Analysis of heat stress in UK dairy cattle and impact on milk yields

Much as humans suffer from heat-stress during periods of high temperature and humidity, so do dairy cattle. Using a temperature-humidity index (THI), we investigate the effect of past heatwaves in the UK on heat-stress in dairy herds. Daily THI data derived from routine meteorological observations show that during the summer, there has been an average of typically 1 day per year per station over the past 40 years when the THI has exceeded the threshold for the onset of mild heat-stress in dairy cattle. However, during the heatwaves of 2003 and 2006, this threshold was exceeded on typically 5 days on average in the Midlands, south and east of England. Most dairy cattle are in the west and north of the country and so did not experience the severest heat. Milk yield data in the south-west of England show that a few herds experienced decreases in yields during 2003 and 2006. We used the 11-member regional climate model ensemble with the A1B scenario from UKCP09 to investigate the possible future change in days exceeding the THI threshold for the onset of mild heat-stress. The number of days where the THI exceeds this threshold could increase to over 20 days yr−1 in southern parts of England by the end of the century.


Introduction
Animals, like humans, suffer in conditions of extreme heat and humidity. For agricultural animals, extreme weather can impair productivity as well impacting animal welfare. The effect on dairy cattle can be relatively easily determined as with twice-daily milking, any drop in yield can be quickly identified, and has an immediate effect on the income generated. We therefore focus on heat-stress in UK dairy herds.
It has been known for over 50 years that dairy cattle are particularly susceptible to heat-stress due to high temperature and humidity (Johnson et al 1963). Measurable physiological consequences include: reduced dry-matter intake, rate of weight gain, fertility of dairy cattle (both sexes) and also milk yield as well as eventual mortality (see e.g. Silanikove 2000, West et al 2003, Jordan 2003, Boonprong et al 2008, O'Brien et al 2010, Crescio et al 2010. These effects reduce the productivity of the herd, with consequences for economic viability (St-Pierre et al 2003). The susceptibility of an individual herd depends on the breed and genetic merit of the herd, with the milk yield of dairy cattle of high genetic merit (usually those selected for high milk yield) being proportionately more affected by heat-stress (e.g. Bryant et al 2007). Some cattle breeds appear to be better acclimatized to warmer temperatures, with some beef cattle adapted to sub-tropical climates better able to withstand prolonged periods of mild heat-stress (Boonprong et al 2008).
In the UK, dairy farming is concentrated in south-west Wales, south-west England, western Midlands and south-west Scotland (figure 1: DEFRA 2008). These hilly, rainy areas are more suited to dairy cattle than to large scale arable farming which dominates the flatter, drier areas in the south and east. Around 90% of dairy cattle are Holstein or Friesian breeds and their crosses. Only about 5% of UK dairy cattle are kept in continuous housing (reared exclusively in barns), but during the winter most dairy herds are kept indoors (British Society of Animal Science (BSAS) 2011). In some areas of the world cattle are kept in barns for longer periods of time, and cooling systems are used to limit heat-stress during the summer (e.g. Smith et al 2012).
A decline of approximately 19% in dairy cattle numbers has been countered by an 18% increase in milk yield per head in the UK between 2001/2002 and 2011/2012, leading to only a 4% decline in total milk production (DairyCo 2012). However this increase yield imposes a higher energy burden on the cow. The cows need to dissipate more heat and hence are more susceptible to heat-stress, along with other health problems (West et al 2003, Oltenacu andBroom 2010).
In this paper we assess the current occurrence of heatstress days for dairy cattle and assess the impacts on milk yields. We then investigate potential future impacts on milk yields in a warming world scenario. We outline the temperature-humidity index (THI) used to monitor heat stress and the hourly meteorological data in sections 2 and 3, and present the results in section 4. Milk yield data for herds in the south-west of England are analysed in section 5, along with climate projections in section 6. A discussion of all the results is in section 7.

THI
A common measure of heat-stress is THI, developed initially for humans by Thom (1958) and extended to dairy cattle by Berry et al (1964). This combines air temperature and relative humidity: lower temperatures at high humidity give similar heat-stress to higher temperatures at lower humidity. Mader et al (2006) extended the THI to include wind speed and solar radiation, to improve its effectiveness as a proxy for the heatstress experienced by (beef) cattle. They found that for each 1 m s −1 increase in wind speed, the THI can be reduced by 1.99 units, and for each 100 W m −2 decrease in solar radiation, the THI can be reduced by 0.68 units. So Gaughan et al (2008) incorporated wind speed, but not solar radiation, into a heat load index and a related accumulated heat load, obtaining better predictions of animal stress than provided by THI and its time-integral. Solar radiation is a particular concern for dark cattle (Robertshaw 1985, Busby and Loy 1996), and Eigenberg et al (2005 have demonstrated the physiological benefits of shading cattle during summer heat. However solar radiation implicitly influences the basic THI because THI and solar radiation are positively correlated (e.g. Mader et al 2006).
