Tracking the future of global N2O gas emissions with data-driven forecasts

Abstract

Forecasting methods are widely used to make accurate decisions before uncertainties or potential problems arise in the future. This research examines the independent performances of traditional statistical Seasonal Autoregressive Integrated Moving Average Model (SARIMA) and deep learning models Long-Short Term Memory Neural Network (LSTM) and Gated Recurrent Unit (GRU) forecasting models in order to forecast the progress of global N₂O (Nitrous Oxide) emissions to 2050. The monthly N₂O emission values between 2001 and 2024 were used to forecast levels up to 2050. The forecast results and actual values were evaluated with R², RMSE, MSE, NSE, MAE and MAPE% error scales. The findings showed that all three methods were successful in forecasting global N₂O gas emissions, but SARIMA model (0.9998 R², 0.011 RMSE, 0.0001 MSE, 1.000 NSE, 0.004 MAE and 0.006 MAPE%) was the method that best fit the available data and produced forecasts with the least error. The results obtained predicted that N₂O emissions could be 8.16 % higher than current levels by 2050. The year 2050 is an important date determined as the global net zero emission target. The models in this study provide a concrete and important contribution to understanding the future course of N₂O emissions and the relationship with the net zero target. It can be used as a guide in the processes of companies to achieve their environmental policies and sustainability goals within the scope of state policies and environmental regulation reporting, when it is desired to increase energy efficiency by reducing emission values, and when it is necessary to calculate climate change risks.