Climate stress and its impact on livestock health, farming livelihoods and antibiotic use in Karnataka, India

Understanding the impact of climate change on livestock health is critical to safeguarding global food supplies and economies. Informed by ethnographic research with Indian farmers, veterinarians, and poultry industry representatives, we evidence that both precipitation and vapour pressure are key climate variables that relate to outbreaks of haemorrhagic septicaemia (HS), anthrax (AX), and black quarter (BQ) across the Indian state of Karnataka. We also identify temperature and maximum temperature to be negatively correlated with the same diseases, indicating that a cooling (but still hot) climate with wetter, humid conditions is a prime risk factor for future outbreaks. Principal component analyses have revealed the SW India monsoon and winter periods to be the most strongly correlated with HS, AX and BQ outbreaks. We identify vapour pressure, a proxy for humidity, as having a positive relationship with these specific livestock diseases. The negative relationship between temperature and these diseases, combined with the positive correlation with rainfall and humidity, allow us to classify climate-associated risk using a combination of gridded meteorological time series and epidemiological outbreak data covering the same region and timespan of 1987–2020. Risk maps were constructed following concerns over the growing impact of climate pressures raised by farmers during ethnographic study. Informed by their insights, we used current climate data and future climate projections as a risk classification tool to assess how disease risk varies in Karnataka in the present and possible future scenarios. Despite a relatively limited epidemiological dataset, clear relationships between precipitation, vapour pressure, and temperature with HS, AX and BQ, along with outbreak high-risk zones were defined. This methodology can be replicated to investigate other diseases (including in humans and plants) and other regions, irrespective of scale, as long as the climate and epidemiological data cover similar time periods. This evidence highlights the need for greater consideration of climate change in One Health research and policy and puts forward a case for, we argue, greater alignment between UNFCCC and One Health policy, for example, within the Tripartite Agreement (between OIE, FOA and WHO) on antimicrobial resistance as disease risk cannot be considered independent of climate change. One Health Impact Statement This research aims to investigate the relationship between factors related to climate (surface temperatures, rainfall, humidity) and outbreaks of livestock-related bacterial diseases. This is especially relevant to the One Health approach as it attempts to integrate findings between not only the science of disease but also the science of climate change as a driver of disease, and address problems that could arise within the public and private sectors (local farming, livestock health, government policy etc.). Providing spatial context to climate-associated disease risk across the Indian state of Karnataka will benefit local farmers that may already be, or transitioning to, more intensive livestock farming along with policy makers and private sector companies who are planning for future investments. This transdisciplinary approach springboards from ethnographic observations of famers’ lived experiences of challenges to their livelihoods and facilitates the use of climate datasets that may not have been primarily collected for or used by disease-related studies to map long-term epidemiological risk. This demonstrates the pragmatic impact that such transdisciplinary projects can have by providing interpretations of observed risks to animal health (highlighted by social scientists during engagement with practitioner communities) that Earth Scientists were then able to quantify, proving links that would be otherwise not have been evidenced. Using disease data sourced from local institutions, including Government of India facilitates as well academic research laboratories, can plan the application of pragmatic solutions to local farmers who are primarily impacted by the findings of the research.

