With limited area meso-scale numerical model under η–coordinate, sensitivity experiments of retrieving model humidity profiles by one-dimensional variational assimilation of digital weather radar-derived precipitation are studied for the abrupt heavy rainfall along the Yangtze River in the east part of Hubei province. Results show as follows.  (1) The smaller the control precision EPS of optimal decline algorithm is, the closer the model forecast precipitation after assimilation is to the real rainfall amount but the iteration times will increase.  (2) In the precipitation assimulation it will not effect the results substantially to ignore to some extent the assimulation of small-valued rainfall, but that of rainfall core or heavy rainfall center is very important, with assimulating it or not leading to the changes of overall forecasting effects. 