Sustainable Precision Dairy Farming: Bridging Animal Welfare and Stakeholder Concerns about the Use of Precision Dairy Technologies

Project Narrative INTRODUCTION The dairy industry must be able to demonstrate how animal welfare concerns from the public are addressed if it is to remain sustainable (von Keyserlingk et al., 2013). Many processors globally (e.g. Danone, Tesco, Saputo) have developed animal welfare standards to which producers within their supply chains are expected to comply. In addition, on-farm animal care assessments such as and proAction in Canada (Dairy Farmers of Canada) have been created as one method to assure consumers that dairy animals are well-cared for. These programs rely on evaluators visiting farms and measuring animal-based outcomes (e.g., body condition, lameness, and injuries) and resource-based outcomes (e.g., access to feed and water) as well as farm protocols (e.g., tail- docking, euthanasia and painful procedures). Although these assessment programs are an important step toward the sustainability of animal agriculture, they come with many challenges. For example, collecting farm and animal-based measures is laborious, time-consuming, and expensive (Vasseur, 2017). In addition, assessments do not account for potential observer biases and variation (Kristensen et al., 2006; Krogh and Enevoldsen, 2012) and are often infrequently performed annually, bi- or triennially (Knierim and Winckler, 2009). Although the vast majority of milk producers in the U.S. are enrolled in the F.A.R.M program, there is evidence of producer distrust in the program (Rink et al., 2019). These programs are also criticized for focusing on the farm-level, whereas animal welfare concerns occur at the individual animal-level. Finally, in-person visits by evaluators are a biosecurity risk for the humans and farm animals involved in the process. Precision dairy technologies (PDTs) may offer solutions to some of the challenges with animal welfare assessments, and may help dairy producers identify and resolve animal welfare concerns before official on-farm assessments occur. Currently, available technologies allow for real-time, continuous recording of animal behavior and other animal-based outcomes at the individual animal-level. However, before these technologies can be useful in aiding dairy producers and evaluators to assess welfare, they must first be validated and integrated into predictive models for common animal welfare concerns. To our knowledge, no research has compared outcomes from standardized animal welfare assessments with data collected from multiple PDTs. Although technology may be useful to help manage animals and identify animal welfare concerns on-farm, dairy producers must be willing to adopt these technologies, see value and trust in these tools, and possess the capacity to make sense of the data. Simultaneously, there is a risk that investment and adoption of novel technologies may be futile if these technologies are ultimately rejected by society. Thus, it is critical that the public is engaged to establish which aspects of these technologies may generate social acceptance or concern. To date, these social aspects of precision animal technologies have not been well-explored by research. To address these gaps in knowledge, our proposed integrated research and extension project aims to bridge PDTs and animal welfare assessments with social aspects of animal welfare. We will validate the use of automated integrated technologies to predict common animal welfare assessment outcomes while simultaneously engaging dairy producers and the public about the role of these technologies on-farm. Our multidisciplinary project will integrate the scientific assessment of animal welfare, artificial intelligence, machine learning, extension, and social science to provide practical recommendations for the future sustainable use of PDTs on dairy farms. 1 Project Narrative Long-term Goal and Supporting Objectives Our long-term goal is to create evidence-based recommendations for the sustainable use of precision technologies on dairy farms, with a focus on improving animal welfare assessment and individual animal management on-farm using PDT data. We intend to accomplish this goal by pursuing 3 specific objectives: Objective 1: Develop and validate the use of PDTs to predict animal-based measurements collected manually from animal welfare assessments (e.g., lameness, injuries, and body condition) using algorithms created from machine learning. We predict that the PDTs, in combination with variables routinely collected from farms, will yield precise and accurate data comparable to manual assessments. Objective 2: Identify key variables collected by PDTs that best predict animal-based outcomes at the herd- and cow-level. We predict that a combination of variables from the PDTs will be able to accurately detect welfare concerns at the herd- and cow-level. Objective 3: Identify support, concerns, and trust relative to PDTs among producers, other industry stakeholders (e.g. processors, veterinarians, nutritionists, consultants, and extension specialists) and the public, as well as disseminate project findings to dairy producers using extension efforts. We predict that increased familiarity and use of data collected by PDTs will improve producer and other industry stakeholder trust in PDTs as a component of management and