Review Reports

Reviewer 

Transcription of suggestions or comments made regarding the reviewed manuscript

Response and Action in the Revised Manuscript     

1

Moving window size. In this study, the authors compute the Rao’s Q index using a fixed 9×9 moving window.Though it has been adopted in some relevant studies, it would be more flexible if users could customize the window size according to their research scenarios. Since window size is a critical parameter affecting the Rao’s Q index, the platform should offer more options to enhance flexibility and interactivity.

We thank the reviewer for this suggestion.

The option to customize the window size has been added to the platform, alongside the ability to select the threshold method through the interface. 

These new functionalities allow users to adapt both parameters to the characteristics of each study area and event type.

1

Equation error. In the Equation 1, the symbol “x” should be replaced with the multiplication sign “×”, so the correct expression is dij​×pi​×pj​.

We thank the reviewer for pointing out this typographical error.

Equation 1 has been corrected accordingly, replacing 'x' with the multiplication sign '×', yielding the correct expression d*ij* × p*i* × p*j*.

1

Threshold determination. The threshold for change detection is determined using a histogram-based method. The authors should provide a clear explanation of how the optimal threshold is selected from the histogram.

We thank the reviewer for this observation, which prompted us to clarify an imprecision in the original manuscript.

The threshold for change detection was established using Otsu's method (Otsu, 1979), which determines an optimal cut-off by iterating over all possible threshold values and selecting the one that maximizes the inter-class variance between the 'change' and 'no change' populations in the difference map. This method was selected given its suitability for bimodal distributions (Sezgin & Sankur, 2004),  a condition expected when comparing snow-covered pre-event surfaces with lava-covered post-event areas. 

The Methods section has been revised to reflect this explicitly. The platform additionally offers the triangle (secant) method as an alternative thresholding approach (Zack et al., 1977), now also mentioned in the Methods section. All reported accuracy metrics remain valid and unchanged.

Reviewer 

Transcription of suggestions or comments made regarding the reviewed manuscript

Response and Action in the Revised Manuscript     

2

The reference list is unbalanced due to a significant number of self-citations (approximately 6-7 references to works by the same group of 63 authors). Additional references to independent groups working on WebGIS solutions and spectral diversity metrics are needed. 

A brief positioning of the platform in relation to modern deep learning-based change detection approaches would also strengthen the introduction.

We thank the reviewer for this suggestion.

To address both points, the reference list has been expanded with recent and independent works on WebGIS solutions and spectral diversity metrics, reducing the proportion of self-citations and providing a more balanced theoretical foundation. 

In addition, a brief positioning of the proposed platform in relation to modern deep learning-based change detection approaches was incorporated into the Introduction to strengthen the methodological context.

Added text (line 105):

"Unlike deep learning-based CD methods, which often require large volumes of labeled training data and substantial computational resources, the proposed approach leverages Rao's Q Index to provide a statistically transparent and interpretable measure of spectral heterogeneity. This makes the platform particularly well-suited for WebGIS applications, enabling efficient and reproducible change detection without the need for model training or specialized hardware."

These revisions broaden the scope of the literature reviewed and better situate the contribution of this work within the current state of the art.

2

Line 106: The abstract mentions "post-fire spatial analysis" in the context of a volcanic eruption study. This appears to be a carried-over error from a previous publication. 

We thank the reviewer for pointing out this inconsistency.

Revised text:

“This system integrates and evaluates the performance of various spectral indices, providing an efficient and accessible environment for conducting post-eruptive spatial analyses.”

This correction ensures that the Introduction accurately reflects the scope of the present study.

2

Lines 102–106: The objective statement contains an incomplete construct ("fast change detection and detection") and should be revised.

We thank the reviewer for this remark.

Previous:

“The general objective of this study was to develop a WebGIS platform capable of rapidly detecting and change detection across distinct insular regions, namely Reykjanes Peninsula, through the implementation of Rao’s Q Index–based change-detection approaches.“

Revised:

“The general objective of this study was to develop a WebGIS platform capable of rapidly detecting land cover changes across distinct insular regions, namely the Reykjanes Peninsula, through the implementation of Rao's Q Index–based change-detection approaches.”

This revision eliminates the redundancy and ensures that the objective is stated clearly and concisely.

2

A 9x9 window size is mentioned but not justified through sensitivity analysis. A brief note on the impact of alternative window sizes on boundary determination is expected.

The triangle method is used without justification for its preference over Otsu's threshold method, particularly given that change detection difference maps may exhibit a bimodal distribution.

Validation points were verified by visual interpretation by the authors themselves, using LMI polygons as the only independent reference. This lack of fully independent ground truth should be explicitly acknowledged as a limitation.

We thank the reviewer for this observation, which prompted us to revisit and clarify this methodological detail. 

Regarding the window size, we acknowledge that the 9×9 moving window was not explicitly justified in the original manuscript. This parameter was adopted following established practice in spectral diversity studies using Rao's Q Index (Rocchini et al., 2017), representing a balance between statistical stability of the local estimator and spatial precision of boundary delineation. We recognize that a formal sensitivity analysis across alternative window sizes would strengthen the methodological robustness, and this is now identified as a priority for future work. Additionally, as noted in response to a related comment, the platform has been updated to allow users to customize the window size.