variables and disease in Indian livestock production and health, which had not originally been 196 intended to be part of either project. We leveraged the lived experience of the stakeholders to 197 identify the ability of the region of India in which the DARPI project was undertaken (sufficiently 198 robust epidemiological data relating to the region in which NEOSTAR was undertaken was 199 not available) to sustain livestock production in future and to predict which other regions may 200 become less or more able to sustain livestock production or may require additional mitigations 201 to enable current activity to continue. Exploring this understanding will help farmers, livestock data are input to the database by farmers who self-report livestock diseases. These data are 212 not followed up for subsequent investigation and we acknowledge the limitations of this. Data 213 were extracted from the database and collated manually; since the earliest available data for 214 Karnataka is 1987, we focus on the subsequent 33-year period of 1987-2020. Data  1987 to 2020 (Fig. 2). Precipitation and vapour pressure fluctuate significantly between 281 specific years, but trend positively overall especially when considering the five-year running 282 average trendline. Surface temperature and maximum temperature also increase, but 283 fluctuate less than precipitation and vapour pressure, remaining relatively stable despite a few 284 anomalous years (e.g., 1998, 2010). Diurnal temperature range fluctuated between 1987-285 2002, after which the records become stable at ~10.85°C. This stability is a clear artefact of 286 the dataset rather than reflecting true values, as the data remains at this value for 18 years 287 which is unlikely. Due to this distortion through the majority of this selected time period, diurnal 288 temperature range is omitted from Fig. 2. 289 Such nuances in climate data are relevant, as many livestock animals homeostatically regulate 290 core body temperature within a narrow range (Biswal et al., 2022). Dairy cattle, for example, 291 need to maintain a constant body temperature of around 38.8±0.5 o C (Ohnstad, 2008). A high 292 ambient temperature above this thermal comfort zone, caused by climate stress, will trigger 293 series of neuroendocrine modulations that are detrimental to the animals' welfare and 294 productivity. In broiler chickens raised for meat production, heat stress (HS) will reduce feed 295 consumption, growth rate, feed digestion and efficiency, immunity, survival rate and overall 296 welfare (Abioja and Abiona, 2021). Dairy cows will experience lethargy, declines in feed intake 297 and thus milk production, reduced fertility and an increase in susceptibility to mastitis (Dahl, 298 2018). Climate change could also impact disease patterns via changing migratory routes for 299 wild birds or other disease vectors: highly pathogenic Avian Influenza is an example that could 300 spread wider, while diseases such as blue tongue have also increased in geographic 301 distribution in recent years due to climate variation (Mayo et al., 2014).   Haemorrhagic septicaemia is most prevalent during the SW monsoon months (June to 331 September), and predominantly low during the summer months (March to May: Fig 3B). 332 Anthrax outbreaks remain relatively low throughout the year compared to the other diseases; 333 however, there is still a subtle increase during the SW monsoon months, and a relatively low 334 through the summer. Black Quarter also peaks during these months; BQ outbreaks are lowest 335 during May. During the summer months ET has a shallow, wider peak in Karnataka before 336 then decreasing to a lower, stable level throughout the remainder of the year at around 17 337 outbreaks per month. All of the bacterial diseases have peaks in January followed by a sharp 338 decrease, indicating distinct change from the winter months into the summer (Fig. 3B). 339 340

341
Although identifying average trends throughput the year is useful, comparing averaged 342 seasonal climate data with disease data for the same period allows for a clearer interpretation 343 of differences in outbreaks per season: first, by comparing peaks between long-term trends, 344 then by identifying potentially more subtle relationships. 345

389