animal welfare assessments. Public acceptance is predicted to associate with the extent to which PDTs are perceived to confer welfare improvements to animals. These objectives directly address two of the Inter-Disciplinary Engagement in Animal Systems (IDEAS) program prio trust around animal agriculture across a diversity of communities, such as producers and cons This project will be the first of its kind to develop models and validate the use of multiple PDTs to assess animal welfare outcomes at the cow- and herd-level. We expect the results of the project to significantly help producers identify and resolve animal welfare concerns in preparation for animal welfare assessments, as well as become more accepting of and engaged in these assessments. We also expect our results to help animal care evaluators collect more accurate and detailed information during assessments. Additionally, this will be the first project to our knowledge to comprehensively assess producer and public attitudes toward PDTs, while at the same time providing essential extension resources to dairy producers about the use of these technologies to support the advancement of farm animal welfare (Figure 1). Finally, while producers get information from many different sources, they prefer to learn from each other and there is a high sense of trust between farmers. Thus, we will in our extension efforts. Background Animal Welfare Assurance Animal welfare can be defined as three general concerns regarding the functioning, freedom from negative emotional states (e.g., pain, distress), as well as their ability to live a reasonably natural life (Fraser et al., 1997). In North America, assessment programs 2 Project Narrative have been created to attempt to measure some aspects of cow welfare on-farm. For example, the National Milk Producers Federation has developed the F.A.R.M. (Farmers Assuring Responsible Management) program in the U.S. and the Dairy Farmers of Canada developed proAction in Canada. These assessment programs have focused on a framework of continuous improvement where producers are expected to improve their animal care over time. Figure 1. Visual representation of data collection from PDTs, farm records, and stakeholder surveys from the proposed project. Briefly, animal welfare assessment programs include a farm visit where evaluators measure animal-based outcomes (e.g., body condition, lameness, hygiene and hock lesions), resource-based outcomes (e.g., social housing, cow comfort, access to pasture, ventilation, etc.) as well as records and protocols (e.g., animal handling, euthanasia decision-making, pain control for painful procedures, etc.). Farms are visited yearly or every few years depending on the program. One criticism of current animal welfare assessments is the challenge of maintaining inter- and intra- observer reliability, or agreement between and within the evaluators (Bokkers et al., 2012). Human error and subjectivity play a role any time a scoring system is used. A second challenge with these assessment programs is producer engagement; as producer trust in outside evaluators has been found to be relatively low (Rink et al., 2019). A recent survey indicated that 45.6% of producers saw little value in the F.A.R.M. program with complaints focused on distrust and perceived incompetence of the program, frustration toward standards and result of participation (Rink et al., 2019). Thus, animal welfare assessments would benefit from an approach that includes more producer engagement and provides producers with detailed, useful feedback. 3 Project Narrative Precision Dairy Technologies Precision dairy technologies may help resolve some of the concerns with current manual animal welfare assessments. Various PDTs have been validated on commercial dairy farms including technologies that measure animal behavior, lameness, and other physical conditions such as body condition. PDTs allow data to be collected automatically and continuously over long periods (e.g., Schulze et al., 2007, Grinter et al., 2019) while eliminating potential bias, error, or inconsistencies compared with human-collected data (Mullins et al., 2019). Many of the currently available PDTs measure animal behavior. For example, tri-axial accelerometer collars have been developed to record the time cows spend ruminating, active (socializing, grazing, commuting, riding) and resting (time stopped without ruminating). These collars have recently been validated by our team with high accuracy when compared to visual observation (Grinter et al., 2019). Newer technologies have also been developed to measure physical attributes of animals, such as body condition score (BCS; fat and muscle content palpable and visible from various areas in the body such as the withers, hooks, tail head, and pins; Wildman et al., 1982). A commercial automated 3D BCS camera system is currently available for dairy cows (DeLaval Body Condition Scoring, BCS DeLaval International AB, Tumba, Sweden) and has been validated to accurately assess BCS similarly to manual scoring by our team (Mullins et al., 2019). The automated BCS system allows cows to be scored accurately down to increments of 0.25. Combining behavioral and physical information collected from PDTs may help identify animals already experiencing poor welfare (e.g., very low BCS and low activity), and those at-risk for poor welfare outcomes such