Digital twins of the Earth require Earth observation data streams that are both spatially continuous and informative about the state of the modelled system. For forests, the standard optical data source is Sentinel-2, whose 10 m multispectral imagery trades spectral detail for spatial resolution; the forthcoming operational imaging spectroscopy missions CHIME and SBG will offer the opposite trade-off with 30-m pixels and hundreds of spectral bands. We quantified the cost of that trade-off using a near-simultaneous EnMAP and Sentinel-2 image pair acquired over the boreal forests in Hyytiälä, central Finland, on 20 August 2023, together with 213 field plots measured by the Finnish Forest Centre. A U-Net convolutional neural network was trained fully convolutionally to predict stand basal area, basal-area weighted mean diameter and mean height. Training had two stages: self-supervised pretraining (spectral reconstruction for EnMAP, masked band prediction for Sentinel-2) followed by supervised fine-tuning on the field plots. On 32 independent test plots, EnMAP resampled to 10 m matched Sentinel-2: R² was 0.65 vs. 0.67 for basal area, 0.80 vs. 0.80 for mean diameter and 0.79 vs. 0.79 for mean height, while mean absolute errors were 9–13 % lower for EnMAP. Predictions obtained from EnMAP at its native 30 m sampling were equally accurate. The nine-fold loss in ground sampling area that accompanies the higher spectral resolution therefore did not degrade the retrieval of forest structural variables, which supports the use of spaceborne imaging spectroscopy as a baseline information source in a forest digital twin.