Title: Computation of self-recruitment in fish larvae using forward and backward in time tracking in a Lagrangian model (SWIM-v2.0) of the simulated circulation of Lake Erie (AEM3D-v1.1.2).

I have carefully reviewed the manuscript by Shi et al., which presents a valuable theoretical and modelling framework for estimating self-recruitment (SR) in larvae by using both forward and backward particle tracking in a biophysical model. The approach is based on a Lagrangian model. This methodology is especially used in the context of population connectivity studies, where the accurate estimation of SR is important for ecological research, conservation planning, and ecosystem management.

The manuscript is clearly written, well-structured, and provides a strong demonstration of the methodology through a case study in Lake Erie. Notably, the authors show how this approach can be linked with genetic studies to validate or complement connectivity estimates, which significantly broadens the applicability of their method.

The study is timely and offers a potentially general tool for assessing self-recruitment, larval dispersal and connectivity in aquatic systems. Overall, I support the publication of this manuscript after minor revisions. My main concern relates to the number of particles used in the simulations and how this might influence the robustness of SR estimates.

Minor Comments:

1. The manuscript would benefit from a more detailed discussion on whether the number of released particles is sufficient to saturate the system, especially given the stochasticity introduced in the simulations. A paragraph in the Discussion section addressing how particle number affects SR estimates, and the potential biases introduced by under-sampling, would strengthen the paper. Ideally, some justification or sensitivity analysis could be added or referenced.
2. It would be valuable to include a short discussion on how larval mortality during dispersal and settlement could influence SR estimates. Furthermore, if the model outputs or the analytical framework could allow for the inference of mortality (e.g., when both larval production and SR are known), this should be briefly discussed.

Specific Comments:

• Lines 15–17: The sentence beginning by “However, various strategies have been employed…” is somewhat confusing. Consider rephrasing for clarity.
• Line 231: The number of larvae released should be explicitly stated in the main text, along with a brief rationale for varying numbers across different release locations.
• Line 232: Mention that the tracking duration corresponds to the pelagic larval duration (PLD) of the target species.
• Line 235: Please clarify why forward tracking is conducted from four regions while backtracking is performed from only two. Does this reflect known settlement areas or observed recruitment patterns in Lake Erie?
• Table 1: Define LR_F_A clearly in the caption. Also, include a brief explanation of whether the particle release numbers are sufficient for system saturation.
• Table 2: Add the term SR_ij to the caption.
• Table 3: The methodology used to derive the larval numbers in this table is not immediately clear. Are these values inferred from combined forward and backward simulations? Please clarify this in the caption with a brief methodological summary.
• Line 339: The phrase “different larval particle release strategies” needs clarification. Does this refer to spatial distribution, timing, number of particles, or something else?

This is a strong and valuable contribution, and with the suggested clarifications and additions, I believe the manuscript will be of high interest to the modelling and marine connectivity communities.

I got very excited by this work reporting the possibility to simulate self-recruitment (SR) without the need of information on larval production, thereby solving a long-time reported issue for comparing models and data of larval connectivity: specifically, biophysical models are appropriate tools to simulate local retention (LR) whereas field studies estimate SR.

Unfortunately, I believe that the proposed method implicitly assumes spatially homogeneous larval production and is therefore no more appropriate for simulating SR than the usual method. The assumption of homogeneous larval production is here implicit, because of the use of backward-in-time tracking, as oppposed to being explicit when using forward-in-time tracking. Indeed, imagine that a given zone is particularly suitable for larval production in the reality. In the model there is no reason why more particles would be advected back to that zone "hydrodynamically". To represent this "biological" reality of enhanced larval production the number of particles advected back to the given zone should be weighted by larval production from that zone in order to calculate SR correctly. If not, then homogeneous larval production is implicitly assumed.

It is true that the quantity simulated using back-tracking will not depend on the number of particles released in a settlement area, as long as this number is high enough to obtain a statistically meaningful value. However that simulated quantity will be comparable to SR as assessed in the field only if larval production is spatially homogeneous. In conclusion I sadly do not see how the proposed method based on back-tracking is an improvement from the one based on forward-tracking. Back-tracking is an approach that may be more efficient to use computationally than forward-tracking when focusing on particles origin rather than destination but both approaches should give the same results.

lines 40-48 I found this paragraph confusing as it looks like the authors decided to define local retention (LR) and theoretical local retention (TLR) oppositely to what was previously used (Hogan et al. 2012, Burgess et al. 2014, Lett et al. 2015).

49: Parentage analysis and/or larval tagging is widely used to estimate SR, not the other quantities, to my knowledge.

69-72: True, but as explained above I believe this issue remains when using back-tracking.

87-91: Same here.

103-4: It has already been shown that LR and TLR are independent on larval production (Burgess et al. 2014, Lett et al. 2015) and there is no need of simulations to prove this.

Equ. (1) this quantity is defined as LR in Lett et al. (2015), why change?

Equ. (2) this quantity is defined as RLR in  Lett et al. (2015), why change?

167-8 This is true, however as explained above that simulated quantity will be SR as assessed in the field only if larval production is spatially homogeneous.

291-2 I don't understand why SR values estimated from back-tracking and forward-tracking would be different. I don't agree that one is "correct" and the other "erroneous". They should be similar as both are based on the same assumption of homogeneous larval production, as explained above. There may be numerical reasons for the reported differences, or other technical reasons that could be explored.