Abstract
OBJECTIVE To determine whether infrared thermographic images obtained the morning after overnight heat abatement could be used as the basis for diagnostic algorithms to predict subsequent heat stress events in feedlot cattle exposed to high ambient temperatures.
ANIMALS 60 crossbred beef heifers (mean ± SD body weight, 385.8 ± 20.3 kg).
PROCEDURES Calves were housed in groups of 20 in 3 pens without any shade. During the 6 am and 3 pm hours on each of 10 days during a 14-day period when the daily ambient temperature was forecasted to be > 29.4°C, an investigator walked outside each pen and obtained profile digital thermal images of and assigned panting scores to calves near the periphery of the pen. Relationships between infrared thermographic data and panting scores were evaluated with artificial learning models.
RESULTS Afternoon panting score was positively associated with morning but not afternoon thermographic data (body surface temperature). Evaluation of multiple artificial learning models indicated that morning body surface temperature was not an accurate predictor of an afternoon heat stress event, and thermographic data were of little predictive benefit, compared with morning and forecasted weather conditions.
CONCLUSIONS AND CLINICAL RELEVANCE Results indicated infrared thermography was an objective method to monitor beef calves for heat stress in research settings. However, thermographic data obtained in the morning did not accurately predict which calves would develop heat stress later in the day. The use of infrared thermography as a diagnostic tool for monitoring heat stress in feedlot cattle requires further investigation.
Feedlot cattle are frequently exposed to high environmental THIs during the summer months within cattle feeding regions of North America. Cattle performance is adversely affected by high temperatures. Cattle housed in unshaded areas have poor dry-matter intakes, high feed-to-gain ratios, and poor average daily gains and therefore weigh less at harvest than their counterparts that have access to shade.1 Additionally, adequate heat abatement overnight is necessary to reduce heat stress the following day.2
The body temperature of an animal can be used to diagnose heat stress3,4; however, few methods are available to accurately and efficiently record the body temperatures of cattle in a feedlot setting. A noninvasive, remotely applied, practical method to identify cattle that did not adequately cool overnight is needed to improve animal welfare and performance. Infrared thermography has been identified as a potential means to identify cattle with bovine respiratory disease5,6 and foot lesions before clinical signs manifest6,7 as well as feedlot cattle with inflammation associated with the SC placement of growth promoting implants on the convex surface of the ear.8 Infrared thermography has also been recognized as a possible indicator of heat production in beef cattle.9
Prediction of which animals in a feedlot are predisposed to heat stress will allow precautionary measures to be established for calves at risk for hyperthermic events, thereby improving animal welfare. The objective of the study reported here was to determine whether infrared thermographic images obtained the morning after overnight heat abatement could be used as the basis for diagnostic algorithms to predict subsequent heat stress events in feedlot cattle exposed to high ambient temperatures. The hypothesis was that use of infrared thermography to obtain profile body surface temperature images of feedlot cattle at 6 am and 3 pm could identify individual cattle that are predisposed to heat stress.
Materials and Methods
Animals
All study procedures were approved by the Kansas State University Institutional Animal Care and Use Committee. Sixty beef heifers with a mean ± SD body weight of 385.8 ± 20.3 kg were used in the study. Calves were housed in groups of 20 in 3 pens (45.7 × 18.3 m) with an earthen surface and no shade. All calves had free access to water and were fed a corn-based step-up ration. A remote weather stationa at a nearby location recorded ambient temperatures and relative humidity readings every hour throughout the observation period. The recorded ambient temperature and humidity were used to calculate the THI as described.10,11 Briefly, THI = (0.81 × ambient temperature) + (relative humidity × [ambient temperature −14.4]) + 46.4.
Study procedures
Profile digital thermal imagesb of individual calves were obtained from immediately outside each pen via convenience sampling during the 6 am and 3 pm hours for 10 days over a 14-day period. The time for the afternoon monitoring period was selected on the basis of the time of day when the rectal temperature of cattle is greatest.10 During the specified hours, an investigator (EMU) walked outside each pen and obtained digital images of calves that were near the periphery of the pen. Digital thermal images were obtained only on days when the temperature was forecasted to exceed 29.4°C.
