Many applications dedicated to urban areas (e.g. land cover mapping and biophysical properties estimation) using high spatial resolution remote sensing images require the use of 3D atmospheric correction methods, able to model complex light interactions within urban topography such as buildings and trees. Currently, one major drawback of these methods is their lack in modeling the radiative signature of trees (e.g. the light transmitted through the tree crown), which leads to an over-estimation of ground reflectance at tree shadows. No study has been carried out to take into account both optical and structural properties of trees in the correction provided by these methods. The aim of this work is to improve an existing 3D atmospheric correction method, ICARE (Inversion Code for urban Areas Reflectance Extraction), to account for trees in its new version, ICARE-VEG (ICARE with VEGetation). After the execution of ICARE, the methodology of ICARE-VEG consists in tree crown delineation and tree shadow detection, and then the application of a physics-based correction factor in order to perform a tree-specific local correction for each pixel in tree shadow. A sensitivity analysis with a design of experiments performed with a 3D canopy radiative transfer code, DART (Discrete Anisotropic Radiative Transfer), results in fixing the two most critical variables contributing to the impact of an isolated tree crown on the radiative energy budget at tree shadow: the solar zenith angle and the tree leaf area index (LAI). Thus, the approach to determine the correction factor relies on an empirical statistical regression and the addition of a geometric scaling factor to account for the tree crown occultation from ground. ICARE-VEG and ICARE performance were compared and validated in the Visible-Near Infrared Region (V-NIR: 0.4–1.0 µm) with hyperspectral airborne data at 0.8 m resolution on three ground materials types, grass, asphalt and water. Results show that (i) ICARE-VEG improves the mean absolute error in retrieved reflectances compared to ICARE in tree shadows by a multiplicative factor ranging between 4.2 and 18.8, and (ii) reduces the spectral bias in reflectance from visible to NIR (due to light transmission through the tree crown) by a multiplicative factor between 1.0 and 1.4 in terms of spectral angle mapper performance. ICARE-VEG opens the way to a complete interpretation of remote sensing images (sunlit, shade cast by both buildings and trees) and the derivation of scientific value-added products over all the entire image without the preliminary step of shadow masking.