Unlike these studies, our work is not a controlled experiment with dedicated meteorological observations: we use routine data from operational weather stations (section 3). These do not in general include solar radiation, and their wind speeds may be unrepresentative of that experienced by dairy cattle, because wind speeds are more dependent on local topography than are temperature and humidity. So we limit our analysis to the effects of temperature and humidity, assuming moderate sunshine and calm conditions. The THI used in this study is:  (NRC 1971), where T is the temperature in°C and RH is the relative humidity in %. Alternative formulae for THI give similar results (Dikmen and Hansen 2009). Those formulae which place more weight on the humidity work better in humid climates, whereas in drier climates, those which place more weight on the temperature work best (Bohmanova et al 2007). Some authors use daily average temperatures and humidities, others use the maximum temperature and the minimum humidity to calculate the most appropriate THI . Some studies use a 3 day running mean of THI, which allows the effect of prolonged periods of heat-stress to be captured (West et al 2003, Jordan 2003. The 3 day mean is based on the measurement for the 24 h in question, along with two 24 h periods preceding it. Work by Holter et al (1996) has suggested that the reduction in dry matter intake depends more on the minimum THI experienced by the cow than the maximum THI. This corresponds to work in humans, where night-time temperatures (T min ) have a greater impact on mortality and morbidity than day-time temperatures (T max ) (e.g. Laaidi et al 2011, Valleron andBoumendil 2004). A number of studies have correlated THI with physiological measures of cattle stress including the level of panting (Mader et al 2006), dry-matter intake, and milk temperatures and yield (West et al 2003) to determine the threshold for the onset of heat-stress. Guidance by the US Government sets thresholds of THI for heat-stress levels (as defined by equation (1)) at 72-78 for mild heat-stress, 78-89 for severe stress and 89-98 for very severe stress. An animal experiencing a THI of 98 or more would not live for very long (Chase 2006). However, the appropriate thresholds will depend on the formulation of the THI. Geographical factors, such as solar radiation intensity, and biological factors such as the different activity levels and breeds of cattle will also affect the thresholds.
There is also evidence for cattle acclimatization. Bryant et al (2007) found that the onset of the effects of heatstress in New Zealand cattle occurred at lower THI values than in the USA. The difference in response of humans to the same weather in different parts of the world (e.g. Duncan andHorvath 1988, Nielsen et al 1993) is well-documented, so acclimatization of cattle is no surprise. Further concerns regarding THI thresholds arise from the continued genetic selection of cattle for milk yield. It is has been suggested that the onset of heat-stress for modern Holstein cattle is at a lower threshold than those in earlier studies, perhaps as low as 65-69 (Bouraoui et al 2002, Bryant et al 2007, Zimbelman et al 2009. In this work, we assume that the response of UK dairy cattle to different THI levels is the same as in the USA. However, further research would need to be done to confirm this assumption, especially given Bryant et al (2007) results.
Some cattle live mainly in large barns rather than outdoors. A number of studies have investigated the relationship between the THI outside and inside the barn, with a view to developing cooling methods in high temperatures. The difference between temperature and relative humidity inside and outside cattle barns changes with the seasons, and also depends on the barns' construction (e.g. Seedorf et al 1998). In general, the temperature is higher indoors (3-5°C for northern Europe), but the relative humidity varies, depending on the external temperature (Seedorf et al 1998, Erbez et al 2010. The resulting effect on THI is therefore complex, but it is on the whole higher indoors than outdoors. In the USA, some barns are equipped with cooling apparatus, including fans and water sprinklers (Morrison et al 1973). The latter option may be counter-productive, as it does not lower the temperature much, but raises the relative humidity considerably (Smith et al 2012). Studies have shown that cattle prefer shade to sprinklers when outdoors despite other benefits of sprinklers (e.g. reducing the annoyance by insects, Schütz et al 2011).

