A mathematical model of HPAI transmission between dairy cattle and wild birds with environmental effects H.O. Fatoyinbo, P. Tiwari, P.O. Olanipekun, I. Ghosh Abstract Highly pathogenic avian influenza (HPAI), especially the H5N1 strain, remains a major threat to animal health, food security, and public health. Recent spillover events in dairy cattle in the United States, linked to wild birds, highlight the critical importance of understanding transmission pathways at the cattle–wild bird–environment interface. In this work, we formulate and analyze a deterministic compartmental model that captures the transmission of HPAI between dairy cattle and wild birds, incorporating both direct and indirect (environmental) routes. The model combines an S ​ E ​ I ​ R SEIR framework for cattle with an S ​ I ​ R SIR structure for wild birds, coupled through an environmental compartment. We derive the basic reproduction number, ℛ 0 \mathcal{R}_{0} , using the next-generation matrix approach, decomposing it into cattle-to-cattle, bird-to-bird, and environmental contributions. Qualitative analysis establishes positivity, boundedness, and global stability of equilibria through Lyapunov functions. Numerical simulations confirm the results of the theoretical analyses, illustrating outbreak trajectories, extinction thresholds, and persistence dynamics. A global sensitivity analysis, based on Latin hypercube sampling and partial rank correlation coefficients, identifies key parameters, particularly transmission among cattle, environmental contamination, and recovery rate as critical drivers of epidemic outcomes. Our results show that disease elimination is achievable when ℛ 0 < 1 \mathcal{R}_{0}<1 , while persistence is inevitable for ℛ 0 > 1 \mathcal{R}_{0}>1 . These findings provide a comprehensive mathematical framework for assessing HPAI risks and offer guidance for biosecurity strategies aimed at mitigating spillover and controlling outbreaks in livestock populations. 1 Introduction Avian influenza A virus (H5N1 strain) can be categorized into two types depending on its pathogenicity in domestic and wild birds: highly pathogenic avian influenza (HPAI) and low pathogenic avian influenza (LPAI) [ 9 ] . HPAI leads to severe respiratory diseases and high mortality rates in infected birds, whereas LPAI causes mild to no symptoms. They can be more accurately classified according to their molecular characteristics. HPAI is often spread by wild birds like migratory waterfowl to domesticated birds like chickens, turkeys, and geese. Infected birds release the HPAI viruses into the environment through salivary, nasal, and fecal discharges. Afterward, the susceptible birds get infected when they come in direct contact with the released viruses from the environment [ 4 ] . The poultry industry is highly vulnerable to this disease causing a major threat to the economy worldwide. Several studies have been conducted to understand the epidemiology of avian influenza virus through genomic, molecular, and statistical modelling approaches [ 29 , 31 , 30 , 28 ] . Genomic surveillance has revealed high rates of mutation and reassortment in HPAI viruses, particularly in wild bird reservoirs, which contribute to their rapid evolution and zoonotic potential. Molecular analyses have helped identify key mutations associated with host adaptation, virulence, and transmission efficiency, informing risk assessments for both animal and public health. Stanislawek et al. [ 27 ] investigated the risk of Avian influenza viruses in wild birds in New Zealand, focusing on the H5 and H7 subtypes combining statistical inference with epidemiology and genomic properties. Recently, Fatoyinbo et al. [ 26 ] implemented a risk-based statistical modelling framework to analyze HPAI H5 outbreaks in wild birds and poultry across Asian subregions, revealing spatial clustering patterns and identifying regions with elevated outbreak risk. These studies demonstrate the importance of integrating molecular and statistical tools to track viral evolution, assess outbreak risk, and inform surveillance strategies across the poultry–wildlife interface. A particular influenza A virus strain endemic to an animal species can occasionally infect another species, for example, the HPAI H5N1 strain has been previously reported in wild foxes in the United States and in other nations [ 4 ] . This poses a particular threat to livestock (cattle, sheep, goats) as well which get infected on coming in direct contact with virus-infested environment. Although HPAI is not a common occurrence in livestock, the H5N1 strain was first confirmed in cattle very recently on March 25, 2024, in the United States [ 5 , 6 ] . Susceptible cattle got exposed to the disease from the environment in the same way susceptible birds do. The epicenter of the disease transmission in the U.S. was reported to be several farms in northern Texas where the wild birds infected the cattle via an environmental spillover event. Movement of these infected cattle further transmitted the disease among other farms in other states, for example, Nevada [ 10 ] . Reported cases of infections in dairy cows in the United States have been associated with mild respiratory symptoms or conjunctivitis [ 11 ] . Prior to 2024, no cases of HPAI H5N1 were reported in cattle. However, in 2007 Kalthoff et al. [ 7 ] conducted an experiment where they infected six Holstein-Friesian calves with the HPAI H5N1 strain and provided a report about symptoms, further spread of the virus, and serologic reactions in the calves. Mathematical models of epidemiology play a huge role in understanding the outbreak dynamics in details. These furthermore play a crucial role in mitigating the disease spread and developing public policies around them. Epidemiological models can be of varying classifications based on whether they are deterministic or stochastic, continuous or discrete time, spatial or non-spatial, and homogeneous or heterogeneous in terms of different agents considered [ 13 ] . Iwami et al. [ 12 ] built a mathematical model to study the transmission of avian influenza between humans and bird. Wang et al. [ 14 ] looked into the transmission of the H5N1 strain within human beings in China. Bourouiba et al. [ 15 ] built a patch type model derived using reaction-advection equations to study how migratory birds spread H5N1 to nonmigratory poultry. Sharma et al. [ 16 ] built a model with two discrete time delays to explore how the H7N9 strain transmits from birds to humans. Pandit et al. [ 17 ] studied the transmission of HPAI H5N1 strain from wild birds to poultry in West Bengal, India, by building a stochastic continuous-time model. duarte et al. [ 18 ] provides a very detailed review of the H5N1 transmission dynamics among wild birds, poultry, and other mammals through an in-depth analysis of contemporary and molecular epidemiology of the H5N1 strain. In a recent study, Hassman et al. [ 9 ] developed a stochastic continuous-time Markov chain model to study LPAI virus spread, where they considered multiple hosts, transmission via the environment, and migration. Furthermore, Ye et al. [ 19 ] explored a fractional-order model with logistic growth studying the transmission of the virus between humans and birds. Liu et al. [ 1 ] studied the transmission dynamics of H5N1 within three classes of birds by proposing a basic ODE model along with its PDE analogs. A