Relative humidity prediction with covariates and error correction based on SARIMA-EG-ECM model

Area of Science:

• Meteorology and Climatology
• Environmental Science

Background:

• Relative humidity (RH) is a key atmospheric variable with significant implications for weather forecasting, climate studies, agriculture, and public health.
• Accurate RH prediction is essential for informed decision-making across various critical sectors.

Purpose of the Study:

• To investigate the influence of covariates and error correction on relative humidity (RH) prediction.
• To propose and evaluate a novel hybrid model for enhanced RH forecasting.

Main Methods:

• Developed a hybrid Seasonal Autoregressive Integrated Moving Average (SARIMA) with Engle-Granger (EG) cointegration and Error Correction Model (ECM), termed SARIMA-EG-ECM (SEE).
• Applied the SEE model to meteorological data from Hailun Agricultural Ecology Experimental Station, China.
• Utilized meteorological variables like air temperature (TEMP), dew point temperature (DEWP), precipitation (PRCP), atmospheric pressure (ATMO), sea-level pressure (SLP), and soil temperature (40ST) as covariates.

Main Results:

• Established a long-term equilibrium relationship (cointegration) between RH and TEMP, DEWP, PRCP, ATMO, SLP, and 40ST.
• Identified significant short-term impacts of DEWP, ATMO, and SLP fluctuations on RH fluctuations via the ECM.
• The SEE model demonstrated slightly decreased performance with extended forecast horizons (6-12 months) but outperformed SARIMA and Long Short-Term Memory (LSTM) models.

Conclusions:

• The SEE model provides a robust framework for RH prediction by integrating long-term equilibrium and short-term dynamic relationships.
• The findings highlight the importance of considering multiple meteorological covariates and error correction mechanisms for accurate RH forecasting.
• The proposed hybrid approach offers a superior alternative to traditional time series and deep learning models for RH prediction in meteorological applications.