The climate data: HadISD
HadISD is a new sub-daily, high-quality dataset created by the Met Office Hadley Centre. It is based on the Integrated Surface Dataset (ISD) held at NOAA's National Climate Data Centre The current release of HadISD contains just over 6100 stations spread all over the world, with hourly or three-hourly measurements since 1973 of temperature, dewpoint, sea-level pressure, wind speed and direction as well as some cloud data. These data have been quality controlled using an automatic suite of tests designed to be optimal in retaining true natural weather extremes while removing erroneous data. For a full description of the data set construction, see Dunn et al (2012).
There are 153 HadISD stations in the UK. From these, stations were selected which had no significant gaps and approximately equal numbers of days with > = four observations per day for temperature and dewpoint (spread over at least 12 h) in the seven 5 year periods from 1975-2010. The final station selection is 68 stations spread around the UK (figure 2). For each day with sufficient observation hours with valid temperature and dewpoint measurements, the daily mean temperature and dewpoint were calculated. Station series were checked by eye for inhomogeneities caused by, for example, station moves or instrument changes, but none was found. From these, the mean RH, and subsequently the mean THI for that day were derived. We use a threshold for the onset of heat stress in dairy cattle of THI > 70, although we note that other studies have found that this threshold could be as low as 65 or as high as 75 (Bryant et al 2007).

UK THI over 1973 to 2012
The average number of days per year with THI > 70 for each of the 68 stations is shown in figure 2. As expected, the stations in the south and east have more high-THI days than stations in the north and west. The stations in London also show urban warming, but most dairy cattle are away from London and in the west of the country (figure 1), and so experience, on average, 1 or no days with THI > 70 in a year, and are therefore unlikely to suffer regular heat stress in the current climate. The average number of days per year where the THI > 70 over all the 68 stations in the UK is 0.8. Even at those stations with at least one instance of THI > 70 over the period of record, the average is around 2 days. This is much lower than the number of high-THI days experienced by cattle elsewhere in the world, e.g. USA, Australia and Israel (e.g. West et al 2003, St-Pierre et al 2003, Berman et al 1985. However, in the last decade there have been a couple of high-profile heatwaves in the summers of 2003 and 2006, which affected human health and infrastructure. Therefore it is plausible that these events also had an effect on some UK dairy cattle herds. Segnalini et al (2011) showed that during the summer of 2003 in the Mediterranean basin, some areas usually favourable for animal production were heavily affected by the extreme temperatures. Figure 3 shows the number of days out of the whole year where THI > 70 in 2003 and 2006. In both these years, the number of high-THI days is much greater than the average for the bulk of England shown in figure 2. Wales, and Scotland along with northern and the far south-west of England escaped the worst effects of the heatwaves.
Areas where both high-THI days and high dairy cattle populations (figure 1) coincided during these two years are the north-west Midlands and also the eastern parts of the south-west (Somerset/Avon and surrounding counties). Dairy cattle in the eastern half of England will have suffered more as a result of these heatwaves than their counterparts further west, but the overall impact will have been limited by the lower intensity of dairy farming.
It is likely that most days with THI > 70 occur during the summer, so we now focus on the days in June-July-August (JJA). The heat waves in the summers of 2003 and 2006 were extreme events in their own right (Beniston 2004, Black et al 2004, but over the last few decades, global average temperatures have been rising (IPCC 2007). To investigate the effect that recent climate change has had on the daily summer THI values at each of the stations, distributions combining all the stations were constructed for 1975-1989 and 1995-2009 (two independent 15 year periods), and fitted with a Gaussian, including skew and kurtosis components (figure 4(a)). The standard deviation and skewness did not change much between the two periods. However, the mean and kurtosis have both increased. As the mean remains on the cool side of THI = 60, we conclude that present-day heat-Environ. Res. Lett. 9 (2014) 064006 R J H Dunn et al stress levels in dairy cattle, in terms of milk productivity, are not of immediate concern for most days in the summer. A similar analysis for the running 3 day THI shows very similar mean, skew and kurtosis values to those for the THI ( figure 4(b)). The standard deviation is a little smaller, as would be expected for this smoothed version of the index. The change in the distribution has increased the frequency of THIs > 70 by around 20-40% (an increase by 17% (40%) from 0.90 to 1.05 (0.95-1.34) days per JJA averaged over all 68 stations using the data (fitted curve)). However, the extremal values of the THI do not appear to have changed, with a maximum of around 76 for both periods (see inset of figure 4(a)). Using the running 3 day THI, the change in the frequency of days >70 is less clear, with a decrease from 0.73 to 0.68 days using the data, but an increase from 0.51 to 0.87 days using the fitted curve. This uncertainty is unsurprising because the data at these high THI values are noisy.