as lameness and illness (e.g., low lying time can increase the risk of lameness; Ito et al., 2010). These technologies may also serve to standardize data collection, records, and analyses for each animal and farm, to be used to compare assessments within farms over time, and between farms nationally. However, a single PDT is not enough to properly assess welfare. Multiple variables studied from different technologies are better used as a group to assess welfare, but further work is needed to evaluate the collation of the best technologies, methodologies, and current farm assessment variables to best assess welfare (Stone, 2017). The Role of Machine Learning The use of multiple PDTs provides the opportunity to train machine learning models and the creation of an algorithm to identify animals and farms at risk for poor animal welfare. Previous studies at the University of Kentucky (Borchers et al., 2017) have applied machine learning models to identify changes of behaviors collected by PDTs and predict calving times. First, we will design relevant features on the collected data. These features will be used to train machine learning models such as decision trees, neural networks, and support vector machines. We will also use statistical tests, such as the Mann-Whitney U-tests, to identify the most relevant features and reduce the complexity of the models. Secondly, we will design deep learning neural networks enriched with convolutional layers. These networks will be given the raw data as input and we will exploit different filters and number of layers to let the model identify the most significant variables. The goal of the machine learning model is to detect small deviations in behaviors and integrate those behavioral changes with other non-behavioral parameters such as environmental measurements, body temperature, and milk yield. These models can be used to 4 Project Narrative are associated with specific welfare issues. Additionally, the identified main contributing variables will be used to drive producer adoption and future technology designs. In addition to finding the best machine learning approach to predict welfare, it is also used to define the sampling strategy, i.e. determine if the entire herd has to be assessed or if a subsample could subsample would reduce the complexity and cost of the proposed approach. A recent study reported that the best sampling strategy to infer an overall lactating herd welfare status can be achieved by assessing the welfare of high producing cows (Van Os et al., 2019). However, no publication to date has reported the best sampling strategy for PDTs. Thus, we will incorporate different sub-samples of the herd into the machine learning algorithm to maintain welfare assessment accuracy. Such subsamples will be identified using different statistical measurements, from pure random sampling to entropy-based sampling based on various parameters, including the features used to train the machine learning models. We, therefore, argue that the development of an automated multivariable animal welfare monitoring is needed by the livestock industry, especially if it enables the continuous detection of at-risk farms in between the current manual assessments. Producer Perspectives of Precision Dairy Technology Precision dairy technology adoption among dairy producers has been slower than expected (Huirne et al., 1997; Gelb et al., 2011) despite its ability to detect estrus and behavior changes associated with diseases (Eckelkamp, 2019). Adoption rates of PDTs are higher among larger dairies; however, even in these large farms PDTs were usually only part of the milking parlor instead of wearable technologies or stand-alone monitoring systems (Gargiulo et al., 2018). Multiple factors contribute to producers being wary about adopting technology. For example, Spahr (1993) stated that even if a PDT reduced time, labor, and difficulty to improve cow performance and welfare, producers would not adopt the technology unless the benefits were obvious and the PDT was easy to use. More recent research has attributed poor adoption rates to lack of producer involvement in PDT creation (Huirne et al., 1997; Wathes et al., 2008; Bewley et al., 2017) which can lead to technologies that do not live up to on-farm needs, are difficult to understand and use, or are cost-prohibitive (Huirne et al., 1997; Yule and Eastwood, 2012; Russell and Bewley, 2013; Borchers and Bewley, 2015). Producers may also perceive technology to have an undesirable cost-to-benefit ratio and provide an overload of information (Russell and Bewley, 2013). Despite these challenges, there is evidence that producers are willing to adopt technologies that meet their needs. For example, producers have been reported to desire simple, affordable PDTs with a good cost-to-benefit ratio and a good technical support system (Borchers and Bewley, 2015). Another appealing aspect of technology is its ability to measure multiple animal behaviors alerts when they noted changes in feeding, activity, or a combination of multiple behaviors. Using PDTs that measure animal-based outcomes that producers already place importance on, such as feeding and activity, could enhance adoption and increase the perceived usefulness of these tools (Eckelkamp and Bewley, 2020). Another concern with using PDTs on-farm is the security of the data. Existing research on big data 5 