comprehensive review of the HPAI H5N1 strain is provided by Charostad et al. [ 8 ] . The focus of this work is to build a mathematical framework that studies an epidemiological compartmental model of the HPAI H5N1 outbreak in livestock (especially cattle), following the recent events of transmission from wild birds to cattle in the U.S. We have developed an eight-compartment model, where the total population is classified into susceptible, exposed, infected, and recovered wild birds, susceptible, infected, and recovered cattle, and the environment itself. The manuscript is organised as follows. In Sect. 2 , we derive the deterministic compartmental model of HPAI transmission between cattle and wild birds. A system of ordinary differential equations (ODEs) is formulated to indicate the disease dynamics of an SEIR cattle –- SIR wild birds–environment type. Then, we reduced the model by removing the recovery classes from both cattle and wild birds. This updated SEI cattle – SI wild birds-environment model is the one we perform all our algebraic and numerical analysis on. In Sect. 3 , we put forward the qualitative properties of the reduced model. We prove positivity and boundedness, compute the disease-free equilibrium and the endemic equilibrium, evaluate the basic reproduction number \mathcal{R}_{0} , and also study the persistence of the disease by performing a stability analysis of both types of equilibria. Additionally, the Latin Hypercube Sampling (LHS) and Partial Rank Correlation Coefficient methods are utilised to perform a global sensitivity analysis of \mathcal{R}_{0} with respect to the model parameters to detect the ones that primarily drive the dynamics of the virus. In Sect. 4 , the dynamical properties of the reduced model are extensively studied using numerics. This section provides a hands-on approach to corroborating the theoretical results achieved in the previous section. Conclusions and future directions are provided in Sect. 5 2 Model formulation To model the spread of highly pathogenic avian influenza (HPAI) between cattle and wild birds, we used a deterministic compartmental model approach. We divide the cattle population into four compartments: susceptible ( S_{c} ), exposed ( E_{c} ), infected ( I_{c} ) and recovered ( R_{c} ), and the wild bird population into three: susceptible ( S_{b} ), infected ( I_{b} ) and recovered ( R_{b} ), and contaminated environment ( B ). For the cattle a Susceptible–Exposed–Infected–Recovered (SEIR) ODE model and for the wild bird a Susceptible–Infected–Recovered (SIR) ODE model are formulated. The flow chart of HPAI transmission dynamics is shown in Fig. 1 . S_{c} E_{c} I_{c} R_{c} S_{b} I_{b} R_{b} B \beta_{c} \sigma_{c} \gamma_{c} \beta_{b} \gamma_{b} \theta_{c} \beta_{bc} \beta_{\text{env}} \beta_{\text{env}} \theta_{b} \mu_{dc} (\mu_{c}+\mu_{dc}) (\mu_{b}+\mu_{db}) (\mu_{c}+\mu_{dc)} \mu_{dc} \mu_{\text{env}} \mu_{db} \mu_{db} \Lambda_{b} \Lambda_{c} Figure 1 : Flow diagram illustrating the transmission routes of the HPAI virus among cattle, wild birds, and the environment. The cattle population is represented by four compartments ( S-E-I-R ), the wild bird population by three compartments ( S-I-R ), and the environment by a single compartment ( B ) Let us formulate the system of ODEs that we can write down from the flowchart of the compartments in Fig. 1 . We represent the rate of change in a compartment by the differential operator \frac{d}{dt} . Let \Lambda_{c} be the rate of recruitment of cattle. This represents the rate at which new cattle are added to the population of cattle. Now \beta_{c} is the rate at which a susceptible cattle S_{c} transforms to the infected class I_{c} . Also \beta_{bc} is the rate at which the virus is transmitted from the infected wild birds I_{b} to the susceptible cattle S_{c} , and \beta_{\rm env} is the rate at which the virus is transmitted to the susceptible cattle S_{c} from the environment B . Finally, a susceptible cattle can pass away with the rate \mu_{dc} . Collecting all these leads to the following differential equation on S_{c} \frac{dS_{c}}{dt}=\Lambda_{c}-\beta_{c}S_{c}I_{c}-\beta_{bc}S_{c}I_{b}-\beta_{\text{env}}S_{c}B-\mu_{dc}S_{c}. The exposed class of cattle is represented by E_{c} . With \sigma_{c} as the progression rate of the exposed class E_{c} to the infected class I_{c} , and \mu_{c} as the rate at which a cattle passes away naturally, along with the fact that the exposed cattle can also pass away (at a rate \mu_{dc} ) from disease-induced mortality, the overall dynamics of the exposed class E_{c} will be \frac{dE_{c}}{dt}=\beta_{c}S_{c}I_{c}+\beta_{bc}S_{c}I_{b}+\beta_{\text{env}}S_{c}B-\sigma_{c}E_{c}-\mu_{c}E_{c}-\mu_{dc}E_{c}. Now, the infected class I_{c} can recover at the rate \gamma_{c} . The infected class can die of natural causes at the rate \mu_{c} and from disease-related complications at the rate \mu_{dc} also. Thus, we will have \frac{dI_{c}}{dt}=\sigma_{c}E_{c}-\gamma_{c}I_{c}-\mu_{c}I_{c}-\mu_{dc}I_{c}. Finally, the recovered class R_{c} can also pass away from disease-induced complications at the rate \mu_{dc} , thus leaving us with \frac{dR_{c}}{dt}=\gamma_{c}I_{c}-\mu_{dc}R_{c}. Next, \Lambda_{b} is the recruitment rate of the wild birds. Given \beta_{b} as the rate of transmittion from the infected bird I_{b} to susceptible bird S_{b} , \beta_{\rm env} as the rate of transmission from the environment B to to the susceptible wild bird S_{b} , and \mu_{db} as the disease-induced mortality in the susceptible birds S_{b} , we will have the dynamics of S_{b} as \frac{dS_{b}}{dt}=\Lambda_{b}-\beta_{b}S_{b}I_{b}-\beta_{\text{env}}S_{b}B-\mu_{db}S_{b}. Similar to the cattle, we have \gamma_{b} representing the recovery rate of the wild bird, \mu_{b} the mortality rate of the bird due to natural cases, and \mu_{db} as the mortality rate of the bird due to the disease. Thus the dynamics of the infected bird I_{b} can be written as \frac{dI_{b}}{dt}=\beta_{b}S_{b}I_{b}+\beta_{\text{env}}S_{b}B-\gamma_{b}I_{b}-\mu_{b}I_{b}-\mu_{db}I_{b}. Finally, the recovered bird R_{b} can also pass away from disease-induced complications at the rate \mu_{db} , thus leaving us with \frac{dR_{b}}{dt}=\gamma_{b}I_{b}-\mu_{db}R_{b}. Note that the environment compartment is denoted by the letter B . The infected cattle I_{c} and the infected birds I_{b} contaminate the environment B at the rate \theta_{c} and \theta_{b} respectively. Finally the virus dies down from the environment at the rate \mu_{\rm env} . Using these three factors we can write the dynamics on B as following: \frac{dB}{dt}=\theta_{c}I_{c}+\theta_{b}I_{b}-\mu_{\text{env}}B. From the above system of ODEs we can infer that S_{c},E_{c}, and I_{c} are free from the effect of R_{c} and similarly S_{b} and I_{b} are not affected by R_{b} . Thus, it is reasonable to reduce the model by removing the R_{c} and R_{b} compartments altogether from the model. The reduced system of ODEs is then given by: \displaystyle\frac{dS_{c}}{dt} \displaystyle=\Lambda_{c}-\beta_{c}S_{c}I_{c}-\beta_{bc}S_{c}I_{b}-\beta_{\text{env}}S_{c}B-\mu_{dc}S_{c}, (2.1) \displaystyle\frac{dE_{c}}{dt} \displaystyle=\beta_{c}S_{c}I_{c}+\beta_{bc}S_{c}I_{b}+\beta_{\text{env}}S_{c}B-\sigma_{c}E_{c}-\mu_{c}E_{c}-\mu_{dc}E_{c}, \displaystyle\frac{dI_{c}}{dt} \displaystyle=\sigma_{c}E_{c}-\gamma_{c}I_{c}-\mu_{c}I_{c}-\mu_{dc}I_{c}, \displaystyle\frac{dS_{b}}{dt} \displaystyle=\Lambda_{b}-\beta_{b}S_{b}I_{b}-\beta_{\text{env}}S_{b}B-\mu_{db}S_{b}, \displaystyle\frac{dI_{b}}{dt} \displaystyle=\beta_{b}S_{b}I_{b}+\beta_{\text{env}}S_{b}B-\gamma_{b}I_{b}-\mu_{b}I_{b}-\mu_{db}I_{b}, \displaystyle\frac{dB}{dt} \displaystyle=\theta_{c}I_{c}+\theta_{b}I_{b}-\mu_{\text{env}}B. A