Improving the Accuracy of the Hyperspectral Model for Apple Canopy Water Content Prediction using the Equidistant...

Area of Science:

• Agricultural Science
• Remote Sensing
• Plant Physiology

Background:

• Accurate monitoring of plant water content is crucial for agricultural management.
• Hyperspectral imaging offers a non-destructive method for assessing plant physiological status.
• Optimizing data sampling is key to developing robust predictive models.

Purpose of the Study:

• To evaluate the effectiveness of the equidistant sampling method in a hyperspectral model for predicting apple tree canopy water content.
• To compare the predictive performance of equidistant sampling against random sampling.
• To establish a stepwise regression model for apple canopy water content prediction.

Main Methods:

• Exploration of the relationship between spectral reflectance and water content.
• Application of equidistant and random sampling for data partitioning.
• Development of stepwise regression models for water content prediction.
• Validation of model performance using calibration and validation sets.

Main Results:

• The equidistant sampling model demonstrated superior prediction ability compared to the random sampling model.
• The equidistant sampling model achieved higher coefficients of determination (0.6599 calibration, 0.8221 validation) than the random sampling model.
• The equidistant sampling method resulted in lower root mean square error (RMSE) and relative error (RE) by 17.23% and 17.09%, respectively.

Conclusions:

• The equidistant sampling method significantly enhances the prediction accuracy of hyperspectral models for apple canopy water content.
• Partitioning calibration and validation sets using equidistant sampling improves model reliability.
• This method provides a more accurate approach for non-destructive water content assessment in apple trees.