Time Series Decomposition and LSTM Neural Networks for Forecasting Transportation CO2 Emissions in Saudi Arabia: Supporting Vision 2030 Climate Objectives

This study develops an advanced forecasting framework by combining time series decomposition with Long Short-Term Memory (LSTM) neural networks to predict CO2 emissions from Saudi Arabia’s transportation sector through 2034, in support of Vision 2030 climate objectives. Using historical data from 1970 to 2024, the results show that LSTM models significantly outperform traditional ARIMA approaches, achieving superior accuracy with an error rate of 6.82% for total emissions compared to 13.51% for ARIMA. The analysis focuses on CO2 emissions in three sectors: road transport, aviation transport, and total transport emissions. Findings indicate that road transport is the dominant source, contributing over 95% of total emissions, with projections rising from 149.8 million tons in 2025 to 172.5 million tons by 2034. Aviation emissions are expected to increase from 3.79 to 6.61 million tons over the same period, while total transport emissions reflect the combined upward trend of both sectors. Despite a gradual decline in annual growth rates, the persistent increase in emissions underscores the urgent need for sustainable transport policies, particularly those that promote electric vehicles and expanded public transit systems. The study’s integration of STL decomposition with LSTM modeling provides a powerful, evidence-based tool for policymakers to guide CO2 mitigation strategies and track progress toward Vision 2030 climate targets.