Vegetation productivity and water balance of Mediterranean regions will be particularly affected by climate and land-use changes. In order to analyze and predict these changes through land surface models, a critical step is to quantify the uncertainties associated with these models (processes, parameters) and their implementation over a long period of time. Besides, uncertainties attached to the data used to force these models (atmospheric forcing, vegetation and soil characteristics, crop management practices...) which are generally available at coarse spatial resolution (>1-10 km) and for a limited number of plant functional types, need to be evaluated. This paper aims at assessing the uncertainties in water (evapotranspiration) and energy fluxes estimated from a Soil Vegetation Atmosphere Transfer (SVAT) model over a Mediterranean agricultural site. While similar past studies focused on particular crop types and limited period of time, the originality of this paper consists in implementing the SVAT model and assessing its uncertainties over a long period of time (12 years), encompassing several cycles of distinct crops (maize, wheat, sorghum, sunflower, peas) and bare soil periods in between crop cycles.

The SVAT model being analyzed in this paper is the ISBA model in its a-gs version which simulates the photosynthesis and its coupling with the stomata conductance, as well as the time course of the plant biomass and the Leaf Area Index (LAI). The model is evaluated over the INRA-Avignon (France) crop site, for which 12 years of energy and water fluxes, soil moisture profiles, vegetation measurements, agricultural practises as well as detailed soil hydrodynamic properties are available. The model is continuously implemented over this site from May 2001 up to November 2012, accounting for the succession of crop and bare soil periods.

The uncertainties in evapotranspiration estimated from ISBA are analyzed according to two simulation scenarios: - In the scenario 1 the simulations are achieved in the context of the operational application of the SVAT model using the SAFRAN re-analysis atmospheric dataset, a global map of texture (BDGSF INRA dataset) and the ECOCLIMAP land surface parameter global database. - In the scenario 2, the model is run using local atmospheric forcing, in situ soil (texture, hydrodynamic properties) and vegetation (LAI time course for each crop cycle) parameters. This scenario aims at evaluating the capacity of the model to accurately describe the physical processes. The analysis mainly focuses on the impact of uncertainties in both soil hydrodynamic and vegetation (photosynthesis, water stress) parameters.

The uncertainties in evapotranspiration and energy flux estimates are quantified from both 12-year trend analysis and selected daily cycles spanning a range of atmospheric conditions and phenological stages. Regarding the operational use of the model, while little differences are observed in the simulations based on the local atmospheric measurements and those using the SAFRAN reanalysis dataset, errors in both the global map of soil texture and the ECOCLIMAP LAI time course lead to larger discrepancies in evapotranspiration estimate. Results of scenario 2 indicate that while the net radiation flux is correctly simulated, the cumulated latent heat flux is under-estimated for both crop and bare soil periods. This latter finding is consistent with the over-estimation of the root-zone soil moisture. The comparison of the soil hydrodynamic properties retrieved from the ISBA pedotransfert function and those derived from in situ measurements highlights that the underestimation of the maximum water content available for the crop is the main source of uncertainties. Besides, discrepancies between the wilting point soil moisture retrieved from the soil water retention curve and that derived from the soil moisture vertical profiles (which substantially varies with crop cycles) as well as the variability in the root-zone-depth, increase the uncertainties in the maximum extractable amount of water by the plant. The impact of the uncertainties in vegetation parameters (mesophyllian conductance, water stress modelling) is less important compared to soil hydrodynamic properties.