Multi-sites fisheries modeling with Itô stochastic differential equations and stochastic optimization problem

Technical Summary: Multi-sites Fisheries Modeling with Itô Stochastic Differential Equations and Stochastic Optimization

Problem Statement
The paper addresses the modeling and optimization of multi-site fisheries, specifically focusing on systems utilizing Fish Aggregating Devices (FADs). While deterministic models have historically dominated population dynamics, they fail to account for natural variability and uncertainties such as climate change and ecological disturbances. The authors propose a framework to model the evolution of fish biomass and fishing efforts across multiple sites ($L$) and an offshore stock ($s$) using stochastic processes. The core challenge involves formulating a system that captures random fluctuations in birth, death, migration, and fishing activities without incurring prohibitive computational costs, particularly regarding the calculation of matrix square roots required for standard stochastic differential equation (SDE) formulations. Furthermore, the paper seeks to solve a stochastic optimization problem to maximize fishing profits under these dynamic, uncertain constraints.

Methodology
The research employs a two-part methodology: stochastic modeling and stochastic optimization.

1. Stochastic Modeling (SDE Formulation):

   • Discrete to Continuous: The authors begin by defining a discrete-time stochastic process for a system with $L$ sites. They enumerate all possible state changes (births, deaths, migrations between sites and the offshore stock, and boat movements) over a small time interval $\Delta t$.
   • Drift and Diffusion: From these discrete transitions, they derive the expectation vector (drift term, $f$) and the covariance matrix ($V$).
   • Equivalent Systems: The standard SDE formulation $dX_t = f(t, X)dt + B(t, X)dW(t)$ requires $B = V^{1/2}$. The authors note that calculating the square root of a large $(2L+1) \times (2L+1)$ matrix is computationally expensive and lacks explicit formulas for large $L$.
   • Matrix Decomposition: To avoid direct matrix square root calculation, the authors utilize an equivalent formulation $dX_t = f(t, X)dt + H(t, X)dW^{**}(t)$, where $V = HH^T$. The matrix $H$ is constructed explicitly using block matrices representing births, deaths, migrations, and boat movements. This approach avoids the need to compute $V^{1/2}$ directly by utilizing the explicit structure of $H$, which is derived from the underlying transition probabilities. For $L \ge 3$, the matrix $H$ has dimensions $(2L+1) \times (L+1)^2$, meaning it has more columns than the state dimension, but its explicit construction bypasses the numerical cost of decomposing the full covariance matrix $V$.
   • Generalization: The model is first detailed for $L=2$ sites and then generalized for $L \ge 3$ sites, incorporating logistic growth for fish populations and specific catchability rates.

2. Stochastic Optimization:

   • Objective Function: The goal is to maximize the expected total profit over a time horizon $[0, T]$, defined as the difference between revenue (catch price $\times$ catchability $\times$ biomass $\times$ effort) and costs (cost per unit effort).
   • Constraints: The optimization is subject to the derived SDE dynamics. A stopping time $\tau$ is defined as the first time the stock reaches zero or time $T$, whichever comes first.
   • Dynamic Programming: The problem is solved using the Dynamic Programming Principle, leading to a Hamilton-Jacobi-Bellman (HJB) equation. The value function $U(x, t)$ represents the maximum expected future profit.
   • Simplification for Control: For the optimization phase, the authors simplify the diffusion matrix $H$ to $H_1$, excluding terms related to boat movement costs and sales, focusing solely on the biological and catch dynamics relevant to the control variable (fishing effort).

Key Contributions

• Equivalent SDE Systems: The paper proposes and establishes the equivalence between the standard SDE formulation (requiring $V^{1/2}$) and an alternative formulation using a matrix $H$ such that $V=HH^T$. This alternative is theoretically sound and numerically more efficient as it avoids the computational burden of calculating matrix square roots for high-dimensional systems by constructing $H$ explicitly from block matrices.
• Multi-Site Stochastic Framework: The authors develop a comprehensive stochastic model for multi-site fisheries that explicitly accounts for fish migration between sites and an offshore stock, as well as the movement of fishing vessels between sites.
• Stochastic Optimization Solution: The study formulates and solves a stochastic optimal control problem for multi-site fisheries. By coupling the SDE dynamics with the HJB equation, the authors derive optimal fishing effort strategies that maximize profit under uncertainty.

Results

• Numerical Validation: The authors performed numerical simulations using the Euler-Maruyama method with 500 sample paths for cases with $L=2$ and $L=10$ sites.
• Equivalence Confirmation: Comparisons between the standard formulation (using matrix $B=V^{1/2}$) and the proposed formulation (using matrix $H$) showed that the expected values for biomass and fishing efforts were nearly identical. The absolute errors between the two methods were low, validating the theoretical equivalence and the utility of the $H$-formulation for reducing computational complexity.
• Optimization Outcomes: Simulations of the HJB equation revealed that the value function increases with stock and biomass levels but decreases with fishing effort. The optimal fishing effort policies derived from the model demonstrate that it is possible to balance economic returns with ecological sustainability even in the presence of stochastic shocks.

Significance and Claims
The paper claims to contribute to the field of fisheries management by providing a rigorous stochastic framework that bridges the gap between theoretical modeling and numerical efficiency.

• Theoretical Contribution: It offers a mathematically equivalent but computationally lighter alternative to standard SDE formulations for multi-site systems, making the modeling of complex, high-dimensional fisheries more feasible.
• Practical Application: The results suggest that stochastic optimization methods are necessary for robust control in modern fisheries, which face inherent uncertainties. The proposed framework supports the development of adaptive harvest strategies and sustainable multi-site policies.
• Limitations and Future Work: The authors modestly acknowledge that their model assumes constant parameters (e.g., migration rates) and does not include age structure or complex species interactions (predator-prey). They propose future work to extend the model to ecosystem-based management, incorporate trophic interactions, and couple fisheries models with climate models. They also suggest exploring advanced numerical schemes (e.g., Milstein, Runge-Kutta) beyond the Euler-Maruyama method used in this study.