Correlations were sought between monthly values for each climate variable and disease. 390 Using the entire dataset (as opposed to peak-to-peak) allows identification of more subtle 391 relationships than the peak-to-peak analysis, whilst mitigating the impact of fluctuation 392 between different months or seasons. 393 394 Using Spearman's rank analysis, Karnataka HS data negatively correlate with maximum 395 temperature, a medium negative correlation with temperature and DTR, and a medium 396 positive relationship with precipitation. There is also a weakly positive relationship with HS and 397 vapour pressure (Fig. 5). Three principal components were determined to account for the majority of variance in the 424 datasets. Principal components 1, 2 and 3 account for 81% of the overall data. 425 Principal Component 1 (Fig. 6A)  ET, indicating a relatively strong negative relationship between these diseases and climate 432 variables (Fig. 6A). Finally, PC3 is contributed to most by ET (32%), then temperature (24%), 433 maximum temperature (16%) and HS (15%). There is a more notable variation of ET away 434 from the other diseases whilst precipitation, vapour pressure, HS, AX and BQ group together 435 as they have similar PC1 values (Fig. 6B). 436 Based on these groupings, PC1 could be a representation of relative climate moisture, with 437 higher values associated with drier periods (summer and winter data grouped more positively) 438 whilst lower values reflect wetter periods (monsoon data grouped in negative PC1 field (Fig.  439 6A). Following this, PC2 may represent relative temperature as lower values are mostly the 440 monsoon and summer periods (typically hotter) and then progressively higher values are the 441 post-monsoon then winter periods (typically cooler) (Fig. 6 A). When considering just PCs 1-442 2, there is no clear relationship between precipitation, vapour pressure and the bacterial 443 diseases; however, they do have similar PC1 eigenvalues, and they become more closely 444 grouped when considering PC1-3 with the exception of ET, which is much more isolated and 445 dominates the PC3 axis. It can therefore be inferred that PC3 may mostly represent the values 446 for ET, rather than a clear seasonal shift as represented by PC1 and PC2. Grouping of HS, 447 AX, BQ, vapour pressure and precipitation ( Fig 6B)

453
The identification of relationships between bacterial disease and specific climate variables is 454 essential for assessing future disease-related risk to livestock potentially affected. Using the 455 defined relationships from Section 3.2, it is possible to define levels of risk for these diseases 456 and project this onto a map of Karnataka to facilitate more spatial interpretation. Even though 457 the disease data collected are at state-level accuracy, we use the defined relationships to 458 project risk at a more granular level, to the resolution provided by the CRU climate data with 459 0.5 x 0.5° grid boxes. 460 461

462
As climate variables can vary significantly throughout the year, it is most important not only to 463 consider the total average for each grid box across Karnataka, but also to focus on the 464 deviation from the mean within each key season that links to the diseases. 465 When considering the mean of the total climate data for each gridbox, the southern region of 466 Karnataka is prone to higher vapour pressure (23 -26 hPa) and precipitation (15 -200 mm) 467 which may indicate a potentially higher risk area to HS, AX and BQ. This is especially relevent 468 when combined with cooler temperature data in the south (24-25°C) and lower maximum 469 temperatures (28-30°C). It is however important to consider how different areas of the state 470 may deviate away from these normal values throughout the average year. Areas that may 471 have lower total mean values may still be higher risk if they frequently deviate more so than 472 others. For example, the northern region of Karnataka has a lower vapour pressure average 473 across the period, but in the monsoon and winter seasons it increases significantly (>5 hPa), 474 which could then provide more possibility for disease when combined with higher rainfall and 475 lower temperatures (Fig. 7). This pattern is especially pronounced along the western coastline  highest-risk areas are the north-western coastline and the southwestern area, with relatively 498 lower risk zones more in the central-eastern areas (Fig. 8A). This risk lowers during the post-499 monsoon period, but there are still multiple zones at high risk (Fig. 8B). The eastern and 500 southern areas have increased risk during the post-monsoon season. Lastly, winter poses 501 very high risk in the north and north-western regions, with medium-low risk areas concentrated 502 more in the central-eastern and south-eastern regions (Fig. 8C) 503 Although useful for identifying seasonal variation in risk, these maps alone are not the best 504 interpretation of risk. The total risk for each grid box needs to be calculated by aggregating 505 the risk per season into one output. This was again classified using a traffic-light system based 506 on percentiles of the data. It was then possible to visualise areas of low risk and high risk 507 annually, providing a clearer picture for farmers and policy makers (Fig. 8D).