The  Bryant et al (2007) studied the effect of the thermal environment on dairy cattle in New Zealand, finding that milk yield and composition varied nonlinearly with 3 day average THI. The variation depended on the breed and on the genetic merit of animals of the same breed. Thresholds for decreasing milk yield ranged between THI = 65 and THI = 75, with similar ranges for milk composition. West et al (2003) found that the milk yield of dairy cattle in Georgia, USA, fell by 0.69 kg day −1 (0.88 kg day −1 ) for each point increase in the concurrent (2 days previous) THI. Ravagnolo et al (2000) found smaller decreases in milk yield of around 0.2 kg per day per point rise in THI. In continental Europe, Renna et al (2010) found measurable changes in the milk yield during 2003 when compared with other, non-heatwave years.

Milk yield and cell count data
We therefore investigated whether recent weather events in the UK have affected dairy cattle health and milk yields. We obtained data from The Cattle Information Service (www. thecis.co.uk) database which contains information on %     protein, % butterfat and the number of white blood cells as well as the milk yield (litres per day) for each animal in a herd at approximately monthly intervals. No more detailed information was available after following multiple lines of enquiry. Although this study focuses primarily on the milk yield of the cattle, the cell count may also be a useful indicator of the animal's health as it correlates with the level of infection. Giesecke (1985) showed that the occurrence of new udder infections and mastitis increased during hot summer months, and so it would be expected that the cell count would rise in dairy cattle that experience heat stress (see also Morse et al 1988).
There are complicating factors in determining whether the change in milk yield is the result of the weather or from physiological factors, such as the number of calves the cow has had (the lactation number or parity) and also how many days since the calf was born (days in milk). These have effects on the milk yield as shown in figure 6.
To avoid the influences of different weather, grazing or feed and routine, we have not pooled data between herds. Each herd has a profile of lactation numbers, but for most herds, there are most measurements from one-calf cows. There is also a sharp drop off in the number of records after around 300 days in milk (DIM). To reduce the impact of these additional factors, we select records for one-calf cows, and split the days in milk into 50 day ranges.
We focus on the August 2003 and July 2006 heatwaves and use data from herds in climatologically warm southern counties with large numbers of dairy cattle (figure 1(c)): Devon, Somerset, Dorset and Wiltshire. The CIS records have 123, 40, 28, and 23 unique herd IDs for these four counties. Milk yields and other measures were recorded approximately monthly, the exact date depending on the herd. The peak temperatures in the two heat waves were on 10 August 2003 and 19 July 2006 respectively. We selected herds with values recorded within 1 day of these peaks (9-11 August 2003 and18-20 July 2006) to sample the strongest effect of the heat. This gave a sub-sample of 17 herds. To protect anonymity, the precise locations of the herds are not given by CIS, so we could not specifically select herds from low-lying land inland sites, which are likely to have been hottest.
We would expect heat stress to lead to a fall in the milk yield during the two events, possibly alongside an increase in white blood-cell count. But only four herds show any indication of a decrease in milk yields during the summers of 2003 and 2006. We illustrate the impacts for two of these herds. For one-calf cows and bins of 0-50, 50-100, and 100-150 DIM, figure 7 shows that the yield for herd 198 is monthly measurement interval may have masked the impacts as the persistence of any effect of heat stress appears to be low. In combination with the smallness of the THI > 70 sample, this has made detection of systematic impacts difficult, and as outlined in section 4, on average there are few days in the UK where cattle would experience heat stress.

Climate projections
To study the effect of any future change in the climate on the number of days with high THI we use the climate projections made for the UK Climate Predication '09 assessment (UKCP09, Murphy et al 2009). An 11-member perturbedphysics ensemble of regional climate model (RCM) runs is available for the UK at 25 km × 25 km resolution on a rotated pole grid. The RCM was driven by (global model used to specify boundary conditions), and takes into account the various factors that influence the climate over the UK, both natural (e.g. volcanic eruptions, variations in solar output) and anthropogenic (e.g. greenhouse gas and aerosol emissions and land use changes), and their likely change under a given scenario when calculating the future climate. The change in anthropogenic emissions is given by the A1B medium emissions scenario (also known as representative climate pathways). The emissions in the A1B scenario are the result of a world with rapid economic growth and a rapid spread of new technologies, but with a drop in population in the last 50 years of the period and a convergent income way of life between regions. There is a balanced emphasis on all energy sources; fossil, and non-fossil. Other scenarios exist which have higher or lower levels of anthropogenic emissions, but these are not available in the 11-member UKCP09 ensemble. The projections include daily temperature and relative humidity for a 150 year run, from 1950-2099. We have combined grid elements together into seven regions (South East, South West, East Anglia, Wales, Midlands, Northern England and Scotland: figure 10).