Panting score evaluation
On days when digital thermal images were obtained, panting scores were assigned to individual calves during the 6 am and 3 pm hours by a trained observer (EMU), who was unaware of the body surface temperature readings obtained from the thermal images. Panting scores were assigned on a 5-point scale modified from a previously described system.12 Briefly, 0 = normal respiration, 1 = increased respiration, 2 = moderate panting or salivation, 3 = open-mouth breathing, and 4 = severe open-mouth breathing with tongue protruding. The primary coat color of each calf was recorded and classified into 1 of 3 categories (black, white, or red).
Thermographic Images
Digital thermal images were imported into a thermographic imaging software programc for analysis. For each calf, an ROI extending from the tail head along the dorsal top line to between the scapulas, along the angle of the shoulder joint and ventral aspect of the abdomen to the caudal portion of the thigh and back to the tail head, was selected (Figure 1) and analyzed to calculate the mean body surface temperature.
Illustration depicting the ROI (shaded area) that was used for infrared pixel analysis on digital thermal images to determine the mean body temperature for each calf of a convenience sample of 60 crossbred beef heifers (mean ± SD body weight, 385.8 ± 20.3 kg) on each of 10 days during the 6 am and 3 pm hours. Calves were housed in groups of 20 in 3 pens without any shade. The observation period was 14 days, and data were obtained only on days when the temperature was forecasted to exceed 29.4°C. During the specified hours on each selected day, an investigator walked outside of each pen and obtained digital thermal images of and assigned panting scores to calves near the periphery of the pen.
Citation: American Journal of Veterinary Research 78, 7; 10.2460/ajvr.78.7.771
The thermographic images were analyzed by use of a standard color scale between 21.1° and 48.8°C, such that the color scheme and temperatures assigned to each color were uniform for all images. Images were then converted to grayscale and uploaded to an image analysis software program,d and the ROI was selected. Each grayscale pixel in the ROI corresponded to a specific temperature and was assigned to a number between 0 and 255. For example, a black pixel equaled 0 and ≤ 21.1°C, a white pixel equalled 255 and ≥ 48.8°C, and all variations of gray represented temperatures between 21.1° and 48.8°C and were assigned values between 0 and 255. A color histogram was created to summarize the number of pixels for each color. The total number of pixels in the ROI was recorded, and the percentage of pixels for each possible value (0 to 255) was calculated. The cumulative percentage of pixels above each possible value was also calculated.
Statistical analysis
Data were imported into a statistical software programe for analysis. A linear mixed model was generated to determine the relationship between the mean body surface temperature and panting scores from the morning and afternoon. The 3-way interaction model included fixed effects for panting score (0 to 4), coat color (black, white, or red), time of day (am or pm), and random effects to account for day (1 to 10) and repeated measures for calves within a pen. Similar linear mixed models were generated to evaluate the respective relationships between THI and panting scores and mean body surface temperature. Those models included a random effect to account for repeated measures for calves within a pen. Backward elimination was used to select the appropriate model. Variables with values of P ≤ 0.05 were considered significant and retained in the final model except for panting score, which was considered significant and retained in the model only at values of P < 0.01. Coat color was retained as a main effect in all models even when it was not significant (P > 0.05) because cattle with dark-colored hides have higher tympanic temperatures than cattle with light-colored hides.13
A separate linear mixed model was generated to evaluate the relationship between morning body surface temperature and afternoon panting score. Only observations that included both a morning image and an afternoon panting score for an individual calf were evaluated in the model. The 2-way interaction model included fixed effects for afternoon panting score and coat color and random effects to account for day and repeated measures for calves within a pen. Backward elimination was used to select the appropriate model. For main effects and interactions, values of P ≤ 0.05 were considered significant. For comparisons within morning and afternoon, values of P < 0.01 were considered significant.
