European beech spring phenological phase prediction with UAV-derived multispectral indices and machine learning...

Area of Science:

• Ecology
• Remote Sensing
• Forestry

Background:

• Climate change impacts forest dynamics, necessitating accurate phenological event data.
• Earth observation (EO) data offers large-scale mapping of forest phenology, but ground truthing remains a challenge.
• Understanding phenological timing is crucial for assessing climate change risks, such as early leaf onset.

Purpose of the Study:

• To explore the feasibility of predicting high-resolution phenological phase data for European beech (Fagus sylvatica).
• To utilize unoccupied aerial vehicle (UAV)-based multispectral indices and machine learning for phenological prediction.
• To identify optimal sensors, vegetation indices, training data, and machine learning models for accurate phenological mapping.

Main Methods:

• Employed a comprehensive feature selection process for UAV-based multispectral indices and machine learning models.
• Utilized Green Chromatic Coordinate (GCC) and Generalized Additive Model (GAM) boosting for phenological phase prediction.
• Derived GCC training data from calibrated visual bands and predicted using uncalibrated RGB sensor data.

Main Results:

• The GCC/GAM boosting model achieved high accuracy in predicting phenological phases on unseen datasets (RMSE < 0.5).
• Demonstrated the potential interoperability of common UAV-mounted sensors, especially low-cost RGB sensors.
• Identified limitations with near-infrared band indices due to oversaturation.

Conclusions:

• UAV-based multispectral indices and machine learning, particularly GCC and GAM boosting, are effective for high-resolution forest phenology prediction.
• Low-cost RGB sensors offer a viable alternative for acquiring phenological data, enhancing accessibility.
• Future research should focus on aligning models with established phenological stages like ICP Forests flushing stages.