513
Overall, the north-western coastline of Karnataka is the highest risk area for the bacterial 514 diseases HS, BQ and AX (Fig. 8D). The northern edge and western region of Karnataka are 515 at high risk with the central area remaining at medium-low risk overall. The lowest risk areas 516 are those that remain low during all three seasons mentioned, primarily in the eastern-central 517 region and a few areas to the south-east around Bengaluru. The highest risk zones are 518 primarily contributed to by all three seasons, whilst the northern region is more impacted by 519 the winter (Fig. 8E). The monsoon season contributes the most to the high-risk areas of the 520 western coast and south-western zones and then finally, the post-monsoon dominates 521 contribution to the eastern and south-eastern areas risk. The majority of the map are 522 contributed to relatively equally by each climate variable (Fig. 8: F Several scenarios have been modelled using the CRU 4.5 precipitation, temperature and 534 vapour pressure data used in this study, and adjusting the data to these future possible values 535 (Harris et al., 2020). 536 In scenario 1, the seasonal RPD from the 1987-2020 mean was adjusted by adding 2°C to 537 each temperature and maximum temperature value, 10% of itself to the precipitation value 538 (Kulkarni et al., 2020;Sanjay et al., 2020), and 0.8 hPa to the vapour pressure value (Shrestha 539 et al., 2020). This was repeated in scenarios 2 and 3, but instead using 20% then 30% rainfall 540 adjustments, and 1 hPa then 1.2 hPa vapour pressure adjustments. The 2°C temperature 541 adjustment was kept the same as this seems the most confident prediction from CMIP6 542 models, while rainfall and vapour pressure are more uncertain. Using these scenario changes in temperature, precipitation, and vapour pressure there is no 548 notable change in risk from the present-day through to the longer-term future (>40 years); 549 therefore the 'Total Risk' panel ( Fig 8D) is an accurate representation of present-day risk and 550 future risk (within these models). The north-western area remains highest risk, and the 551 northern and southwestern areas are at high risk. In scenarios 2 and 3, there is a change in 552 one grid box near Bengaluru, to the southwest, from medium to low risk; however, that is the 553 only change, despite a modelled increase in precipitation and vapour pressure.  Greru et al., 2022). New tools are required that will enable them to better predict and 562 mitigate those changes or give them confidence that their livelihoods will not face increased 563 challenges in the short-to-mid-term future. Our use of correlative statistics and PCA has 564 proven effective in identifying relationships between climate variables and the bacterial 565 diseases HS, AX and BQ. Using peak-to-peak correlations alone cannot identify accurate 566 relationships (e.g., in Fig. 4, it appears the climate variables can both positively and negatively 567 correlate with disease, depending on the year). The main use of the peak-to-peak correlations 568 is to identify certain anomalous spikes in disease and climate trends, such as summer 2008 569 where ET outbreaks appear to spike significantly (>150). Although the overall trend of HS and 570 BQ appear decrease from the long-term running average (Fig. 3), this may be the result of 571 targeted vaccination programmes instigated by Government of India agencies such as the 572 Department of Animal Husbandry and Dairying (Rathod, Chander and Bangar, 2016; Basu, 573 2020). These programmes seem to be working (Kushram et al., 2020), but could be even more 574 useful if targeted to areas that have been identified at higher risk using the climate data. The preferential seasons for outbreaks of HS, AX and BQ seem to be the monsoon, post-581 monsoon, and winter periods. This is logical when considering the likely climate variable 582 contributions within each season, and the ideal parameters for outbreak (higher precipitation 583 and vapour pressure, lower temperatures). During the monsoon season, these optimal 584 conditions are met, especially along the Karnataka coastline which is what leads to this area 585 being the highest risk during the June-September period. During the post-monsoon season, 586 there is much lower precipitation; however, vapour pressure remains relatively high along the 587 coast. Temperatures are, however, close to the period average, leading to a decline in risk 588 level across the western region. Winter variable deviation is rather similar to those in the 589 monsoon, with higher precipitation and vapour pressure to the western and northern regions, 590 and cooler maximum / average surface temperatures, leading to increased risk again in the 591 west coast and the highest risk to be in the north. The variation of risk between seasons is 592 critical to defining action taken by farmers and government to mitigate the potential for disease 593 outbreaks all year round. 594 It is also important to note the way in which risk fluctuates into the long-term future, as 595 establishing new farms or businesses in regions that may increase in risk is impractical. 596 According to our scenario models, there is no significant change in climate-related risk across 597 Karnataka in the long-term future (i.e., 2040-69), a good indication of stability of both local 598 and state-level economies. Increasing rates of climate change may still affect this, however, 599 and consistent updating with the latest climate predictions should be conducted to ensure risk 600 assessment remains as accurate as possible. One further consideration is the level of 601 expertise in extracting and analysing data that is currently required to use our models, which local farmers are unlikely to meet; co-automating the process with local partners is therefore 603 an important next stage of the process. 604 605