Algorithm development
In addition to traditional statistical analyses, data-mining techniques were used to evaluate the accuracy of potential classification methods to identify heat stress events in the afternoon on the basis of the morning data. The cumulative percentages of the pixel data for the morning thermographic images were matched with afternoon panting scores for individual calves and imported into a commercial data-mining software packagef for analysis by use of methods similar to those used for other training, testing, and validation data sets.14 Data regarding calf coat color, predicted high ambient temperature and humidity obtained from the local newspaper on each monitoring day, and actual ambient temperature and relative humidity at 6 am on each monitoring day were imported into the program. Panting scores were divided into 2 categories (< 3 and ≥ 3 [ie, clinical signs of moderate to severe heat stress]) to evaluate heat-stress events. Data were then partitioned into training (50%), testing (25%), and validation (25%) data sets. The training data set was used to build automatically generated decision trees, decision stumps, random trees, random forests, logistic model trees, naïve Bayes classifiers, naïve Bayes trees, logistic bases, base-first trees, Bayesian logistic regressions, and artificial neural networks to predict a panting score category (< or ≥ 3) based on the known panting score category from the training data set. The test data set was then incorporated into each model to generate a predicted panting score. The predicted panting score category and actual pant-score category were then compared to evaluate the overall predictive accuracy for each model.
Algorithm refinement
Correlations among all inputs were calculated, and the overall accuracy of each model was assessed in a stepwise manner. To begin, variables with a correlation of 1 were removed from the model, and the accuracy was reassessed. Then, variables with a correlation > 0.95 were removed from the model, and the accuracy was reassessed. The process was continued for variables with a correlation > 0.90 and > 0.85. The correlation setting with the highest κ was selected for each learning model. By changing linear correlations and the number of breaks, models were able to learn and increase pruning ability. After generation of final classification models, the validation data set was evaluated and used to generate diagnostic sensitivity, specificity, overall accuracy, and κ for each model type. Finally, thermographic data were removed from each algorithm to assess whether model accuracy was affected.
Results
All calves remained in the study throughout the observation period. Sunrise was at approximately 6 am and sunset was at approximately 9 pm throughout the study. The environmental conditions for each monitoring day were summarized (Table 1). The maximum ambient temperature did not reach 29.4°C on days 7 (28°C) and 8 (29.1°C). The cumulative frequency distribution of panting scores for both the morning and afternoon were also summarized (Table 2).
Mean, minimum, and maximum ambient temperature, relative humidity, and THI on each of 10 days during which a convenience sample of 60 crossbred beef heifers (mean body weight, 385.8 kg) underwent infrared thermographic imaging and were monitored for panting during the 6 am and 3 pm hours.
| Day | Mean temperature (°C) | Minimum temperature (°C) | Maximum temperature (°C) | Mean humidity (%) | Minimum humidity (%) | Maximum humidity (%) | Mean THI | Minimum THI | Maximum THI |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 28.38 | 22.5 | 33.4 | 73.67 | 59 | 91 | 79.24 | 72.00 | 84.66 |
| 2 | 30.01 | 24.4 | 34.8 | 68.21 | 57 | 80 | 81.08 | 74.16 | 86.62 |
| 3 | 31.53 | 24.4 | 38.3 | 70.08 | 53 | 86 | 83.45 | 74.76 | 90.09 |
| 4 | 30.44 | 24.4 | 37.3 | 70.29 | 43 | 92 | 81.59 | 75.36 | 88.26 |
| 5 | 30.20 | 23.7 | 35.8 | 67.75 | 48 | 90 | 81.10 | 73.97 | 86.13 |
| 6 | 27.19 | 21.6 | 32.9 | 78.25 | 60 | 94 | 77.93 | 70.66 | 84.33 |
| 7 | 23.48 | 17.7 | 28.0 | 72.46 | 53 | 91 | 71.58 | 63.74 | 76.83 |
| 8 | 23.57 | 16.8 | 29.1 | 76.71 | 60 | 94 | 71.95 | 62.26 | 78.94 |
| 9 | 26.55 | 20.2 | 32.6 | 80.46 | 62 | 94 | 77.16 | 68.10 | 84.70 |
| 10 | 31.37 | 24.9 | 38.4 | 69.13 | 47 | 91 | 82.81 | 74.76 | 89.85 |
Ambient temperature and relative humidity were recorded hourly by a remote weather station near the pens where the calves were housed. The THI was calculated as (0.81 × ambient temperature) + (relative humidity × [ambient temperature – 14.4]) + 46.4.
