Predicting Lightning from Near-Surface Climate Data in the Northeastern United States: An Alternative to CAPE

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Predicting Lightning from Near-Surface Climate Data in the 2 Northeastern United States: An Alternative to CAPE

Charlotte Uden, Patrick Clemins, & Brian Beckage

Summary:

Authors use a straightforward approach to parameterize seasonal flash rates over the northeastern U.S.  in terms of multiple surface variables.  The framework they develop can be used by others to parameterize flash rates over other regions, especially for regions and times when upper air data are unavailable. 

General:

Training at seasonal scales.   Discuss the advantages and disadvantages of training at seasonal scales as opposed to at higher temporal resolution – say hourly.

Northeast U.S. focus.   How generalizable are these results to other domains within the U.S. and the world.

Specific:

L49: lighting -->  lightning

L65-66: Why did you choose to use these 6 near-surface variables as opposed to other near-surface variables? Was it determined by the variables available in ERA5 or was some weight given to which can be reconstructed from paleoclimate data.  The reader might be interested in how various terms are reconstructed from paleo data and for what time perios they can be …

L96-98: What was the detection efficiency of the lightning network and did it change from 2005 to 2010? How does the network distinguish CG flashes from IC flashes?

Explicitly state that you are modeling CG flashes and not the total of CG and IC flashes.

L137: What do you mean by the baseline models “were fitted without standardization?

L120-121: 3246 observations – does that mean there were 3246 grid boxes within New England and New York? or would there be 3246 / 6 years grid boxes? i.e., be clear as to what that number means.

L187: How did you determine which of the surface variables to leave out as you moved from N13 to N12 to N11 etc.   Did you try all possible combinations and then choose the best fits or other.

L270: Figure 2: Adding an additional y-axis on the RHS with abbreviated labels (1-13) would be helpful.

L291 permuted --> permuted (shuffled)

L300-302: The fact that CAPE outperforms individual surface predictors should be mentioned in abstract or conclusions.

L307: The poor performance of precipitation as a predictor is somewhat surprising given that convective precipitation rates are sometimes used as a predictor of flash rates.   Is this because stratiform precipitation dominates the distribution, due to the fact that you are training on a seasonal versus an hourly scale or other reason?

L309-310: Add temporal correlations or some other variable to support your contention of limited temporal tracking.

L315 and elsewhere: More context is needed as to what constitutes a good RMSE.   Would it make more sense to show the normalized RMSE or coefficient of variation.

L349: Could you include the spatial correlations on Figure 5 and discuss.  Most readers are too lazy to go back to Figure 2.

L370-372: Add some statistics (ACC & RMSE) that show much performance degrades when surface pressure is added to the mix. 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study develops statistical models using only near-surface climate variables to predict lightning in the Northeastern US, offering an alternative to CAPE-based methods and evaluating linear, gamma GLM, and Bayesian approaches. This manuscript presents a valuable and timely investigation into predicting lightning activity using solely near-surface climate variables, with the explicit goal of enabling long-term paleoclimate reconstructions where upper-air data are unavailable. However, the manuscript would be significantly strengthened by addressing the following major concerns, which primarily revolve around the scope and generalizability of the findings.

General comments

（1） The combination of a very short study period (6 years) and a single, low-lightning region (Northeastern US) raises questions about the stability and transferability of the derived parameter estimates and model relationships. The conclusions would be more powerful if the models were at least validated against a held-out period or a different geographic region to demonstrate robustness beyond the specific spatiotemporal window used for training.

(2) The study would benefit from a deeper discussion of the physical reasons why these near-surface variables are effective proxies for the convective processes typically captured by CAPE. While shortwave radiation is correctly highlighted, a more thorough physical interpretation linking all key predictors (e.g., why surface pressure is important but degrades temporal skill) would elevate the discussion from a purely statistical correlation to a more process-based understanding.

Specific comments

Line 26: Monthly aggregation ignores storm-scale processes, restricting utility for short-term forecasting.

Line 47: Please check whether the format of the references cited is accurate. This gives the impression of "piling up literature" to support a weak argument. Reviewers might question: How strong is the consensus that the fire risk in the northeastern United States is increasing?

Lines 281-285: This paragraph only describes the performance of the benchmark model, but does not explain or comment on it. Why is the spatial correlation of the model based on CAPE "limited" and the anomaly correlation "weak"?

Line 305: When reporting the Bayesian model B4, only its spatial correlation (r=0.20) was given, which is much lower than that of other models. This is actually an early sign of a major flaw in the Bayesian model, but the author merely listed it as a data point without emphasizing its severity.

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

OK. No further comments. Thanks.