Limitations of Light Use Efficiency Models for Upscaling the CO2 Flux from Tower to Region

Light use efficiency (LUE) models estimate the gross primary productivity (GPP) or the net primary productivity (NPP) according to the total photosynthetically active radiation absorbed by the canopy. These models are simple and have acceptable accuracy for some applications if the LUE values are accurately determined. LUE models are therefore widely used for mapping GPP or NPP using remote sensing data and for upscaling from tower to region. The MODIS GPP product serves this purpose very well. However, it is also well known that the LUE of shaded leaves is much larger than that for sunlit leaves. Since LUE models do not differentiate between sunlit and shaded leaves, it would be necessary to evaluate the possible range of errors caused by not separating these two groups of leaves. In process-based canopy-level photosynthesis modeling, it has been long recognized that big-leaf models are conceptually flawed in representing the flow of CO2 from the atmosphere to the site of carboxylation inside of leaves in multiple layers and two-leaf models with sunlit and shaded leaf separation can avoid this problem. In fact, LUE models may be regarded as a simplified form of big-leaf models. In our current study, we first evaluate a two-leaf and a big-leaf model against flux measurements at 9 sites, and find that GPP simulated by the big-leaf model is negatively biased by 48-67% if correct leaf-level parameters are used, while these biases are reduced to less than 10% with the two-leaf model. We then further compare GPP maps produced using the two-leaf model against MODIS GPP maps over conterminous USA. In this comparison, we find that MODIS GPP varies much less spatially and seasonally relative to results of the two-leaf model. The spatial and temporal variations of GPP produced by the MODIS LUE algorithm are dampened because it uses fixed maximum LUE values that cannot take into account of the variable and considerable contributions from shaded leaves. If the MODIS LUE is correctly determined from tower flux data, it represents the mean conditions. At locations (times) where the leaf area index (LAI) is larger than the average, the contribution of shaded leaves is underestimated, causing GPP underestimation. This is because the shaded leaf fraction increases with LAI and at high LAI, the shaded leaf fraction is larger than the average condition. At locations (times) where LAI is smaller than the average, the contribution of the shaded leaves is overestimated, causing GPP overestimation. According to these research results, we propose a two-leaf LUE model to avoid the problem of big-leaf LUE models. Considerable improvement of this two-leaf LUE model over a big-leaf LUE model is demonstrated against tower flux data.