Upon careful revision, we recognize that the original manuscript description was imprecise: while the platform developed in this study offers three thresholding approaches, Otsu's method, the triangle (secant) method, and the 95th percentile quantile, the results presented throughout the manuscript were obtained exclusively using Otsu's method (Otsu, 1979). 

The Methods section has been revised to reflect this explicitly, and to justify the choice of Otsu's method for the present analysis: given that the difference maps compare snow-covered pre-event surfaces with lava-covered post-event areas, a bimodal distribution of pixel values is expected, precisely the condition for which Otsu's method is theoretically well-suited. 

The availability of alternative thresholding methods within the platform is now mentioned in the Methods section as a feature intended to accommodate different event types and land cover conditions. 

All reported accuracy metrics remain valid and unchanged, as they correspond to the Otsu-derived binary maps throughout.

Regarding the validation points, we would like to respectfully clarify that the validation procedure did not rely on visual interpretation by the authors.

The random sampling scheme was adopted, with 50 points generated inside the official lava flow polygons and validated using the GIS information provided by the Icelandic Institute of Natural History (IINH) and 50 points generated outside these polygons.

Original text:

“A stratified random sampling scheme was adopted, comprising 100 points generated in QGIS 3.10, equally distributed between the two classes: 50 points assigned to the "changed" category and 50 to the "no change" category. The validation was performed using the official lava flow data provided by the National Land Survey of Iceland [63], ensuring an independent reference for the accuracy assessment.”

This ensured an equal distribution between the "changed" and "no change" classes and provided an independent spatial reference for the accuracy assessment, rather than a subjective visual interpretation.

2

The near-random performance of Band 12 in isolation (OA = 46%) is explained by sensitivity to snowpack moisture, but the rationale for its inclusion as a separate configuration—given that snow dominance was a predictable condition on both survey dates—is not addressed. The authors should either justify this choice or reformulate it as an intentional negative control.

The multivariate RaoMD result (OA = 83.0%, PA = 84.0%, UA = 82.4%) is a key finding and should be more explicitly recommended as the default configuration for operational emergency mapping.

Claims regarding the generalizability of the results to forest fire, flood, and coastal erosion contexts are repeated repeatedly, but are based on a single case studied under favorable, spectrally homogeneous conditions. These claims should be appropriately qualified.

We thank the reviewer for this suggestion and agree that this finding deserved greater emphasis. 

The multivariate Rao's Q configuration (B11+B12) was designed precisely to combine the thermal sensitivity of B11 with the structural information of B12, mirroring the spectral logic of the NHI index used as the primary thermal indicator in this study. The balanced accuracy profile obtained (OA = 83.0%, PA = 84.0%, UA = 82.4%), the only configuration achieving simultaneously PA > 80% and UA > 80%, reflects this complementarity and supports its recommendation as the default configuration for operational lava flow mapping. The manuscript has been revised accordingly, while retaining the note that the univariate B11 configuration remains preferable in contexts where minimizing false alarms is the dominant operational criterion.

The inclusion of Band 12 as a standalone configuration was intentional: given that the NHI index, the primary thermal indicator used in this study, is defined as a ratio involving both Band 11 and Band 12, evaluating each band individually alongside the multivariate combination was a natural diagnostic step to assess their respective contributions to the detected signal. The near-random performance of Band 12 in isolation (OA = 46%, PA = 16%) provides empirical evidence of its limited discriminative capacity under snow-dominated conditions and contextualizes the added value of Band 11 and the multivariate approach. The manuscript has been revised to make this rationale explicit.

Regarding the generalizability, the following text was added:

Revised (lines 569-575):

"It should be noted that the results presented here derive from a single case study conducted under relatively favorable and spectrally homogeneous conditions. Therefore, the applicability of the proposed approach to other environmental change contexts, such as forest fires, floods, or coastal erosion, should be regarded as a promising research direction rather than a demonstrated outcome. Future work involving diverse case studies and more spectrally heterogeneous environments will be essential to confirm the generalizability of the methodology."

2

The Sentinel-1 SAR integration should be elevated from a brief mention to a stated research priority, given its immediate relevance for cloud and smoke scenarios—precisely the conditions under which optical approaches fail. 

The authors should also clarify whether future versions of the platform plan to provide users with the ability to select the window size and threshold method.

We thank the reviewer for this valuable suggestion.

Regarding the SAR integration, the following text was added:

Revised text (lines 576-584):

"Future developments of the platform will prioritize the integration of Sentinel-1 SAR data, given its immediate relevance for scenarios in which optical sensors are compromised by cloud cover or volcanic smoke. Nevertheless, it is important to note that, during the course of an eruption, SAR data do not provide direct information on the location of thermally active areas, such as active lava flows or eruptive vents. Its contribution is therefore largely complementary, serving primarily as an indicator of land cover changes rather than as a tool for identifying ongoing volcanic activity. For this reason, the combined use of SAR and optical data is envisioned as the most effective strategy for ensuring continuous monitoring under adverse atmospheric conditions."

Regarding the window size and threshold method, we would like to clarify that the platform already provides users with the ability to select both the window size and the threshold method through its interface. 

This functionality is available in the current version and allows users to adjust these parameters according to the characteristics of each study area.