Project Narrative (Wiseman et al., 2019). Specifically, producers were concerned with data privacy and service providers profiting off their data (Wiseman et al., 2019). It has also been suggested that issues of trust and transparency will discourage participation in technologies (Jakku et al., 2019). Therefore, it is critical to understand what aspects of data security dairy producers are most worried about, and relieve some of their concerns by engaging them in the process. Public Perspectives of Precision Dairy Technology It is essential to ensure that technologies employed by the dairy industry are socially sustainable, and satisfy societal concerns about the treatment of dairy cattle (Croney and Anthony, 2011; von Keyserlingk et al., 2013). Public influence on agricultural practices is increasingly relevant to the as satisfied consumers sustain markets and voice concerns about management practices by voting on state-level referendums or ultimately in consumption patterns. Thus, it is important that the industry engages with public concerns and as a first step there is a need to better understand public views toward practices within the dairy industry. Others have previously explored public attitudes toward several dairy management practices (Widmar et al., 2017), including but not limited to: pasture access (Schuppli et al., 2014), dehorning (Robbins et al., 2015), tie-stall housing (Robbins et al., 2019), calf housing (Perttu et al., accepted), and cow-calf separation (Ventura et al., 2013). Many of these studies suggest that public acceptance or rejection of specific practices hinges on the degree to which these practices are perceived to allow cattle to live natural lives (e.g. be housed in spacious, naturalistic environments and have behavioral freedom). Public views toward PDTs have not yet been explored. Public trust in science and technology (especially as applied to the production of food) is complex (Funk, 2017). Moreover, direct human- animal connection on farms is valued by many members of the public (Ventura et al., 2016; Perttu et al., accepted), so any technology that may be perceived to replace direct stockmanship may elicit concern. It is also possible that public acceptance of the use of different technologies on farms may shift depending on how these technologies are perceived to benefit the animals. For example, members of the public were more receptive to the use of genetic modification to improve disease resistance and create hornless cattle (removing the need for painful disbudding) when rationales for these technologies focused on welfare improvements for the animals (Ritter et al., 2019). It is likely that PDTs may likewise be more accepted if their uses are perceived to improve animal welfare. Related Work of the Investigators and Preliminary Data The research team at the University of Kentucky Dairy Program, led by PD Costa, has ample experience in animal welfare, applied ethology, and validating and using PDTs to record, analyze and interpret behavior or physiological parameters from large data sets on both individual and group level (e.g. Cantor et al., 2018; Cantor et al., 2020; Grinter et al., 2019; Mullins et al., 2019; review by Costa et al., in press). The experience of working with the validation of different PDTs and how to integrate those behavioral and physiological data into a manageable format is extremely valuable in the scope of this proposal. For example, in 2017, the growing interest and importance of accurate BCS led the research group to collaborate with DeLaval International AB to field test and validate a commercially available automatic BCS system. We also validated the automated collection of BCS using a camera in 6 Project Narrative comparison to manual scores, whereby 343 cows were scored by the automated system and by 3 manual scorers (Mullins et al., 2019). Through this project we were able to show that the automated BCS camera was able to accurately determine BCS compared to manual scoring (Table 1). Table 1. Regression analysis of cow body condition scores (BCS, scale 1 to 5, low to high, n = 343) for camera versus manual scores (P < 0.05 indicates that there is no difference between BCS score collected from the camera and manual assessment). Dataset Assessment N Minimum Mean Maximum p-value Equivalent 343 Difference Difference Difference Camera v. Manual Equivalent 199 (overall) -1.05 -0.11 0.60 <0.0001 Camera v. Manual -1.05 -0.10 0.50 <0.001 BCS: 3.00 to 3.75 We recently conducted a study using the BCS camera on a large commercial farm in Indiana, USA where almost 3,000 lactations were recorded. We showed that only 1.4% of the daily records were below a BCS of 3 and just 0.01% were below a score of 2.5. Many factors impacted BCS, such as disease (Figure 2), lactation number, and BCS at calving (Truman et al., in Review). Our team has recently validated a commercially available collar for monitoring cow rumination time, heat detection, feeding-, and resting-behaviors in lactating dairy cows. Our results indicated that the collar was able to accurately and precisely measure these behaviors in lactating dairy cattle (Grinter at al., 2019). Figure 2. Mean (95% CI) automated body condition score of dairy cattle collected using a 3D camera system at a commercial dairy in Indiana, USA. Data presented across days in milk to 300 days in milk (DIM) stratified by disease