Assessing Parameter Importance of the Weather Research and Forecasting Model Based On Global Sensitivity Analysis Methods

The effectiveness and efficiency of two state-of-the-art global sensitivity analysis (SA) methods, the Morris and surrogate-based Sobol' methods, are evaluated using the Weather Research and Forecasting (WRF) model, version 3.6.1. The sensitivities of precipitation and other related meteorological variables to 11 selected parameters in the new Kain-Fritsch Scheme, WRF Single-Moment 6-class Scheme, and Yonsei University Scheme are then investigated. The results demonstrate that (1) the Morris method is effective and efficient for screening important parameters qualitatively, and with recommended settings of levels p = 8 and replication times r = 10 only 10 x (D + 1) WRF runs are required, where D is the dimension of parameter space; (2) Gaussian process regression (GP) is the best method for constructing surrogates, and the GP-based Sobol' method can provide reliable quantitative results for sensitivity analysis when the number of WRF runs exceeds 200; and (3) the sensitivity index in the Morris method is closely related to the Sobol' index S-T, and even for qualitative sensitivity analysis, the GP-based Sobol' method is more efficient compared to the Morris method. The SA results show that larger values of the downdraft-related parameter x(1), entrainment-related parameter x(2), and downdraft starting height x(3) significantly decrease rainfall, while the maximum allowed value for the cloud ice diameter x(6) has a moderate decreasing effect on precipitation. This work is useful for further tuning of the WRF to improve the agreement between the climate model and observations.