature and humidity is weighted in the respective estimators. Bohmanova et al. (2007) proposed linking the different weighting schemes to specific environmental conditions. The authors suggest giving less weight to humidity, when THI is used to measure heat stress in areas of relatively low humidity. On the contrary, for areas with high relative humidity an index giving more weight to humidity is considered more appropriate.

With global climate change, traditional areas of cattle production may turn into areas at risk for increased heat stress, making animal production a risky business. The identification of high THI-value areas could help reduce the vulnerability of the dairy sector with respect to climate change.

The objective of this study was to investigate changes and trends in the number of heat stress days in Hungary due to global climate change. Firstly, meteorological time series were observed and analyzed. Secondly, short-term projections were compared to a reference period based on a regional climate model. All analyses paid special attention to spatial or geographic heterogeneity in the number of heat stress days.

Materials and methods

Data

Observed time series

Daily meteorological data archived in the Global Surface Summary of the Day (GSOD) dataset were retrieved from the National Climatic Data Centre (NCDC) at National Oceanic and Atmospheric Administration (NOAA). This dataset is free and allows unrestricted use in research, education, and other non-commercial activities (NCDC 2009).

The GSOD dataset contains data from 9,000 meteorological stations distributed world wide. Of those, 45 stations are located in Hungary. However, 19 stations continuously recorded daily mean temperature and mean dew point temperature data from 1973 onwards and are thus included in the present study. The daily time series data were stored in a relational database (SQLite; Owens 2006).

Climate scenarios

The climate scenarios used in this study were based on high resolution regional climate models (RCMs), where the initial and boundary conditions were taken from global climate (or general circulation) models (GCMs) utilizing different greenhouse gas (GHG) emission scenarios. The GCMs describe the interactions of the components of the whole climate system: the atmosphere, the oceans, the terrestrial and marine biospheres, the cryosphere and the land surface. The different GCMs vary in their parameterization of physical processes. The Intergovermental Panel on Climate Change (IPCC) Special Report on Emissions Scenarios (SRES) provides a comprehensive set of 40 scenarios for GHG emission sorted into four storylines. Each storyline was developed to describe the relationships between the emission driving forces. The four storylines represent different demographic, social, economic, technological and environmental developments (Nakicenovic and Swart 2000).

In this study, climate scenario data were projected from various combinations of global climate models, future greenhouse gas emission scenarios and regional models (Table 1). The GCMs used were: ARPEGE (Déqué et al. 1994), BCM (Furevik et al. 2003), ECHAM5 (Roeckner et al. 2003), HadCM3Q0 (Collins et al. 2006). In one model (ECHAM5) the A2 emission scenario was used, while all other models were based on the A1B model (Nakicenovic and Swart 2000). The following regional climate models were used to produce projected climate scenarios: CLM (Böhm et al. 2006), HIRHAM (Haugen and Haakenstad 2006), RAMCO (de Bruijn and van Meijgaard 2005), RegCM (Pal et al. 2005), RCA (Kjellström et al. 2005) and REMO (Jacob 2001).
Table 1

Properties of regional model-based climate scenarios, the results of which were used in the analyses. GCM Driving global climate model, GHG scen. greenhouse gas emission scenario, RCM regional climate model

Model No.

GCM

GHG scen.

RCM

Institute

Spatial resolution

Time interval

1

ARPEGE

A1B

HIRHAM

Danish Meteorological Institute (DMI)

25 km

1951–2100

2

BCM

A1B

HIRHAM

Norwegian Meteorological Institute (MET.NO)

25 km

1951–2050

3

BCM

A1B

RCA

Swedish Meteorological and Hydrological Institute (SMHI)

25 km

1961–2100

4

ECHAM5

A1B

RAMCO

Royal Netherlands Meteorological Institute (KNMI)

25 km

1950–2100

5

ECHAM5

A1B

RCA

Community Climate Change Consortium for Ireland (C4I)