status in the first 60 days of lactation. Our team also has extensive experience working with machine learning modeling. In a partnership project with Dr. Proudfoot, we have already used this approach to predict calving time with a combination of behavioral changes (rumination and lying time) recorded from PDTs (Borchers et al., 2017). Co-PD Silvestri has a wealth of experience utilizing this approach in various 7 Project Narrative applications, including the use of drone images and other agriculture-related projects (e.g., Sorbeli et al., 2018; Silvestri et al., 2018; Khamesi et al., 2020). Co-PD Eckelkamp has also conducted machine-learning analyses examining combinations of PDT-captured behaviors to detect diseases such as hyperketonemia, hypocalcemia, and metritis (Eckelkamp et al., 2017). Our team also has experience assessing producer perspectives and the use of PDTs, and disseminating gained knowledge through Extension activities. For example, Co-PD Eckelkamp recently researched they placed on PDT-generated health alerts (Eckelkamp and Bewley, 2020). Producers evaluated 21% of PDT-generated alerts, with a preference for eating and activity alerts in high- risk cows, such as fresh and early lactation. As the dairy extension specialist for TN, Eckelkamp works with the dairy producer community providing on- and off-farm support as well as ongoing outreach programs using county-based meetings, webinars, fact sheets, and newsletters targeting solutions to improve dairy farm efficiency and sustainability. Her flagship program is the Tennessee Master Dairy Producer Program which provides continuing education for dairy producers several times throughout the year. Co-I Schexnayder has engaged dairy producers in three recent projects focused on perceptions, attitudes, practices, and resources that contribute to dairy farm sustainability. She co-led the effort of the Southeast Quality Milk Initiative using qualitative and quantitative methods to understand management on their dairy farms (DeLong et al., 2017; Ellis et al., 2020; Lee et al., 2020) and has extensive experience in survey design and implementation focused on agricultural Keyser et al., 2019). Schexnayder and Eckelkamp are currently collaborating to analyze the results of a mixed- and where best to focus TN state resources for dairy farm sustainability (Sen et al., 2020). Our team also has experience assessing public perspectives about various animal welfare topics. Co-I Ventura has applied both qualitative and quantitative methodologies to describe multi- stakeholder perspectives related to the care and welfare of dairy cattle. For example, Ventura et al. (2015, 2016) used focus groups to engage producers, veterinarians, and other dairy industry stakeholders in discussions about priority welfare concerns and perceived challenges and solutions in the dairy cattle industries. Ventura also has experience in survey design and implementation to capture public views about specific practices in the dairy industries (Ventura et al., 2016; Perttu et al., accepted) and has written more broadly about social sustainability challenges facing the dairy industry (Weary et al., 2015; Ventura and Croney, 2018). Our multidisciplinary research team has the practical experience in deploying a variety of PDTs on commercial farms for validation, the ability to use established statistical methodologies in how to validate the corresponding technologies, analyzing and interpreting large quantities of behavioral and physiological data, evaluate producer use of PDTs to improve cow welfare and assess multi-stakeholder views on specific practices within the dairy industry. Rationale and Significance Animal welfare assessments are becoming the norm in the livestock industry as the public becomes increasingly concerned over the welfare of farmed animals (Deemer and Lobao, 2011). However, 8 Project Narrative reliable methods for measuring animal welfare on-farm are not well established, as current assessment programs rely on manual, infrequent, and subjective measurements of welfare. A drastic increase in the number of animals per farm presents an even greater challenge to measure welfare at both the individual cow- and herd-level. Additionally, many dairy producers are untrusting of animal welfare assessments, and do not feel they provide useful information for them to continually improve their herd management. Despite an increase in the number of farms participating in animal welfare assessments, there is a need to provide dairy producers with more frequent and useful information about their herd management, and how it compares to others in their region. Producers are required to make day- to-day decisions based on many unknowns regarding the welfare of the herd. However, as welfare assessments are infrequent (every year or more), these assessments alone may be inadequate to help producers tackle welfare challenges throughout the year. This dilemma, alongside the public growing scrutiny of the dairy industry, could lead to increased friction between producers and the public. We propose that the integration of technology into animal welfare assessments will help resolve many of the current challenges with these assessments. Using direct feedback from producers and a sophisticated machine learning modeling approach, we will create a reliable tool to help producers tackle day-to-day animal welfare