| 12 Sep 2025
Geospatial Analysis of Fault–Epicenter Dynamics in Bangladesh and Adjacent Regions Using Remote Sensing and Statistical Modeling
Abstract. Bangladesh and its adjacent regions are situated at the junction of several tectonic plates and are hence highly susceptible to earthquakes. This study investigates the spatial dynamics between fault lines, their classifications, and earthquake epicenters in Bangladesh and its neighboring countries. With a high-density population, absence of urban planning, and inter-border seismic hazards, identifying the way fault types interact with seismic activity is crucial for an effective estimation of the hazard in this area.
Landsat 8 Thermal Infrared Sensor (Band 10) satellite imagery was used for the detection of faults, followed by the extraction of lineaments using PCI Geomatica and spatial analysis in ArcMap 10.8. Fault lines were identified as four principal types: normal, reverse, left-lateral, and right-lateral, based on geometric and spatial features. Earthquake epicenter data between 1924 and 2024 were derived from the USGS Earthquake Catalog. Spatial autocorrelation analysis (Moran's I), Kruskal-Wallis test, Dunn's test, and multinomial logistic regression were used to examine fault-epicenter relationships.
Approximately 40,000 fault lineaments were identified. Moran's I index (0.298, p<0.000001) confirmed significant spatial clustering of epicenters and fault lines. Dunn's test demonstrated that reverse faults significantly differ from the others in terms of proximity to epicenters. Multinomial logistic regression revealed that earthquakes tend to be closer to normal (p = 0.042) and left-lateral faults (p = 0.016), whereas reverse faults (p = 0.676) did not exhibit significant differentiation based on proximity.
This work highlights the crucial need to incorporate fault types and epicenter spatial relationships into seismic hazard models. The results offer practical insight into regional earthquake risk mitigation, infrastructure design, and transboundary disaster preparedness in Bangladesh and adjacent regions.
This study investigates the spatial relationships between fault and earthquake locations in Bangladesh. First, faults are mapped using satellite imagery and an automated workflow that extracts lineaments. Earthquake locations are taken from the USGS catalog. Statistical analysis are then performed, which indicate that earthquake locations in Bangladesh are not randomly distributed but tend to cluster near faults.
The statistical analysis is impressive, and the study is well written. However, as outlined below, I have significant reservations about the underlying data used in this analysis. In particular, the robustness of fault mapping, and whether their automated workflow is reliably identifying active faults and correctly classifying their kinematics. In addition, the high (5-10 km) earthquake location uncertainties in Bangladesh severely limit the spatial analysis, as does the fact that it is performed in 2D (i.e.., it doesn’t consider the depths at which the earthquakes occur, or the down-dip projection of faults).
I appreciate the author’s extensive work analysing fault-earthquake spatial relationships in Bangladesh. However, given the above points, my recommendation is that they instead just focus on using the satellite imagery to develop a robust active fault map for Bangladesh. I have provided some ideas for how this could be achieved below. As the authors correctly point out, Bangladesh is highly vulnerable to future earthquakes, and I think a study that provides new insights into the distribution of its active faults would form an important and interesting article in Solid Earth.
Major comments
1.) Fault mapping: section 2.3 describes how faults in Bangladesh were mapped from lineaments identified in satellite imagery using PCI Geomatica’s lineament extraction software. Given the implementation of an automated workflow to map faults, I strongly recommend some testing for whether the extracted lineaments do represent faults and/or more documentation is provided for why these lineaments are interpreted as active faults (see for example Scott et al 2025). For example, are the lineaments consistent with faults identified in geologic maps, hillshade rendering of digital elevation models, and other active fault compilations in Bangladesh (e.g., Hossain et al 2020, Styron and Pagani 2020)? Is it possible that some of the lineaments are other geomorphic or structural features (e.g., joints?) The very high number of short (<10 km) diffuse faults in Figures 3 and 4 is unlike most other fault compilations in that: (1) the distribution of fault lengths tends to follow a power-law distribution with an exponent of 2 (Zou and Fialko 2024), and (2) most active fault traces (or more specifically, earthquake surface ruptures) are >5 km long (Christophersen et al 2015). Some of these comments are addressed in the discussion (Lines 276-284), but additional work is needed to demonstrate that these lineaments represent active faults.
Secondly, this compilation is specifically for active fault traces (e.g., Line 234). Hence, it is necessary that this study provides details on how active and inactive faults are distinguished. This is important as there is no universal definition for what constitutes an ‘active’ fault (see for example, Styron and Pagani 2020, Williams et al 2022), and without these details, I cannot be confident that inactive faults are being excluded in the statistical analysis of earthquake-fault relationships.