25 km

1951–2099

6

ECHAM5

A2

RCA

Community Climate Change Consortium for Ireland (C4I)

25 km

1951–2050

7

ECHAM5

A1B

RegCM

Eötvös Loránd University (ELU)

10 km

1961–1990

2021–2050

2071–2100

8

ECHAM5

A1B

REMO

Max Planck Institute (MPI)

25 km

1951–2100

9

HadCM3Q0

A1B

CLM

Swiss Federal Institute of Technology (ETHZ)

25 km

1951–2099

One of these climate scenario datasets came from projections made by the Eötvös Loránd University, Hungary (ELU); the others were downloaded from the Regional Climate Model database of ENSEMBLES project (ENSEMBLES 2009) developed by different institutes (Table 1). Datasets used in the present study had complete spatial grid coverage over Hungary. Datasets with missing data between April and October were excluded from analysis. The climate parameters were available on a grid with 25 km spatial resolution except for the model of ELU, which has a 10 km resolution grid. From the downloaded datasets the daily mean temperature, dew point temperature and relative humidity at 2 m above ground level were used in the calculations prepared by climate data operators (CDO, Schulzweida et al. 2009).

Methods

THI calculation

Following the proposal of Bohmanova et al. (2007) it was shown (Reiczigel et al. 2009) that the most sensitive indicator of heat stress on cattle milk production in Hungary is the THI published by Bianca (1962) with a higher humidity weight:
$$ THI = \left( {0.{15} \times {T_{db}} + 0.{85} \times {T_{wb}}} \right) \times {1}.{8} + {32}, $$
where Tdb is the dry bulb temperature (°C) and Twb is the wet bulb temperature (°C). The datasets used in the current study contained the dry bulb temperature, but not the wet bulb temperature. The wet bulb temperature can be calculated in a number of ways. In the current study, the following formula (based on Magnus equation; New et al. 2000) was applied:
$$ {T_{{wb}} = {{{\left( {3 \times T_{{dp}} } \right)} + {\left( {2 \times T_{{db}} } \right)}} \over 5},} $$
where the Tdp is the dew-point temperature (°C). Because the ELU dataset does not contain dew-point temperature, for this dataset this parameter was calculated based on relative humidity (Sonntag 1990):
$$ {T_{dp}} = \frac{{\lambda \times \left( {ln\left( {\frac{{RH}}{{100}}} \right) + \frac{{\beta \times {T_{db}}}}{{\lambda + {T_{db}}}}} \right)}}{{\beta - \left( {ln\left( {\frac{{RH}}{{100}}} \right) + \frac{{\beta \times {T_{db}}}}{{\lambda + {T_{db}}}}} \right)}}, $$
where RH is the relative humidity (%). In the range −45°C to 60°C, the Magnus parameters are given as β = 17.62 and λ = 243.12°C.

Heat stress days

Following the results of Reiczigel et al. (2009), a THI threshold of 68 was applied in this study. Each day with a THI above this threshold was considered to be a heat stress day.

The number of heat stress days were summarized for each year (YNHD) of the study period for every spatial point of the data sources. For historical data, the spatial reference points were the meteorological observational units. For RCM datasets, native model gridpoints were used.

Historical trends

To analyze trends in the change of YNHD on the meteorological observational points a Bayesian binomial-normal hierarchical model was fitted:
$$ {y_i} \sim {\hbox{Binomial}}\left( {{n_i},{p_i}} \right) $$
$$ logit\left( {{p_i}} \right) = {\alpha_{j\left[ i \right]}} + {\beta_{{ }j\left[ i \right]}}{{\hbox{T}}_i} + {\varepsilon_{\rm{i}}}, $$
where j denotes observational points, i denotes years, ni is the number of days in i-th year.
Assumed distributions were:
$$ {\alpha_{\rm{j}}} \sim {\hbox{N}}\left( {0,\sigma_\alpha^2} \right) $$
$$ {\beta_{\rm{j}}} \sim N\left( {0,\sigma_\beta^2{ }} \right) $$
$$ {\varepsilon_{\rm{i}}} \sim N\left( {0,\sigma_\varepsilon^2{ }} \right) $$

When σε = 0 the model reduces to binomial, otherwise it is overdispersed (Gelman and Hill 2006).

