Univariate approaches for forecasting of commercial octopus (Octopus vulgaris Cuvier, 1797) landings: A comprehensive analysis of traditional time series models, machine learning algorithms and a deep learning algorithm

Identificadores
URI: http://hdl.handle.net/10498/39281
DOI: 10.1016/J.FISHRES.2025.107622
ISSN: 0165-7836
Estadísticas
Métricas y Citas
Metadatos
Mostrar el registro completo del ítemFecha
2026Departamento/s
BiologíaFuente
Fisheries Research - 2026, Vol. 293, Article number 107622Resumen
In the Gulf of Cádiz and the Alboran Sea, the common octopus (Octopus vulgaris Cuvier, 1797) represents a key economic resource for the Andalusian artisanal fleet, which employs both small-scale fishing gear and bottom trawling techniques. Due to the biological characteristics of this species, its high environmental sensitivity, and its economic importance, it is essential to implement management strategies based on accurate projections and robust models. This study analyzes monthly commercial landing data from the Andalusian Fisheries Marketing and Production Information System (IDAPES) from 2000 to 2022, combining traditional time series models (ARIMA and ETS) with machine learning algorithms (NNAR, MLP and ELM) and a deep learning architecture (LSTM) to characterize and project the fishery's dynamics from a univariate perspective. Historically, landings have increased in both areas, with 57 % originating from the Gulf of Cádiz, where the strength of seasonality was moderate (0.4), and 43 % from the Alboran Sea, where the strength of seasonality was strong (0.7). Wavelet analysis identified variability cycles of 6 years in the Gulf of Cádiz and 7 years in the Alboran Sea. Traditional models detected trends and autoregressive patterns, while machine learning algorithms captured complex and nonlinear patterns. Regarding deep learning, the LSTM showed strong but context-dependent performance, competitive under the Alboran Sea's stable seasonality, but not superior under the Gulf of Cádiz's pulse-driven dynamics. One- and two-year projection scenarios revealed different dynamics. In the Gulf of Cádiz, ARIMA and ELM forecasted peak landings between November and January, whereas MLP indicated a progressive increase until November, the LSTM aligned with a moderate November–December peak. Consequently, the ensemble approach showed a pattern of higher landings from November to January, accompanied by a minor rise between June and August, followed by a gradual decline throughout the rest of the year. In the Alboran Sea, all models, along with their ensemble projection, indicated peak landings in June, followed by a decline towards the end of the year. This study highlights the effectiveness of combined approaches that integrate traditional statistical techniques with machine learning and deep learning to enhance fishery landing projections. These tools provide valuable insights for fishery strategic planning and contribute to the sustainability of one of the most economically significant marine resources in southern Spain.
Materias
Modelling fisheries; Time series; Marine resource; Artificial intelligenceColecciones
- Artículos Científicos [12077]
- Articulos Científicos Biología [883]






