Air temperature time series modelling and prediction using neural network and SARIMA model

Accurate modelling and forecasting of air temperature play a vital role in many facets of human life and environmental endeavours. This study specifically focuses on determining the superior method for modelling and predicting air temperature in Warri, a city located in the southern region of Delta State, Nigeria. The dataset used in this research was provided by NiMet (2000–2020). To accomplish the modelling and prediction task, we employed two prominent methods: the Seasonal Autoregressive Integrated Moving Average (SARIMA) and the Neural Network Time Series Autoregressive (NNETAR) models. Initially, the dataset exhibited non-stationary behaviour, as evident from the time plot series and confirmed by conducting the Augmented Dickey-Fuller (ADF) test. However, after applying the first difference, the data was transformed into a stationary series. Further analysis using the Hegy and Canova-Hasen tests revealed the presence of seasonality in the series, with a seasonal order of 1. Upon achieving stationarity, we proceeded to evaluate various models, and both the SARIMA (2,1,2) (1,0,1) [12] and NNAR (22,1,12) [12] models demonstrated the best performance. However, when assessing the accuracy measures, the NNAR (22,1,12) [12] model outperformed the SARIMA (2,1,2) (1,0,1) [12] model. Having identified the superior model, we utilised the fitted NNAR (22, 1, 12) [12] model to forecast the air temperature for the next five years. This approach provides valuable insights into the future climate trends of Warri City and supports informed decision-making in various sectors.

Adhikari, R. and Agrawal, R.K. (2013). An introductory study on time series modeling and forecasting, Lambert Academic Publishing, arXiv preprint arXiv:1302.6613.

Ahmar, A.S. and Boj, E. (2021). Application of neural network time series (NNAR) and ARIMA to forecast infection fatality rate (IFV) of COVID-19 in Brazil, International Journal on Informatics Visualization, 5(1), 8-10.

Balibey, M. and Turkyilmaz, S. (2015). A time series approach for precipitation in Turkey, Gazi University Journal of Science, 28, 549-559.

Box, George, E.P., Gwilym, M.J., Gregory, C.R. and Greta, M.L. (2015). Time Series Analysis: Forecasting and Control, John Wiley & Sons.

Chen, P., Niu, A., Liu, D. and Jiang, W. (2018). Time series forecasting of temperatures Using SARIMA: An example from Nanjing, IOP Conference Series: Materials Science and Engineering, 394(5), 052024.

Di, C., Yang, X. and Wang, X. (2014). Hybrid neural network models for hydrologic time series forecasting, Plus ONE, 9, e104663.

Dimri, T., Ahmad, S. and Sharif, M. (2020). Time series analysis of climate variables using seasonal ARIMA approach, Journal of Earth System Science, 129, 149.

Dwivedi, D.K., Kelaiya, J.H. and Sharma, G.R. (2019). Forecasting monthly rainfall using autoregressive integrated moving average model (ARIMA) and artificial neural network (ANN) model: A case study of Junagadh, Gujarat India, Journal of Applied and Natural Science, 11(1), 35-41.

Green, M. A. (2003). General temperature dependence of solar cell performance and implications for device modeling. Progress in Photovoltaics: Research and Applications, 11, 333-340.

Grigonyte, E. and Butkeviciuye, E. (2016). Short-term wind speed foresting using ARIMA model, Energetika, 62.

IPPC (2018). Summary for policymakers in global warming of 1.5oC. An IPCC special report on the impact of global warming of 1.5oC above pre-industrial levels and related global greenhouse gas emission pathways, in the context of strengthening the global. Word Meteorological Organization, Geneva, Technical Report.

Jain, A. and Kumar, A.M. (2007). Hybrid neural network models for hydrologic time series forecasting, Applied Soft Computing, 7, 585-592.

Khandelwa, I., Adhikari, R. and Verma, G. (2015) Time series forecasting using hybrid ARIMA and ANN models based on DWT decomposition, Procedia Computer Science, 48, 173-179.

