1st Page

Journal of Animal Science - Animal Production

Modeling response to heat stress in pigs from nucleus and commercial farms in different locations in the United States1


This article in JAS

  1. Vol. 94 No. 11, p. 4789-4798
    Received: Apr 07, 2016
    Accepted: Aug 18, 2016
    Published: October 13, 2016

    2 Corresponding author(s):

  1. B. O. Fragomeni 2*,
  2. D. A. L. Lourenco*,
  3. S. Tsuruta*,
  4. S. Andonov,
  5. K. Gray,
  6. Y. Huang and
  7. I. Misztal*
  1. * Animal and Dairy Science Department, University of Georgia, Athens 30602
     Faculty of Agricultural Sciences and Food, Ss. Cyril and Methodius University, Skopje, Republic of Macedonia
     Smithfield Premium Genetics, Rose Hill, NC 28458


The purpose of this study was to analyze the impact of seasonal losses due to heat stress in different environments and genetic group combinations. Data were available for 2 different swine populations: purebred Duroc animals raised in nucleus farms in Texas and North Carolina and crosses of Duroc and F1 females (Landrace × Large White) raised in commercial farms in Missouri and North Carolina; pedigrees provided links between animals from different states. Traits included BW at harvest age for purebred animals and HCW for crossbred animals. Weather data were collected at airports located close to the farms. Heat stress was quantified by a heat load function, defined by the units of temperature-humidity of temperature–humidity index (THI) greater than a certain threshold for 30 to 70 d before phenotype collection. Heat stress responses were quantified by a linear regression of phenotype on heat load. The greatest coefficient of determination occurred with a length of 30 d before phenotype measurements for all states and genetic groups. In the crossbreed data, THI thresholds were 67 in Missouri and 72 in North Carolina. For pure breeds, heat load had the best fit for THI thresholds greater than 70 in North Carolina, although differences in coefficient of determinations were negligible. On the other hand, no optimal THI threshold existed in Texas. In this study, heat stress had a greater impact in commercial farms than in nucleus farms and the effect of heat stress on weight varied by year and state.


Seasonal impacts in livestock production due to heat stress are observed in different species all over the world. The pork industry is especially affected because pigs are not physiologically adapted to dissipate all the heat by sweating or respiration (Renaudeau et al., 2011). In the U.S. pork industry alone, heat stress causes an estimated $299 million loss (St-Pierre et al., 2003). Heat stress can also negatively impact animal welfare, increasing disease incidence and causing physiological changes. Losses due to heat stress occur for several traits including decreased carcass value and poor reproductive performance. Additionally, heat-stressed animals have lower feed intake (Collin et al., 2001a,b), less muscle, and increased fat deposition (Bridges et al., 1998), which decreases carcass value.

Despite advances in cooling strategies for pig farms, pigs cannot dissipate the energy they produce, resulting in production losses associated with heat stress (St-Pierre et al., 2003; Baumgard and Rhoads, 2013). Moreover, it has been reported that the problem can worsen with genetic selection: an improved genetic line produces more heat than an unselected line (Brown-Brandl et al., 2001).

The implications of thermal challenges can be better understood if the consequences of heat stress in practical conditions are known, and strategies can be developed to improve production in harsh environments (Renaudeau et al., 2011). Zumbach et al. (2008a) developed a heat load function, which accounted for the effect of heat stress prior to slaughter on BW. It was the first step toward a genetic evaluation for a small commercial pig population (Zumbach et al., 2008b). Heat load functions might vary depending on the population and specific environmental conditions. Therefore, the purpose of this study was to establish and evaluate heat load functions in 2 different genetic groups (pure breed and crossbreed) and in different states.


Animal Care and Use Committee approval was not obtained for this study because data were from an existing database.


Data were available for purebred Duroc animals from nucleus farms and for crossbred animals from commercial farms. Crossbred animals were progeny of Duroc sires and F1 Landrace × Large White dams. Animals were removed from the data set because of conflicts in data (such as wrong state, litter, or batch), outlier phenotypes (greater than 4 SD), and no weather information available for the weigh date. Furthermore, contemporary groups (concatenation of state, year, and week of data collection) were removed if groups had less than 10 observations or within-group SD greater than 11 kg. In total, 27 out of 413 contemporary groups were removed.

