Analyzing Independent LFMC Empirical Models in the Mid-Mediterranean Region of Spain Attending to Vegetation Types and Bioclimatic Zones

Round 1

Reviewer 1 Report

This article describes empirical models equipped with a stepwise multiple linear regression procedure for estimating the moisture content of living fuel in an area with Mediterranean vegetation where several species of shrubs and trees coexist. As predictors, we used spectral indices extracted from Sentinel-2 images and their averages for the studied period of time at each site with an accuracy of 10 m, interpolated meteorological data, topographic and static seasonal variables. The overall aim of this study is to calculate and evaluate empirical models for shrub plots on one side and mixed tree plots on the other over a wide area of the Valencian region in the central Mediterranean region of Spain. Field measurements were used to fit the models, obtained over a period that includes data for all seasons of the year and two-year periods of dry seasons. This study is an extension of the study described earlier over a longer study period that included dry and wet seasons, considering a wider set of sampling points in shrub areas, but also applied to areas with trees as the dominant species. The proposed models were adapted to seasonal changes and maps were generated to visualize the spatial evolution in the months leading up to a major wildfire.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

General comment:

The study appears to provide valuable insights into the development of empirical models to estimate Live Fuel Moisture Content (LFMC) in a Mediterranean area. The approach of using stepwise linear regression to differentiate models for shrub species and tree dominant plots is commendable. The inclusion of topographic variables and spectral indices to mark spatial differences in LFMC and reduce model errors also adds value to the study's findings. However, the study has several limitations and deficiencies that should be considered when interpreting its results. Firstly, the study is limited to a specific geographical region, and the results may not be applicable to other regions with different climatic and ecological conditions. So, the title should be revised to specify the region. Secondly, the study only considers a limited number of species and fuel types, which may not represent the diversity of vegetation present in the area. Thirdly, the study relies on field measurements of LFMC, subject to measurement errors, and it is unclear how these errors were addressed in the analysis. Fourthly, the study does not account for the effects of other environmental factors such as wind, humidity, and solar radiation on LFMC. Finally, while cross-validation was performed to evaluate the performance of the models, the study did not perform an independent validation using data from a different time period or geographical location, which could provide more robust evidence of the models' predictive power. Furthermore, the study only covers a period of 1.5 years, which may not capture long-term trends in LFMC variability and may not be sufficient to establish robust models. The study did not consider the effects of anthropogenic factors such as land use change and fire suppression on LFMC, which could influence the accuracy of the models. While the conclusions section of the study does not explicitly address the limitations of the study or ways for improvement, it is important to consider potential limitations and deficiencies when interpreting the findings and designing future research.

Some specific/minor comments:

1. Ensure that the abstract accurately summarizes the key findings of the study.

2. The study may benefit from additional literature review. A thorough literature review can help identify gaps in existing knowledge and provide a better understanding of the current state of research in the field. You might find the following work useful:

- Coates, T. A., & Ford, W. M. (2022). Fuel and vegetation changes in southwestern, unburned portions of Great Smoky Mountains National Park, USA, 2003–2019. Journal of Forestry Research, 33(5), 1459-1470.

- Ellis, T. M., Bowman, D. M., Jain, P., Flannigan, M. D., & Williamson, G. J. (2022). Global increase in wildfire risk due to climate‐driven declines in fuel moisture. Global Change Biology, 28(4), 1544-1559.

3. Please provide a brief overview of the methods used in the study in the introduction section.

4. Please clearly state the contributions of your study in the introduction section.

5. Use active voice instead of passive voice where possible to improve readability.

6. There are some inconsistencies with the MDPI style. Ensure consistent formatting and style throughout the manuscript.

7. For describing the data used, please provide more context and background information where necessary to help readers understand the study's significance.

8. Figures 2 and 4 must be improved. The legends are illegible.

9. Please include a separate section discussing the limitations and deficiencies of the study and ways for improvement.

10. Provide a clear conclusion that summarizes the main findings of the study.

Minor editing is required.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

General comments
I enjoyed reading this paper. While the study is not novel, it is interesting, obviously involved a lot of careful work, and potentially provides some confirmation and advances on previous findings. However, I am a concerned about the regression analyses for the following reasons:

1. The data are repeated measures for sites so I would have expected a site random-effect to be included in each model to account for this, i.e. a mixed model rather than a simple linear model.

2. The authors mention in several parts of the text that spatial effects are important, but the models did not include any terms for site location, distance from the coast etc., and no check for spatial auto-correlation of model residuals is presented.

3. LFMC can only be positive and, I think, one would expect its variance to increase with the mean. There seems to be some indication of increasing variance in several of the panels in figure 4, although it is only slight.

Given these concerns, I would really like to see generalized additive models (GAMs, e.g. with the ‘mgcv’ R package) fitted to these data, with a gamma or similar error distribution that is more appropriate for positive-only data with likely increasing variance. If this was done, spatial effects could readily be included via a 2D smooth function of site locations (e.g. centred map coordinates), which is quite a standard approach. For example, see Estes et al. 2017: https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecs2.1794 (note: I have no connection to these authors). Including a spatial term could also lead to more precise estimates for the other model terms. The predictor for day of year could be represented as a cyclic cubic spline (bs = "cc") rather than using the a priori sine and cosine transformations, allowing the models to explore the potential shape of the fitted trend more flexibly. It is straightforward to include site random effects in a GAM by including a term such as: s(site, bs="re"). Finally, generating predictions from fitted GAMs and calculating bounds on those predictions is no harder than when working with linear models.

