Spatial predictions of tree density and tree height across Mexico forests using ensemble learning and forest...

Area of Science:

• Forestry science
• Geospatial analysis
• Machine learning applications

Background:

• Mexico's National Forest and Soils Inventory (INFyS) faces spatial data gaps due to field survey limitations.
• These gaps introduce bias and uncertainty into forest management decision-making.
• Accurate spatial data on forest attributes is crucial for effective conservation and management.

Purpose of the Study:

• To develop accurate spatial predictions for tree height and tree density across all Mexican forests.
• To address data gaps in the National Forest and Soils Inventory (INFyS).
• To provide reliable data for informed forest management decisions.

Main Methods:

• Utilized ensemble machine learning models for wall-to-wall spatial predictions on 1-km grids.
• Incorporated remote sensing imagery and geospatial data (precipitation, temperature, canopy cover) as predictor variables.
• Trained models using over 26,000 sampling plots from the 2009-2014 INFyS data cycle.

Main Results:

• Models showed better performance for predicting tree height (r²=0.35) than tree density (r²=0.23).
• Highest predictive accuracy for tree height was in broadleaf forests (~50% variance explained); for tree density, it was in tropical forests (~40% variance explained).
• Tree height predictions had low uncertainty (<60%) except in arid/semiarid ecosystems (>80%); tree density predictions generally had high uncertainty (>80%).

Conclusions:

• Ensemble machine learning effectively predicts spatial distribution of tree height and density, addressing INFyS data gaps.
• The open science approach is replicable and scalable, supporting INFyS decision-making.
• Highlights the need for advanced analytical tools to maximize the utility of Mexican forest inventory data.