Estimating Above Ground Forest Biomass Using High-Resolution NAIP Imagery and Deep Learning

Accurate estimation of forest aboveground biomass (AGB) is critical for a range of applications, including the development of regional carbon policies, bioenergy production, and sustainable forest management. This study leverages high-resolution (0.3 m) imagery from the National Agriculture Imagery Program (NAIP), integrated with field measurements and topographic variables, to estimate AGB in Drew County, Arkansas, USA. Feature variables extracted from the data were utilized as inputs for three deep learning (DL) algorithms: Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN). The models were evaluated across 15 combinations of feature variables. Among the models, the DNN achieved the highest prediction accuracy (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}=0.65, \text{RMSE}=48.14 \mathrm{t} / \text{ha}$</tex>), outperforming both the CNN and RNN models, which exhibited moderate to lower accuracies. Pearson correlation analysis revealed strong associations between AGB and specific spectral indices. Furthermore, SHapley Additive exPlanations (SHAP) analysis identified key variables influencing model predictions, with elevation emerging as a significant predictor. This study demonstrates the potential of integrating high-resolution remote sensing imagery with advanced DL techniques to enhance AGB estimation. The findings offer a scalable framework for biomass mapping, utilizing cost-effective NAIP imagery and DL algorithms for broader environmental and resource management applications.