Sensitivity analysis (SA) is a commonly used approach for identifying important parameters that dominate model behaviors. We use a newly developed software package, a Problem Solving environment for Uncertainty Analysis and Design Exploration (PSUADE), to evaluate the effectiveness and efficiency of ten widely used SA methods, including seven qualitative and three quantitative ones. All SA methods are tested using a variety of sampling techniques to screen out the most sensitive (i.e., important) parameters from the insensitive ones. The Sacramento Soil Moisture Accounting (SAC-SMA) model, which has thirteen tunable parameters, is used for illustration. The South Branch Potomac River basin near Springfield, West Virginia in the U.S. is chosen as the study area. The key findings from this study are: (1) For qualitative SA methods, Correlation Analysis (CA), Regression Analysis (RA), and Gaussian Process (GP) screening methods are shown to be not effective in this example. Morris One-At-a-Time (MOAT) screening is the most efficient, needing only 280 samples to identify the most important parameters, but it is the least robust method. Multivariate Adaptive Regression Splines (MARS), Delta Test (DT) and Sum-Of-Trees (SOT) screening methods need about 400–600 samples for the same purpose. Monte Carlo (MC), Orthogonal Array (OA) and Orthogonal Array based Latin Hypercube (OALH) are appropriate sampling techniques for them; (2) For quantitative SA methods, at least 2777 samples are needed for Fourier Amplitude Sensitivity Test (FAST) to identity parameter main effect. McKay method needs about 360 samples to evaluate the main effect, more than 1000 samples to assess the two-way interaction effect. OALH and LPτ (LPTAU) sampling techniques are more appropriate for McKay method. For the Sobol' method, the minimum samples needed are 1050 to compute the first-order and total sensitivity indices correctly. These comparisons show that qualitative SA methods are more efficient but less accurate and robust than quantitative ones

Despite the widespread use of the latest community Noah with multiparameterization (Noah‐MP) land surface model, it has not been rigorously evaluated over the complex Tibetan Plateau. This study assessed uncertainties in Noah‐MP simulations of a cropland site using observations from the 2008 Joint International Cooperation program field campaign. Such an assessment was conducted in the context of performing a total number of 4608 Noah‐MP physics ensemble simulations using two analysis methods: the natural selection approach and Tukey's test, where the impacts of uncertainties in atmospheric forcing conditions, vegetation parameters, and subprocesses on model simulations were identified. Uncertainty in precipitation data exerts greater influence on the general behavior of Noah‐MP ensemble simulations than that in the leaf area index (LAI). However, using a more realistic seasonal LAI improves the seasonal variations of surface heat fluxes. Combining a better precipitation forcing data set and Moderate Resolution Imaging Spectroradiometer monthly LAI significantly reduces the uncertainty range of the ensemble mean of surface heat fluxes. The uncertainty analysis results using the natural selection method are largely similar to that from Tukey's test but show some subtle differences. Both methods reveal greater uncertainties in the following subprocess schemes: canopy resistance, soil moisture threshold for evaporation, runoff and groundwater, and surface‐layer parameterization for this cropland site. The uncertainty analysis identifies the parameterization schemes that demonstrably degrade model performance. The uncertainties in ensemble simulations were significantly reduced when those schemes were excluded, and it was possible to configure an optimal combination of parameterization schemes to obtain similar performance to the ensemble mean of the “best” ensemble experiment.

