Due to the lack of hydrological information, the river roughness calibration algorithm usually is unstable, and the calibrated roughness may deviate from the true value seriously. In order to treat this problem, a model, which utilized the prior information to regularize the under-determinate inverse problem, was proposed. The method could improve the stability and reliability of inverse solution. In this inverse model ,both the error minimization and the empirical information were took account of. A river network case was used to test the effectiveness of the model. Its numerical stability, uniqueness of solution and reliability of results were discussed. At last, this model was applied in the real case of creek Kootenai. The result of these cases showed that the present model can calibrate river roughness effectively with high numerical stability and reliable results.
