Terrestrial ecosystems are a key component for the study of Earth climate. There are evidences that land surface processes affect climate and weather in a range of time scales, from seconds to million of years. Land surface models (LSMs) are important tools for research on climatic simulations and weather prediction, generally working coupled to general circulation models, and acting as sub models calculating the fluxes throughout the biosphere- atmosphere interface. To obtain good LSMs simulations for a given region, it is necessary to optimize the parameters and calibrate the model against observed data for the region of study. The main objective of this work is to develop a calibration methodology that allow to calibrate all processes simulated by the model Integrated Biosphere Simulator (IBIS). The study is based on systems hierarchy and on the calibration of all IBIS simulated variables. The calibration is performed at temporal hierarchy order, from the fastest process (radiative fluxes) to the slowest process (carbon allocation). We proposed a relative efficiency index, D, to evaluate the multiobjective and hierarchical calibrations, based on single objective calibrations of each simulated variable. The single objective calibrations are considered the best possible ones, and are used as a reference to evaluate the proposed method. Initially, we develop a sensitivity analysis (SA) that efficiently select the most important parameters for each model output. The SA is based on the Morris method, with modifications in the sensitivity calculations, to adjust the method to work with LSMs. Then, we develop a software for multiobjective automatic calibration of the IBIS model, called Optis. Optis is based on the multiobjective genetic algorithm NSGA-II (Nondominated Sorted Genetic Algorithm II). This software allows the adjustment of multiple statistics and many model outputs simultaneously. Finally, we proposed a hierarchical methodology for LSM calibration, that performs the calibration of all processes simulated by the model. The SA is efficient to select only the most important model parameters to be considered in calibration, thus eliminating out of the search process those parameters with low influence on model output. Optis was effective to solve all the multiobjective calibration tests performed, reaching more than 90% of the efficiency obtained by the single objective calibration. Optis also is efficient in the use of computational resources, with total efficiency in the code parallelization. The hierarchical procedure proposed was tested in two experimental sites in the Amazon rainforest, Flona Tapajós and Sinop. In both sites, the procedure demonstrated to be efficient in calibrating all the model processes, with D greather than 0.80 in both sites. This calibration methodology allows a more realistic calibration of the model, as all simulated variables are optimized.
