The authors have declared that no competing interests exist.
Dairy farming is one the most important sectors of United Kingdom (UK) agriculture. It faces major challenges due to climate change, which will have direct impacts on dairy cows as a result of heat stress. In the absence of adaptations, this could potentially lead to considerable milk loss. Using an 11-member climate projection ensemble, as well as an ensemble of 18 milk loss estimation methods, temporal changes in milk production of UK dairy cows were estimated for the 21st century at a 25 km resolution in a spatially-explicit way. While increases in UK temperatures are projected to lead to relatively low average annual milk losses, even for southern UK regions (<180 kg/cow), the ‘hottest’ 25×25 km grid cell in the hottest year in the 2090s, showed an annual milk loss exceeding 1300 kg/cow. This figure represents approximately 17% of the potential milk production of today’s average cow. Despite the potential considerable inter-annual variability of annual milk loss, as well as the large differences between the climate projections, the variety of calculation methods is likely to introduce even greater uncertainty into milk loss estimations. To address this issue, a novel, more biologically-appropriate mechanism of estimating milk loss is proposed that provides more realistic future projections. We conclude that South West England is the region most vulnerable to climate change economically, because it is characterised by a high dairy herd density and therefore potentially high heat stress-related milk loss. In the absence of mitigation measures, estimated heat stress-related annual income loss for this region by the end of this century may reach £13.4M in average years and £33.8M in extreme years.
Global consumption of milk is increasing in most parts of the world, driven by population and income growth, urbanization and changes in diets [
Projected changes in climate will directly impact the dairy cow, mainly as a result of heat stress, but also through the indirect effects climate change will have on pasture yield and quality, and the length of the growing and grazing season [
As heat stress is likely to be a direct effect of climate change on dairy cows, the overall aim of the present study was to apply a modular approach to investigate the potential outcomes. In the first analysis, we considered only the impact of heat stress on milk losses from dairy cows, assuming no mitigation measures are taken. We recognise that other factors, such as cow fertility, disease and mortality rate [
In the framework of the UKCP09 project [
Heat waves (frequency and length) were the focus of particular attention in these climate projections because this information is required for milk loss methods (model M5 and M6 in
# | THI calculation | Milk Loss (ML) equation | Time step | Reference for THI method | Reference for ML method |
---|---|---|---|---|---|
THI = T + 0.36×Tdew + 41.2 | ML = 0.0695×(THImax−THIthr)2×D | sub-daily | [ |
[ |
|
THI = 1.8×T+32–(0.55–0.0055×RH)×(1.8×T– 26) | ML = 0.0695×(THImax−THIthr)2×D | sub-daily | [ |
[ |
|
THI = T + 0.36×Tdew + 41.2 | ML = max(THI−THIthr, 0)×0.37 | daily | [ |
[ |
|
THI = 1.8×T+32–(0.55–0.0055×RH)×(1.8×T– 26) | ML = max(THI−THIthr, 0)×0.39 | daily | [ |
[ |
|
M1 on heat wave days |
M1 on heat wave days |
mixed | [ |
[ |
|
M2 on heat wave days |
M2 on heat wave days |
mixed | [ |
[ |
T, Tdew, RH, THImax and THIthr denote temperature [°C], dew point temperature [°C], daily maximum of THI [] and the threshold THI [], respectively. D denotes the THI load, the duration of time the cows are experiencing heat stress in a day. T, Tdew, and RH denote daily and hourly averages in case of daily and sub-daily models, respectively.
The daily milk loss values (kg/cow) were calculated for each grid cell by using six methods (2 sub-daily step, 2 daily step, and 2 mixed) described in
In European studies, the THI threshold (THIthr) used to calculate the risk of heat stress varies with the production system, with values ranging from 60 [
These models were used to estimate milk loss in each grid cell without taking into account the type of dairy farming system (at pasture vs indoors). It was assumed that temperature and relative humidity were the same for all systems, and that no mitigation practices were implemented. We also assumed that cattle were not significantly different from the current UK breed types, even though breeding for heat stress tolerance is one of the proposed measures to mitigate effects of climate change on dairy farms [
The annual milk loss per cow (AML, kg/cow/y) value was used to assess the projected impact of climate change on milk production and was calculated using each model as the summation of predicted daily losses for each year. The 11 climate projections and the 18 calculation methods resulted in 1980 AML values for every grid cell for every decade from the 2010s to the 2090s. The uncertainty of the calculated AML values was characterised with the coefficient of variation (CV, standard deviation (SD) divided by the mean) for each grid cell. The uncertainty of the AML figures originates from three major sources: 1) Year effect: caused by the interannual variability of temperature and humidity patterns within a decade; 2) Climate Projection effect: caused by the differences in the climate model projections; 3) Method effect: caused by the differences in the milk loss calculation methods. The contribution of these three factors to the overall uncertainty of AML was quantified as follows: 1) Year effect: for every year the average of AMLs obtained for each climate projection and method combination was calculated (average of 198 values). Then, the coefficient of variation was calculated across the years (CV of 10 values). 2) Climate Projection effect: for every climate projection the average of AMLs obtained for each year and method combination was calculated (average of 180 values). Then, the coefficient of variation was calculated across the climate projections (CV of 11 values). 3) Method effect: for every method the average of AMLs obtained for each year and climate projection combination was calculated (average of 110 values). Then, the coefficient of variation was calculated across the methods (CV of 18 values).