The posterior distribution of the parameters was obtained from Markov Chain Monte Carlo simulations (MCMC). The convergence of chains was checked by the Gelman-Rubin convergence diagnostic method (Gelman and Rubin 1992).

Model fitting was performed by WinBUGS (Lunn et al. 2000) connected with R (R Development Core Team 2008) using R2WinBUGS (Sturtz et al. 2005) and coda (Plummer et al. 2008) packages.

Prediction of future changes

To assess the impact of future climate change on the evolution of YNHD one could compare historical observations with future model projections. However, global and regional climate simulations suffer from systematic biases even in the most important meteorological parameters (temperature, precipitation, humidity: see e.g., Ines and Hansen 2005; Misra 2007; Christensen et al. 2008). Consequently, it is not possible to estimate future tendencies with the combination of unbiased historical observations and biased model projections. However, following the widely used methodology of impact studies (e.g., Diffenbaugh et al. 2007), changes can be assessed within the models using historical predictions and future projections. As the model errors are expected to be similar in the present and in the future, the trends and tendencies derived purely from model data are much more realistic than those estimated based on the observations and biased model predictions. In this study, averages over a 30-year interval were compared. YNHD was predicted for each climate change scenario, and each grid point in Hungary, for the time periods 1961–1990 and 2021–2050.

All computational work was performed in R-environment (R Development Core Team 2008) using the following additional R-packages: beanplot (Kampstra 2008), gstat (Pebesma 2004), RNetCDF (Michna 2006), sp (Pebesma and Bivand 2005), splancs (Rowlingson and Diggle 2008), and SQLiteMap (Solymosi et al. 2008).

Results

Historical

The overall minimum, mean and maximum YNHD in Hungary was calculated from the observed meteorological dataset (GSOD) using parameters from the 17 observational sites (Fig. 1). The results show that the trend in YNHD is increasing. Fitting a Bayesian hierarchical binomial-normal model showed that the increase in YNHD in consecutive years ranged between 3.2% and 5.6% among the 17 observational points. The countrywide average increase of YNHD was 4.1% per year. The spatial heterogeneity of increase in YNHD is presented in Fig. 2.
https://static-content.springer.com/image/art%3A10.1007%2Fs00484-009-0293-5/MediaObjects/484_2009_293_Fig1_HTML.gif
Fig. 1

The countrywide minimum, mean and maximum of number of heat stress days per year (YNHD) calculated over the 17 sites of observed meteorological data [Global Surface Summary of the Day (GSOD)]

https://static-content.springer.com/image/art%3A10.1007%2Fs00484-009-0293-5/MediaObjects/484_2009_293_Fig2_HTML.gif
Fig. 2

The spatial distribution of YNHD increase (%) in consecutive years estimated using a hierarchical binomial-normal model based on the observed meteorological data (GSOD)

The increase in magnitude between yearly maximum and minimum suggests that not only is YNHD increasing but its variability will also increase over time (Fig. 1). This tendency is supported by data presented in Fig. 3, where the minimum, mean and maximum YNHD was calculated for every observational sites; countrywide decade averages and standard deviations were determined from these values (Fig. 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs00484-009-0293-5/MediaObjects/484_2009_293_Fig3_HTML.gif
Fig. 3

Countrywide decade average (bar) and standard deviation (whisker) of minimum, mean and maximum of YNHD at observational sites

