The flood that would result from the greatest depth of precipitation “meteorologically possible”, or Probable Maximum Precipitation (PMP) is used in the design of dam spillways and other high-risk structures. Historically, PMP has been estimated by scaling precipitation depth-area-duration relationships obtained from severe historical storms. Over the last decade, numerical weather prediction models have been leveraged to instead predict precipitation resulting from the addition of moisture at the model boundaries (called relative humidity maximization, or RHM). Despite the major improvement this represents, model-based methods have not yet been applied to produce operational PMP estimates. Several sources of uncertainty still limit their applicability: (i) model uncertainty, (ii) uncertainty regarding whether maximum storm efficiency (moisture conversion to precipitation) was achieved by observed storms and (iii) uncertainty caused by a lack of physically-constrained guidance on moisture amplification. These uncertainties result in differences in model-based PMP estimates among studies that used different model setups, datasets and moisture amplification approaches, which limits confidence in the estimates: could PMP be much larger than a given study indicates? Or are we over-designing for unrealistically large events? Focusing on mountainous watersheds in the coastal western U.S. affected by atmospheric river (AR) storms (the Feather and Willamette River basins among others), I seek to identify the largest sources of uncertainty affecting model-based PMP estimates and ways to reduce these uncertainties. Using the Weather Research and Forecasting (WRF) regional atmospheric model, I performed experiments in which I reconstructed and amplified several severe storms. First (Chapter 2), I produced an ensemble of simulations to characterize model uncertainty (parametrization and propagation of model error). I found that the spread in PMP totals among ensemble members was modest, ranging from +/- 7% of the ensemble mean. Next (Chapter 3), I simulated artificial storms selected from a large sample (~1500 years) produced by a global climate model (the CESM2 large ensemble, or CESM2-LE). I found that the largest PMP estimates they produced was only 8% larger than estimates obtained from observed storms. Finally (Chapter 4), I found that the way moisture maximization is implemented has the largest impact on PMP estimates (with saturation of the atmosphere resulting in up to 20% larger precipitation totals than other approaches). I as a result developed a method that uses climatology to objectively determine the magnitude of moisture amplification. The understanding of the uncertainties in model-based PMP estimates and the tools I developed to characterize and reduce these uncertainties provide key information for the development of more reliable model-based PMP estimation guidance.