| 22 Sep 2026
Daily Precipitation-Frequency Estimates under Climate Oscillations across the Conterminous United States
Abstract. Precipitation-frequency data are essential for design storms, flood risk assessment, infrastructure design, and erosion and sediment control plans, but remain uncertain due to regional heterogeneity and climate variability. Existing precipitation-frequency products are generally based on stationary estimates and provide limited information on how extreme-precipitation frequency varies with large-scale climate conditions. Accounting for these spatial and climate variability effects in models over large domains is computationally intensive, which has limited the generation of such precipitation-frequency products. In this study, we developed a 0.1° gridded precipitation-frequency dataset, named PreXFOCUS (Precipitation eXtreme Frequency under Oscillations in Climate across the Conterminous United States; Takallou et al., 2026), using Bayesian hierarchical modeling of annual maximum daily precipitation. The dataset is available on Zenodo at https://doi.org/10.5281/zenodo.21979711. Four model formulations were compared, ranging from independent site-specific estimation to spatially varying extreme-value parameter fields with Gaussian- copula dependence among annual maxima. Spatial pooling across all GEV parameters combined with data-level spatial dependence provided the best goodness of fit. The selected model was then used for conditional spatial simulation at held-out sites, where the resulting return levels were validated against estimates obtained from direct model fitting. These return-level estimates were subsequently incorporated into spatially varying Bernoulli occurrence models to quantify climate-conditioned changes in their exceedance probabilities. Adding MEI, NAO, and AMO to these occurrence models improved model fit relative to the time-invariant baseline and shifted exceedance probabilities by up to 40 % in some regions. The resulting products provide fine-resolution precipitation-frequency design estimates and climate-informed return levels across the CONUS, supporting hydrologic modeling, infrastructure planning, and regional flood-risk assessment.
General assessment:
This manuscript presents PreXFOCUS, a gridded dataset of stationary and climate-conditioned precipitation-frequency estimates across the CONUS. The dataset could be useful for extreme precipitation analysis and for applications considering climate variability. The data are publicly available, with README files and instructions for using the reference and climate-conditioned products. However, when inspecting the released dataset, I noticed a potential issue with the uncertainty ranges of the nonstationary estimates that needs further investigation.
Overall, the manuscript provides a useful description of the dataset, although some aspects of the methodology, validation, and interpretation of the results could be clarified.
Specific comments:
PreXFOCUS dataset
Nonstationary estimates: In the released dataset, I found negative values in the 90% interval range for the nonstationary estimates around Texas. If this range is calculated as the upper bound minus the lower bound, it should always be non-negative. Could the authors check this?
Introduction
Line 53: “BHMs must represent two distinct forms of spatial structure.” Not all BHMs need to explicitly model both forms of spatial structure simultaneously.
Line 70: Please clarify what is meant by “decrease identifiability.”
Line 70: The rationale for using the Bernoulli approach is not sufficiently clear. Although the authors mention potential identifiability issues and increased uncertainty with the nonstationary GEV approach, they do not discuss the limitations of the two-stage Bernoulli approach, particularly the lack of uncertainty propagation from the first-stage GEV estimates. Please clarify why this approach is preferred.
Methods
Line 105: The 20% missing-data threshold seems a bit loose for annual maximum precipitation analysis. I would recommend a sensitivity analysis using a stricter threshold, such as 5% missing data.
Section 2.1: Please clarify whether the precipitation estimates in PreXFOCUS represent constrained 1-day precipitation or unconstrained 24-hour precipitation. NOAA Atlas 14 provides unconstrained 24-hour estimates, so this distinction should be considered when comparing the two products.
Section 2.2: I would recommend including a summary table comparing M1–M4 and N1–N2 to clarify the differences. A flowchart summarizing the overall experimental design, including the model selection procedures for M1–M4 and N1–N2 would also make the methodology easier to follow.
Line 228: I’d be cautious about using differential entropy to evaluate precision, as it measures the spread of the fitted distribution, and a lower or higher spread does not necessarily indicate better precision.
Results
Figure 2(c–d): It would be helpful to discuss the differences in the spread of CRPSS among the four models, and what these differences suggest about their predictive performance across sites.
Line 397: It would be helpful to explain more about why M4 is preferred over M2 and M3, given that it does not perform best across all validation metrics.
Section 3.2: Maps showing the differences between M4 and NOAA Atlas 14 estimates at the gauge locations would be helpful for seeing the spatial patterns of the differences.
Line 472: Since NOAA Atlas 14 is based on a different dataset (time period till 2017), I would be cautious about concluding that M4 performs better based on validation against the observations used in this study.
Figure 8: Additional maps showing the differences between the climate-conditioned estimates and the stationary M4 estimates would be helpful for seeing the spatial patterns of change.