2.) Fault Classification: Following the identification of fault lineaments, a slip type is assigned to a fault to based ‘on the assessment of its length, curvature, intersection patterns, clustering, and orientation of the faults (Lines 135-137).’ This is highly unusual as fault kinematics should instead be defined by offset markers (e.g., geologic units, geomorphic features). Is there any indication from these features for what the kinematics of Bangladesh’s faults are?
Alternatively, it’s noted that fault orientations are used to infer kinematics, and this is defensible For example, if it was performed by applying the Andersonian theory of faulting to Bangladesh’s regional stress state and/or comparison to earthquake focal mechanisms in this region? (https://www.globalcmt.org/CMTsearch.html). In this context, it is worrying that Figure 4 indicates that there are reverse and normal faults adjacent to each other and left lateral faults that strike at 90º to each other, and the same for right-lateral faults. In addition, there is an along-strike sharp transition from right- to left-lateral faulting around Rangpur without any change in fault orientation. This implies very small-scale stress rotations. Do the authors think these are realistic?
3.) Earthquake data: The analysis of earthquake locations in this study was conducted using M>3 events between 1924-2024 in the USGS earthquake catalog (Section 2.5). However, it should be noted that due to uncertainties in picking earthquake arrivals, sparse station spacing, and seismic velocity models, the earthquake locations in this catalog have a horizontal and depth location uncertainty of 5-10 km (and that’s only for events after 2014 when these uncertainties are reported).
Ideally, statistical analyses between earthquake locations and faults should be performed using high resolution earthquake catalogs (e.g. Hauksson et al 2012). If this is not possible for the Bangladesh earthquake catalogs, then I recommend a sensitivity analysis for whether the earthquake location uncertainties influence the earthquake-fault spatial analysis in Section 2.6 (notwithstanding my next comments below). For example, by repeating this analysis with randomly perturbed earthquake locations.
4.) Earthquake-fault relationships: The statistical tests described in Section 2.6 between fault lines and earthquake epicentres is essentially a 2D analysis. It therefore neglects that earthquakes occur within 3D space and that faults are 2D planes that project down-dip through the crust. In other words, the Euclidean relationship between earthquake locations and faults should be considered in 3D (except in case of vertically dipping faults where this simplification is acceptable).
Minor comments
Lines 40-42: Suggest removing the reference that earthquakes “claimed an average of over 25,000 years lives annually” as this implies that there is some regularity to the number of earthquake casualties each year when this was not the case. Or in other words, most earthquake casualties in the 20th century can be attributed to a few destructive events, and so in most years there are relatively few earthquake casualties (see for example Figure 1 in Holzer and Savage 2013).
Lines 65-68: Although fault damage zones undoubtedly influence earthquake rupture and seismic hazard (see also Biegel and Sammis 2004), I suggest remove these sentences from the introduction, as fault damage zones and their influence on earthquake hazards in Bangladesh are not discussed elsewhere in the manuscript.
Line 75: What does TEC mean?
Figure 1: It would he helpful if tectonic plate boundaries around Bangladesh were depicted on this map
Line 117: Although a 30 m resolution image is acceptable for remotely mapping faults with prominent scarps (heights >5 m, see for example Hodge et al 2019), inevitably, some smaller faults would be missing from this mapping. I therefore recommend adding some commentary on how this bias would influence the completeness of their fault mapping.
Line 125: I recommend adding some references here for examples of where PCI’s Geomatica software lineament extraction analysis has been used to map geological features.
Lines 279-281: It would be helpful to elaborate here with examples of where lineaments are associated with earthquake clusters and/or lineaments are favourably oriented for reactivation in this region’s stress state. For example, what are the principal stress orientations in this region?
Lines 348-352: I disagree with the assertion that the correlation between earthquake locations and faults can be used to constrain a fault’s seismic hazard. Ultimately, the key factor that determines a fault’s earthquake potential is its slip rate and area, as these in turn, influence its seismic moment rate. Notably, there are several examples of ‘locked’ faults which have very little seismicity associated with them, but are inferred from geologic studies to have high slip rates and seismic hazard (e.g. New Zealand’s Alpine Fault, Norris and Toy 2014).Furthermore, it is expected that a considerable amount of seismicity occurs away from mapped faults (Zou and Fialko 2024).
References