In order to obtain comparable CV values that are calculated from samples having considerably different sizes (10 or 11 versus 18) the jackknife resampling method [
A detailed assessment of the sub-daily (M1-2) and daily (M3-4) methods was performed to reveal the most important cause of the differences between the results of the two method types. This analysis was carried out using all three THI thresholds (68, 70 and 72) but only the results obtained with THIthr = 70 were presented for a selected grid cell. The number of days affected by heat stress (THId > THIthr) as well as the number of days characterised by THImax > THIthr and THId < THIthr was determined for each grid cell and for every year of the 2010–2100 period. The latter indicates the days when the daily step methods predict no heat-stress and no milk loss while sub-daily step methods predict a considerable milk loss. In general, conditions when THId > THIthr represent greater severe heat stress potential than at other times.
The characteristics of trends in AML and number of heat stress days from years 2010–2100 were investigated by regression analysis in STATISTICA 12.0 [
The financial aspect of heat stress related milk loss was estimated for each of the NUTS-1 regions of the UK. The Nomenclature of Territorial Units for Statistics (NUTS) system is a geocode standard for referencing the subdivisions of EU member countries for statistical purposes [
Income losses were calculated for average years (when RAML is the average of AMLs) as well as for extreme years (when RAML equates to the 90th percentile of AMLs).
The vertical bars denote the range between the minimum and the maximum values predicted by the 11 climate projections. Baseline period: 2010s.
CV of 1980 values per each cell.
NUTS-1 region | Heat stress days | Milk loss (kg/cow/y) | ||||||
---|---|---|---|---|---|---|---|---|
Day1 | Day2 | Max | Mean | StDev | Median | 90th percentile | ||
0.6 | 6.4 | 191.0 | 1.8 | 6.4 | 0.0 | 4.7 | ||
2.0 | 13.9 | 240.4 | 5.0 | 11.9 | 0.8 | 13.9 | ||
6.6 | 21.3 | 650.2 | 15.6 | 33.9 | 3.6 | 41.5 | ||
1.0 | 10.0 | 164.4 | 2.7 | 7.7 | 0.2 | 6.9 | ||
3.5 | 20.8 | 212.3 | 7.8 | 16.3 | 2.4 | 21.4 | ||
11.6 | 32.5 | 515.7 | 25.8 | 43.0 | 10.6 | 67.7 | ||
2.5 | 14.8 | 310.6 | 7.1 | 16.9 | 1.4 | 19.3 | ||
7.3 | 28.0 | 428.6 | 19.4 | 34.1 | 6.9 | 53.5 | ||
17.2 | 36.7 | 900.3 | 48.1 | 74.1 | 20.4 | 126.5 | ||
2.0 | 15.4 | 153.6 | 5.5 | 12.0 | 1.0 | 14.9 | ||
7.0 | 29.5 | 275.2 | 16.0 | 25.8 | 6.0 | 43.7 | ||
16.9 | 38.4 | 546.6 | 42.1 | 59.0 | 19.7 | 109.0 | ||
3.4 | 21.5 | 217.5 | 9.2 | 17.9 | 2.4 | 27.0 | ||
10.4 | 37.2 | 387.9 | 26.2 | 40.9 | 11.5 | 73.2 | ||
22.6 | 45.2 | 757.6 | 61.2 | 80.4 | 31.0 | 159.0 | ||
7.4 | 32.6 | 391.9 | 21.7 | 34.5 | 7.6 | 59.3 | ||
19.7 | 47.8 | 605.5 | 56.5 | 75.1 | 29.4 | 154.4 | ||
35.8 | 52.4 | 1129.5 | 112.9 | 137.1 | 64.9 | 297.7 | ||
7.6 | 34.2 | 363.8 | 22.2 | 35.4 | 8.2 | 60.0 | ||