For each region, we calculate the number of days each grid box is above the THI threshold. We then calculate the average number of days across the region, for each year and for each ensemble member, resulting in 11 THI-curves for each region. We show these for the south-west region in figure 11, and for the remaining six regions, in figure 12 in the appendix.
As can be seen in figure 11, within a single ensemble there is a high inter-annual variability, but across all ensemble members, there is a steady rise in the number of days per year where a grid box has a THI > 70. In the south-west of the UK, Most dairy cattle in the UK are in the west of the country (figure 1), in south-west and northern England, the Midlands, Wales and Scotland (figure 10). Of these, the south-west and Midlands appear to be most susceptible to having a large number of days where dairy cattle could be suffering from heat-stress by the end of the century. However the severity is likely to be higher for the few dairy cattle kept in the south-east.
Studies in Hungary (Solymosi et al 2010) and South Africa (Nesamvuni et al 2012) show similar results, with increasing levels of heat stress in dairy cattle expected over time. Using different emissions scenarios, Nesamvuni et al (2012) unsurprisingly found that severe stress would be more common under maximum daily climate conditions in South Africa. Solymosi et al (2010) found that the number of heat stress days increased in all from 1961-1990 to 2021-2050, using a number of different GCMs for the same emissions scenario (A1B for all but one model which was A2). However the amount of increase varied from model to model, and the baseline value also had a large range between models. Seven out of the nine models showed an increase in the number of heat-stress days by a factor of four over at least 80% of the   country (for 2021-2050). This is similar in magnitude to the change expected in the southern regions of England and Wales (see figures 11 and 12).

Discussion
Currently the climate of the UK results in few days during an average year when the THI rises above the threshold for the onset of heat-stress in dairy cattle. Also, the distribution of dairy cattle farming in the UK is such that there are comparatively few dairy farms in the south-east which is most susceptible to high THI days in the summer. However there are some indications that during the heatwave event of 2006 at least one herd in the south-west of England did have decreased milk yields of around 30%. RCM projections of the future change in the number of days exceeding the THI threshold for the onset of heat stress indicate that for southern parts of the UK this could increase from on average 1-2 per year to over 20 per year by 2100, with correspondingly more during heatwave events.
In the USA, the projected continued gain in milk yield per head is expected to offset milk production lost due to heat-stress in future warmer summers, and similar could be expected for the UK. However, the reduction in milk yield may be as high as 0.9 kg of milk production per cow per day for each percentage point the THI lies above the heat-stress threshold (West et al 2003), though other studies find a reduction of 0.2 kg day −1 . At present, south-west England, south Wales and the Midlands are the regions with high densities of dairy cattle which are most at risk from any rise in the number of days with high THI, as they are closest to the areas which have already experienced substantial numbers of high THI days during heat waves. The cost of a reduction in milk yield or running cooling mechanisms to improve the herd's welfare could have a large impact on the livelihood of the farmers in these regions. Moreover, if a short-term, high THI period is part of a longer weather trend, then there can be compounded problems, especially if drought and hot periods affect the supply and quality of feed (Bryant et al 2007). The changes observed in the milk yields for the UK herds studied here could be the result of drought associated with the hot weather reducing the pasture quality as well as effects from heat stress in the animals.
Furthermore, it is important that further research is invested in setting an exact threshold or indeed another UKtailored measure related to THI, indicating the onset of bovine heat-stress, since the projection of the direct impact on dairy cattle farming varies so much between location and breed, and evidence-based decisions are required. Future dairy farms may become more intensive where cattle are predominantly kept indoors (POST-NOTE 2012). Heat-stress in these environments may be more of an issue in the summer, but in these cases regulating the temperature with cooling systems is practical, albeit expensive. The difference in the temperature and humidity between the interior of barns and the outside has been studied, both in closed and open-sided structures (Seedorf et al 1998, Erbez et al 2010. Temperatures were always higher inside barns, as was the THI, but humidity offsets varied (Erbez et al 2010). Therefore it is possible, in future heatwave events, permanently barned cattle may suffer more heat-stress than those which graze outside depending on the cooling systems installed. Heat-stress affects all animals, not just dairy cattle, and so the potential impact of any future investigations of this nature is large.