Future

Based on regional climate scenarios, the countrywide yearly mean of YNHD calculated on the gridpoints for the two study periods were visualized using beanplots (Fig. 4). The vertical shifts in the distributions and the 30-year averages suggest that YNHD may increase in the near future. This tendency is in accordance with aforementioned results based on observed meteorological parameters. The differences and ratios of 30-year means of YNHD for the periods 2021–2050 and 1961–1990 are presented in Table 2. The increase in YNHD shows considerable variation between the climate scenarios; e.g., Model 2 predicts less then 2 days, and for Model 5 the expected increase is over 26 days.
https://static-content.springer.com/image/art%3A10.1007%2Fs00484-009-0293-5/MediaObjects/484_2009_293_Fig4_HTML.gif
Fig. 4

Beanplots representing the distributions and averages (horizontal black lines) of countrywide yearly mean of YNHD in period 1961–1990 (dark) and 2021–2050 (light) according to the climate scenarios used

Table 2

Changes in 30-year average of number of heat stress days per year (YNHD) between the two periods (2021–2050 vs 1961–1990)

Model no.

Overall mean increase

Area affected with n increased days (% of the country)

n = 1

n = 4

n = 8

n = 21

n = 28

Days

%

1

22.95

166.21

100

100

100

100

 

2

1.69

301.31

86

    

3

11.97

192.83

100

100

99

  

4

8.03

213.21

100

99

53

  

5

26.63

187.23

100

100

100

100

21

6

5.41

184.73

100

84

   

7

2.92

175.60

100

1

   

8

7.38

153.66

100

100

32

  

9

18.92

175.64

100

100

100

27

 
According to various scenarios, specific spatial distributions of differences in 30-year mean heat stress days are mapped in Fig. 5 to compare the periods 2021–2050 to 1961–1990. The proportion of the area in which increase of YNHD exceeds some threshold values (1, 4, 8, 21, 28 days) is presented in Table 2.
https://static-content.springer.com/image/art%3A10.1007%2Fs00484-009-0293-5/MediaObjects/484_2009_293_Fig5_HTML.gif
Fig. 5

Spatial distributions of differences in 30-year mean YNHD (days) comparing the period 2021–2050 to 1961–1990, based on the different regional climate scenarios

Discussion

Analysis based on observed meteorological data showed that YNHD increased between 1973 and 2008. The variability in YNHD over the last two decades also indicates a growing tendency. Taken together, these results mean that extreme events occur more frequently nowadays than in previous decades.

Although the magnitude of future increases is uncertain, the trend in YNHD based on regional climate models is also increasing. Using the same forcing GCM and the same SRES scenario but applying different RCMs can lead to significant differences (around 60%) in YNHD over a 30-year long period (Table 2). The best case scenario is No. 2, which predicts less then a 2-day increase in the 30-year mean of YNHD. However, scenario No. 5—worst case scenario—implies an increase that will extend over 26 days. More than half of the scenarios (No. 1, 3, 4, 5, 9) project a minimum of eight additional heat stress days in the near future compared to the past. However, as the future evolution of climate is poorly constrained, one model cannot be stated to be better or worse than another. Instead, analyzing several RCM outputs in parallel may be a more reliable method to predict future trends (such as in the case of state-of-the-art numerical weather prediction).

The increase in YNHD has spatial heterogeneity, which varies according to different scenarios. One scenario (No. 5) predicted that 21% of the area will be affected by at least a 4-week increase in YNHD. For two scenarios (No. 1 and 5), the measure of increase over the whole country is more than 3 weeks. According to seven scenarios (No. 1, 3, 4, 5, 6, 8, 9), at least 80% of the county will be affected by a 4-day increase in YNHD. More than half the scenarios (No. 1, 3, 4, 5, 9) show that at least half of the country will have at least an eight-day increase. All scenarios showed a similar pattern, with the south and east of Hungary being impacted more by an increase in heat stress days than the rest of the country. This is important, because the affected area has the highest cattle density in Hungary.