Limitations of Bacterial Disease Data -NADRES v2
606 Success criteria rely on the quality of both climate and disease data being used. The NADRES 607 v2 database was the source of disease outbreak data; although it proved useful for studying 608 Karnataka, the data do not seem as robust for all regions of India. For example, disease data 609 for the state of Assam are limited to the year 2000 onwards, and even within this period data 610 seems full of artefacts resulting from the original collection process (e.g., anomalous spikes 611 then years of no data). Data are also restricted to outbreaks alone; no data are available for 612 exact numbers of cases or deaths; assumptions had to be made against general trends and 613 assumptions that farmers reported uniformly across all years, so that a year in which a high 614 number of outbreaks were recorded was actually indicative of more outbreaks rather than just 615 more reported ones. 616 The interface itself is also not the most user-friendly with respect to data extraction, as 617 automatically generated graphs are missing the axis values or are limited to set parameters. 618 Manual collection is therefore the only way to extract data, using the GIS interface, which itself 619 has numerous usability issues. The number of diseases with data available is limited and the 620 long-term range of the data is restricted to earliest in 1987, making longer-term studies 621 impossible without other data sources. More open-source data with improved user access is 622 critical to future investigations using livestock disease data in India. Although the diurnal 623 temperature range data was not useable for Karnataka, the overall CRU TS 4.5 series of data 624 proved excellent in both quality and accessibility. Epidemiological data matching the 625 granularity and timespan of the CRU dataset would be ideal for future data-integrated 626 investigations, and would address noted concerns with the current lack of standardised data 627 over long-term (i.e. decades, centuries and even millennia) available for rigorous testing of socioecological system resilience, making such long-term predictions difficult (Allen et al., 629 2014

632
The risk maps generated from this study use climate variables alone as a classifying 633 parameter. Risk calculations were based on relative percent differences of seasonal averages 634 from total period (1987-2020) averages. Classification of risk then uses sequential 0.2 635 percentile ranges of the RPD values to assign a risk category to the specific grid box. This 636 system is therefore sensitive to RPD values, which may not indicate a significant change from 637 the period mean or are all either negative or positive. Using this system requires user 638 interpretation of the initial RPD values and raw data to ensure risk assignment is objective. 639 For example, our surface temperature RPD values are lower than precipitation RPD values, 640 as rainfall fluctuates more than temperature. These values require user verification of real 641 temperature anomalies, and that the percentile ranges reflect the increase/decrease in 642 deviation. Without manually checking, it is possible that the 'very high risk' category using the 643 0.8 percentile could match with a negative number instead of values >0. This risk classification 644 system also gives no preferential weighting to any particular climate variable. Future work 645 should be geared towards defining relationships between climate and these diseases. Once 646 clearer thresholds are established, preferential weightings could be applied to the risk 647 assignment e.g., if vapour pressure were to have a significantly greater impact on HS outbreak 648 than temperature, it should contribute a higher weighting to risk classification. 649 650

651
Mitigating the impact of climate on livestock is a long-term problem and therefore requires 652 advanced planning with effective long-term solutions. Our recommendations are three-fold. 653 First, future farming and livestock policies need to be implemented that fully respect the longevity of the impact of climate on disease outbreak and mitigate this effectively. Secondly, 655 farmers themselves need to be better informed and able to make local decisions to address 656 the particular variable(s) that may impact them the most. Thirdly, future research should 657 continue to further define these meteorological-epidemiological relationships and classify 658 distinct thresholds further. 659 Our recommendations address and acknowledge the quality of disease data collection. 