AUTHOR=Bade Abdirisak Osman , Shire Dahir Muse , Hassan Saleban Mohamed , Muse Abdisalam Hassan 
  
TITLE=Forecasting pastoralist red meat production trajectories in a fragile and climate-vulnerable state: a comparative evaluation of classical, state-space, long-memory, and neural network models in Somalia
  
JOURNAL=Pastoralism: Research, Policy and Practice
  
VOLUME=Volume 16 - 2026
  
YEAR=2026
  
URL=https://www.frontierspartnerships.org/journals/pastoralism-research-policy-and-practice/articles/10.3389/past.2026.17161
  
DOI=10.3389/past.2026.17161
  
ISSN=2041-7136
  
ABSTRACT=BackgroundThe national economy of Somalia is significantly reliant on the production of pastoralist red meat, a sector that is increasingly debilitated by severe and recurrent climate-related shocks. Accurate forecasting of supply trajectories is crucial for proactive food security planning. However, empirical evaluations of mathematical and machine learning models in fragile states are notably scarce.ObjectivesThis study aimed to model, evaluate, and project Somalia’s aggregate pastoralist red meat output over a 10-year horizon (2025–2034) using a diverse set of econometric and computational forecasting models.MethodsUtilizing a historical dataset spanning 64 years (1961–2024), the out-of-sample validation accuracy for the period 2015–2024 was assessed across seven forecasting frameworks. These frameworks included classical models (ARIMA, Theta), state-space models (ETS, TBATS, BATS), long-memory models (ARFIMA), and autoregressive neural networks (ARNN). Stationarity diagnostics were systematically performed using the Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) unit-root tests.ResultsThe stationarity diagnostics confirmed that the historical series is integrated of order one, I(1). Over the out-of-sample window, the long-memory ARFIMA model substantially outperformed all other frameworks, achieving the lowest predictive errors (sMAPE = 3.11%, RMSE = 10,836.89, MASE = 0.85, and Theil’s U = 0.69). The non-linear ARNN ranked as the second-best model (sMAPE = 4.49%), while classical and state-space models suffered from systematic bias, predicting overly smoothed trajectories that completely missed real-world shocks. Decadal projections (2025–2034) generated by the fully fitted ARFIMA model forecast a slow, steady decline and stabilization in red meat production, dropping from 175,332.80 to 168,884.30 tons.ConclusionThis projected plateau indicates that Somalia’s traditional pastoral system could be nearing dynamic ecological and structural limits under escalating environmental pressures. Transitioning from reactive, crisis-driven aid to proactive climate resilience requires targeted investments in digital index-based livestock insurance, commercial fodder value chains, and mobile veterinary infrastructure.