19.9 | 48.6 | 628.3 | 57.6 | 75.8 | 29.5 | 153.7 | ||
35.8 | 52.7 | 1116.8 | 111.3 | 132.1 | 65.4 | 292.1 | ||
4.4 | 18.7 | 424.5 | 11.4 | 24.2 | 2.9 | 31.3 | ||
12.0 | 33.5 | 613.6 | 32.6 | 51.6 | 13.0 | 83.0 | ||
24.8 | 40.6 | 1222.7 | 70.5 | 105.4 | 33.3 | 187.6 | ||
10.7 | 40.2 | 379.7 | 29.8 | 43.4 | 13.3 | 79.2 | ||
26.3 | 51.4 | 727.0 | 75.7 | 88.5 | 40.4 | 194.6 | ||
43.8 | 52.7 | 1257.8 | 136.4 | 149.4 | 81.8 | 348.1 | ||
8.6 | 32.0 | 459.0 | 22.9 | 39.0 | 7.8 | 62.6 | ||
23.2 | 46.0 | 656.1 | 63.4 | 83.5 | 31.7 | 160.7 | ||
41.3 | 49.5 | 1270.2 | 130.6 | 153.3 | 71.1 | 329.5 | ||
13.6 | 42.5 | 469.9 | 37.9 | 50.8 | 17.3 | 101.3 | ||
31.8 | 51.7 | 741.8 | 92.8 | 108.6 | 52.6 | 235.4 | ||
51.0 | 52.0 | 1310.3 | 171.9 | 178.1 | 105.7 | 432.9 |
Day1: number of days when THId>70; Day2: number of days when THId<70 but THImax>70. Greater London was merged with the SE England when statistics were calculated. Both the minimum and the 10th percentile are practically zero for all the NUTS-1 regions in the UK.
Except for the 2030s, the CV associated with the milk loss calculation method effect was consistently higher than that associated with methods of calculating climate projection and inter-annual variability for the investigated future time slots (
Decades | |||||
---|---|---|---|---|---|
2010s | 2030s | 2050s | 2070s | 2090s | |
51.4 | 77.3 | 42.0 | 31.0 | 26.8 | |
54.6 | 68.3 | 38.6 | 38.8 | 24.7 | |
54.3 | 63.2 | 45.9 | 41.0 | 35.9 |
Sources of uncertainty: inter-annual variability (Year effect); differences between the climate projections (CP effect); different milk loss estimation methods (Method effect).
Both the daily and the sub-daily step methods showed an exponential increase in AMLs. The sub-daily step methods (M1-2), however, project a much more substantial rise in AML.
Form of the fitted exponential curve: AML = a×eb×(y-2010); M1-2: a = 27.86, b = 0.0234, SEa = 1.68, SEb = 0.00074, R2 = 0.989, NRMSE = 7.4%; M3-4: a = 6.13, b = 0.0251, SEa = 0.456, SEb = 0.0009, R2 = 0.986, NRMSE = 6.6%; M5-6: a = 17.19, b = 0.0229, SEa = 1.11, SEb = 0.00079, R2 = 0.986; NRMSE = 6.8%. U test showed significant difference (P < 0.001) between the milk loss calculation methods: M1-2 different from M3-4 and both pairs different from M5-6.
The exponential increase in milk loss (
Triangles (and the fitted dotted linear) denote the days with heat stress that are detected only by the sub-daily methods (THId<THIthr and THImax>THIthr). Circles (and the fitted exponential curve) denote the days with heat stress that are detected by both the sub-daily and daily methods (THId>THIthr). Circles: number of heat stress days = a×eb×(y-2010), a = 9.69, b = 0.0199, SEa = 0.523, SEb = 0.00077, R2 = 0.91, NRMSE = 16.2%; Triangles: number of heat stress days = a×(y-2010)+b, a = 0.23, b = 38.28, SEa = 0.0184, SEb = 0.95, R2 = 0.639; NRMSE = 15.7%.
The presented values are the average of 11 UKCP09 SCP climate projections.
