General comments:
The revised version of the manuscript significantly improved compared to the previous version. The methodology as well as the data set is now explicitly described and splitting the model calibration in validation in two chapters makes it easier for the reader to follow the study design. Therefore, I would recommend accepting the manuscript for publication after the following minor corrections:
P. 5 L1-3: It is unclear why the authors in a first step use geo-referencing to transfer the address into a point coordinate to visually inspect if the address is within the building footprint. As a person not familiar with the Italian house number system, I would assume that every damage entry for an address (incl. house number) also belongs to a particular building. Please provide reasons for your approach.
P6 L10-20, P.7 L21-26: According to P.7 L22-25 the authors are using a 3-fold cross validation, where for each of the three iterations two thirds of the data is used for calibration and one third is used for validation. However, reading the previous chapter P.6 L10-20, one could think that the full 613 observations are used for calibration. I would recommend mentioning the data splitting procedure in P.6 and that in fact three models are calibrated with the three different data sets. It is also not 100% clear to me, if the bootstrap approach described on P. 6, is performed before or after splitting the dataset for cross validation.
Technical Corrections:
Abstract L1: amount in million Euros instead of dollars
P. 3 L26: 2.5 thousand to 25,000
P. 9 L17 -22: can be "made" instead of "found". Shorter sentences
P. 10 L15: “maximum damage as a percentage and the starting elevation for damage”: please specify percentage of …; please clarify “starting elevation damage”
Fig 5 and 6: Class limits of third class are probably wrong 14-60 should probably 41-60 |