Cumulative frequency distribution of panting scores assigned to the calves described in Table 1.
| Panting score | |||||
|---|---|---|---|---|---|
| Time | 0 | 1 | 2 | 3 | 4 |
| 6 am | 520 | 24 | 0 | 0 | 0 |
| 3 pm | 25 | 121 | 177 | 85 | 31 |
Calves were housed in groups of 20 in 3 pens without any shade. During the specified hours on each of 10 days, an investigator walked outside of each pen and obtained digital thermal images of and assigned panting scores to calves near the periphery of the pen. Panting score was assigned on a 5-point scale, where 0 = normal respiration, 1 = increased respiration, 2 = moderate panting or salivation, 3 = open-mouth breathing, and 4 = severe open-mouth breathing with tongue protruding.
See Table 1 for remainder of key.
The final multivariable model for the association between body surface temperature and panting scores in the morning and afternoon included fixed effects for the panting score, coat color, time of day, and interaction between panting score and time of day and a random effect for repeated measures for calves within a pen. In general, body surface temperatures in the afternoon were greater than those in the morning (Figure 2). No calf was assigned a panting score > 1 during the morning, and the mean morning body surface temperature did not vary significantly between pant-score categories 0 and 1. In the afternoon, the mean body surface temperature increased significantly for each increase in pant-score category from 0 to 3; however, the mean body surface temperature for calves assigned a panting score of 3 did not differ significantly from that of calves assigned a panting score of 4. A similar pattern was observed for the association between THI and pant-score category (Figure 3). There was also a significant (P < 0.01) positive association between THI and body surface temperature.
Mean ± SE body surface temperature for each of 5 categories (0 = normal respiration [white bars], 1 = increased respiration [gray bars], 2 = moderate panting or salivation [striped bars], 3 = open-mouth breathing [dotted bars], and 4 = severe open-mouth breathing with tongue protruding [black bars]) of panting scores that were assigned during the 6 am and 3 pm hours on each of 10 days to the calves of Figure 1. Mean body surface temperature was modeled by use of a linear mixed model that included fixed effects for panting score, coat color (black, white, or red), time of day (am or pm), and the interaction between panting score and time of day and random effects to account for day and repeated measures for calves within a pen. The interaction between panting score and time of day was significant (P = 0.01). Within a time of day, panting score categories with different lowercase letters differ significantly (P < 0.01). See Figure 1 for remainder of key.
Citation: American Journal of Veterinary Research 78, 7; 10.2460/ajvr.78.7.771
Mean ± SE THI by pant-score category (A) and mean body surface temperature (B) for the calves of Figure 1. The respective associations between THI and pant-score category and body surface temperature were modeled by use of separate linear mixed models that included fixed effects for the independent variable of interest and coat color and random effects to account for day and repeated measures for calves within a pen. The THI was significantly (P < 0.01) associated with both pant-score category and body surface temperature. See Figures 1 and 2 for remainder of key.
Citation: American Journal of Veterinary Research 78, 7; 10.2460/ajvr.78.7.771
The final multivariable model for the association between afternoon panting score and morning body surface temperature included fixed effects for mean morning body surface temperature and coat color and random effects for day and repeated measures for calves within a pen. However, afternoon panting score was not significantly associated with mean morning body surface temperature (P = 0.57) or coat color (P = 0.40; Figure 4).
Mean ± SE morning body surface temperature by afternoon pant-score category for the calves of Figure 1. Morning body surface temperature was modeled by use of a linear mixed-effects model that included fixed effects for panting score and coat color and random effects to account for day and repeated measures for calves within a pen. Morning body surface temperature was not significantly associated with afternoon pant-score category. See Figures 1 and 2 for remainder of key.
Citation: American Journal of Veterinary Research 78, 7; 10.2460/ajvr.78.7.771
The diagnostic performance of the predictive classification algorithms evaluated was summarized (Table 3). The decision tree algorithm had the highest overall accuracy (80.2%; 95% CI, 71.1% to 87.1%) and K (0.55; 95% CI, 0.38 to 0.72). The decision tree algorithm had 3 branches (Figure 5). The first cut point was a morning ambient temperature of 24.6°C. When the morning ambient temperature was ≤ 24.6°C, a panting score < 3 was assigned, and when the morning ambient temperature was > 24.6°C, the data moved to the next learning step. The second cut point was a morning relative humidity of 76%. When the morning relative humidity was < 76%, a panting score of < 3 was assigned, and when the morning relative humidity was > 76%, the data moved to the next learning step. The third cut point was 98.6% of the pixels in the ROI ≤ or > 30.5°C at 6 am (ie, morning thermographic data). When 98.5% of the pixels in the ROI were ≤ 30.5°C, a panting score < 3 was assigned, and when 98.5% of the pixels in the ROI were > 30.5°C, a panting score ≥ 3 was assigned. Of the 211 observations in the validation data set, 39 (18%) were misclassified (16 at the first learning step, 6 at the second learning step, and 17 at the third learning step). When the morning thermographic data were removed from the decision tree algorithm, accuracy dropped to 78.3% (95% CI, 690% to 85.5%). The sensitivity and K were both 0 for the decision stump, neural network, and Bayesian logistic regression algorithms, which indicated that those algorithms had no diagnostic value.
