Improving confidence in model-based Probable Maximum Precipitation : Assessing sources of model uncertainty in storm reconstruction and maximization

We present an analysis of uncertainty in model-based Probable Maximum Precipitation (PMP) estimates. The focus of the study is on “model-based” PMP derived from WRF (Weather Research and Forecasting) model reconstructions of severe historical storms and amplified by the addition of moisture in the boundary conditions (so-called Relative Humidity Maximization technique). Model-based PMP offers numerous advantages over the currently-used approach that is described in NOAA Hydrometeorological Reports. By scaling moisture and producing the resulting precipitation according to model formulation, the model-based approach circumvents the need for linearly scaling precipitation. Despite the significant improvement this represents, model-based PMP retains some degree of uncertainty that precludes its use in operational settings until the uncertainty is rigorously evaluated. This paper presents an ensemble of PMP simulations that samples recognized sources of uncertainty: (1) initial/boundary condition error, (2) choice of physics parametrizations and (3) model error due to unresolved subgrid processes. To our knowledge, this is the first uncertainty analysis conducted for model-based PMP. We applied this ensemble approach to the Feather River watershed (Oroville dam) in California. We first carried out in-depth evaluation of model reconstructions and found that the performance of some storm reconstructions that underlie the PMP estimate is not ideal, though the lack of uncertainty information about observations currently prevents us from identifying “well-reconstructed” storms or performing bias correction. That being said, our ensemble indicates that the 72-hour maximized precipitation totals used for PMP estimation do not differ greatly (110% at most) from the single-value estimate when model uncertainty is considered. We emphasize that model-based PMP estimates should always be presented as a range of values that reflects the uncertainties that exist, but concerns about model uncertainty should not hinder the development of model-based PMP.