For purebred animals, data was collected from 4 farms in North Carolina and 1 farm in Texas. Farms in North Carolina were located within a radius of 63 km; therefore, they were all considered as being from the same region. The contemporary group effect did include farm because data collection was performed in batches within farms. For crossbred animals, data was collected in 1 packing plant for each state. Similar to pure breeds, packing plants were included in the contemporary groups. The number of phenotypes was uniformly distributed across months in all of the combinations of breed/state. As data were collected for genetic evaluation purposes, no details were available regarding temperature inside pens, exact cooling technologies implemented, and stocking densities.

Descriptive statistics of data across states and genetic groups are shown in Table 1. Purebred animals were evaluated at farms in North Carolina and Texas. Phenotypes were available for BW collected at a mean of 170 d (SD 5.19) in North Carolina from 2003 to 2014 and a mean of 168 d (SD 5.97) in Texas from 2005 to 2014. Crossbred animals were raised and measured in packing plants in North Carolina and Missouri. Phenotypes were available for HCW collected at mean ages of 189 d (SD 13.8) in North Carolina from 2009 to 2014 and 181 d (SD 11.7) in Missouri from 2012 to 2014. Within both genetic groups, pedigrees linked animals on different farms, and 30% of pigs had phenotypes available for siblings from a different state.

View Full Table | Close Full ViewTable 1.

Descriptive statistics for purebred Duroc animals in North Carolina and Texas and crossbred Duroc × F1 (Landrace × Large White) in North Carolina and Missouri

Purebred Duroc
Crossbred Duroc × F1
Item North Carolina Texas North Carolina Missouri
No. 151,336 55,897 141,756 86,435
Weight,1 kg 117.3 (±13.0) 115.0 (±13.0) 92.68 (±9.5) 95.1 (±8.2)
Age, d 169.9 (±5.2) 168.3 (±5.9) 188.63 (±13.9) 180.5 (±10.7)
No. THI > 702 75,457 41,122 64,319 22,855
No. THI > 782 22,897 10,759 31,723 5,858
1Weight for purebred animals was live BW and for crossbred animals was HCW.
2No. represents the number of animals; THI = temperature–humidity index. No. THI > 70 and No. THI > 78 were the number of animals with data collected on days with THI values greater than these numbers.

Weather Data

The R package “WeatherData” (Narasimhan, 2014) was used to collect daily information from weather stations near the farms. Weather information was also available from the High Plains Regional Climate Center (Lincoln, NE) for Texas and State Climate Office of North Carolina (Raleigh, NC) for North Carolina. As correlations of temperatures or humidity between the 2 sources were greater than 0.99, only data from the “WeatherData” package were used.

In the “WeatherData” package, the weather information can be either from public weather stations or from airports. As correlations between information from public stations and airport were always greater than 0.98 (0.99 on average for 30 d intervals) and measurements from some airports were more complete, only information from airports with data available from the whole period of research was used in this study. Airports selected were Wilmington, NC, for North Carolina farms; Pampa, TX, for Texas farms; and Des Moines, IA, for Texas farms. The average distance between farms and airports was 132 km. There were airports closer to farms but with less weather data available. As the correlation between data from the closest airports and the selected ones were 0.99, we decided to use airport stations with more complete data.

The temperature–humidity index (THI) was calculated for each day using the which t was the observed maximum temperature in degrees Celsius, rh was the observed minimum humidity on a 0 to 100 scale (as in Bohmanova et al., 2007)), and THI was the THI for a given day.

Once THI was computed, a heat load (HL) function was calculated using the formulain which THI was the observed THI for a given day, THIT was a threshold THI value, and HL was the HL value. The HL function was intended to measure the amount of heat exceeding a given threshold. This threshold value can be interpreted as the maximum range of the thermal neutral zone. This HL definition is broadly used for genetic evaluation for heat stress in pigs (Zumbach et al., 2008a,b) and in dairy cattle (Ravagnolo and Misztal, 2000; Bohmanova et al., 2007). For testing purposes, thresholds from 60 to 77 units of THI were evaluated.

Heat Load Regression

After HL were calculated, a linear regression was fitted to evaluate the relationship between the HL and phenotype. Data were divided by breeds (pure breed and crossbreed) and then by states. Next, mean weight (or HCW) was calculated for each observation date and combined with HL. Temperature–humidity index threshold values from 55 to 80 units were fitted for calculating HL function. Mean HL within 30, 40, 50, 60, and 70 d before weigh (or slaughter) date were considered to account for cumulative effects of heat stress immediately prior to data collection.

