Spatially varying coefficients improve discrete choice models for tuna purse seine fisheries in the Western–Central Pacific ICES Journal of Marine Science

The discrete choice model (DCM) is commonly used to analyze fishing behavior and model fishing location choices based on choice attributes such as expected revenue, cost, and previous effort. However, traditional or mixed DCMs treat parameters among each fishing location as independent and fail to account for spatial autocorrelation among fishing grounds. To address this limitation, we extend traditional DCM by incorporating spatial autocorrelation and spatially varying coefficients (SVCs) to account for latent processes linked to environmental conditions, referred to as spatial DCM. We first develop a diffusion-taxis movement simulation model to simulate fishing vessel behavior, where spatial preferences are influenced by tuna density and oceanographic indices such as the El Niño-Southern Oscillation (ENSO). ENSO is incorporated in the simulation as a time-varying climate index that is multiplied by an SVC, modeling how fishermen adapt fishing strategies in response to regional oceanographic conditions. The simulation testing shows the spatial DCM effectively estimates the spatial preference generated by the movement simulation model through the incorporation of SVCs. Finally, we suggest that spatial DCM can be a useful tool to analyze and forecast fishing behavior for tuna purse seine fisheries in the Western and Central Pacific Ocean (WCPO). The application results showed that the spatial DCM can identify baseline fishing preferences, seasonal spatial variations, and spatially varying responses to environmental conditions beyond the utility predicted from covariates such as expected catch (previous year catch value) and cost (previous year effort and distance to port). Specifically, by incorporating SVCs, the spatial DCM reveals that El Niño events enhance fishing activity in the western WCPO (Papua New Guinea and Federated States of Micronesia), while La Niña events increase fishing activity in the eastern WCPO (Kiribati), presumably representing how fishers adapt to changes in tuna distribution and catch efficiency driven by shifts in oceanographic conditions associated with climate events. We therefore conclude that the spatial DCM is a useful approach to account for spatial autocorrelation and latent oceanographic influences.

离散选择模型（discrete choice model, DCM）是分析捕捞行为、基于预期收益、成本及前期作业量等选择属性构建捕捞选址选择模型的常用工具。然而，传统或混合DCM将各捕捞位点的参数视为独立变量，未能考量渔场间的空间自相关效应。为解决这一局限，我们将空间自相关与空间变系数（spatially varying coefficients, SVCs）纳入传统DCM，以刻画与环境条件相关的潜在过程，由此构建空间离散选择模型（spatial DCM）。

我们首先构建扩散-趋化运动模拟模型，用于模拟渔船行为——该模型中，空间偏好受金枪鱼密度及厄尔尼诺-南方涛动（El Niño-Southern Oscillation, ENSO）等海洋学指数影响。研究将ENSO作为时变气候指数纳入模拟，通过与空间变系数相乘，刻画渔民如何根据区域海洋环境条件调整捕捞策略。

模拟测试结果表明，通过引入空间变系数，空间DCM可有效估计由运动模拟模型生成的空间偏好。最后，我们提出空间DCM可作为分析、预测中西太平洋（Western and Central Pacific Ocean, WCPO）金枪鱼围网渔业捕捞行为的有效工具。

应用结果显示，相较于仅基于预期渔获量（上年渔获值）与成本（上年作业量及到港距离）等协变量得到的效用预测结果，空间DCM可识别基准捕捞偏好、季节性空间变异及对环境条件的空间异质性响应。具体而言，通过纳入空间变系数，空间DCM揭示了：厄尔尼诺事件会增强中西太平洋西部（巴布亚新几内亚与密克罗尼西亚联邦）的捕捞活动，而拉尼娜事件则会提升中西太平洋东部（基里巴斯）的捕捞活跃度，这大致反映了渔民如何适应由气候事件相关海洋环境变化驱动的金枪鱼分布与捕捞效率变动。

综上，我们认为空间DCM是一种可有效刻画空间自相关与潜在海洋环境影响的有效分析方法。