Parameter optimisation for a better representation of drought by LSMs: inverse modelling vs. sequential data assimilation

Abstract. Soil maximum available water content (MaxAWC) is a key parameter in land surface models (LSMs). However, being difficult to measure, this parameter is usually uncertain. This study assesses the feasibility of using a 15-year (1999–2013) time series of satellite-derived low-resolution observations of leaf area index (LAI) to estimate MaxAWC for rainfed croplands over France. LAI interannual variability is simulated using the CO₂-responsive version of the Interactions between Soil, Biosphere and Atmosphere (ISBA) LSM for various values of MaxAWC. Optimal value is then selected by using (1) a simple inverse modelling technique, comparing simulated and observed LAI and (2) a more complex method consisting in integrating observed LAI in ISBA through a land data assimilation system (LDAS) and minimising LAI analysis increments. The evaluation of the MaxAWC estimates from both methods is done using simulated annual maximum above-ground biomass (B_ag) and straw cereal grain yield (GY) values from the Agreste French agricultural statistics portal, for 45 administrative units presenting a high proportion of straw cereals. Significant correlations (p value  <  0.01) between B_ag and GY are found for up to 36 and 53 % of the administrative units for the inverse modelling and LDAS tuning methods, respectively. It is found that the LDAS tuning experiment gives more realistic values of MaxAWC and maximum B_ag than the inverse modelling experiment. Using undisaggregated LAI observations leads to an underestimation of MaxAWC and maximum B_ag in both experiments. Median annual maximum values of disaggregated LAI observations are found to correlate very well with MaxAWC.