challenges. By engaging producers along the way, we expect these technologies will be more likely to be adopted. In addition, this tool will also help animal welfare evaluators make more accurate and detailed recommendations to producers for improving their animal management. A novel and important aspect of the proposed research is the inclusion of other stakeholders and the public in the discussion about technology and animal welfare assessment. Dairy producers in the U.S. hold the social license to farm, and are reliant on the trust of the public to remain a viable business. Engaging dairy producers, other stakeholders (e.g., veterinarians, nutritionists, consultants) and the public in a dialogue about the role of technology on-farm and for the assessment of welfare can help us develop socially sustainable, long-lasting and evidence-based solutions. APPROACH All procedures in this proposal will be conducted according to the Guide for the Care and Use of Agricultural Animals in Research and Teaching, as well as approved Institutional Animal Care and Use Committee (IACUC) and Institutional Review Board (IRB) protocols from the University of Kentucky and University of Tennessee. Objective 1: Develop and validate the use of PDTs to predict animal-based measurements collected manually from animal welfare assessments (e.g., lameness, injuries, and body condition) using algorithms created from machine learning. Facilities and Recruitment. To reach this objective, 50 dairy farms distributed in the Southeast and Midwest of the U.S. will be enrolled in one study. Recruitment will be performed by our team with the support of a subset of farms (n = 15, see support letters) that have already been approached and have agreed to participate for immediate data collection. The remaining farms needed for the completion of data collection will be identified via the study team, extension agents, veterinarians, KDDC) and other 9 Project Narrative dairy industry experts throughout year 1 and 2 (see support letters). Producers will be contacted directly by a member of the research team by telephone at which time, if the producer agrees, an appointment for a visit will be scheduled. Participation in this study is voluntary and producers will not be remunerated for participation in the study beyond receiving access to the PDTs and the data yielded from them. Animals and Study Design. A subset of 40 dairy cows from each farm will be selected following criteria based on Van Os et al. (2019). Cows will be enrolled at least 20 days before expected calving dates, and data will be collected from d 0 to 305 after calving. Two validated PDTs (a collar on each cow and a 3D camera installed on a walkway exiting the milking parlor) will be provided to each farm. The collar will measure activity, rumination, and the time devoted to resting (time stopped without ruminating; Grinter at al., 2019), and the 3D camera will measure body condition scores (1-5 scale, 0.25 BCS increments; Mullins et al., 2019). Data will be collected from the PDTs twice per day for BCS (at milking), and hourly for activity, rest, and rumination time. The research group, with the aid of service representatives of the corresponding technology, will be available for troubleshooting or replacement of failed technology throughout the project. In addition to data collected from the PDTs, the research team will also collect other data routinely collected on-farm from each of the 40 cows (milk yield, milk components, and health records). Milk and PCDart), and milk components will be measured by taking a milk sample from each cow and evaluating the components in-house during the evaluation visits. During the first farm visit and every 4 months thereafter until 305 d (months 0, 4, and 8), two trained evaluators from the study team will conduct an animal care assessment adopted from the entire lactating herd, including hospital and special needs pens, will be scored. Training of the research personnel will be done in the first semester of the project and will follow the methodology of Vasseur et al. (2013) and Gibbons et al. (2012). An inter- and intra-observer reliability test will be calculated to ensure a and Kock, 1977) and a total agreement of at least 90% to reduce subjectivity and bias associated with the Manual Assessment. Although the F.A.R.M. program also includes calves and heifers, we will use only the adult cow measurements for this assessment. If successful, future work can build upon this project to include all dairy animals. The Manual Assessment will include the collection of the following animal-based measurements: locomotion/lameness (using a 3-pt scale), hock/knee lesions (using a 3-pt scale), broken tails (yes/no), and BCS (using a 5-pt scale). From these data, each farm will be categorized as improvem (e.g., 95% or more of the lactating and dry dairy herd score 2 or less for lameness; 95% or more of the lactating and dry dairy herd score 2 or less for hock/knee lesions; 99% or more of all age classes of animals have a body condition score of 2 or greater; 95% or more of all age classes of animals do not have broken tails). If a farm falls below the threshold for one of these animal- i Model Development. Thresholds for animal-based outcomes used by the F.A.R.M program will be applied as acceptance thresholds indicating farms that require improvement to their management 10