Early Warning System Against Rainfall-Induced Landslide in Sri Lanka

Abstract

Based on solid evidence, scientists attribute the global warming trend observed since the mid-twentieth century to the human expansion of the “greenhouse effect.” Extreme rainfall events have become more frequent worldwide, resulting in hydro-meteorological hazards creating more deaths and devastation. One of the most remarkable disasters of rain-induced rapid long-traveling landslides (RRLL) in Sri Lanka took place at Aranayake, 70 km east of Colombo, in 2016 (JICA Survey Team (2016), Survey results of Aranayake Disaster, JICA. URL: https://www.jica.go.jp/srilanka/english/office/topics/c8h0vm00006ufwhl-att/160720.pdf [Last accessed: April 14, 2020]). The fluidized landslide mass ran over an about 2 km distance claiming the lives of 125 people. This tragic event highlighted the importance of reliable early warning and disaster management mechanisms even more than ever because the presence of these hidden unstable soil masses, as well as their run-out distances, are hardly predicted. Once they start sliding, it is almost impossible to stop them. Since 2020 (after the preceding period of 2019), both the National Building Research Organization, Sri Lanka (NBRO) and the International Consortium on Landslides (ICL) have jointly started a new 5-year research project, “Development of early warning technology of Rain-induced Rapid and Long-travelling Landslides (Project RRLL),” within the framework of SATREPS. SATREPS, standing for “Science and Technology Research Partnership for Sustainable Development,” is a Japanese government program promoting international joint research. This article reports on the outline of the project, including its background, goals, plans of plots for developing critical technologies for the early warning system, etc.

1 Introduction

As an isolated island within the Northern Indian Ocean, Sri Lanka experiences extreme rainfalls in two monsoon seasons in response to the bi-annual reversing monsoonal winds (Department of Meteorology, Sri Lanka 2020). Particularly the south-western monsoon from May to September brings rain to the southwest mountainous area of Sri Lanka. Out of 25 administrative districts in Sri Lanka, ten districts, approximately 30% of the total land area of the Island, are the most prone to landslides. Landslides in Sri Lanka, mainly due to natural causative factors, have long been remote and isolated events. However, these landslide-prone areas have become the primary area of tea and cinnamon plantations; thus, about 35% of the population of Sri Lanka live in these areas nowadays.

Moreover, post-civil-war Sri Lanka has been attracting tourists with 5 of 7 UNESCO cultural and natural world heritages of Sri Lanka located in landslide-prone areas. Studies have revealed that human-induced interventions such as rapid urbanization, population growth toward the foot of mountains, inappropriate land management, deforestation in steep slopes, etc., have influenced nearly 70% of the landslides in Sri Lanka. One of the most remarkable disasters to mention took place at Aranayake in Kegalle District, 70 km east of Colombo (JICA Survey Team 2016). The fluidized landslide mass ran over a 2 km distance killing 125 people. These hidden unstable soil masses are difficult to find in advance, and their long run-out distances are hard to predict. Once they start sliding, it is almost impossible to stop them. Thus, this tragic event at Aranayake highlights the importance of sophisticated early warning and efficient disaster management protocols.

Since 2020 (after the preceding period of 2019), both the National Building Research Organization, Sri Lanka (NBRO hereafter) and the International Consortium on Landslides (ICL hereafter) have jointly started a new 5-years project for Sri Lanka titled “Development of early warning technology of Rain-induced Rapid and Long-travelling Landslides (Project RRLL hereafter).” The project is within the framework of SATREPS. SATREPS, “Science and Technology Research Partnership for Sustainable Development,” is a Japanese government program with two funding organizations, Japan International Cooperation Agency (JICA hereafter) and the Japan Science and Technology Agency (JST hereafter), that promote international joint research. JICA has proactively contributed to building a rational landslide disaster mitigation framework for Sri Lanka. This project aiming to develop cutting-edge technologies for one-day-in-advance forecasts of the occurrence of RRLLs is indeed complementary to all efforts that JICA has been deploying in this country. This article outlines the background story that led up to the project, the project’s framework, and some noteworthy developments in the technology for implementing the early warning protocols.

2 Background Story

2.1 Recent Landslides in Sri Lanka

The technologies to stabilize reactivated and creeping landslide masses have progressed much because we can identify their locations. However, the number of recent tragic events in Sri Lanka shows a soaring trend in general, with some remarkable spikes in 2003, 2007, 2011, 2014, 2016, and 2017, as shown in Figs. 1 and 2. These events highlight the difficulty in coping with the devastations caused by RRLLs. We can identify neither their locations nor early signs of movement in advance. Therefore, implementing advanced and feasible technologies for early warnings of RRLLs is crucial.

