Summary

The paper presents a new approach to model and quantify infrastructure resilience to flooding. The paper uses an ecological conceptualisation of resilience defined as the ability of a system to maintain functionality during a shock event and to recover to a state of equilibrium in a timely manner. The study focus primarily focusses on measuring the time it takes a system to recover using a constrained project scheduling problem (RCPSP) model. This approach breaks the recovery process down into individual steps that are needed to repair or replace infrastructure after a shock event. Identifying the length of each step and whether the step is dependent on a previous one, the model estimates either the total recovery time or to what degree a system has recovered for pre-defined time steps (e.g. 300 days) for hazard events of different severities. This conceptual model is then applied to a neighbourhood in Accra, Ghana to estimate their resilience, i.e. the time to recover for statistical flood events with return periods of 5, 50 and 500 years. The results are presented as both the mean recovery time in days as well as the mean recovery state in percent 300 days after the event.

General comments

The paper addresses the very relevant question on how to measure and quantify the resilience of a system or community after a shock event. The study clearly defines what resilience means in this context and makes the argument that more complex conceptualisations of resilience might better reflect reality but are too complex to model and therefore often become a road block in operationalising resilience in disaster risk management planning. The paper is generally well written and structured. My two main points of criticism are:

1. The authors describe in line 95ff that agent-based models have been used to model resilience, but have been criticized for their lack of transparency, difficulties to assess and little consensus about model complexity. While the authors argue that their model is an improvement to previous approaches, I struggle to see how the disadvantages of ABMs mentioned do not apply to RCPSP. From reading the manuscript my understanding of RCPSP is that it is very similar to an ABM in terms of specific assumptions of how different building classes are damaged and how specific tasks during the recovery process are carried out. I would argue that this creates the same challenges as using ABM as there is also little consensus about how these tasks are carried out, whether it would be affected by local supply chain issues etc. While I can see the author’s point that the model itself is more linear and therefore more transparent, my understanding is that it is still based on a large number of rules and values that are largely based on assumptions or best guesses. Would be great if the authors could discuss how their model and results compare to other approaches modelling resilience.
2. It would be great if the authors could say a bit more about how decision makers and planners can use the outcomes of their model. Especially in regard to the 300-day resilience: what does a value of 0.69 mean for planning and decision making? In my view this information is only useful if the outcomes would be presented alongside specific thresholds above which the building is functional/inhabitable again (with restrictions). My impression is that this would be possible to include into the scenarios and could be helpful when planning for shelters, temporary accommodation etc. as it would mark the point when people and businesses can move back.

Specific comments

L9ff: I am not sure how useful the numbers are in the abstract as they are not very intuitive to interpret. I would recommend to replace this with a more qualitative statement about what the results mean.

L265: How do you come up with 300 days as a suitable assessment period. It seems a bit arbitrary and probably also dependent on the magnitude and size of the event, whether this is a sensible value to present. Was wondering how this metric is more useful than simply reporting the average number of days until full recovery. As mentioned in the general comments I am wondering how discretising the days it takes to recover is useful for decision making.

L284f: Would be good if the authors could mention what the metric for the benchmark performance is.

L302: A threshold of 100sqm seems large for footprints of informal buildings. Is there a reason for the threshold being this high?

L328ff: Would be could to briefly describe how the damage functions were developed and for which region instead of only referring to another paper.

I would like to start with emphasizing that I think the manuscript  is well-written and was very interesting to read. There is indeed a gap in the literature when it comes to the assessment of recovery, and this manuscript presents a very interesting approach to fill some of this gap. I have a few comments to improve the manuscript:

• On line 212 (in assumption 1): what type of tasks are you referring too?
• On line 403: have you checked the completeness of land-use in OSM? That data is not always fully complete, and may impact your results.
• I would start the results section with section 4.2 (the inundation and damage), and then present the first results of the recovery (now section 4.1). That feels, to me, a bit more logical to read.
• Linked to the comment above, I actually miss a discussion sections (also directly the most important comment of my review). The ‘results and discussion’ section is actually just a ‘results’ section. As such, I would strongly suggest to include a separate discussion section to actually discuss the results and the methods.
• I would suggest to include a discussion section that includes the following elements: (i) put the method in perspective with other recovery studies (it is limited of what has been done, but good to discuss how this model has improved the field/is different from others), (ii) put the results in perspective (can we validate some of the results, even if its just anecdotal evidence)? And implications of this work.
• Finally, one of the most important issues: why is the data and code not available? It is not really of this day and age anymore to not publish the code along-side an article in which a method is being presented. And most of the data is based on open data anyways (e.g. the data extracted from OpenStreetMap)? If the code becomes available, other researchers can also benefit much more of this work. You can just license the work to make sure people use it the way you want them to use it.