A physics-informed hybrid deep learning model for spatiotemporal rice disease prediction using multi-source data
Accurate, reliable, large-scale disease predictions are essential to ensure rice production. Existing disease prediction models often face a trade-off between interpretability and predictive capability, necessitating the integration of mechanistic knowledge and data-driven learning within a modelling framework. Accordingly, we propose a physics-informed hybrid gated recurrent unit (PI-HGRU) model…
Accurate disease predictions are crucial for sustaining rice production. Existing models often struggle with balancing interpretability and predictive power, calling for a blend of mechanistic knowledge and data-driven learning within a single framework. This research introduces the physics-informed hybrid gated recurrent unit (PI-HGRU) model for predicting the dynamic spread of sheath blight disease in rice, a fungus-induced ailment.
The PI-HGRU model merges differential equations depicting disease transmission dynamics into a hybrid gated recurrent unit (HGRU) structure, acting as mechanistic constraints to facilitate a collaborative modeling process between epidemiological processes and data-driven learning. The model was trained and assessed using a comprehensive dataset collected over 17 years (2000-2016) from 16 key rice-producing regions in southern China.
This dataset comprises spatiotemporally synchronized field disease observations, remote sensing data, meteorological data, and soil property data.
To tackle the issues of irregular sampling intervals and varying sequence lengths in disease survey data, the study employed a sliding time-window-based prediction framework. Additionally, a time-window sensitivity analysis was carried out to determine optimal input time window and lag time configurations, allowing the model to capture the cumulative and delayed impacts of environmental factors.
The PI-HGRU framework notably outperformed the purely data-driven HGRU baseline, boosting the squared Pearson correlation coefficient (r2) by 22.8%, while simultaneously decreasing the root mean square error (RMSE) and mean absolute error (MAE) by 10.2% and 18.0%, respectively. Further investigation into the model's intermediate variables revealed that the transmission rate (β)(t) displayed interpretable correlations with environmental conditions within the input time window, establishing a process-related connection between environmental drivers and the modeled disease transmission dynamics.
In summary, this research shows that incorporating epidemiological mechanisms into deep learning models in a physics-informed manner can enhance predictive accuracy, model stability, and interpretability. This approach holds great promise for large-scale disease forecasting and precision disease management in rice production.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