Sensitivity, specificity, accuracy, and κ values by algorithm type for prediction of afternoon heat stress in cattle as determined on the basis of morning infrared body surface temperature, coat color, predicted maximum ambient temperature and relative humidity, and actual ambient temperature and relative humidity at 6 am.
| Algorithm | Sensitivity (95% CI) | Specificity (95% CI) | Accuracy (95% CI) | κ (95% CI) |
|---|---|---|---|---|
| Decision tree | 63.2 (46.0–77.7) | 89.7 (79.3–95.4) | 80.2 (71.1–87.1) | 0.55 (0.38–0.72) |
| Random tree | 50.0 (32.6–67.3) | 91.0 (81.8–96.0) | 80.2 (71.1–87.1) | 0.45 (0.25–0.64) |
| Logistic model trees | 67.7 (48.5–82.7) | 84.0 (73.3–91.1) | 79.2 (70.1–86.2) | 0.51 (0.33–0.69) |
| Naïve Bayes | 74.2 (55.1–87.5) | 80.0 (68.9–88.0) | 78.3 (69.0–85.4) | 0.51 (0.34–0.68) |
| Naïve Bayes tree | 80.6 (61.9–91.9) | 77.3 (65.9–85.9) | 78.3 (69.0–85.5) | 0.53 (0.36–0.69) |
| Logistic base | 61.3 (42.3–77.6) | 81.3 (70.3–89.1) | 75.5 (66.0–83.1) | 0.42 (0.23–0.61) |
| Random forest | 25.0 (11.4–45.2) | 93.6 (85.0–97.6) | 75.4 (66.0–83.1) | 0.23 (0.03–0.43) |
| Base-first tree | 29.0 (14.9–48.2) | 93.3 (84.5–97.5) | 74.5 (65.0–82.3) | 0.27 (0.07–0.46) |
| Decision stump | 0.0 (0.0–15.0) | 100.0 (94.2–100.0) | 73.6 (64.0–81.5) | 0.00 (—) |
| Neural network | 0.0 (0.0–13.7) | 100.0 (93.9–100.0) | 70.8 (61.0–79.0) | 0.00 (—) |
| Bayesian logistic regression | 0.0 (0.0–13.7) | 100.0 (93.9–100.0) | 70.8 (61.0–79.0) | 0.00 (—) |
A heat stress event was defined as an afternoon panting score ≥ 3.
— = Not calculated.
See Table 2 for remainder of key.
Branches for variables included in the final decision tree algorithm used to predict afternoon panting score for the calves of Figure 1. For this algorithm, panting score was classified into 2 categories (< 3 and ≥ 3 [eg, heat stress event]). Values in parentheses represent the number of observations that were misclassified at each step. Use of this algorithm resulted in misclassification for 39 of 211 observations. Temp = Temperature.
Citation: American Journal of Veterinary Research 78, 7; 10.2460/ajvr.78.7.771
Discussion
Results of the present study indicated that infrared thermography was useful for measurement of heat stress events in beef calves in a research setting. In the beef industry, panting scores are commonly used to determine whether cattle are undergoing heat stress and evaluate the extent of heat-related animal discomfort and are directly associated with environmental heat load.12 In another study15 involving beef cattle, the panting score increased as the body temperature increased. The major limitation of panting scores is that they are subjective. In the present study, mean body surface temperature as determined by infrared thermography was significantly associated with panting score, and the mean body surface temperature increased as panting score increased. Compared with a pant-score system, infrared thermography has the advantage of being an objective measure. Thus, infrared thermography can be used to objectively measure heat load in cattle.
