Nationwide Native Forest Structure Maps for Romania based on Field 3D Laser Scanning and Remotely Sensed Metrics

Mapping forest structure variables for large areas requires combining field-based measurements with remotely sensed data. However, a large number of field plots is required to develop the remotely sensed proxies. Our goal was to develop a methodology that combines field-based 3D laser scanning (to determine tree biometric variables) with metrics from Sentinel-2 and topography variables to derive forest structure maps at 10-m spatial resolution. Our study area is Romania (~238,000 km2) where we measured 1,861 field plots of 1,000 m2 in 2021. At each plot we collected position, diameter, and height of each tree using a 3D mobile scanner (Figure 1a). From these data we used automated tree segmentation to obtain four variables: diameter at 1.30 m (DBH, in cm), basal area (m2) height (in m), and volume (m3). As independent variables, we calculated six metrics from Sentinel-2 (NDVI-normalized vegetation index, EVI-enhanced vegetation index, SI-shadow index, SAVI-soil adjusted vegetation index, BI-bare soil index and NDII-normalized difference infrared index) and two metrics from the shuttle radar topography mission-SRTM (slope and aspect). We applied random forest regression to model and map the four forest attributes across Romania (Figure 1b). For validation, we computed the root-mean square error (RMSE). Our models predicted forest structure attributes moderately well (RMSE lower than 50%). The height model performed best, reaching absolute and relative RMSEs of 3.07 m and 17.45%, respectively (Figure 1c). Our methodology yielded reliable predictions of forest structure attributes based on consistent 3-D laser scanning field-based methodology, providing a strong foundation for forest management and conservation planning in Romania and elsewhere.