Within each subgroup, a linear model was fitted on HL nested in year. Additionally, data were summarized by mean weight (or HCW) based on HL integer values, and linear and quadratic regressions of weight on HL were fitted to analyze the relationship between weight (HCW for crossbreeds) and HL. The HL groups were between 0 and maximum HL for each data set, and a different group was specified for every for every unit of HL. The goodness of fit was measured by R2.


In purebred animals, small weight fluctuations were observed and larger proportions of animals with THI on day of data collection greater than 70 and 77 were observed in Texas when compared with North Carolina (Table 1). For crossbreed data, animals in North Carolina tended to be slaughtered older than in Texas even though HCW was lighter in North Carolina. Proportionally, more animals experienced hotter THI in North Carolina than Texas. Figures 1 and 2 show the frequency of THI values in each state for purebred and crossbred animals, respectively.

Figure 1.
Figure 1.

Histograms illustrating distributions of records based on temperature–humidity index (THI) for purebred Duroc animals in North Carolina and Texas.

Figure 2.
Figure 2.

Histograms illustrating distributions of records collected based on temperature–humidity index (THI) for crossbred Duroc × F1 (Landrace × Large White) animals in North Carolina and Missouri.


Figure 3 shows the HL defined by a threshold THI of 70 and a 30-d period for the 3 states (data from North Carolina before 2009 was omitted to enable better visual inspection of this plot). Figure 4 shows the HL values in North Carolina based on varying definitions of threshold THI (60, 65, 70, and 75) and number of days (30, 50, and 70 d).

Figure 3.
Figure 3.

Daily heat load, defined as the 30-d average units of temperature-humidity index greater than 70, for purebred animals in North Carolina and Texas and for crossbred animals in North Carolina and Missouri.

Figure 4.
Figure 4.

Daily heat loads for the North Carolina state weather station with heat load definitions differing by length (30, 50, and 70 d) and temperature–humidity index (THI) threshold (60, 65, 70, and 75).


Coefficients of determination for all combinations of period length and THI thresholds for calculating HL are presented in Fig. 5 for pure breed data and in Fig. 6 for crossbreed data. The best fit occurred for the 30-d period in all the subgroups. The best THI threshold varied in each analysis; however, the thresholds for the greatest R2 ranged from 67 to 72. Figure 7 displays the actual and predicted phenotype of purebred animals in each state and with HL defined for 30 d and varying THI thresholds. Similarly, Fig. 8 shows predicted and observed HCW for crossbred animals for the same HL definitions.

Figure 5.
Figure 5.

Coefficient of determination for different heat load functions, depending on the threshold for temperature–humidity index and the number of days before the weigh date, for purebred Duroc animals in North Carolina and Texas.

Figure 6.
Figure 6.

Coefficient of determination for different heat load functions, depending on the threshold for temperature–humidity index and the number of days prior to the weigh date, for crossbred Duroc × F1 (Landrace × Large White) pigs in North Carolina and Missouri.

Figure 7.
Figure 7.

Daily observed (gray) and predicted (black) BW by regression of phenotype on heat load (HL) nested in year for purebred Duroc animals in Texas and North Carolina. Heat load was calculated with different threshold (HL = 60, 65, 70, or 74) and averaged for 30 d.

Figure 8.
Figure 8.

Daily observed (gray) and predicted (black) HCW by regression of phenotype on heat load (HL) nested in year, across time for crossbred Duroc × F1 (Landrace × Large White) animals in Missouri and North Carolina. Heat load was calculated with different threshold (HL = 60, 65, 70, or 75) and averaged for 30 d.


Results from regressions of BW on HL (calculated for THI above 70 averaged for 30 d) are shown in Fig. 9 for purebred animals. Heat load and weight had a positive relationship in Texas and a negative relationship in North Carolina. Figure 10 shows regressions of HCW on HL (also calculated for THI above 70 averaged for 30 d) for crossbred animals. In this case, the relationship was inversely proportional in both Texas and North Carolina. Additionally, in the crossbred populations (Fig. 10), the weight decrease with increasing HL was greater than in the pure breeds (Fig. 9).

Figure 9.
Figure 9.

Observed and predicted BW by regression of weight on heat load for purebred Duroc animals in Texas and North Carolina. Heat load in the x-axis was calculated with temperature–humidity index above 70 averaged for 30 d.

Figure 10.
Figure 10.

Observed and predicted HCW by regression of phenotype on heat load for Duroc × F1 (Landrace × Large White) animals in Missouri and North Carolina. Heat load in the x-axis was calculated with temperature–humidity index above 70 averaged for 30 d.



