A machine learning approach to fill gaps in dendrometer data.

Area of Science:

• Dendrochronology and ecological monitoring.
• Application of machine learning in environmental science.

Background:

• Dendrometer data time-series analysis is often hindered by data gaps caused by device failures.
• Existing gap-filling methods have limitations, including dependence on smaller gaps, external tree data, or climate parameters.

Purpose of the Study:

• To evaluate machine learning algorithms for filling gaps in individual tree dendrometer data.
• To identify the most effective algorithm for bridging data gaps without external dependencies.

Main Methods:

• Testing eight machine learning algorithms on dendrometer data from urban and non-urban trees.
• Utilizing artificially created gaps to assess algorithm performance.
• Focusing on gap-filling capabilities for individual trees.

Main Results:

• Extreme gradient boosting (XGB) exhibited the highest skill in bridging artificial data gaps.
• XGB models successfully filled gaps up to 30 consecutive days, performing well at the beginning and end of the growing season.
• The method proved independent of climate variables and data from neighboring trees.

Conclusions:

• Machine learning, specifically XGB, offers a valid and effective approach for filling gaps in individual tree dendrometer data.
• This method is particularly valuable in scenarios with limited inter-tree data correlation, such as urban environments.
• The findings support the use of ML for improving the robustness of dendroecological time-series analyses.