This is a well-designed study that combines long sub-daily USGS streamflow records, reanalysis/remote sensing data, and two hydrological models to disentangle the relative roles of precipitation, temperature and land cover change on flash flood statistics across CONUS. The multi-method design (trend analysis + RF + scenario simulations) is a strong point. My main concerns are about (i) clarity and transparency of some methodological choices, and (ii) how far some key statements can be generalized given the data and model limitations. I also note several places where the wording could be clearer or more precise.

Major comments

1. Throughout the paper, terms like flood, flash flood, POT floods, and flashiness  are central, but the hydrological definition is not completely transparent from the text. In the Methods section where AMF, POTF, POTM, POTFL and MDF are defined, please clearly state: The exact thresholds for POT events (e.g. percentile, exceedance criteria); The minimum inter-event time (if any) used to separate consecutive events; How you handle overlapping events and multi-peak events. It would also help to explicitly say whether your “flash floods” are defined purely from streamflow response time and hydrograph shape, or whether you also enforce a rainfall duration criterion (e.g. rainfall concentrated within ≤6 h). I suggest adding a short paragraph early in the Methods section with a operational definition (1–2 sentences) and a reference.
2. You use 294 “small and medium-sized catchments” across CONUS, mostly with areas <3000 km². The broad selection rules are mentioned, but some important details are not fully clear.I suggest clarifying filtering rules & missing data: What is the minimum record length and completeness required (e.g. at least X years of hourly data, no more than Y % missing)? How are gaps handled (e.g. is a year with missing sub-daily data excluded from the trend analysis)? Do you exclude gauging stations with clear signs of strong regulation (e.g. upstream reservoirs) or use all available?
3. You use a rich set of precipitation indices (AFHP1h/6h/12h/24h, AMP-N, RX1D, RX6H etc.), which is excellent. However, the connection between these indices and the typical response time of your catchments could be explained more clearly.Please explain why you chose these particular time windows (1 h, 6 h, 12 h, 24 h) and how they relate to: Median concentration time or lag time of the basins. The typical duration of the POT events you identify in streamflow. For basins with longer response times, 24 h rainfall might be more relevant. Do you see different patterns there? A few sentences in the Results section (or Discussion) would help.
4. The Random Forest (RF) analysis is an important part of the study, but some methodological details are currently too brief. It would be helpful to provide basic RF settings (number of trees, maximum depth, minimum samples per leaf, random seed) and to indicate whether any cross-validation or out-of-bag error estimates were used. In addition, the procedure for converting continuous importance values into discrete dominance classes (e.g., “precipitation-dominated”, “temperature-dominated”) should be explained—particularly the threshold used and the rationale behind it. Given that many predictors are correlated (e.g., PET–T, various precipitation indices), acknowledging potential biases in RF importance and, if feasible, adding a simple sensitivity check such as permutation importance or removing one of two highly correlated predictors would be valuable.(Or add some discussion)
5. The “P+T-dominated” category appears especially meaningful but is not yet fully explored. A short summary of the climatic and land-cover characteristics of these basins, and how they differ from purely precipitation-dominated basins, would help the reader interpret why both drivers matter simultaneously.The two DRIVE temperature scenarios (“dynamic temperature” vs. “static 1981 temperature”) are central to the conclusions, but the construction of these scenarios needs further clarification. Please specify exactly which meteorological variables differ between DRIVE-DT and DRIVE-ST—whether only air temperature is modified, or whether entire 1981 meteorological fields are reused.
6. The land-cover analysis is important, particularly the conclusion that LULC change has limited hydrological impact in most basins. Providing more methodological detail would strengthen this argument. For example, please describe how GLC_FCS30 is reclassified into the land-cover categories required by DRIVE and TS-DUH, whether fractional cover or dominant classes are used, and how TS-DUH incorporates land cover (e.g., roughness, routing speed, retention). In the Results, when reporting that most basins show <3–5% hydrological change, a brief summary of the actual land-cover transitions in the 294 basins (e.g., forest–cropland, cropland–urban) would help readers assess whether the small hydrological response reflects limited LULC change or limited model sensitivity. For the urban case study (e.g., Atlanta), reporting the baseline impervious fraction and the absolute magnitude of the peak-flow increase would provide useful context.
7. Because the study applies Mann–Kendall tests and Poisson regressions across nearly 300 basins and multiple flood and climate indices, it would be appropriate to briefly discuss the potential for false positives arising from multiple hypothesis testing. Even a short note on whether any false-discovery-rate (FDR) considerations were made—or how many significant trends might be expected by chance—would be helpful. Clarifying how sensitive the key conclusions (e.g., “67% of basins show significant increases”) are to the chosen p-value threshold would further improve transparency. For the temperature and LULC scenario results (e.g., the reported 3.6%, 8%, and 20% changes), showing the distribution across basins (e.g., boxplots or histograms) would allow readers to understand variability and uncertainty rather than rely on single summary statistics.