660 Available disease data via NADRES have insufficiently high spatial resolution; however, we 661 have identified parameters that may relate to increased outbreak risk in certain bacterial 662 diseases on a state level (i.e., precipitation, vapour pressure). By disaggregating these 663 parameters to the resolution of the climate data (0.5 x 0.5°), risk mapping can be conducted 664 at a higher resolution than the original disease data provided. However, if disease data were 665 to be made more easily available at a much more granular level and frequency, improved 666 interpretations, and accuracy of the levels of risk would result. 667 Typical risk assessments follow hazard-orientated procedures; similarly, we identify and 668 model these complex relationships and define critical relationships. A definition of critical 669 thresholds per disease and per potentially impacted livestock, however, would be far more 670 beneficial. This risk assessment provides an insight into larger regions and into long-term 671 planning more than specifically providing disease critical thresholds per climate variable. 672 Further investigations should be carried out to define quantitative thresholds for precipitation 673 and vapour pressure to which disease outbreak is related. 674 One possible solution to mitigating the impact of climate change on livestock is the wider 675 introduction of environmentally controlled sheds (Ambazamkandi et al., 2015), along with the 676 energy infrastructure needed to support them; such infrastructure is currently insufficient in 677 many rural and even peri-urban areas (Greru et al., 2022). A second consideration is a shift 678 from livestock and poultry rearing to aquaculture, as is already being seen in northeast India, 679 in regions likely to become more prone to heavy rainfall and flooding (Sarkhel,  We have identified a clear awareness of climate change impacts amongst those whose 692 livelihoods depend on farming and have developed a system through which risk can be better 693 understood and predicted. Our efforts evidence a relationship between average and maximum 694 surface temperature, precipitation and vapour pressure, and several livestock bacterial 695 diseases. There is a modest positive relationship between precipitation and vapour pressure 696 with HS, AX and BQ, followed by a negative relationship between temperature and maximum 697 temperature with the same diseases over a period of 30 years. There is no identified 698 relationship between ET and diurnal temperature range and these climate variables. 699 Based on these relationships, we find that the north-western coast of Karnataka is the highest-700 risk area for HS, AX and BQ, irrespective of other factors that may also govern outbreaks. The 701 western coastline and northern regions are at high risk of outbreak, while the central-eastern 702 and south-eastern regions are the lowest risk. These risk levels are not predicted to change 703 in the next 50 years, even with increased temperatures, and changing spatiotemporal patterns 704 of precipitation and vapour pressures following CMIP6 modelled values. This may not, 705 however, be true of other regions of India, or globally, where changing climate conditions over 706 the coming decades are likely to shift the climate parameters of currently low-risk regions into higher-risk states. This suggests that the ability to predict climate fluctuations and long-term 708 changes will become increasingly important in coming decades and may require greater 709 consideration of climate science within policy intended to protect and improve animal health, 710 such as increased joining up of the UNFCCC and the Global Action Plan on AMR (known 711 colloquially as the 'Tripartite Agreement') agreed by the World Health Organization (WHO), 712 Food and Agriculture Organization of the United Nations (FAO) and the World Organization 713 for Animal Health (OIE) This will be particularly in regions such as India that are at the sharp 714 edge of that change (Rajesh, 2021). In short, we argue that animal health cannot be 715 considered independently of climate change. Such considerations may also help to converge 716 fields that approach challenges to global health from slightly different angles, such as One 717 Health and Planetary Health, and offers to unite them behind common goals. 718 Epidemiological data and interpretations were restricted to Karnataka; data for other states of 719 interest (i.e., Assam, where NEOSTAR was conducted) were too limited in both time and 720 space to provide insight. This is due to the poor data availability through the NADRES vs 721 database. Our work could add to the NADRES online system by providing long-term predictor 722 maps for India livestock disease; the existing maps provide only two-months' notice of 723 increased risk. 724 The techniques used here can be applied to analogous projects for multi-purpose use, 725 particularly for those where climate and epidemiological data cover matching time frames at 726 equal resolution (e.g., monthly averages). The use of this workflow to generate long-term risk 727 maps can also be applied elsewhere in the world. Our future intentions are to automate the 728 risk map production process and then test the model on epidemiological datasets that are 729 equally robust and granular as the climate data. 730 731