Kreuzer, D., Munz, M. and Schluter, S. (2020). Short-term temperature forecasts using a convolutional neural network - An application to different weather stations in Germany. Machine Learning with Applications, 2, 100007.

Kuhn, M. and Johnson, K. (2018). Applied predictive modeling, First edition, Springer, New York.

Lee, S., Lee Y. and Son, Y. (2020). Forecasting Daily temperature with different time interval data using deep neural networks, Applied Sciences, 10,       1609. https://doi.org/10.3390/app10051609.

Li, Y. and Dzombak, D.A. (2020). Use of the autoregressive integrated moving average (ARIMA) model to forecast near-term regional temperature and precipitation. Weather and Forecasting, 35(3), 959-976.

Mitu, K.N. and Hasan, K. (2021). Modeling and forecasting daily temperature time series in Memphis, Tennessee, International Journal of Environmental Monitoring and Analysis, 9(6), 214-221.

Mushtaq, R. (2011). Augmented Dickey-Fuller Test, https://ssrn.com/abstract=1911068 or http://dx.doi.org/10.2139/ssrn.1911068.

Nishimwe, A.M.R. and Reiter, S. (2022). Using artificial intelligence models and degree days method to estimate the heat consumption evolution of a building stock until 2050: A case study in a temperate climate of the northern part of Europe. Cleaner and Responsible Consumption, 5, 100069.

Raicharoen, T., Lursinap, C. and Sanguanbhokai, P. (2003). Application of critical support vector machine to time series prediction, International Symposium on Circuits and Systems, 5, 741-744.

Ray, S., Das, S.S., Mishra, P.A.L. and Khatib, A.M.G. (2021). Time series SARIMA modeling and forecasting of monthly rainfall and temperature in the South Asian countries, Earth systems and Environment, 5(3), 531-546.

Reza, A. and Debnath, T. (2020). An approach to make comparison of ARIMA and NNAR models for forecasting price of commodities. Towards Data Science, https://www.researchgate.net/publication/342563043.

Romilly, P. (2005). Time series modeling of global mean temperature for managerial decision-making Journal of Environmental Management, 76, 61-70. DOI: 10.1016/j.jenvman.2005.01.008.

Sardans, J., Penuelas, J. and Estiarte, M. (2018). Warming and drought alter soil phosphatase activity and soil availability in Mediterranean shrubland, Plant Soil, 289, 227-238.

Shad, M., Sharma, Y.D. and Singh, A. (2022). Forecasting of monthly relative humidity in Delhi, Indian using SARIMA and ANN models, Modeling Earth Systems and Environment, DOI:10.1007/s.40808-022-01385-8.

Sharma, N., Sharma, P., Irwin, D. and Shenoy, P. (2011). Predicting solar generation from weather forecasts using machine learning, Proceedings of 2011 IEEE International Conference on Smart Grid Communications, 528-533.

Shih, H. and Rajendran, S. (2019). Comparison of time series methods and machine learning algorithms for forecasting Taiwan blood services foundation’s blood supply, Journal of Healthcare Engineering, 613-745.

Smith, D.M., Cusack, S., Colman, A.W., Folland, C.K., Harris, G.R. and Murphy, J.M. (2007). Improved surface temperature prediction for the coming decade from a global climate model. Science, 317, 796-799.

Tylkowski, J. and Hojan, M. (2019). Time decomposition and short-term forecasting of hydrometeorological conditions in the south Baltic coastal zone of Poland, Geosciences, 9, 68.

Wang, W., Van Gelder, Vrijling, J. and Ma, J. (2006). Forecasting daily streamflow using hybrid ANN models, Journal of Hydrology, 324, 383-399.

Wanishsakpong, W. and Owusu, B.E. (2019). Optimal time series model for forecasting monthly temperature in the southwestern region of Thailand, Modeling Earth Systems and Environment, 6, 525-532.