A multivariate soil temperature interval forecasting method for precision regulation of plant growth environment

Area of Science:

• Agricultural Science
• Environmental Monitoring
• Data Science

Background:

• Foliage plant cultivation requires precise environmental control, with soil temperature being a critical factor.
• Accurate short-term soil temperature forecasting is challenging due to non-linear variations and time lags.

Purpose of the Study:

• To develop a robust multivariate forecasting method for short-term soil temperature prediction.
• To improve the accuracy and stability of multi-step soil temperature forecasts.

Main Methods:

• Gradient Boosting Tree model to identify key environmental factors influencing soil temperature.
• A forecasting model combining Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS) and a Gaussian likelihood function.
• Multi-objective optimization algorithm for optimal initial parameter selection.

Main Results:

• The proposed model demonstrated superior predictive performance compared to existing methods like LSTM.
• Achieved significant reductions in Mean Absolute Error (MAE) for 20, 60, and 120-minute soil temperature forecasts.
• Provided stable forecasting intervals, addressing instability in multi-step point forecasts.

Conclusions:

• The developed model offers a scientific approach for precise regulation and early warning systems in controlled cultivation environments.
• Enhances the reliability of soil temperature forecasting for agricultural applications.