Land surface processes and their coupling to the atmosphere over the Tibetan Plateau (TP) play an important role in modulating the regional and global climate. Therefore, identifying and quantifying uncertainty in these land surface model (LSM) processes are essential for improving climate models. The specifications of land cover and soil texture types, intertwined with the uncertainties in associated vegetation and soil parameters in LSMs, are significant sources of uncertainty due to the lack of detailed land survey in the TP. To differentiate the effects of land cover or soil texture specifications in the Noah with Multiple Parameterizations (Noah‐MP) LSM from the effects of uncertainties in the model parameters, this study first identified the most sensitive vegetation and soil parameters through global sensitivity analysis and then conducted parametric ensemble simulations using two land cover data sets and two soil texture data sets over the central TP to estimate their corresponding impacts on the overall model responses. The distinction level and the Kolmogorov‐Smirnov test were then applied to assess the differences between the results from parametric ensemble simulations using different land cover or soil texture data sets. The results show that the simulated energy and water fluxes over the central TP are dominated by soil parameters. The canopy height is the most sensitive vegetation parameter, and the Clapp‐Hornberger b parameter (the exponent in the function that relates soil water potential and water content) is the most sensitive soil parameter. Relative to the background parametric uncertainties, the Noah‐MP LSM could not sufficiently distinguish the effects of changes between forested types or soil texture types, which highlight the need for further quantifying and reducing the parametric uncertainties in LSMs. Further analysis shows significant sensitivities of the distinction level and changes in model response to annual precipitation and vegetation fraction. This work provides a scientific reference for assessing the impacts of land cover or soil texture changes on Noah‐MP simulations under future climate change conditions.

A global sensitivity analysis method was used to identify the parameters of the Weather Research and Forecasting (WRF) model that exert the most influence on precipitation forecasting. Twenty‐three adjustable parameters were selected from seven physical components of the WRF model. The sensitivity was evaluated based on skill scores calculated over nine 5 day precipitation forecasts during the summer seasons from 2008 to 2010 in the Greater Beijing Area in China. We found that eight parameters are more sensitive than others. Storm type seems to have no impact on the list of sensitive parameters but does influence the degree of sensitivity. We also examined the physical interpretation of parameter sensitivity. This analysis is useful for further optimization of the WRF model parameters to improve precipitation forecasting.

The community Noah land surface model with multiparameterization options (Noah‐MP) provides a plethora of model configurations with varying complexity for land surface modeling. The practical application of this model requires a basic understanding of the relative abilities of its various parameterization configurations in representing spatiotemporal variability and hydrologic connectivity. We designed an ensemble of 288 experiments from multiparameterization schemes of six physical processes to assess and reduce the structural uncertainty for land surface modeling over 10 hydrologic regions in China for the period 2001–2010. The observed latent heat (LH) was well reproduced by the ensemble. Meanwhile, most experiments underestimated sensible heat (SH) throughout the year and overestimated the cold season but underestimated the warm season terrestrial water storage anomaly (TWSA). The sensitive processes and best‐performing schemes varied not only with regions but also among variables. The LH and SH were most sensitive to runoff‐groundwater (RUN), surface heat exchange coefficient (SFC), and radiation transfer (RAD). The TWSA was dominated by RUN and RAD while largely influenced by soil moisture factor for stomatal resistance (BTR) and frozen soil permeability (INF) over some limited regions. By contrast, supercooled liquid water (FRZ) had little influence on all variables. Our optimization for individual variables produced high mean Taylor skill scores that ranged from 0.95–0.99 for LH, 0.82–0.99 for SH, and 0.63–0.95 for TWSA depending on regions. The simultaneous optimization made trade‐off among the three variables, which improved TWSA performance at the cost of reducing the skill for LH and SH over a few regions.

Weather forecasting skill has been improved over recent years owing to advances in the representation of physical processes by numerical weather prediction (NWP) models, observational systems, data assimilation and postprocessing, new computational capability, and effective communications and training. There is an area that has received less attention so far but can bring significant improvement to weather forecasting—the calibration of NWP models, a process in which model parameters are tuned using certain mathematical methods to minimize the difference between predictions and observations. Model calibration of the NWP models is difficult because 1) there are a formidable number of model parameters and meteorological variables to tune, and 2) a typical NWP model is very expensive to run, and conventional model calibration methods require many model runs (up to tens of thousands) or cannot handle the high dimensionality of NWP models. This study demonstrates that a newly developed automatic model calibration platform can overcome these difficulties and improve weather forecasting through parameter optimization. We illustrate how this is done with a case study involving 5-day weather forecasting during the summer monsoon in the greater Beijing region using the Weather Research and Forecasting Model. The keys to automatic model calibration are to use global sensitivity analysis to screen out the most important parameters influencing model performance and to employ surrogate models to reduce the need for a large number of model runs. Through several optimization and validation studies, we have shown that automatic model calibration can improve precipitation and temperature forecasting significantly according to a number of performance measures.