Revalidation Technique on Landslide Susceptibility Modelling: An Approach to Local Level Disaster Risk Management in Kuala Lumpur, Malaysia

Authors’ Response

Reviewer 1

1. Please provide the compass icon in Figure 1

Figure 1 has been modified to show the compass icon in page 4, section “2.1 Study Area’.

2. Please discuss the result of Model 1, Model A, and Model B with the map of population density in Kuala Lumpur, Malaysia, so that we can get the urgency of the landslide susceptibility for land use planning and development control in Kuala Lumpur, Malaysia.

The authors would like to thank the reviewer on the helpful suggestion. We agree that it is important to compare the models with the population density to show the urgency for land use planning and development control. The available population density map is limited to the district level or parliamentary zone where the data is obtained from the 2020 census. The findings show that the population density map of district level produced using the 2020 census data indicates the highest population per km² are distributed at the northwest region found in Wangsa Maju followed by the second highest class represented by Setiawangsa, Batu and Seputeh in southwest region (Figure 6). With reference to the landslide susceptibility model, we recognize that part of the hilly areas in Wangsa Maju, Setiawangsa and Seputeh located in high to very high susceptibility class, requiring urgent effort in hazard mitigation, disaster preparedness and stringent actions for development. 

This has been added to page 13 in section “4.Discussion”,  paragraph 5 in lines 457 to 463.

3. Model 1 and Model A have the SRC values > 0.90 and the same PRC values. In your opinion, what are the possible factors that generate this phenomenon? Please discuss your answer in your manuscript.

The similar SRC values of Model 1 and Model A indicates the goodness-of-fit is excellent and able to categorize historical landslides within the highest susceptibility classes. The difference of 0.01 may not be statistically significant. However, one of the possible factors is the higher distribution of landslides within zone A subset (58%), generated a model with the closest SRC value to that of Model 1. In the case of same PRC values, the authors suggest that the small sample size of the 2021 landslide events and similar landslide distribution in both models could explain the finding.

The explanation has been added to page 11 in section “3.3 Model Evaluation and Validation”, paragraph 1 in lines 378 to 382, and paragraph 3 in lines 392 to 394.

Reviewer comment

Author’s Response

Reviewer 2

(1) According to the seven main influencing factors described in the paper, the analytic hierarchy process (AHP) of geological hazard evaluation is also a common method. It is suggested that the author use the analytic hierarchy process (AHP) to evaluate and compare with this method.

The authors would like to thank the reviewer for the valuable suggestion.

Employing the AHP method for the study now would defeat the purpose of highlighting the retrospective validation method as this new model will post-date landslide events. In addition, the AHP method requires substantial technical work and it is not possible to conduct this within 7 days. In this study we have incorporated input from expert judgement in the selection of parameters before conducting bivariate statistical analysis. This is to overcome limitations in the AHP approach with respect to variability of expert knowledge and insufficient information on the mechanism of past landslides.

We have incorporated this suggestion for future work, as part of the effort to compare models using other methods. This has been added to page 14, in section “4.Discussion”,  paragraph 6 in lines 482 to 485.

(2) The relationship between the seven selected elements and the statistical model established by the author is not clear;

The selection of the seven parameters were based on high local representation. The association of the parameters in model analysis is as follows (please refer to Supplementary Table S1 (a) for further details):

Slope gradient:  Slope gradient is usually regarded as the primary causative factor for the onset of slope failure associated to shear strength of the material on slope (Dai & Lee, 2002).  The analysis of the parameter indicates that the landslide density and weightage values increase with increasing slope gradient.

Surface geology: Most of the ground is covered by surface geology and shows a strong influence towards landslide initiation compared to bedrock geology, caused by the deep weathering profile which generates thick residual soil over most area (Paramananthan et al., 2021). The residual soil of Kenny Hill formation holds the highest weightage of 1.00 with 70% of total landslide, due to the interlayering of metasedimentary rocks of variable strength and properties while the granite represents 127 (20%) landslides with a weightage value of -0.18.

Elevation:  Elevation of the slope is not a conditioning factor by itself but affects the overall surface of the terrain and topographic features which controls the vegetation distribution (Saadatkhah et al., 2015c). The highest weightage is represented by the 100 - 150 m class with a value of 0.55 and has 10% of the total landslides while 50 -100 m class represents 68.8% of the total landslides with a lower weightage of 0.49 due to the larger area.

Distance to lineament: The stability of the surrounding area is reduced by induced regional perturbations in the fracturing occurrence and enhanced weathering in the rocks (Yusof & Pradhan, 2014). The distance to lineament factor and landslide shows direct correlation with the highest weightage (0.93) is within 0 - 250 m proximity.

Distance to road: Slope which has closer proximity to roads especially near cut and fill slope could affect the stability by increase of stress in its base and accumulation of water from nearby slope (Yalcin,2008). The highest weightage of 0.18 produced from the 25-50 m class interval representing 25% of total landslides while the majority of landslides (47%) occurs within 0-25 m distance from a roadway yields a low weightage of 0.02 due to larger class area.

TPI:  TPI gives an indication to where the landslide point is located with reference to the topographical position or sometimes expressed as landscape position (Jebur et al., 2014). The highest percentage and weightage of landslides is within the middle slope with a value of 61.17% and 1.13 respectively with the upper slope class having a weightage of -0.01 with 15.08% of total landslides.

Surface roughness: Surface roughness generally correlates to excavation, surface erosion and the density of vegetation cover and could signify the type of land cover of that particular area and has been classified into high, moderate and low values (Yusof & Pradhan, 2014).  Half of the landslides (53 %) are categorized as having moderate surface roughness with a weightage of 0.07 however the highest weightage of 0.29 represented by slope with high roughness due to smaller areal class.

This has been added to page 5-6, in section “2.3 Landslide Conditioning Factors”, paragraph 4, lines 199 to 231.

(3) Lines No. 267, why 1 km is used as the standard, and what is the basis?

The smallest geological unit in Kuala Lumpur measures approximately 1 km in width, thus assigning the spatial partitioning into 1 km wide zone is able to represent the feature in the assessment.

This has been added to page 9, in section “2.5 Model Evaluation and Validation”, paragraph 3, lines 301 to 303.

(4) In addition to statistical methods, it is suggested that the author further consider the mechanism of surface landslide for analysis.

The authors would like to thank the reviewer on this insightful input.

The mechanism of landslide is not available in the landslide inventory provided by JMG. However, the authors observed that most landslides in the study area are shallow rotational landslide, which occur in the Kuala Lumpur granite, and rotational to translational landslide observed in the Kenny Hill Formation. They commonly occur together and categorized by Varnes (1978) as slides in the same movement type.

Since the mechanism is not available in the inventory and only two main surface landslide mechanisms were observed, we chose not to differentiate them for statistical analysis.

This has been added to page 10 in section “3.1 Landslide Inventories”, paragraph 1, lines 324 to 327.