The histograms in Fig. 1 and Fig. 2 and the HL graphs (Fig. 3) indicated periods of possible heat stress during the summer in all investigated states and years. Bohmanova et al. (2007) described THI affecting production in dairy cattle and had a distribution similar to swine production. Huynh et al. (2005) showed physiological changes in fattening pigs starting at 22°C, and Quiniou et al. (2001) found a lower critical temperature of 24°C. These studies did not account for humidity, which makes comparisons difficult; however, a THI of 70 would be equal to 21°C with 100% relative humidity.

When the length for calculating the HL function was longer (70 d; Fig. 4), the HL curve shifted to the right, indicating a delay on the effects of heat stress. With a lesser THI threshold (THI = 60), a longer heat stress period was observed. With a greater threshold (THI = 75), the heat stress period was shortened. Longer periods of HL resulted in smoother curves, because more data points were used to calculate the mean, resulting in less fluctuation.

For commercial animals in both states, the HL function achieved the best R2 with HL defined for 30 d (Fig. 6). Additionally, the threshold with the largest R2 differed for each state: 72 in North Carolina and 67 in Texas. Yet R2 differences were small, indicating the threshold THI was not critical. No differences were expected among states because the genetics were similar across states. The R2 values were smoother in data from North Carolina, probably because more data were available.

For nucleus animals, the shorter length also provided the best prediction equation of weight by the HL function; however, the differences related to the length of heat stress were much less than in the commercial farms animals’ data (Fig. 5), especially in North Carolina. Coefficients of determination in Texas were much less and decreased as HL increased. In Texas, including HL in the equation led to lesser R2 when compared with adjusting only for year effects (R2 = 0.37; results not shown). When predicted phenotypes were plotted with actual phenotypes in pure breeds (Fig. 7), the HL effect was positive in some occasions and negative in others; the direction of the HL effect was not consistent. Additionally, Fig. 9 shows no overall weight loss with increasing HL in Texas and only a small decrease in North Carolina. One explanation for the unexpected results for Texas pure breeds was confounding of HL with other factors.

Purebred nucleus animals are usually raised in a better environment when compared with commercial animals, and differences in environments include but are not limited to better cooling properties in the building and more space per animal. Such factors have implications for heat stress mitigation because heat-stressed pigs tend to stay in sternal lying position and avoid contact with other individuals (Huynh et al., 2005). Such behavior is more favorable in less populated farms (Kerr et al., 2005). Even though data from stock density was not available, one can expect a lower density in nucleus farms: industry standards for commercial pig density is somewhere between 0.04 to 0.09 m2 per animal less than what is found on the nucleus herds (K. Gray, Smithfield Premium Genetics, Rose Hill, NC, personal communication). Furthermore, swine farms usually use evaporative cooling systems, which tend to perform better in dry climates (Ryan et al., 1992; (Bohmanova et al., 2007). In particular, Texas has a drier climate (WeatherSpark, 2013), and in this case, the difference between THI calculated with data from weather station and from data collected inside the barn would be more different.

The response to heat stress was clearer in the commercial farms than the nucleus farms, which could be a result of differences in industry standards of management associated with stocking density and cooling systems used for each group. The use of different traits to assess the effect of heat stress might also have an influence; however, it was expected that a decrease in live weight would occur if animals from nucleus farms were affected. The age difference between the 2 populations was not a concern, because both populations were within the age window of “finishing” and, therefore, were expected to have similar comfort zones (Myer and Bucklin, 2012). Figure 8 shows a weight decrease in almost every summer for both states with crossbreeds. The R2 values in Fig. 6 show effects of heat stress especially in North Carolina; for this state, Zumbach et al. (2008a) found that the best HL was with THI greater than 65 for a period of 70 d, which was moderately different from the present study.

Results from any new studies on heat stress in pigs are affected by yearly environmental changes. In particular, the management of heat stress evolves, genetics of pigs change because of constant selection within breeding programs, feeding is affected by pricing fluctuations, and actual THI differ every year with the possible impact of climate change. In this study, THI from on-site data might have provided some information about the efficiency of cooling strategies.


The effect of heat stress in the swine industry can be quantified by a HL function calculated from public weather information. The best function in this study was units of THI above 70 to 75 for 30d prior to data collection. Heat stress is more evident in commercial farms than in nucleus farms and varies by state. The extent of heat stress is likely to vary across time due to changing management practices and genetics.




Be the first to comment.

Please log in to post a comment.
*Society members, certified professionals, and authors are permitted to comment.