Using Structural Class Pairing to Address the Spatial Mismatch Between GEDI Measurements and NFI Plots

Abstract

The Global Ecosystem Dynamics Investigation (GEDI) mission can significantly enhance multisource national forest inventories (MSNFI) by improving the spatio-temporal resolution of forest attributes while preserving the statistical relevance of the design-based inference approach. The main challenge is the lack of systematic spatial alignment between GEDI footprints and National Forest Inventory (NFI) plots, which is necessary to accurately link in situ forest measurements with GEDI data. In this study, we aim to tackle the aforementioned issue by introducing a methodology for interpolating GEDI measurements to NFI plots, enabling the calibration of GEDI data using localized NFI estimates. Our proposed method incorporates clustering, classification, and regression techniques, and utilizes GEDI and NFI data, along with Sentinel-2 images, land-use information, and topographic data. Beginning with the prediction of profile structural classes and shapes on NFI plots, the proposed method ultimately projects actual measurements onto the NFI plot sites through profile pairing within the predicted structural classes. The method is conceived and validated using the data acquired across the mountainous area of $\sim $500 kha, covered by >500 NFI plots. Our validation framework shows that the method is able to project relative height profiles at NFI plots, allowing to partly interpolate the lower part of the profile and not only the canopy top height. This enables the construction of models that efficiently relate GEDI profiles and wood volume, demonstrating the importance of incorporating lower relative height values when linking forest attributes and lidar measurements ($R^{2}=0.65$, $MBE=2.31$ m$^{3}$/ha).

Keywords

• Global Ecosystem Dynamics Investigation (GEDI)
• machine learning
• modeling
• National Forest Inventory (NFI)
• Sentinel-2
• wood volume