Compared to current the UK average annual dairy farm business income (£80,000) the heat stress-related income loss was projected to be less than 7% even in the most affected southern UK regions towards the end of the century (
NUTS-1 region | Period | Average Year | Extreme Year | ||
---|---|---|---|---|---|
scenario_1 | scenario_2 | scenario_1 | scenario_2 | ||
86 | 86 | 217 | 217 | ||
232 | 401 | 649 | 1122 | ||
727 | 1787 | 1936 | 4760 | ||
72 | 72 | 187 | 187 | ||
210 | 475 | 578 | 1308 | ||
696 | 2452 | 1826 | 6431 | ||
235 | 235 | 638 | 638 | ||
641 | 1301 | 1768 | 3588 | ||
1591 | 4867 | 4181 | 12788 | ||
183 | 183 | 494 | 494 | ||
530 | 1075 | 1444 | 2930 | ||
1390 | 4252 | 3602 | 11019 | ||
305 | 305 | 893 | 893 | ||
867 | 1760 | 2421 | 4913 | ||
2021 | 6182 | 5254 | 16070 | ||
717 | 717 | 1958 | 1958 | ||
1866 | 3787 | 5103 | 10356 | ||
3732 | 11415 | 9839 | 30096 | ||
733 | 733 | 1985 | 1985 | ||
1905 | 3866 | 5079 | 10306 | ||
3680 | 11255 | 9652 | 29525 | ||
350 | 350 | 966 | 966 | ||
1007 | 2116 | 2560 | 5382 | ||
2175 | 6970 | 5790 | 18555 | ||
984 | 984 | 2619 | 2619 | ||
2503 | 5079 | 6430 | 13048 | ||
4509 | 13792 | 11505 | 35192 | ||
758 | 758 | 2070 | 2070 | ||
2097 | 4255 | 5310 | 10776 | ||
4317 | 13205 | 10889 | 33308 | ||
1253 | 1253 | 3346 | 3346 | ||
3068 | 6226 | 7779 | 15787 | ||
5682 | 17382 | 14307 | 43763 |
Average year (milk loss = average of AMLs). Extreme year (milk loss = 90th percentile of AMLs). Scenario_1: no more centralisation of herds (constant AHS); Scenario_2: continuous growth of AHS with a rate that was observed in the past two decades (increasing AHS).
A relatively low rate of occurrence of heat stress in UK dairy cows in the current climate (2010s) was estimated by all the methods used in the present study. Similarly, Dunn et al (2014) and Hill and Wall (2015) reported an average of one day of heat stress conditions per year [
British dairy farming is heavily reliant on pasture use [
Although average UK temperature increases are estimated to have relatively minor impact in many regions, our analysis predicted that heat waves could lead to severe heat stress in dairy cows with projected AMLs greater than 1,200 kg/cow by the end of the century in high-risk areas. These areas are Wales, South West, South East England, and East of England, although potential total milk losses in Wales and the South West are likely to be higher than in the South East and East of England because of the higher concentrations of dairy cattle in the west of the UK [
The variability of the AML projections was disaggregated into three major components and the uncertainty originating from, 1) the different climate projections, 2) the inter-annual variability of the weather, and 3) the different milk loss calculation methods. These were estimated by using the coefficient of variation of the AML values. Despite the considerable inter-annual variability of the AML as well as the large differences between the climate projections, the variety of the calculation methods may introduce even larger uncertainty in the milk loss projections for the future. This finding is in line with previous climate change impact assessments. Inter-annual variability was predicted to increase slightly at higher temperatures (toward the end of the century) but this effect was generally less than inter-model variability [
In general, THImax is used for the sub-daily methods whereas THId is used for the daily step methods. This methodological difference explains why there was a greater difference in the projected number of days of heat stress between the daily and sub-daily methods of calculation for the 2010s compared to the 2090s. The increased T towards the end of this century increased the number of days when THId > THIthr, equalizing the number of days when THImax > THIthr (
The sub-daily method [
In conclusion, we have developed a modelling framework to estimate potential effects of climate change on milk production of pasture-based dairy cattle using the UK as an example. We estimated relatively low AML that can be mitigated by implementing current practices for heat stress relief of cows on pasture. However, we detected specific regions of current dairy farming importance, where AML were projected to reach 17% of current annual milk yield in extreme years due to an increased frequency, duration and severity of heat waves. For these regions, the application of sophisticated technologies should be implemented to reduce projected losses. The choice of different THI threshold values made a large difference to projected milk loss. This observation alone emphasises the need for more intensive research seeking to determine the most biologically relevant THIthr values of milk loss estimation methods and exploring the factors that influence this parameter. While this remains a challenging and complex issue [
(DOCX)