Infrared thermography could be an asset in future heat stress trials. It can be used to provide an objective assessment of the heat load of study cattle, which should allow heat load to be more uniformly described among trials and reduce observer bias and error that are associated with subjective measures such as panting scores. Constraints associated with the use of infrared thermography in research settings include the time and expertise required to upload thermographic images and calculate mean body surface temperatures, as opposed to the fairly quick and simple method of assigning subjective panting scores. Given currently available technology, it does not appear that infrared thermal images obtained several hours before a heat-stress event (eg, after overnight heat abatement and before peak THI for the day is achieved) will be beneficial for predicting heat-stress risk in commercial feedlot settings unless heat-stress screening is being done on a regular basis.
In the present study, the mean morning body surface temperature as determined by infrared thermography varied by < 2°C throughout the observation period and provided little additional value to weather information for predicting the risk of an afternoon heat-stress event in individual calves. In short, thermographic images of calves obtained in the morning were poor predictors of afternoon heat-stress events as determined by the use of artificial learning models. The decision tree algorithm had the highest κ for the extent of agreement beyond chance between the predicted pant-score category and actual panting score. The naïve Bayes tree algorithm had the highest sensitivity and would be the preferred model if the goal was to identify cattle that are most likely to experience heat stress. However, the wide 95% CI associated with that model suggested that false-positive results would be fairly common, and reliance on that model could be costly if the intervention was expensive. If the goal was to identify cattle that will not develop clinical signs (ie, moderate or severe panting) of a heat stress event, then an algorithm with a 100% specificity such as the decision stump, neural network, or Bayesian logistic regression models would be preferred; however, those 3 algorithms had 0% sensitivity, which indicated that they consistently classified all cattle as having a panting score < 3, and thus had no diagnostic value. The algorithm with the highest specificity and sensitivity was the random forest model. It is important to note that the high specificities calculated for the algorithms evaluated in this study may be misleading because the data used to calculate those values were obtained from a population in which the proportion of calves that did not develop clinical signs of heat stress was substantially greater than the proportion of calves that did develop clinical signs of heat stress.
Although the decision tree model for predicting heat-stress events had the highest κ and overall accuracy, it did not entirely meet the primary objective of the present study, which was to determine whether a diagnostic algorithm that included infrared thermography data was a practical method for predicting which calves were at risk of undergoing a heat stress event several hours prior that event. Interestingly, the learning process for the decision tree model required 3 branches to yield the final predicted values; however, the first 2 branches consisted of morning weather conditions (ambient temperature and relative humidity; Figure 5) rather than thermographic data. Analysis of the learning order for the model indicated that it would be fairly accurate with the morning weather data alone, and the thermographic data provided minimal additional information for prediction. When thermographic data were excluded from the algorithm, the overall accuracy decreased by only 1.9%, and the decision tree cut points were morning ambient temperature, morning relative humidity, and forecasted high ambient temperature.
The present study was not without limitations. We did not have the means to measure the solar radiation or wind speed that calves were exposed to on a daily basis. Wind speed and radiation are important factors associated with heat load in cattle.12 Also, only body surface data were obtained for the calves; body core or rectal temperatures were not obtained. The convenience sampling method used allowed the study to be performed in a commercial feedlot setting without disturbing the calves, but the results may not be representative of the entire study population because data were obtained only for calves near the periphery of the pens, and the individual calves from which data were collected at each designated time varied. Consequently, we were unable to determine whether the body surface temperatures of heat-stressed calves were effectively lowered overnight. Moreover, the mean morning body surface temperature had a narrow range (< 2°C), which precluded assessment of differential overnight cooling. In the present study, coat color was not significantly associated with body surface temperature. Investigators of another study3 reported that the tympanic temperature of black cattle was significantly greater than that of their white counterparts and the interaction between coat color and time of day was significant. Additional data are necessary to determine whether mean body surface temperature is correlated with core body temperature in cattle. Another factor that would be interesting to assess is whether the amount of mud on the coat of cattle is associated with body surface temperature and heat stress. There was higher than usual rainfall during the observation period of this study, and many of the calves were coated in mud. The study calves had a mean body weight of 385 kg and tended to weigh less than calves generally considered at high risk of developing heat stress in a feedlot. Feedlot cattle typically considered at greatest risk of developing heat stress are those closest to harvest and generally weigh > 600 kg. Finally, the actual peak ambient temperature exceeded 29.4°C (the minimum predicted ambient temperature required for data acquisition days) on only 8 of the 10 data acquisition days. Had the actual peak ambient temperature exceeded 29.4°C more consistently throughout the observation period, it is likely more calves would have developed a heat stress event, which would have provided more helpful data for training the artificial learning systems.