Minor comments

1. Some expressions in the abstract could be clearer. For example, the phrase “mitigate this effect” is slightly vague—please make clear which effect is meant (e.g., “mitigate the increase in flash-flood frequency associated with heavier precipitation”). The final sentence is long and contains several ideas (spatial–temporal variation, urbanization, flood-risk management). Splitting it into two sentences would improve readability.
2. The manuscript uses both “land cover” and “land-cover.” Please choose one style and apply it consistently. A common approach is to use “land cover” as the noun and “land-cover change” when used as an adjective.
3. Please ensure all abbreviations are defined at their first appearance in the main text—not only in tables. Some that may need checking include POTFL, MDF, AFHP, AMP-N. Key abbreviations should also be briefly defined in figure captions so that each figure can be understood on its own.
4. Please check for consistent unit formatting throughout the manuscript—for example “mm d⁻¹” vs “mm/day” and “°C” vs “degC.” Make sure the units for PET and temperature-related indices are given at their first mention and shown clearly in figure axes.
5. In places where “flood intensity” is used, please specify whether this refers to peak flow, specific discharge, or flood volume.
6. When describing the effect of warming, please consider adding qualifiers such as “in our model experiments” or “under the scenario where only temperature changes.” 

The manuscript "Drivers of Flash Flood Frequency and Intensity in the United States: A Quantitative Analysis of Hydrometeorological Interactions" presents a comprehensive analysis of flash flood trends across the Contiguous United States (CONUS) over a 30-year period. By integrating observational data with the DRIVE hydrological model and the TS-DUH method, and utilizing Random Forest for attribution, the authors provide valuable insights into the non-linear relationship between precipitation and flood response. The study’s finding that increased heavy precipitation frequency does not uniformly translate to increased flood frequency is important and aligns with current hydrological debates regarding the non-stationarity of flood generation mechanisms. However, several key aspects of the methodology and interpretation require clarification and strengthening.

Major Comments:

1. The authors selected data from May to September for analysis (or was this criterion solely used for station screening? Please clarify). While this period encompasses the high-incidence season for flash floods, this strategy inevitably excludes spring snowmelt floods and autumn/winter storm floods. Such seasonal truncation may lead to the oversight of specific hydrological processes. It is suggested that the authors elaborate in the Discussion section on the potential impact of this data screening strategy on the generalizability of the conclusions.
2. The study utilizes NLDAS-2 precipitation data with a resolution of approximately 12 km. For the many small catchments mentioned in the text (e.g., L213, with an area of 78.8 km²), this resolution implies that the catchment is covered by very few grid points, which makes it difficult to capture the spatial variability of localized convective precipitation events that cause flash floods. I recommend that the authors discuss the influence of this limitation on simulation accuracy.
3. Figure 5 indicates that some catchments are dominated by “Temperature,” and the text attributes the flood attenuation caused by warming to “rising temperature and evapotranspiration.” The current explanation is somewhat generalized. Please further clarify the specific physical mechanisms involved in this process.
4. Figure 5 displays the spatial distribution of the primary drivers. It would be beneficial to supplement this with a chart (such as a stacked bar chart or table) to quantitatively present the proportion of catchments dominated by distinct single drivers versus combined drivers (e.g., P&T, P&Land Cover), along with a brief discussion on the spatial distribution characteristics of these combined driving patterns.
5. The abstract mentions that LUCC has remained stable over the past 40 years, which may be attributed to the fact that the 294 selected stations are mostly located in upstream areas with less human interference. Please add a restrictive description of the study station attributes in the abstract. Furthermore, the statement that “our results show 10% increase in impervious surfaces could lead to 20% peak flow increase” appears to be based on specific cases like Peachtree Creek rather than the general analysis. The wording in the abstract needs adjustment to explicitly state that this conclusion applies to specific contexts (e.g., rapidly urbanizing areas).
6. The current description of the Random Forest regression is too brief. Key details, such as hyperparameter settings and the division of training/testing sets, should be added to the Methodology section.
7. The TS-DUH model operates at a 90 m resolution, while the DRIVE model runs at 0.125° (approx. 12 km). How was this significant resolution discrepancy resolved during the coupling process? In particular, the processing of land cover data within DRIVE when extracting vegetation parameters requires a more detailed explanation.
8. The study analyzes the long-term trend of soil moisture (ASM) and finds a downward trend. However, in the rainfall analysis, the Random Forest utilizes the annual mean ASM. For events like flash floods, the soil moisture prior to the event is a more direct factor determining runoff response. It is recommended that the authors explore the role of “antecedent conditions” more deeply in the Discussion/Results section.
9. A paragraph regarding model uncertainty could be added to the Discussion section. For example, discuss the uncertainty of DRIVE model parameters and the potential impact of selecting different atmospheric forcing data (e.g., NLDAS-2 vs. MERRA-2) on the results.

Minor Comments

1. Please unify the format of separators between keywords.
2. In the abstract, please specify the exact temporal precision of the “sub-daily data.” Additionally, indicate the statistical test method and significance level used for the reported 17.3% increase in flash flood frequency.
3. When citing Ivancic and Shaw (2015), I suggest adding recent literature regarding “Flash Drought leading to Flash Flood” to provide a more comprehensive context.
4. There is a discrepancy between Line 334 (90.8%) and Line 361 (90.3%) regarding the percentage of observation points with discharge changes within ±3%. Please verify.
5. Provide the full name or a brief description when “GLC_FC30” first appears.
6. The manuscript totals 25 pages, with the reference section accounting for 6 pages. authors are recommended to streamline content, focusing on optimizing references to balance core research and citations.