Leaf traits are expressed either as area-based content (g m−2) or mass-based concentration (mg/g or %) and subsequently scaled to canopy level using leaf area index (LAI) and foliage biomass, respectively. Despite their wide use, it remains unclear whether the two trait expressions and scaling options (by LAI or foliage biomass) yield the same prediction accuracy using multispectral remote sensing data and additional site variables. In this study, we used LAI (m2 m−2) and foliage biomass (kg) to upscale area-based content [leaf mass per area (LMA), equivalent water thickness (EWT), nitrogen (N area) and potassium (K area)] and mass-based concentration [leaf dry matter content (LDMC), leaf water content (LWC), N mass , and K mass)], respectively in two contrasting study sites (coastal vs inland) for an ecologically important yet distinct regional species (red spruce , Picea rubens Sarg.) in Maine, USA. We used partial least squares regression (PLSR) and random forest (RF) to examine the effect of these metrics on model performance and prediction accuracy at both leaf- and canopy-scales using Sentinel-2 satellite bands, vegetation indices, as well as several topographical variables and depth to water table (DWT). The best-performing models (based on cross-validated normalized root mean square error: nRMSE) were used to generate high-resolution canopy trait maps. Results demonstrated that prediction accuracies of area-based leaf traits were generally higher (R2 = 0.41–0.65, nRMSE = 0.14–0.20) than the mass-based concentration (R2 = 0.42–0.65, nRMSE = 0.15–0.25) for both sites and modelling techniques. We also observed that canopy traits upscaled by LAI yield better prediction accuracies (R2 = 0.44–0.70, nRMSE = 0.17–0.20) compared to a foliage biomass upscaling approach (R2 = 0.41–0.57, nRMSE = 0.19–0.24). Among site variables, DWT was a significant variable in modeling canopy traits and showed robust behavior across the contrasting study sites. Overall, our study highlights the importance of trait expression and how it affects foliar trait retrieval accuracy at both leaf- and canopy-scale using satellite multispectral data and site variables. These findings have implications on leaf trait modelling especially in the context of monitoring dynamics in forest health and productivity from space. [ABSTRACT FROM AUTHOR]