table listing all the model parameters and their corresponding descriptions is provided in Table 1 . From now on, we perform all model analysis on the reduced system of ODEs ( 2.1 ). Table 1 : Model parameters and their descriptions. Parameter Description \Lambda_{b} Recruitment rate of wild birds. \Lambda_{c} Recruitment rate of cattle. \beta_{b} Transmission rate among wild birds. \beta_{c} Transmission rate among cattle. \beta_{bc} Cross-species transmission (birds to cattle). \beta_{\text{env}} Environmental transmission rate. \sigma_{c} Progression rate in cattle. \mu_{c} Natural mortality rate of cattle. \mu_{b} Natural mortality rate of birds. \mu_{dc} Disease-induced mortality in cattle. \mu_{db} Disease-induced mortality in birds. \mu_{\text{env}} Decay rate of pathogens in environment. \gamma_{b} Recovery rate of birds. \gamma_{c} Recovery rate of cattle. \theta_{b} Contamination of environment by birds. \theta_{c} Contamination of environment by cattle. 3 Qualitative properties of the model In this section we establish the basic qualitative properties of ( 2.1 ) in terms of positivity and boundedness, computation of disease-free equilibrium (DFE) and endemic equilibrium (EE), calculation of the basic reproduction number \mathcal{R}_{0} using the next generation matrix method, and finally their stability analysis to study virus persistence. We also discuss performing a global sensitivity analysis of the basic reproduction number with respect to the model parameters. The methods used are the widely accepted Latin Hypercube Sampling (LHS) and Partial-Rank Correlation Coefficient (PRCC) [ 3 ] . 3.1 Positivity and Boundedness Because ( 2.1 ) represents a model that involves a population of cattle, wild birds, and the HPAI virus, it should be ensured that the model does not generate a negative value for every compartment. To verify the positivity and boundedness of ( 2.1 ), we will analyze the differential equations for each compartment and determine under what conditions the solutions for the state variables remain positive and bounded over time. The variables S_{c}(t),E_{c}(t),I_{c}(t) represent the number of cattle in the compartments of the reduced system ( 2.1 ), as S_{b}(t),I_{b}(t) represents those of birds, while B(t) represents the number of viruses from the environment. Thus, we expect these values to be at least zero at all times t\geq 0 . Let \mathcal{E}_{c}:=\{(S_{c},E_{c},I_{c})\in\mathbb{R}_{+}^{3}:N_{c}\leq\frac{\Lambda_{c}}{\mu_{dc}}\} and \mathcal{E}_{b}:=\{(S_{b},I_{b})\in\mathbb{R}_{+}^{2}:N_{b}\leq\frac{\Lambda_{b}}{\mu_{db}}\} . We will show that the feasible region for the model is \mathcal{E}:=\mathcal{E}_{c}\times\mathcal{E}_{b}\times B\subset\mathbb{R}_{+}^{3}\times\mathbb{R}_{+}^{2}\times\mathbb{R}_{+} . The next two theorems show that the solutions are positive and the region \mathcal{E} is a positive invariant. Theorem 3.1 . Suppose that the initial conditions for the model are S_{c}(0)\geq 0,E_{c}(0)\geq 0,I_{c}(0)\geq 0,S_{b}(0)\geq 0,I_{b}(0)\geq 0 and B(0)\geq 0 . The solutions (S_{c}(t),\cdots,B(t)) of the model are positive for all time t\geq 0 . Proof. The first equation of the model ( 2.1 ) \displaystyle\frac{dS_{c}}{dt}+(\beta_{c}I_{c}+\beta_{bc}I_{b}+\beta_{\text{env}}B+\mu_{dc})S_{c} \displaystyle=\Lambda_{c} can be written as \displaystyle\frac{d}{dt}\mathopen{}\mathclose{{\left(S_{c}(t)\exp\mathopen{}\mathclose{{\left(\int_{0}^{t}(\beta_{c}I_{c}(z)+\beta_{bc}I_{b}(z)+\beta_{\text{env}}B(z)+\mu_{dc})dz}}\right)}}\right) \displaystyle=\Lambda_{c}\exp\mathopen{}\mathclose{{\left(\int_{0}^{t}(\beta_{c}I_{c}(z)+\beta_{bc}I_{b}(z)+\beta_{\text{env}}B(z)+\mu_{dc})dz}}\right) giving \displaystyle S_{c}(t) \displaystyle=\exp\mathopen{}\mathclose{{\left(-\int_{0}^{t}(\beta_{c}I_{c}(z)+\beta_{bc}I_{b}(z)+\beta_{\text{env}}B(z)+\mu_{dc})dz}}\right)S_{c}(0) \displaystyle+\exp\mathopen{}\mathclose{{\left(-\int_{0}^{t}(\beta_{c}I_{c}(z)+\beta_{bc}I_{b}(z)+\beta_{\text{env}}B(z)+\mu_{dc})dz}}\right) \displaystyle\quad\times\int_{0}^{t}\Lambda_{c}\exp\mathopen{}\mathclose{{\left(\int_{0}^{s}(\beta_{c}I_{c}(z)+\beta_{bc}I_{b}(z)+\beta_{\text{env}}B(z)+\mu_{dc})dz}}\right)ds \displaystyle>0. By a similar token, the second equation of the model ( 2.1 ) can be written as \displaystyle\frac{d}{dt}\mathopen{}\mathclose{{\left(E_{c}(t)\exp\mathopen{}\mathclose{{\left(\int_{0}^{t}(\sigma_{c}+\mu_{c}+\mu_{dc})\,dz}}\right)}}\right) \displaystyle=(\beta_{c}S_{c}I_{c}+\beta_{bc}S_{c}I_{b}+\beta_{\text{env}}S_{c}B)\exp\mathopen{}\mathclose{{\left(\int_{0}^{t}(\sigma_{c}+\mu_{c}+\mu_{dc})dz}}\right) giving \displaystyle E_{c}(t) \displaystyle=\exp[-(\sigma_{c}+\mu_{c}+\mu_{dc})t]E_{c}(0) \displaystyle+\exp[-(\sigma_{c}+\mu_{c}+\mu_{dc})t] \displaystyle\quad\times\int_{0}^{t}\mathopen{}\mathclose{{\left[\beta_{c}S_{c}(s)I_{c}(s)+\beta_{bc}S_{c}(s)I_{b}(s)+\beta_{\text{env}}S_{c}(s)B(s)}}\right]\exp[(\sigma_{c}+\mu_{c}+\mu_{dc})s]\,ds \displaystyle>0. Similar computations yield I_{c}(t)>0 , S_{b}(t)>0 , I_{b}(t)>0 , and B(t)>0 for all t>0 . ∎ Theorem 3.2 . The model populations are bounded from above, given that the initial values are positive. Proof. Let the total cattle population N_{c} be given by N_{c}(t)=S_{c}(t)+E_{c}(t)+I_{c}(t). Using the first three equations of the model ( 2.1 ) we have \displaystyle\dfrac{dN_{c}(t)}{dt} \displaystyle=\dfrac{dS_{c}(t)}{dt}+\dfrac{dE_{c}(t)}{dt}+\dfrac{dI_{c}(t)}{dt} \displaystyle=\Lambda_{c}-\mu_{c}(E_{c}(t)+I_{c}(t))-\gamma_{c}I_{c}-\mu_{dc}N_{c}(t). (3.2) Theorem 3.1 guarantees E_{c}(t)>0 , I_{c}(t)>0 for all t>0 , thus \displaystyle\dfrac{dN_{c}(t)}{dt} \displaystyle\leq\Lambda_{c}-\mu_{dc}N_{c}(t). (3.3) Noting that E_{c}(t)+I_{c}(t)\leq N_{c}(t) , we have from ( 3.2 ) the estimate \displaystyle\dfrac{dN_{c}(t)}{dt} \displaystyle\geq\Lambda_{c}-\gamma_{c}I_{c}(t)-(\mu_{c}+\mu_{dc})N_{c}(t) \displaystyle\geq\Lambda_{c}-\gamma_{c}N_{c}(t)-(\mu_{c}+\mu_{dc})N_{c}(t). (3.4) Combining ( 3.3 ) and ( 3.4 ) gives \Lambda_{c}-(\gamma_{c}+\mu_{c}+\mu_{dc})N_{c}(t)\leq N_{c}^{\prime}(t)\leq\Lambda_{c}-\mu_{dc}N_{c}(t). This is a double differential inequality whose solution satisfies the estimates \displaystyle\frac{\Lambda_{c}}{\gamma_{c}+\mu_{c}+\mu_{dc}}+\exp(-(\gamma_{c}+\mu_{c}+\mu_{dc})t)\mathopen{}\mathclose{{\left(N_{c}(0)-\frac{\Lambda_{c}}{\gamma_{c}+\mu_{c}+\mu_{dc}}}}\right) \displaystyle\leq N_{c}(t)\leq\frac{\Lambda_{c}}{\mu_{dc}}+\exp(-\mu_{dc}t)\mathopen{}\mathclose{{\left(N_{c}(0)-\frac{\Lambda_{c}}{\mu_{dc}}}}\right). Thus, \displaystyle\frac{\Lambda_{c}}{\gamma_{c}+\mu_{c}+\mu_{dc}}\leq\limsup_{t\to+\infty}N_{c}(t)\leq\frac{\Lambda_{c}}{\mu_{dc}}. Considering the population of birds given by N_{b}(t)=S_{b}(t)+I_{b}(t) , an argument similar to the one above yields the estimate \displaystyle\Lambda_{b}-(\gamma_{b}+\mu_{b}+\mu_{db})N_{b}(t)\leq\frac{dN_{b}(t)}{dt}\leq\Lambda_{b}-\mu_{db}N_{b}(t) which is a double differential inequality whose solution satisfies \displaystyle\frac{\Lambda_{b}}{\gamma_{b}+\mu_{b}+\mu_{db}}\leq\limsup_{t\to+\infty}N_{b}(t)\leq\frac{\Lambda_{b}}{\mu_{db}}. Next, we consider how the birds and cattle interact with the “B”-compartment. Let t>0 , denote by N(t):=N_{c}(t)+N_{b}(t) the total population of cattle and wild birds. We find that \displaystyle\dfrac{dB}{dt} \displaystyle=\theta_{c}I_{c}(t)+\theta_{b}I_{b}(t)-\mu_{\textnormal{env}}B(t) \displaystyle\leq\theta_{c}N_{c}(t)+\theta_{b}N_{b}(t)-\mu_{\textnormal{env}}B(t) \displaystyle\leq(\theta_{c}+\theta_{b})N(t)-\mu_{\textnormal{env}}B(t) \displaystyle\leq(\theta_{c}+\theta_{b})\mathopen{}\mathclose{{\left(\frac{\Lambda_{c}}{\mu_{dc}}+\frac{\Lambda_{b}}{\mu_{db}}}}\right)-\mu_{\textnormal{env}}B(t). This differential inequality has the solution B(t)\leq\frac{(\theta_{c}+\theta_{b})}{\mu_{\textnormal{env}}}\mathopen{}\mathclose{{\left(\frac{\Lambda_{c}}{\mu_{dc}}+\frac{\Lambda_{b}}{\mu_{db}}}}\right)+\mathopen{}\mathclose{{\left[B(0)-\frac{(\theta_{c}+\theta_{b})}{\mu_{\textnormal{env}}}\mathopen{}\mathclose{{\left(\frac{\Lambda_{c}}{\mu_{dc}}+\frac{\Lambda_{b}}{\mu_{db}}}}\right)}}\right]e^{-\mu_{\textnormal{env}}t}. Note that B also satisfies the following refined estimate B(t)\leq\frac{\theta_{c}\frac{\Lambda_{c}}{\mu_{dc}}+\theta_{b}\frac{\Lambda_{b}}{\mu_{db}}}{\mu_{\textnormal{env}}}+\mathopen{}\mathclose{{\left[B(0)-\frac{\theta_{c}\frac{\Lambda_{c}}{\mu_{dc}}+\theta_{b}\frac{\Lambda_{b}}{\mu_{db}}}{\mu_{\textnormal{env}}}}}\right]e^{-\mu_{\textnormal{env}}t} which is more appropriate for our model. ∎ Corollary 3.3 . Suppose that the model parameters satisfy \alpha:=\max\{\theta_{c},\theta_{b}\}\leq\mu_{\textnormal{env}} , then the following estimate holds B(t)\leq\frac{\Lambda_{c}}{\mu_{dc}}+\frac{\Lambda_{b}}{\mu_{db}}+\mathopen{}\mathclose{{\left[B(0)-\frac{\Lambda_{c}}{\mu_{dc}}-\frac{\Lambda_{b}}{\mu_{db}}}}\right]e^{-\alpha t}. 3.2 Equilibrium analysis An equilibrium point of a dynamical system is a point on the state space at which the system reaches an equilibrium, that is, the dynamics do not change further. The first step towards interpreting the intricate properties of a dynamical system is to analyse its equilibrium points. An epidemiological model has two equilibrium points of two types: (i) a disease free equilibrium point, where the number of infected class I is zero, and (ii) an endemic equilibrium point, where the number of infected class is non-zero. First, we compute the disease free equilibrium (DFE), which occurs when there is no infection in the population. This implies that all compartments related to infection E_{c},I_{c},I_{b} and B are zero. Therefore the DFE, denoted \psi_{\text{DFE}} , for the model ( 2.1 ) is \psi_{\text{DFE}}(S_{c}^{*},E_{c}^{*},I_{c}^{*},S_{b}^{*},I_{b}^{*},B^{*})=\mathopen{}\mathclose{{\left(\frac{\Lambda_{c}}{\mu_{dc}},0,0,\frac{\Lambda_{b}}{\mu_{db}},0,0}}\right). (3.5) We will use \psi_{\rm DFE} to compute the basic reproduction number \mathcal{R}_{0} . This number represents the number of new cases transmitted by an infected individual in a population of all susceptible individuals. Depending on its value, one can measure whether the disease in a population persists or dies down after a certain time. In order to compute \mathcal{R}_{0} , we look into the system of ODEs that represent the infected classes \displaystyle\frac{dE_{c}}{dt} \displaystyle=\beta_{c}S_{c}I_{c}+\beta_{bc}S_{c}I_{b}+\beta_{\text{env}}S_{c}B-\sigma_{c}E_{c}-\mu_{c}E_{c}-\mu_{dc}E_{c}, (3.6) \displaystyle\frac{dI_{c}}{dt} \displaystyle=\sigma_{c}E_{c}-\gamma_{c}I_{c}-\mu_{c}I_{c}-\mu_{dc}I_{c}, \displaystyle\frac{dI_{b}}{dt} \displaystyle=\beta_{b}S_{b}I_{b}+\beta_{\text{env}}S_{b}B-\gamma_{b}I_{b}-\mu_{b}I_{b}-\mu_{db}I_{b}, \displaystyle\frac{dB}{dt} \displaystyle=\theta_{c}I_{c}+\theta_{b}I_{b}-\mu_{\text{env}}B. The infection matrix \mathcal{F} describes the flow of new infections. It corresponds to the terms that generate new infections in the equations for E_{c},I_{c},I_{b} and B . At \psi_{\rm DFE} , we have \mathcal{F}=\begin{pmatrix}0&\beta_{c}\frac{\Lambda_{c}}{\mu_{dc}}&\beta_{bc}\frac{\Lambda_{c}}{\mu_{dc}}&\beta_{\text{env}}\frac{\Lambda_{c}}{\mu_{dc}}\\ 0&0&0&0\\ 0&0&\beta_{b}\frac{\Lambda_{b}}{\mu_{db}}&\beta_{\text{env}}\frac{\Lambda_{b}}{\mu_{db}}\\ 0&0&0&0\\ \end{pmatrix} (3.7) The transition matrix \mathcal{V} describes the flow of individuals out of infected compartments due to recovery or other processes. The transition matrix is derived from the negative terms in the equations. At \psi_{\rm DFE} we have \mathcal{V}=\begin{pmatrix}\sigma_{c}+\mu_{c}+\mu_{dc}&0&0&0\\ -\sigma_{c}&\gamma_{c}+\mu_{c}+\mu_{dc}&0&0\\ 0&0&\gamma_{b}+\mu_{b}+\mu_{db}&0\\ 0&-\theta_{c}&-\theta_{b}&\mu_{\text{env}}\\ \end{pmatrix} To obtain the basic reproduction number \mathcal{R}_{0} , we use the next generation matrix (NGM) approach. We denote the NGM as \xi_{\text{NGM}}=\mathcal{FV}^{-1} . We have \mathcal{V}^{-1}=\begin{pmatrix}\frac{1}{\sigma_{c}+\mu_{c}+\mu_{dc}}&0&0&0\\ \frac{\sigma_{c}}{(\sigma_{c}+\mu_{c}+\mu_{dc})(\gamma_{c}+\mu_{c}+\mu_{dc})}&\frac{1}{\gamma_{c}+\mu_{c}+\mu_{dc}}&0&0\\ 0&0&\frac{1}{\gamma_{b}+\mu_{b}+\mu_{db}}&0\\ \frac{\sigma_{c}\theta_{c}}{(\sigma_{c}+\mu_{c}+\mu_{dc})(\gamma_{c}+\mu_{c}+\mu_{dc})\mu_{\text{env}}}&\frac{\theta_{c}}{(\gamma_{c}+\mu_{c}+\mu_{dc})\mu_{\text{env}}}&\frac{\theta_{b}}{(\gamma_{b}+\mu_{b}+\mu_{db})\mu_{\text{env}}}&\frac{1}{\mu_{\text{env}}}\end{pmatrix} Then the NGM is given by \xi_{\text{NGM}}=\begin{pmatrix}\dfrac{\Lambda_{c}\sigma_{c}(\beta_{c}\mu_{\text{env}}+\beta_{\text{env}}\theta_{c})}{\mu_{dc}\mu_{\text{env}}D_{c}}&\dfrac{\Lambda_{c}(\beta_{c}\mu_{\text{env}}+\beta_{\text{env}}\theta_{c})}{\mu_{dc}\mu_{\text{env}}(\gamma_{c}+\mu_{c}+\mu_{dc})}&\dfrac{\Lambda_{c}(\beta_{bc}\mu_{\text{env}}+\beta_{\text{env}}\theta_{b})}{\mu_{dc}\mu_{\text{env}}(\gamma_{b}+\mu_{b}+\mu_{db})}&\dfrac{\Lambda_{c}\beta_{\text{env}}}{\mu_{dc}\mu_{\text{env}}}\\[10.0pt] 0&0&0&0\\[10.0pt] \dfrac{\Lambda_{b}\beta_{\text{env}}\sigma_{c}\theta_{c}}{\mu_{db}\mu_{\text{env}}D_{c}}&\dfrac{\Lambda_{b}\beta_{\text{env}}\theta_{c}}{\mu_{db}\mu_{\text{env}}(\gamma_{c}+\mu_{c}+\mu_{dc})}&\dfrac{\Lambda_{b}\beta_{b}\mu_{\text{env}}+\Lambda_{b}\beta_{\text{env}}\theta_{b}}{\mu_{db}\mu_{\text{env}}(\gamma_{b}+\mu_{b}+\mu_{db})}&\dfrac{\Lambda_{b}\beta_{\text{env}}}{\mu_{db}\mu_{\text{env}}}\\[10.0pt] 0&0&0&0\end{pmatrix}, where D_{c}=(\sigma_{c}+\mu_{c}+\mu_{dc})(\gamma_{c}+\mu_{c}+\mu_{dc}) . We then find the spectral radius, \rho , of \xi_{\text{NGM}} to obtain the basic reproduction number, \mathcal{R}_{0} given by \mathcal{R}_{0}=\rho(\mathcal{FV}^{-1})=\mathcal{R}_{c}+\mathcal{R}_{b}+\mathcal{R}_{B}, (3.8) where, \displaystyle\mathcal{R}_{c} \displaystyle=\frac{\beta_{c}\Lambda_{c}\sigma_{c}}{\mu_{dc}(\sigma_{c}+\mu_{c}+\mu_{dc})(\gamma_{c}+\mu_{c}+\mu_{dc})}, \displaystyle\mathcal{R}_{b} \displaystyle=\frac{\beta_{b}\Lambda_{b}}{\mu_{db}(\gamma_{b}+\mu_{b}+\mu_{db})}, \displaystyle\mathcal{R}_{B} \displaystyle=\frac{\beta_{\text{env}}\Lambda_{c}\theta_{c}}{\mu_{dc}(\sigma_{c}+\mu_{c}+\mu_{dc})(\gamma_{c}+\mu_{c}+\mu_{dc})\mu_{\text{env}}}+\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{db}(\gamma_{b}+\mu_{b}+\mu_{db})\mu_{\text{env}}}. Next, we compute the endemic equilibrium of the model ( 2.1 ) which is be represented as \psi_{\text{EE}}=(S_{c}^{**},E_{c}^{**},I_{c}^{**},S_{b}^{**},I_{b}^{**},B^{**}). Note that in this case, the infected classes are non-zero. Setting the right-hand side of ( 2.1 ) to \mathbf{0} we will obtain i) B^{**}=\frac{\theta_{c}I_{c}^{**}+\theta_{b}I_{b}^{**}}{\mu_{\rm