A key problem in the use of physically-based models of landslide hazards is how to parameterize the representation of soil properties. We applied a physically-based model for the topographic control on shallow landsliding (SHALSTAB) to two catchments in Rio de Janeiro. In particular we investigate the accuracy of the model results in relation to parameterization of soil properties to address the relevance of the use values derived from laboratories tests to the field problem and the trade-offs inherent in model parameterization when such data doesn't exist. From this previous research we established the relation range of possible cohesion, bulk density and friction angle values and run the model for all of possible discrete combinations. Many comparisons were made between model results and mapped landslides scars. Through the number of the pixels that corresponds to be predicted unstable for each set parameter was built a rank sorted from the best to worse model results performance. To optimize and to better visualize this methodology we submit the data rank to a robust regression, to find an equation that represents all the parameters combinations. With this equation we found a surface that represents the combination of the best parameterization for the best rank results. Our analysis suggested a trade off such that there isn't a unique best model and also showed the importance of to calibrate such model to field data. The results attest that the application of the model to an area where the soil properties are not well known requires a previous calibration of these parameters with the landslide map. Only after this step, the model can be used as a predictive tool.