In the present study, infrared thermography was identified as an objective method to monitor beef calves for a heat stress event in a research setting. Unfortunately, infrared thermographic data obtained in the morning did not accurately predict which calves would develop heat stress later in the day. Further research is necessary to determine whether infrared thermography can be used as a diagnostic tool for monitoring heat stress in feedlot cattle.
Acknowledgments
This manuscript represents a portion of a thesis submitted by Ms. Unruh to the Kansas State University College of Veterinary Medicine Department of Clinical Science as partial fulfillment of the requirements for a Master of Science degree.
ABBREVIATIONS
| CI | Confidence interval |
| ROI | Region of interest |
| THI | Temperature-humidity index |
Footnotes
WS-2812, La Crosse Technology, La Crosse, Wis.
Ti110 camera, FLUKE Corp, Everett, Wash.
SmartView Software, FLUKE Corp, Everett, Wash.
Image J, National Institutes of Health, Bethesda, Md.
RStudio Team (2015), RStudio: Integrated Development for R, RStudio Inc, Boston, Mass.
Knime Analytics, KNIME AG, Zurich, Switzerland.
References
1. Gaughan JB, Bonner S, Loxton I, et al. Effect of shade on body temperature and performance of feedlot steers. J Anim Sci 2010; 88: 4056–4067.
2. Gaughan JB, Mader TL, Holt SM, et al. A new heat load index for feedlot cattle. J Anim Sci 2008; 86: 226–234.
3. Davis MS, Mader TL, Holt SM, et al. Strategies to reduce feedlot cattle heat stress: effects on tympanic temperature. J Anim Sci 2003; 81: 649–661.
4. Mader TL. Environmental stress in confined beef cattle. J Anim Sci 2003; 81(suppl 2):E110–E119.
5. Schaefer AL, Cook NJ, Church JS, et al. The use of infrared thermography as an early indicator of bovine respiratory disease complex in calves. Res Vet Sci 2007; 83: 376–384.
6. Rekant SI, Lyons MA, Pacheco JM, et al. Veterinary applications of infrared thermography. Am J Vet Res 2016; 77: 98–107.
7. Alsaaod M, Büscher W. Detection of hoof lesions using digital infrared thermography in dairy cows. J Dairy Sci 2012; 95: 735–742.
8. Spire MF, Drouillard JS, Galland JC, et al. Use of infrared thermography to detect inflammation caused by contaminated growth promotant ear implants in cattle. J Am Vet Med Assoc 1999; 215: 1320–1324.
9. Gomes RA, Busato KC, Ladeira MM, et al. Technical note: relationship between infrared thermography and heat production in young bulls. J Anim Sci 2016; 94: 1105–1109.
10. Theurer ME, Anderson DE, White BJ, et al. Effects of weather variables on thermoregulation of calves during periods of extreme heat. Am J Vet Res 2014; 75: 296–300.
11. Hahn GL. Dynamic responses of cattle to thermal heat loads. J Anim Sci 1999; 77(suppl 2):10–20.
12. Mader TL, Davis MS, Brown-Brandl T. Environmental factors influencing heat stress in feedlot cattle. J Anim Sci 2006; 84: 712–719.
13. Mader TL, Holt SM, Hahn GL, et al. Feeding strategies for managing heat load in feedlot cattle. J Anim Sci 2002; 80: 2373–2382.
14. Amrine DE, White BJ, Larson RL. Comparison of classification algorithms to predict outcomes of feedlot cattle identified and treated for bovine respiratory disease. Comput Electron Agric 2014; 105: 9–19.
15. Gaughan JB, Mader TL. Body temperature and respiratory dynamics in un-shaded beef cattle. Int J Biometeorol 2014; 58: 1443–1450.