env}} , ii) I_{b}^{**}=\frac{\beta_{\rm env}S_{b}^{**}B^{**}}{\mu_{db}+\mu_{b}+\gamma_{b}-\beta_{b}S_{b}^{**}} , iii) S_{b}^{**}=\frac{\Lambda_{b}}{\beta_{b}I_{b}^{**}+\beta_{\rm env}B^{**}+\mu_{db}} , iv) I_{c}^{**}=\frac{\sigma_{c}E_{c}^{**}}{\gamma_{c}+\mu_{c}+\mu_{dc}} , v) E_{c}^{**}=\frac{\beta_{c}S_{c}^{**}I_{c}^{**}+\beta_{bc}S_{c}^{**}I_{b}^{**}+\beta_{\rm env}S_{c}^{**}B^{**}}{\sigma_{c}+\mu_{c}+\mu_{dc}} , and vi) S_{c}^{**}=\frac{\Lambda_{c}}{\beta_{c}I_{c}^{**}+\beta_{bc}I_{b}^{**}+\beta_{\rm env}B^{**}+\mu_{dc}} . Now utilising the last three equations (iv) – (vi), we first obtain I_{c}^{**}=\frac{\sigma_{c}\Lambda_{c}}{(\gamma_{c}+\mu_{c}+\mu_{dc})(\sigma_{c}+\mu_{c}+\mu_{dc})}\bigg[\frac{\beta_{c}I_{c}^{**}+f_{1}(I_{c}^{**},B^{**})}{\beta_{c}I_{c}^{**}+f_{1}(I_{c}^{**},B^{**})+\mu_{dc}}\bigg], (3.9) where \displaystyle f_{1}(I_{c}^{**},B^{**}) \displaystyle=\beta_{bc}I_{b}^{**}+\beta_{\rm env}B^{**} \displaystyle=\bigg[\frac{\beta_{bc}\Lambda_{b}\theta_{b}}{(\mu_{db}+\mu_{b}+\gamma_{b})(\beta_{b}\mu_{\rm env}B^{**}-\beta_{b}\theta_{c}I_{c}^{**}+\theta_{b}\beta_{\rm env}B^{**}+\mu_{db})-\beta_{b}}+1\bigg]\beta_{\rm env}B^{**}, (3.10) utilising (i)–(iii). This means we have a nonlinear coupled equation in I_{c}^{**} and B^{**} given by Eq. ( 3.9 ). Now let f_{2}(I_{c}^{**},B^{**})=(\mu_{db}+\mu_{b}+\gamma_{b})(\beta_{b}\mu_{\rm env}B^{**}-\beta_{b}\theta_{c}I_{c}^{**}+\theta_{b}\beta_{\rm env}B^{**}+\mu_{db})-\beta_{b}, meaning function f_{1} can be rewritten as f_{1}(I_{c}^{**},B^{**})=\frac{\beta_{bc}\Lambda_{b}\theta_{b}+f_{2}(I_{c}^{**},B^{**})}{f_{2}(I_{c}^{**},B^{**})}\beta_{\rm env}B^{**}, meaning Eq. ( 3.9 ) can be written as \displaystyle I_{c}^{**}\{f_{3}(I_{c}^{**},B^{**})+\mu_{dc}\}-Af_{3}(I_{c}^{**},B^{**})=0, (3.11) where A=\frac{\sigma_{c}\Lambda_{c}}{(\gamma_{c}+\mu_{c}+\mu_{dc})(\sigma_{c}+\mu_{c}+\mu_{dc})} and f_{3}(I_{c}^{**},B^{**})=\beta_{c}I_{c}^{**}f_{2}(I_{c}^{**},B^{**})+\{\beta_{bc}\Lambda_{b}\theta_{b}+f_{2}(I_{c}^{**},B^{**})\}\beta_{\rm env}B^{**}. Because of the nonlinearity in the equation, it is impossible to come up with a closed form solution of \psi_{\rm EE} . So we leave the algebra up to the point where we have ( 3.11 ). To numerically compute \psi_{\rm EE} for specific parameter values, we will need to use a standard numerical solver. 3.3 Global stability analysis of equilibria The next matter at hand is to perform a global stability analysis of both equilibria, which gives us an analytical overview of the trajectories of the disease dynamics. We take the Lyapunov function approach for both. We begin with the following Lemma. Lemma 3.4 . Let a_{1}>0 and a_{3}>0 be two real numbers satisfying -\frac{C}{D}a_{1}\leq a_{3}\leq-\frac{V}{W}a_{1} with constants V<0 , D<0 , W>0 and C>0 , then Va_{1}+Wa_{3}\leq 0\quad\quad\text{and}\quad\quad Ca_{1}+Da_{3}\leq 0. Proof. Since a_{3}\leq-\frac{V}{W}a_{1} and W>0 , we have Wa_{3}\leq-Va_{1} which gives Va_{1}+Wa_{3}\leq 0 . Similarly, since -\frac{C}{D}a_{1}\leq a_{3} and D<0 , we have -Ca_{1}\geq Da_{3} which gives Ca_{1}+Da_{3}\leq 0 . Note that the quantities -\frac{C}{D} and -\frac{V}{W} are both positive. ∎ We define the Lyapunov function for \psi_{\rm DFE} as \mathcal{L}=a_{1}E_{c}+a_{2}I_{c}+a_{3}I_{b}+a_{4}B. Differentiating the Lyapunov function and substituting \frac{dE_{c}}{dt},\frac{dI_{c}}{dt},\frac{dI_{b}}{dt} and \frac{dB}{dt} using ( 2.1 ), we have \displaystyle\frac{d\mathcal{L}}{dt} \displaystyle=a_{1}\mathopen{}\mathclose{{\left(\frac{\beta_{c}\Lambda_{c}}{\mu_{dc}}I_{c}+\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}}I_{b}+\frac{\beta_{\text{env}}\Lambda_{c}}{\mu_{dc}}B-(\sigma_{c}+\mu_{c}+\mu_{dc})E_{c}}}\right) \displaystyle+a_{2}\mathopen{}\mathclose{{\left(\sigma_{c}E_{c}-(\gamma_{c}+\mu_{c}+\mu_{dc})I_{c}}}\right) \displaystyle+a_{3}\mathopen{}\mathclose{{\left(\frac{\beta_{b}\Lambda_{b}}{\mu_{db}}I_{b}+\frac{\beta_{\text{env}}\Lambda_{b}}{\mu_{db}}B-(\gamma_{b}+\mu_{b}+\mu_{db})I_{b}}}\right) \displaystyle+a_{4}\mathopen{}\mathclose{{\left(\theta_{c}I_{c}+\theta_{b}I_{b}-\mu_{\text{env}}B}}\right). Grouping the terms of E_{c},I_{c},I_{b} and B , we get \displaystyle\frac{d\mathcal{L}}{dt} \displaystyle=(-a_{1}\mathopen{}\mathclose{{\left(\sigma_{c}+\mu_{c}+\mu_{dc})+a_{2}\sigma_{c}}}\right)E_{c}+\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{c}\Lambda_{c}}{\mu_{dc}}-a_{2}(\gamma_{c}+\mu_{c}+\mu_{dc})+a_{4}\theta_{c}}}\right)I_{c} \displaystyle+\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}}+a_{3}\frac{\beta_{b}\Lambda_{b}}{\mu_{db}}-a_{3}(\gamma_{b}+\mu_{b}+\mu_{db})+a_{4}\theta_{b}}}\right)I_{b} \displaystyle+\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{\text{env}}\Lambda_{c}}{\mu_{dc}}+a_{3}\frac{\beta_{\text{env}}\Lambda_{b}}{\mu_{db}}-a_{4}\mu_{\text{env}}}}\right)B. (3.12) The scaling factors a_{2} and a_{4} are calculated in terms of a_{1} and a_{3} by equating the coefficients of E_{c} and B to zero. We get \displaystyle a_{2} \displaystyle=a_{1}\mathopen{}\mathclose{{\left(\frac{\sigma_{c}+\mu_{c}+\mu_{dc}}{\sigma_{c}}}}\right), (3.13) \displaystyle a_{4} \displaystyle=\frac{1}{\mu_{\text{env}}}\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{\text{env}}\Lambda_{c}}{\mu_{dc}}+a_{3}\frac{\beta_{\text{env}}\Lambda_{b}}{\mu_{db}}}}\right). Substituting the values of scaling factors and collecting the inflow and outflow terms together to simplify the derivative of Lyapunov function, we get \displaystyle\frac{d\mathcal{L}}{dt}=a_{1}I_{c}\Bigg[\frac{\beta_{c}\Lambda_{c}}{\mu_{dc}}-\frac{(\sigma_{c}+\mu_{c}+\mu_{dc})(\gamma_{c}+\mu_{c}+\mu_{dc})}{\sigma_{c}}+\frac{\theta_{c}\beta_{\text{env}}\Lambda_{c}}{\mu_{\text{env}}\mu_{dc}}\Bigg]+a_{3}I_{c}\mathopen{}\mathclose{{\left(\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}}{\mu_{\text{env}}\mu_{db}}}}\right) (3.14) \displaystyle+a_{1}I_{b}\mathopen{}\mathclose{{\left(\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}}+\frac{\beta_{\text{env}}\Lambda_{c}\theta_{b}}{\mu_{\text{env}}\mu_{dc}}}}\right)+a_{3}I_{b}\Bigg[\frac{\beta_{b}\Lambda_{b}}{\mu_{db}}-(\gamma_{b}+\mu_{b}+\mu_{db})+\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{\text{env}}\mu_{db}}\Bigg]. Recall that D_{c}=(\sigma_{c}+\mu_{c}+\mu_{dc})(\gamma_{c}+\mu_{c}+\mu_{dc}) . We simplify equation ( 3.14 ) to get \displaystyle\frac{d\mathcal{L}}{dt}=\frac{a_{1}D_{c}I_{c}}{\sigma_{c}}\Bigg[\frac{\beta_{c}\Lambda_{c}\sigma_{c}}{\mu_{dc}D_{c}}-1+\frac{\beta_{\text{env}}\Lambda_{c}\sigma_{c}\theta_{c}}{\mu_{\text{env}}\mu_{dc}D_{c}}\Bigg]+\frac{a_{3}I_{c}D_{c}}{\sigma_{c}}\mathopen{}\mathclose{{\left(\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}\sigma_{c}}{\mu_{\text{env}}\mu_{db}D_{c}}}}\right)+a_{1}I_{b}\mathopen{}\mathclose{{\left(\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}}+\frac{\beta_{\text{env}}\Lambda_{c}\theta_{b}}{\mu_{\text{env}}\mu_{dc}}}}\right) (3.15) \displaystyle+a_{3}(\gamma_{b}+\mu_{b}+\mu_{db})I_{b}\Bigg[\frac{\beta_{b}\Lambda_{b}}{\mu_{db}(\gamma_{b}+\mu_{b}+\mu_{db})}-1+\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{\text{env}}\mu_{db}(\gamma_{b}+\mu_{b}+\mu_{db})}\Bigg], or equivalently \displaystyle\frac{d\mathcal{L}}{dt}=\frac{a_{1}D_{c}I_{c}}{\sigma_{c}}\Bigg[\frac{\beta_{c}\Lambda_{c}\sigma_{c}}{\mu_{dc}D_{c}}-1\Bigg]+I_{c}\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{\text{env}}\Lambda_{c}\theta_{c}}{\mu_{\text{env}}\mu_{dc}}+a_{3}\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}}{\mu_{\text{env}}\mu_{db}}}}\right)+a_{1}I_{b}\mathopen{}\mathclose{{\left(\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}}+\frac{\beta_{\text{env}}\Lambda_{c}\theta_{b}}{\mu_{\text{env}}\mu_{dc}}}}\right) (3.16) \displaystyle+a_{3}(\gamma_{b}+\mu_{b}+\mu_{db})I_{b}\Bigg[\frac{\beta_{b}\Lambda_{b}}{\mu_{db}(\gamma_{b}+\mu_{b}+\mu_{db})}-1\Bigg]+a_{3}I_{b}\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{\text{env}}\mu_{db}}. For the sake of clarity, let us set \displaystyle\frac{d\mathcal{L}}{dt}=U_{1}+U_{2}+U_{3}+U_{4}+U_{5}. (3.17) Observe that \displaystyle U_{2} \displaystyle:=I_{c}\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{\text{env}}\Lambda_{c}\theta_{c}}{\mu_{\text{env}}\mu_{dc}}+a_{3}\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}}{\mu_{\text{env}}\mu_{db}}}}\right) \displaystyle=I_{c}D_{c}\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{\text{env}}\Lambda_{c}\theta_{c}}{\mu_{\text{env}}\mu_{dc}D_{c}}+a_{3}\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}}{\mu_{\text{env}}\mu_{db}D_{c}}}}\right) \displaystyle=D_{c}I_{c}\mathopen{}\mathclose{{\left(a_{1}\mathcal{R}_{B}-a_{1}\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{db}(\gamma_{b}+\mu_{b}+\mu_{db})\mu_{\text{env}}}+a_{3}\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}}{\mu_{\text{env}}\mu_{db}D_{c}}}}\right) \displaystyle=D_{c}I_{c}\mathopen{}\mathclose{{\left(a_{1}(\mathcal{R}_{B}-1)+a_{1}-a_{1}\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{db}(\gamma_{b}+\mu_{b}+\mu_{db})\mu_{\text{env}}}+a_{3}\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}}{.}{\mu_{\text{env}}\mu_{db}D_{c}}}}\right) On the other hand, we set D_{b}:=(\gamma_{b}+\mu_{b}+\mu_{db}) so that \displaystyle U_{3}+U_{5} \displaystyle:=a_{1}I_{b}\mathopen{}\mathclose{{\left(\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}}+\frac{\beta_{\text{env}}\Lambda_{c}\theta_{b}}{\mu_{\text{env}}\mu_{dc}}}}\right)+a_{3}I_{b}\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{\text{env}}\mu_{db}} \displaystyle=D_{b}I_{b}\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}D_{b}}+a_{1}\frac{\beta_{\text{env}}\Lambda_{c}\theta_{b}}{\mu_{\text{env}}\mu_{dc}D_{b}}+a_{3}\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{\text{env}}\mu_{db}D_{b}}}}\right) \displaystyle=D_{b}I_{b}\mathopen{}\mathclose{{\left(a_{1}\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}D_{b}}+a_{1}\frac{\beta_{\text{env}}\Lambda_{c}\theta_{b}}{\mu_{\text{env}}\mu_{dc}D_{b}}+a_{3}(\mathcal{R}_{B}-1)+a_{3}-a_{3}\frac{\beta_{\text{env}}\Lambda_{c}\theta_{c}}{\mu_{dc}D_{c}\mu_{\text{env}}}}}\right) We set V:=1-\frac{\beta_{\text{env}}\Lambda_{b}\theta_{b}}{\mu_{db}D_{b}\mu_{\text{env}}} , W:=\frac{\theta_{c}\beta_{\text{env}}\Lambda_{b}}{\mu_{\text{env}}\mu_{db}D_{c}} , C:=\frac{\beta_{bc}\Lambda_{c}}{\mu_{dc}D_{b}}+\frac{\beta_{\text{env}}\Lambda_{c}\theta_{b}}{\mu_{\text{env}}\mu_{dc}D_{b}} and D:=1-\frac{\beta_{\text{env}}\Lambda_{c}\theta_{c}}{\mu_{dc}D_{c}\mu_{\text{env}}} with V\leq 0 , D\leq 0 and demand that a_{3}\in\mathopen{}\mathclose{{\left[-\frac{C}{D},-\frac{V}{W}}}\right] . Using Lemma 3.4 we obtain the following estimates \displaystyle U_{2} \displaystyle=D_{c}I_{c}(a_{1}(\mathcal{R}_{B}-1)+Va_{1}+Wa_{3}) \displaystyle\leq D_{c}I_{c}a_{1}(\mathcal{R}_{B}-1) \displaystyle0 for u\neq u^{**} , so \mathcal{L} is strictly convex and positive definite. All the scaling parameters a_{j}=(a_{1},a_{2},a_{3},a_{4})>0 . Differentiating equation ( 3.22 ) by using the identity \frac{d}{dt}\mathopen{}\mathclose{{\left(\frac{u}{u^{**}}-\ln\frac{u}{u^{**}}-1}}\right)=\mathopen{}\mathclose{{\left(1-\frac{u^{**}}{u}}}\right)\cdot\frac{1}{u^{**}}\cdot\frac{du}{dt}. (3.23) We get the derivative of the Lyapunov function as follows: \frac{d\mathcal{L}}{dt}=a_{1}\mathopen{}\mathclose{{\left(1-\frac{E_{c}^{**}}{E_{c}}}}\right)\frac{\dot{E}_{c}}{E_{c}^{**}}+a_{2}\mathopen{}\mathclose{{\left(1-\frac{I_{c}^{**}}{I_{c}}}}\right)\frac{\dot{I}_{c}}{I_{c}^{**}}+a_{3}\mathopen{}\mathclose{{\left(1-\frac{I_{b}^{**}}{I_{b}}}}\right)\frac{\dot{I}_{b}}{I_{b}^{**}}+a_{4}\mathopen{}\mathclose{{\left(1-\frac{B^{**}}{B}}}\right)\frac{\dot{B}}{B^{**}}. (3.24) For the sake of simplicity, let us assume that \frac{d\mathcal{L}}{dt}=\tau_{1}+\tau_{2}+\tau_{3}+\tau_{4}. (3.25) Substituting the values of \dot{E}_{c} , \dot{I}_{c} , \dot{I}_{b} , and \dot{B} using model ( 2.1 ) in ( 3.24 ) and using the endemic equilibrium identities \dot{E_{c}}^{**}=\dot{I_{c}}^{**}=\dot{I_{b}}^{**}=\dot{B}^{**}=0 , to calculate and simplify the terms \tau_{1},\tau_{2},\tau_{3} , and \tau_{4} as follows: \tau_{1}=a_{1}\mathopen{}\mathclose{{\left(1-\frac{E_{c}^{**}}{E_{c}}}}\right)\frac{\dot{E}_{c}}{E_{c}^{**}}=a_{1}\mathopen{}\mathclose{{\left(1-\frac{E_{c}^{**}}{E_{c}}}}\right)\mathopen{}\mathclose{{\left(\frac{\beta_{c}S_{c}I_{c}+\beta_{bc}S_{c}I_{b}+\beta_{\text{env}}S_{c}B-(\sigma_{c}+\mu_{c}+\mu_{dc})E_{c}}{E_{c}^{**}}}}\right). (3.26) Now at endemic equilibrium, using the identity \dot{E}_{c}^{**}=0 , i.e., \beta_{c}S_{c}^{**}I_{c}^{**}+\beta_{bc}S_{c}^{**}I_{b}^{**}+\beta_{\text{env}}S_{c}^{**}B^{**}-(\sigma_{c}+\mu_{c}+\mu_{dc})E_{c}^{**}=0. (3.27) We get \displaystyle\tau_{1} \displaystyle=a_{1}\mathopen{}\mathclose{{\left(1-\frac{E_{c}^{**}}{E_{c}}}}\right)\mathopen{}\mathclose{{\left[(\sigma_{c}+\mu_{c}+\mu_{dc})-(\sigma_{c}+\mu_{c}+\mu_{dc})\frac{E_{c}}{E_{c}^{**}}}}\right] \displaystyle=a_{1}(\sigma_{c}+\mu_{c}+\mu_{dc})\mathopen{}\mathclose{{\left[\mathopen{}\mathclose{{\left(1-\frac{E_{c}^{**}}{E_{c}}}}\right)\mathopen{}\mathclose{{\left(1-\frac{E_{c}}{E_{c}^{**}}}}\right)}}\right]. We can rewrite this as \tau_{1}=-\frac{a_{1}}{u_{1}}(\sigma_{c}+\mu_{c}+\mu_{dc})(u_{1}-1)^{2},\\ (3.28) where u_{1}=\frac{E_{c}}{E_{c}^{**}} . Similarly, at endemic equilibrium, using \displaystyle\dot{I}_{c}^{**} \displaystyle=0\Rightarrow\sigma_{c}E_{c}^{**}=(\gamma_{c}+\mu_{c}+\mu_{dc})I_{c}^{**}, \displaystyle\dot{I}_{b}^{**} \displaystyle=0\Rightarrow\beta_{b}S_{b}^{**}I_{b}^{**}+\beta_{\text{env}}S_{b}^{**}B^{**}=(\gamma_{b}+\mu_{b}+\mu_{db})I_{b}^{**}, \displaystyle\dot{B}^{**} \displaystyle=0\Rightarrow\theta_{c}I_{c}^{**}+\theta_{b}I_{b}^{**}=\mu_{\text{env}}B^{**}, we get \displaystyle\tau_{2} \displaystyle=-\frac{a_{2}}{u_{2}}(\gamma_{c}+\mu_{c}+\mu_{dc})(u_{2}-1)^{2}, (3.29) \displaystyle\tau_{3} \displaystyle=-\frac{a_{3}}{u_{3}}(\gamma_{b}+\mu_{b}+\mu_{db})(u_{3}-1)^{2}, (3.30) \displaystyle\tau_{4} \displaystyle=-\frac{a_{4}}{u_{4}}\mu_{\text{env}}\>(u_{4}-1)^{2}, (3.31) where u_{2}=\frac{I_{c}}{I_{c}^{**}},u_{3}=\frac{I_{b}}{I_{b}^{**}},\text{and}\>u_{4}=\frac{B}{B^{**}}. Combining all the terms of \tau_{1},\tau_{2},\tau_{3} and \tau_{4} in ( 3.24 ), we get \displaystyle\frac{d\mathcal{L}}{dt}=-a_{1}(\sigma_{c}+\mu_{c}+\mu_{dc})\mathopen{}\mathclose{{\left(\frac{E_{c}}{E_{c}^{**}}-1}}\right)^{2}-a_{2}(\gamma_{c}+\mu_{c}+\mu_{dc})\mathopen{}\mathclose{{\left(\frac{I_{c}}{I_{c}^{**}}-1}}\right)^{2} (3.32) \displaystyle-a_{3}(\gamma_{b}+\mu_{b}+\mu_{db})\mathopen{}\mathclose{{\left(\frac{I_{b}}{I_{b}^{**}}-1}}\right)^{2}-a_{4}\>\mu_{\text{env}}\mathopen{}\mathclose{{\left(\frac{B}{B^{**}}-1}}\right)^{2}. Since all of the parameters and scaling factors are positive, the right hand side of the above expression is strictly negative except at equilibrium, i.e., \frac{d\mathcal{L}}{dt}\leq 0 . Thus by LaSalle’s Invariance Principle, the endemic equilibrium is globally asymptotically stable within the biologically relevant region. 