Inclusion of additional emission scenarios (B1, B2) would shed more light on the results. Since these scenarios assume lower GHG emissions, they are likely to predict a smaller increase in temperature than the A SRES storylines (A1, A2; Nakicenovic and Swart 2000). In this sense, the B storylines project a more favorable climate than the studied scenarios. Worst case scenario (No. 5) is very serious, with a doubling of average YNHD for the period 2000–2008. We cannot reject this scenario if we take into consideration the results of Raupach et al. (2007), who state that the emissions growth rate since 2000 was greater than for the most fossil-fuel intensive (e.g., A1FI) of the IPCC emissions scenarios developed in the late 1990s. The worst case emission scenario (A1B) used in this study is under A1FI. Based on these data, the expected future may be worse than predicted by scenario No. 5.

Given the predicted increasing number of YNHD, growing variability, as well as extreme events and uncertainty, there is a need to monitor likely heat stress conditions and, on the basis of such monitoring, to develop a forecast system to help reduce projected production losses in the dairy industry. Besides monitoring, it is also important to develop adaptation strategies to keep track of changing environmental conditions. Among the adaptation opportunities available is that of cooling the cows’ environment (Armstrong 1994; Correa-Calderon et al. 2004; Avendaño-Reyes et al. 2006; Smith et al. 2006; Adin et al. 2009). Examination of the adaptability of various cooling methods could be the subject of a further study.

Since the severity of the impact caused by heat stress of equal magnitude depends primarily on genotype (Ravagnolo and Misztal 2000), selection of heat-tolerant cattle lineages could be useful in adapting to climate change. Cattle bred in tropical climates (Bos taurus types including N’dama, Senepol, Romosinuano, Carora; Bos indicus types with zebu genetic background) possess traits that allow for the regulation of body temperature while maintaining production and reproduction during times of heat stress (Hansen 2004). At the same time, thermotolerant cattle breeds have not been subjected to selection for milk and meat production traits at the same rate as European and North American cattle breeds. For this reason, crosses between tropical and subtropical breeds improve heat tolerance in the offspring, but at the same time usually reduce production traits. Breeds adapted to tropical climates sometimes possess undesirable traits such as bad temper, late onset of reproduction, poor meat quality and shorter lactation curves (Gaughan et al. 1999; Seif et al. 1979; Hammond et al. 1996; Kosgey et al. 2005). In the European (B. taurus) and zebu type (B. indicus) cattle breeds, several genes (polygenes) are fixed, which results in different responses to environmental heat stress (Finch 1986; Malayer and Hansen 1990). Heat tolerance of the two genotypes is based on differing physiological and cellular responses. Regulation of body temperature is based on several physiological pathways. Among these genetic pathways, recent studies have highlighted the role of the HSP70 gene and gene products and differences in its expression level, as well as the role of the slick hair gene, as key elements in the cellular regulation of heat tolerance and in heat stress. However, the molecular genetic basis of heat tolerance in cattle is not well understood.

Conclusion

Apart from the indirect effects of global climate change on food animal production, e.g., through emergence of infectious diseases, further serious direct effects on animal health and production are expected from heat stress. The dairy industry is expected to be strongly affected. Based on observational data and regional climate change model output, this study found that YNHD has increased in recent decades and will continue to increase in Hungary in the near future. Several climate change scenarios investigated in this study project that at least half of Hungary will see the dairy industry affected by, on average, an additional 1 week of heat stress for the period 2021 to 2050 as compared to 1961 to 1990. However, the estimated impact of climate change on the dairy industry has a heterogeneous spatial distribution in Hungary. The largest impact is expected in those areas where dairy cattle production is most intensive at present. Food animal production, and especially the cattle industry, in Hungary should investigate adaptation opportunities in order to be prepared for the challenges of climate change with respect to increasing heat stress.

Acknowledgments

In memory of Prof. Zsolt Harnos, who initiated our work. Observation-based meteorological data were obtained from the National Climatic Data Center. Regional climate model based data (except scenario No. 7) the ENSEMBLES data were used in this work funded by the EU FP6 Integrated Project ENSEMBLES (Contract number 505539) whose support is gratefully acknowledged. Research leading to results of scenario No. 7 has been supported by the CECILIA project of the European Union Nr. 6 program (contract no. GOCE-037005).

Copyright information

© ISB 2009