[1] We present a two-criteria inverse modeling approach to analyze the effects of seasonal drought on ecosystem gas exchange at three Mediterranean sites. The three sites include two nearly monospecific Quercus ilex L. forests, one on karstic limestone (PuĂŠchabon), the other on fluvial sand with access to groundwater (Castelporziano), and a typical multispecies shrubland on limestone (Arca di Noè). A canopy gas exchange model Process Pixel Net Ecosystem Exchange (PROXEL NEE ), which contains the Farquhar photosynthesis model coupled to stomatal conductance via the Ball-Berry model, was inverted in order to estimate the seasonal time course of canopy parameters from hourly values of ecosystem gross carbon uptake and transpiration. It was shown that an inverse estimation of leaf-level parameters was impossible when optimizing against ecosystem H 2 O or CO 2 fluxes alone (unidentifiable parameters). In contrast, a criterion that constrained the optimization against both H 2 O and CO 2 fluxes yielded stable estimates of leaf-level parameters. Two separate model inversions were implemented to test two alternative hypotheses about the response to drought: a reduction in active leaf area as a result of patchy stomatal closure or a change in photosynthetic capacities. In contrast to a previously tested hypothesis of classical (uniform) stomatal control, both hypotheses were equally able to describe the seasonality of carbon uptake and transpiration on all three sites, with a decline during the drought and recovery after autumn rainfall. Large reductions of up to 80%, in either active leaf area or photosynthetic capacities, were necessary to describe the observed carbon and water fluxes at the end of the drought period. With a threshold-type relationship, soil water content was an excellent predictor of these changes. With the drought-dependent parameter changes included, the canopy model explains 80-90% of the variance of hourly gross CO 2 uptake (root mean squared error (RMSE): 1.1-2.6 mmol m Ă2 s Ă1 ) and 70-80% of the variance of hourly transpiration (RMSE: 0.02-0.03 mm h Ă1 ) at all sites. In addition to drought effects, changes in leaf photosynthetic activity not related to water availability, i.e., high spring activity, were detected through the inverse modeling approach. Moreover, our study exemplifies a kind of multiconstraint inverse modeling that can be profitably used for calibrating ecosystem models that are meant for global applications with ecosystem flux data.