Net ecosystem exchange (NEE) measurements using the eddy covariance technique have been widely used for calibration and evaluation of carbon flux estimates from terrestrial ecosystem models as well as for remote sensing-based estimates across various spatial and temporal scales. Therefore, it is vital to fully understand the land surface characteristics within the area contributing to these flux measurements (i.e. source area) when upscaling plot-scale tower measurements to regional-scale ecosystem estimates, especially in heterogeneous landscapes, such as mixed forests. We estimated the source area of a flux tower at a mixed forest (Harvard Forest in US) using a footprint model, and analyzed the spatial representativeness of the source area for the vegetation characteristics (density variation and magnitude) within the surrounding 1- and 1.5-km grid cells during two decades (1993–2011). Semi-variogram and window size analyses using 19 years of Landsat-retrieved enhanced vegetation index (EVI) confirmed that spatial heterogeneity within the 1-km grid cell has been gradually increasing for leaf-on periods. The overall prevailing source areas lay toward the southwest, yet there were considerable variations in the extents and the directions of the source areas. The source areas generally cover a large enough area to adequately represent the vegetation density magnitude and variation during both daytime and nighttime. We show that the variation in the daytime NEE during peak growing season should be more attributed to variations in the deciduous forest contribution within the source areas rather than the vegetation density. This study highlights the importance of taking account of the land cover variation within the source areas into gap-filling and upscaling procedures.