3.4 Global sensitivity analysis In this study, we used Latin Hypercube Sampling (LHS) and Partial Rank Correlation Coefficient (PRCC) methods to perform a global sensitivity analysis. The LHS technique generated N parameter sets by sampling probabilistic variables P uniformly over predefined ranges, thereby incorporating parameter uncertainty into the analysis. Subsequently, the PRCC method quantified the strength and direction of monotonic relationships between each input parameter P and the corresponding model output Q , while adjusting for the effects of other parameters to isolate the influence of each P on Q . Let P_{i} and Q_{i} denote the sampled input and output values, with means \bar{P} and \bar{Q} , respectively, and N the total number of samples. The PRCC value r , representing the partial correlation between P and Q , is computed as: r=\frac{\mathrm{Cov}(P_{i},Q_{i})}{\sqrt{\mathrm{Var}(P_{i})\,\mathrm{Var}(Q_{i})}}=\frac{\sum_{i=1}^{N}(P_{i}-\bar{P})(Q_{i}-\bar{Q})}{\sqrt{\sum_{i=1}^{N}(P_{i}-\bar{P})^{2}\sum_{i=1}^{N}(Q_{i}-\bar{Q})^{2}}}. (3.33) The PRCC values lie in the range [-1,+1] , where values near \pm 1 indicate stronger monotonic relationships. A positive PRCC ( r>0 ) signifies a direct relationship, meaning that increases in P lead to increases in Q . Conversely, a negative PRCC ( r<0 ) indicates an inverse relationship, where increases in P correspond to decreases in Q . 4 Numerical simulation and results In this section, we examine the dynamical behaviour of the model ( 2.1 ) using numerical simulations. All simulations were conducted using the Runge–Kutta solver, ode45 in MATLAB2025 [ 22 ] . The initial conditions were specified as S_{c}(0)=1500 , E_{c}(0)=0 , I_{c}(0)=0 , S_{b}(0)=499 , I_{b}(0)=1 , and B(0)=0 . The parameter values are provided in Table 2 . Table 2 : Parameters of model 2.1 , their values, and sources. Parameter Value Source Parameter Value Source Parameter Value Source \Lambda_{b} 10 Assumed \mu_{c} 0.02 Assumed \gamma_{b} 0.2 [ 1 ] \Lambda_{c} 30 Assumed \mu_{b} 0.02 [ 34 ] \gamma_{c} 0.2 [ 32 , 33 ] \beta_{b} 0.00005 [ 34 ] \mu_{dc} 0.02 [ 25 ] \theta_{b} 0.1 Assumed \beta_{c} 0.00008 [ 2 ] \mu_{db} 0.02 [ 25 ] \theta_{c} 0.1 Assumed \beta_{bc} 0.00008 Assumed \mu_{\text{env}} 0.5 Assumed \beta_{\text{env}} 0.00005 Assumed \sigma_{c} 0.25 [ 33 ] Using the values of the parameters in Table 2 and the threshold quantity defined in equation 3.8 , the reproduction numbers were estimated as follows: \mathcal{R}_{c}=0.4306 , \mathcal{R}_{b}=0.1044 , \mathcal{R}_{B}=0.2362 , and \mathcal{R}_{0}=0.7712 . Since \mathcal{R}_{0}<1 indicates the inability of the disease to invade, the result \mathcal{R}_{0}=0.7712 suggests HPAI will eventually be eliminated from the population. The epidemiological insight of these results indicates that the transmission from cattle to cattle ( \mathcal{R}_{c} ) is the primary driver of the dynamics of HPAI, followed by environmental transmission ( \mathcal{R}_{B} ) and, to a lesser extent, bird-to-cattle transmission ( \mathcal{R}_{b} ). When a single case is introduced into a fully susceptible population, the long-term outcome, persisted or eliminated, is determined by whether \mathcal{R}_{0}>1 or \mathcal{R}_{0}<1 . 4.1 Dynamics of the state variables of model 2.1 Figure 2 shows the time series of HPAI dynamics in cattle illustrating the temporal progression of susceptible, exposed, and infected compartments in 365 days. In panel (a), the susceptible cattle population decreases steadily from an initial value of 1,500 to 340 cattle on day 200 , reflecting the depletion of susceptibles as the outbreak progresses. In panel (b), the exposed compartment increases sharply during the early phase of the outbreak, reaching a peak of approximately 370 cattle on day 75 . This peak reflects the accumulation of cattle in the incubation phase before progressing to active infection. The infected compartment as seen in panel (c) exhibits a pronounced rise following the exposed peak, with infections reaching a maximum of approximately 330 individuals on day 81 . As time progresses, the number of infected cattle declines due to recovery or disease-induced mortality. Figure 2 : Dynamics of state variables for cattle with \beta_{c}=0.0008 , other parameters remain fixed as in Table 2 Figure 3 : Dynamics of state variables for wild birds and environment with \beta_{c}=0.0008 , other parameters remain fixed as in Table 2 The transmission dynamics of wild birds and their environment over a 365-day period are shown in Figure 3 . The susceptible bird population remains relatively stable, fluctuating slightly around 490 individuals, suggesting that disease-induced mortality has minimal impact on the overall population size during the simulation period. In panel (b), the number of infected birds decreases from 10 to about 2 by day 54, then begins to rise again, peaking at approximately 7 individuals around day 90. Following this peak, the number of infected birds declines sharply as they either recover or die from the disease. As shown in panel (c), the shedding of HPAI virus into the environment peaks around day 90, reaching approximately 65 units. 4.2 Extinction of disease Different initial values are used for the model state variables ( 2.1 ), to illustrate the global asymptotic behaviour of the disease free equilibrium ( \psi_{\text{DFE}} ) given in ( 3.5 ). It is observed from Fig. 4 that irrespective of the value of the initial size of the state variable, the solution will always converges toward \psi_{\text{DFE}} , which implies that HPAI eventually dies out in the population when the threshold quantity \mathcal{R}_{0}<1 . This numerical result illustrates the result of ( 3.21 ). (a) Stability plot for exposed cattle (b) Stability plot for infectious cattle (c) Stability plot for infectious wild birds (d) Stability plot for contaminated environment Figure 4 : Global stability of the disease-free equilibrium (DFE) when \mathcal{R}_{0}=0.7712<1 . Solutions starting from different initial conditions all converge to the DFE 4.3 Persistence of disease To investigate the stability behaviour of the endemic equilibrium ( \psi_{\text{EE}} ), we use the different initial sizes of the population to depict the convergence of solution trajectories in Fig. 5 . This validates the global stability result of \psi_{\text{EE}} in ( 3.32 ). We observed that irrespective of the value of the initial size of the state variable, the solution will always converge toward \psi_{\text{EE}} , which implies that HPAI persists in the population when the threshold quantity \mathcal{R}_{0}>1. (a) Stability plot for exposed cattle (b) Stability plot for infectious cattle (c) Stability plot for infectious wild birds (d) Stability plot for contaminated environment Figure 5 : Global stability of the endemic equilibrium (EE) when \mathcal{R}_{0}=5.5860>1 . Solutions starting from different initial conditions all converge to the EE 4.4 Global sensitivity analysis In this section, we used the LHS in conjunction with the PRCC of Section 3.4 to explore the influence of the model parameters on the threshold quantity \mathcal{R}_{0} . We used a sample size N=1000 , and the results of the sensitivity analysis are shown in Figure 6 . Figure 6 : Global sensitivity analysis of the basic reproduction number \mathcal{R}_{0} with respect to model ( 2.1 ) parameters using Partial Rank Correlation Coefficients (PRCC) and Latin Hypercube Sampling (LHS) The longer bars indicate that those parameters have a significant impact on \mathcal{R}_{0} . A positive PRCC value signifies a direct relationship, which means that increasing the parameter leads to an increase in \mathcal{R}_{0} . For example, \beta_{c},\beta_{b},\beta_{\rm env} and \theta_{c} are positive, therefore, increasing any of these parameters would increase the value of \mathcal{R}_{0} . This implies that increasing these parameters contributes to the burden of infection. In contrast, these parameters, \mu_{dc},\gamma_{c}, and \sigma_{c} , have a negative PRCC value, indicating an inverse relationship. Therefore, increasing any of these parameters corresponds to decreasing the basic reproduction number, \mathcal{R}_{0} . Then increasing these parameter values contributes to a reduction in the number of infections. (a) (b) (c) (d) (e) (f) Figure 7 : Effects of varying significant parameters, including the transmission and recovery rates, on the number of infected cattle, wild birds, and the environment over time The results of this sensitivity analysis can be seen in simulations. For example, the effects of varying significant parameters, including the transmission and recovery rates, on the number of infected cattle and wild birds over time are shown in 7 . As seen in Fig. 7(a) , when \beta_{c}=0.0004 the number of infected cattle is 160 , while for \beta_{c}=0.0015 the number of infected cattle is 450 . This shows that an increase in the transmission rate between cattle results in a significant increase in the number of infected cattle, showing the sensitivity of disease spread to contact transmission dynamics. Similar behaviour is observed when \beta_{b} and \beta_{\rm env} are varied, see Figs. 7(c) and 7(d) . These dynamics clearly highlight the importance of implementing appropriate biosecurity measures to reduce the spread of avian influenza between cattle, wild birds, and the environment. Figure 7(b) shows the dynamics of infected cattle for three values of the cattle recovery rate, \gamma_{c} . For a low recovery rate of \gamma_{c}=0.05 , the highest infected cattle in the population is approximately 750 compared to when \gamma_{c}=0.4 where the highest infected cattle are approximately 150 . This indicates that a slower recovery rate results in a higher burden of infection. A similar behaviour is observed in Fig. 7(f) , where varying the decay rate of pathogens in the environment, \mu_{\rm env} , significantly affects the infection dynamics: slower decay (i.e., longer persistence of the pathogen) leads to higher and more sustained levels of infection. These findings emphasize the role of recovery rates and longevity of environmental pathogens in shaping the course of avian influenza outbreaks in cattle populations and the environment. Next, we simulate the impact of varying two model parameters simultaneously on the threshold quantity \mathcal{R}_{0} . The results of three combinations of two parameters are shown in Fig. 8 , in each plot the magenta line indicates the threshold line that is when \mathcal{R}_{0}=1 . Fig. 8(a) shows how \mathcal{R}_{0} varies with the transmission rate between cattle \beta_{c} and the recovery rate of cattle \gamma_{c} . We see that in the upper left region, high \beta_{c} and low \gamma_{c} implies high \mathcal{R}_{0} , this is because infected cattle stay infectious longer when \gamma_{c} is small. The bottom right region corresponds to low \beta_{c} and high \gamma_{c} , resulting in low \mathcal{R}_{0} . A similar result is observed for the (\beta_{b},\gamma_{b}) plane in Fig. 8(b) . Finally, the result of the variation of \mathcal{R}_{0} as the rate of environmental transmission, \beta_{\rm env} , and the rate of environmental contamination by cattle, \theta_{c} are simultaneously varied is shown in Fig. 8(c) . We observe that these two parameters are highly correlated and have a similar combined effect on \mathcal{R}_{0} . In the top right region of the plot, the high value of \beta_{\rm env} and \theta_{c} corresponds to the high value \mathcal{R}_{0} , indicating a high risk of outbreak. In contrast, in the bottom left region, low values of \beta_{\rm env} and low \theta_{c} result in low \mathcal{R}_{0} , suggesting effective disease control. This result aligns with the result in Fig. 6 , where increasing either \beta_{\rm env} or \theta_{c} leads to an increase in \mathcal{R}_{0} , indicating the critical role of environmental contamination in the spread of avian influenza. (a) (b) (c) Figure 8 : The contour plots of the basic reproduction number \mathcal{R}_{0} as function of (a) the transmission rate among cattle ( \beta_{c} ) and recovery rate of cattle ( \gamma_{c} ), (b) the transmission rate among wild birds ( \beta_{b} ) and recovery rate of wild bird ( \gamma_{b} ). (c) the environmental transmission rate ( \beta_{\rm env} ) and rate of environmental contamination by cattle ( \theta_{c} ) 5 Conclusion Avian influenza, particularly the highly pathogenic avian influenza (HPAI) virus, poses a significant threat to both animal and public health, with severe ecological and economic consequences. Traditionally associated with outbreaks in poultry and wild bird populations, recent spillover events have increasingly implicated domestic livestock, particularly cattle [ 23 ] . These complexities require a theoretical framework to understand and predict the interplay between environmental, wildlife, and livestock factors that drive disease spread [ 24 ] . This study developed a mathematical model to examine the transmission of HPAI between dairy cattle and wild birds, which incorporates both direct contact and indirect environmental pathways. The model captures essential characteristics of transmission dynamics and provides insight into the conditions that lead to disease elimination or persistence, depending on the value of the basic reproduction number \mathcal{R}_{0} . Explicit closed conditions were formulated to analytically diagnose the persistence of the disease ( \mathcal{R}_{0}>1 ) or the death of the disease ( \mathcal{R}_{0}<1 ). Global stability for both equilibria was also confirmed. Numerical simulations illustrated the dynamics over time and sensitivity analysis identified key parameters, such as cattle-to-cattle transmission, environmental contamination, and recovery rates, as the most influential in shaping outbreak outcomes. This model offers a foundation for understanding HPAI outbreaks in mixed livestock–wildlife systems and highlights the importance of environmental management and contact reduction in disease control. From a policy perspective, these findings support interventions aimed at improving farm-level biosecurity, monitoring environmental contamination, and controlling interactions between wildlife and livestock. Future work could extend this model to include transmission between herds, incorporate age structure within cattle populations, and account for seasonal effects in wild bird migration and environmental persistence. These extensions would improve the ability of the model to support targeted surveillance and control strategies in real world conditions. Code Availability The code used to generate the results and figures in this study is available at: https://github.com/hamfat/HPAI-Cattle_Wildlife_Environment