Review of essd-2022-475 by Miguel Cruz

I commend the authors for their work on developing a significant wildfire behaviour dataset. I found the work of great interest to the fire research community, and can see their methods and data being used by many in the I have few major comments (below) and a number of small comments that are in the attached pdf.

1. The manuscript reads well, but there are a number of sentences that seem incomplete or are somewhat incoherent. I note those in the attached pdf. Authors should also be more direct and concise in their writing. The manuscript is quite long, and I made note of content that could be left out as it is not necessary.
2. This is a wildfire spread database which is quite relevant. The authors should nonetheless mention clearly that no collation of data on weather, fuels, topography, etc was conducted within the context of this study. This is never mentioned. Although the authors mentioned later in the manuscript that the data can be used for better understanding of wildfire drivers, model evaluation, etc. this cannot be conducted unless other data is present. As it is, the spread data in isolation does not allow for much of an analysis.
3. There are important limitations of the satellite data. Some of them are mentioned, others are not. An important point that should be made is that just because the satellite or someone says the fire is at a certain location at a given time, it does not mean that the fire just arrived there at such time. The fire might have arrived hours before, and hence the average rate of spread for a burning period is a value that is diluted, combining periods of rapid spread and no spread. This is quite relevant as fire behaviour is highly nonlinear, and averaged values over larger time periods can be misleading. These aspects should be noted in the manuscript.
4. It is not clear how combining different methods to map a fire location are integrated. Do some methods have prevalence over others? Photographic evidence vs satellite information? Is there a process that is followed? If yes, this should be described.
5. I have two main technical comments.
   1. The proposal of a fire behaviour classes based on your data is fraught with error. It is ok to explore the distribution of your data, but to propose such distribution (which you said, was biased to large fires) to derive a fire behaviour classification is wrong. The classification class threshold have no physical meaning, and you can realise that if you check into a number of fire behaviour and danger classifications developed from fire behaviour – operational implications. A proof that your proposed classification is meaningless, is the fact that if in the next two fire seasons you add 40 new wildfires all burning under moderate to high fire danger (lets say it is a mild fire season), your new fire behaviour classification classes will change drastically. What is the point then? I strongly suggest this is removed from the manuscript.
   2. The authors use their dataset and make a ‘finding’ that area burned is mostly a function of fire growth rate rather than rate of fire spread. This result is obvious by several reasons, the simplest one being that the rate of spread is only related to the area burned for the initial stages of a fire growing from a point source. From the moment a fire is affected by topography, fuels, and burn over several burn periods and days, it is the area growth rate that is linked with the fire area, not the rate of fire spread. I do not see this a finding, whatsoever. Of course, a 2 dimensional area growth metric is going to be more related to the final burned area than a one dimensional metric of fire propagation (ROS). As with the previous point, I strongly suggest that this is removed from the manuscript.

It is not clear why the authors depart from their main focus of the study, describing how the database was assemble, to do a spurious analysis of the data and come up with these findings, that, in my view, are not really findings. If the authors want to explore those aspects of fire behaviour, then they should do so in a different piece of work, with proper basis and analysis.

Other minor comments:

What is the certainty in the intermediate perimeters (isochrones) in figure 3? I cannot imagine you were able to collect data across all the perimeter to make such a nice polygon. Were they interpolated? If so, they might provide a false sense of certainty.

Benali et al. have developed a valuable wildfire behaviour dataset that includes the spread of 80 large wildfires in Portugal between 2015 and 2021. The authors have combined data from multiple sources, which helps to reduce gaps and uncertainties in the wildfire data collection. This open data has the potential to improve the simulation of wildfire in the context of changing environments and to better manage wildfires. The manuscript is well-written and the authors have done a commendable job of presenting the data. I think it is publishable if several minor issues can be addressed.

1. The Input Data section could benefit from being more concise. Consider presenting the information in tables, especially when discussing data sources, to improve clarity for readers.
2. In section 3.1 Overview of the PT-FireSprd database, the authors provide a comparison of ROS and FGR and give some preliminary results. However, this appears to detract from the main focus of the paper, and as a result, the section feels disjointed. Consider revising the section to more clearly tie to the central argument of the paper or move the comparison to an appendix to maintain focus on the main topic.

Other minor comments:

Line 508: It should be 3.2 not 2.2.

Review: The Portuguese Large Wildfire Spread Database

This manuscript details the creation of the first truly multi-proxy high spatial and temporal resolution fire progression archive of its kind. Combining data from multiple remote sensing sources leveraging the full range of available temporal and spatial resolution, combined with empirical observations, written records, and photographic evidence, this is the most complete crosswalk of multi-proxy data sources I have seen.  The dataset has significant potential to improve the understanding of fire activity, resource use effectiveness, fire climatology, and other fields in Portugal.

                The data are catalogued in three phases, the first in my option being the largest contribution and most important- a structured approach to reconstructing fire progression at the finest resolution possible.  The second phase develops three derived descriptor variables that include some straightforward calculations under the stated assumptions (rate of spread and rate of fire growth) and some more questionable calculations (average fire radiative power) given the nature of the data sources.  The third phase leverages the derived characteristics to ascribe a fire behavior class to each fire growth polygon.  I’m not clear on how thresholds were determined to differentiate fire behavior classes or how they would be used operationally but if determined in consultation with fire managers I can see this as a useful way to present the data.

Major concerns:

                The treatment of Fire Radiative Energy as an additive measurement or something that can divided by an area is problematic.  By definition, fire radiative energy is an instantaneous and constantly varying measure of energy release for a given area of measurement (Zhang et al 2018).  Dividing FRE by an additive area metric to divvy the instantaneous measure by the area burned assumes that the area burned occurred in that instant.  The area burned is a function of the free burning rate per 30-minute period. This is mixing average and instantaneous data sources. Unless the MSG_SEVIRI sensor does this different from MODIS, I don't think it can be applied to an area polygon.

Minor concerns:

                The manual methods are quite labor intensive and leave room for standardization and automation that would alleviate some concerns I have about repeatability with other data sources and in other geographies.

Other comments:

I agree that and change detection algorithm applied to the 30-minute FRE, ROS, or FRG could be an interesting way to determine the important phases of fire growth and then relate these phases to other environmental data (e.g. weather, fuel matrix, suppression resources use, etc.).  I hope the research group is able to continue updating the dataset and is able to adapt systems for standardizing quality control of records and automating generation of polygons and derived metrics.  As it is the dataset serves as a valuable series of inputs for future analysis and keeping it up to date will ensure continued use and applications.