The patterns in the data are clearly strong enough for the statistically invalid (in my opinion) linear regressions to still have convincing and probably useful predictive power. Also, it is true that fitting a GAM can make checking for multi-collinearity more complicated compared to using VIF with a linear model (e.g. see the discussion at https://stats.stackexchange.com/a/371043/133387) but I don’t believe this should be a major obstacle. For example, the authors could choose to represent all predictors other than DOY, site locations and the site random effect as parametric (i.e. non-smooth) terms in order to make checking for multi-collinearity easier.

After weighing up the strong patterns in the data on the one hand, against the need for statistical rigour on the other, I’m afraid that I have to recommend the authors re-do the analyses. This really shouldn’t be a big job (a day or two perhaps) and it will result in a much better paper.

My other general comment is that the discussion section presently seems a bit weak, with some paragraphs essentially just restating methods and results rather than discussing their implications for ecological understanding and fire risk assessment.

Specific comments

Line 48: fuel cargo accumulation

I’m unfamiliar with this term. Does it refer to some specific portion of forest fuel?

Lines 123-124: they were weighted in wet, then oven-dried in the laboratory at 100 °C and weighted again

Minor edit: change “weighted” to weighed

Lines 130-131: LFMC values per plot were calculated as a weighted average of LFMC values of the dominant species, considering the FCC (fraction of canopy cover) of each species as weights

This seems like a reasonable approach, but does it assume that the fraction of projected canopy cover is a good proxy for the fraction of biomass, and therefore on the different species being sampled having reasonably similar plant structure?

Lines 169-171: In addition, the mean values of these spectral indices (period: June 2020-November 2021) were also calculated to obtain information about intersite differences in the vegetation.

I am unclear on what was being averaged here. Does it mean that a single mean value was calculated for each index over all satellite dates in that time window? If so, how does that help to characterize inter-site differences? Please explain a little more.

Line 171: de EVI average

Change to: the EVI average

Lines 202-204: Moreover, a LFMC model for the Rosmarinus officinalis species was also calculated using LFMC data of that species in shrub plots where this was the dominant species.

Why was this separate model was fitted whereas one was one not fitted for Ulex, for example? Was Ulex never a dominant species in the sites?

Lines 232-233: In addition, there is a spatial variation of LFMC in the same shrub species, even between plots of the same bioclimatic group

Please see general comments about the need for an explicit spatial term in the models.

Figure 2:

Please use a more colour-blind friendly palette for the lines.

Figure 3:

Please put dates on the X-axis, similar to figure 2. Also, show bounds (e.g. based on central 50% or 90% quantiles of site values, or perhaps standard deviations) for each of the average lines. Without bounds, it is hard for the reader to judge the contrast between the two groups.

Lines 283-285: Differences between plots were captured using the average of spectral indices or topographic factors such as elevation or slope

I was not sure what this refers to.

Figure 4 caption: (se=standard error)

Why is this text included in the caption?

Figure 4:

There seems to be a suggestion of the variance in observed LFMC values increasing with the mean, especially in panels b, d, and f. Please see general comments referring to this.

Figure 5 and figure 6a:

Please use a more colour-blind friendly palette for the lines and show bounds on the prediction line, e.g. 95% central interval.

Line 352: The errors in various dates of 2022 were within the error range of the model

See previous comment – without bounds on the graphed line the reader cannot judge this.

Figure 6b:

This would be a lot clearer if, instead of a bar plot, the observed and predicted values for each site were shown as vertically aligned points, with bounds (e.g. central 95% interval) on the predictions.

Figure 7:

Please use a more colour-blind friendly palette (e.g. viridis or cviridis) for these maps.

Lines 397-399: Ulex parviflorus, on the other hand, is a species that dries out very quickly and does not retain as much moisture, which makes it very dangerous in terms of fire spread.

Given this, and as per previous comment, why wasn’t a separate model fitted for Ulex whereas one was fitted for Rosmarinus?

Lines 404-406: Thus, sampled species in each plot (table A2) were averaged to represent the LFMC for the same date and location, using the FCC information for each species

Do the authors think that such averages will be the best measure of potential fire behaviour? Or would it sometimes be better to focus on individual common species which dry out quickly such as Ulex? Some thoughts about this might be useful in the discussion section.

Lines 416-418: On the other hand, the highest errors occurred in the estimation of the LFMC of Rosmarinus officinalis species, in which the temporal variability is much greater, with LFMC values above 200% in the humid season, but below 60% in the dry season.

Errors were (I assume) based on the difference between observed and mean predicted values. A more informative and practically useful measure might be the proportion of observations that were within the 50% and/or 90% bounds on the predictions. This could be calculated in a sliding window over the LFMC range, or alternatively within intervals of LFMC.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The manuscript has been thoroughly revised. As far as I am concerned, it can be accepted for publication.

Reviewer 3 Report

I thank the authors for their considerable efforts in revising the manuscript which is now much improved.