Estimating Above Ground Forest Biomass Using High-Resolution NAIP Imagery, Machine Learning, and Google Earth Engine

Accurate Forest Aboveground Biomass (AGB) mapping is essential for understanding carbon storage in ecosystems, global carbon cycles, and effective forest management. Remote sensing (RS) technologies and advanced machine learning algorithms (MLAs) provide a reliable way to estimate AGB. This study explores using National Agriculture Imagery Program (NAIP) data, field measurements, and topographic variables for estimating AGB in Drew County, Arkansas, USA. The study used features derived from NAIP imagery, such as vegetation indices and texture metrics, as well as data from digital elevation models (DEMs) as input variables for three ML algorithms: Random Forest (RF), Gradient Tree Boosting (GTB), and Classification and Regression Trees (CART), tested across 15 different combinations of feature variables. Of the models tested, RF performed best, achieving the highest prediction accuracy <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\mathrm{R}^{2}=0.77, \text{RMSE}=34.85 \ \mathrm{t}/\text{ha})$</tex>, outperforming GTB and CART. The use of multiple feature variables significantly improved the accuracy of predictions in all the algorithms, with the most significant contributions to model performance coming from vegetation indices and texture metrics derived from NAIP imagery. In contrast, relying only on elevation-based variables resulted in lower accuracy, underscoring the importance of using various data sources. The study also found that AGB was overestimated in non-forested areas when only topographic variables were used, highlighting the need for a balanced integration of features. This research provides a scalable and reproducible framework for continuous AGB monitoring at high spatial resolutions, utilizing the strengths of ML and advanced RS data.