Fig. 1

Number of landslides and deaths from 2003 to 2021 (Credit NBRO)

Fig. 2

Locations of landslides from 2003 to 2017 (Credit NBRO)

2.2 EWS in Sri Lanka

Landslide Early Warning System (EWS) plays a significant role as a non-structural risk reduction method on rainfall-induced landslides. The National Building Research Organization (NBRO) is the authorized body for landslide Early Warning (EW) in Sri Lanka. It has provided two kinds of EWS since 2007: the first one is the Regional Landslide Early Warning (RLEW) based on the rainfall thresholds (NBRO 2022a). The other is the Site-Specific Landslide Early Warning based on sophisticated measurement instruments or community-based landslide warning (NBRO 2022a).

The first EWS (RLEW) started in 2007. The warning on a regional scale goes off based on predefined rainfall threshold limits under three levels, i.e., watch, alert, and evacuation respectively at 75 mm/day, 100 mm/day, and 150 mm/day or 75 mm/h. The NBRO uses an automated network of 325 rain gauges (Fig. 3) established in the landslide-prone areas of the country to obtain real-time rainfall data, based on which the early warnings are issued. The messages were notified directly to local people through not only the Web of NBRO, but also radio communication, television, telephone, and bells on sites from local agencies, which have the responsibilities for the safety of local people. But some problems have been emerging through the implementation process of the system. Warning messages do not always efficiently reach local people, and local people often ignore the message even when they have adequately spread to the people. Moreover, there have been many false-negative and false-positive predictions of landslide occurrences, making people think that the EW messages are just like the tale of “The Boy Who Cried, Wolf.”

Fig. 3

Distribution of automatic rain gauges in Sri Lanka (Credit NBRO)

NBRO, given official development assistance from JICA, started implementing the second method, site-specific EWS and the Community-Based Landslide Early Warning (CBLEW) approach for the landslide vulnerable communities identified in the country from 2016 to solve the problems mentioned above. Site-specific information is necessary for a more reliable warning for a specific slope. It was thus a practical solution to save more lives in line with the four priorities of the Sendai Framework for Disaster Risk Reduction 2015–2030.

The CBLEW approach aims to empower the communities and establish systematic preparedness plans for timely self-evacuation of communities in case of a landslide event. In this course, the communities get education on preparedness and self-decision based on the rainfall data obtained by manual rain gauges distributed in the village. The CBLEW approach consists of the following:

• (1) Site selection by NBRO based on its preliminary landslide investigation reports and landslide hazard zonation maps (NBRO 2022b),

• (2) Awareness programs for local government officers to familiarize them with the project, and

• (3) Awareness programs for local communities to give knowledge about the following:

  • What a landslide is,

  • What pre-symptoms people can observe,

  • Risk reduction techniques, and

  • How the community people can create a community risk map (Fig. 4, next page), install manual rain gauges (Fig. 5, next page) and extensometers (Fig. 6, next page), and sirens in the community.

    Fig. 4

    

    Example of community hazard map (Credit NBRO)

    Fig. 5

    

    Hand-made rain gauge (Credit NBRO)

    Fig. 6

    

    Handmade extensometer (Credit NBRO)

However, fatal landslides/slope failures that continue happening in the country remind us of the current need for a more sophisticated early warning system. At first, developing novel approaches such as precipitation nowcast and forecast techniques is mandatory for improving landslide early warnings. Behind this, we have the drastic change in monsoon and cyclone patterns in the Indian Ocean, responsible for unpredictable extreme events such as intense and localized rain impacts. Secondly, we need to examine the zoning for the EWS, which currently follows the administrative zoning. It is perhaps more rational to issue a warning considering the catchment boundaries. Thirdly, site-specific EWS will be more accurate given more reliable geotechnical parameters.

As mentioned above, RRLLs are the most troublesome among all types of landslides because the movements of soil masses are associated with liquefactions of whole masses and along sliding surfaces. Their motions are swift and extremely dangerous. Therefore, a pressing need is to develop an effective early-warning system that predicts RRLL occurrences one day in advance. Advancements in technology for landslide disaster mitigation are remarkable in the international world of the academy, with the ICL as the core organization. They include forecasting localized precipitation events, early detection of ground movements, and relaying timely early warning to the last mile, namely, residents at landslide risk. Among them, the critical technologies for the development of RRLL EWS are:

• (1) Time prediction of heavy rainfalls and pore water pressure build-ups,

• (2) Site prediction of landslide initiations and motions, and

• (3) Effective risk communication and public education.

In light of this situation, we have started a new 5-year project between ICL on the Japanese side and NBRO on the Sri Lankan side in 2020 (Japan Science and Technology Agency 2020). The project is entitled “Development of Early Warning Technology of Rain-Induced Rapid and Long-Travelling Landslides,” in short, “Project RRLL.” As said, this 5-year project, which officially started in 2020, is one of the SATREPS projects.

3 Project RRLL

3.1 Planning and Kickoff

The “Sendai Framework for Disaster Risk Reduction 2015–2030” is an international document adopted by the United Nations member states between 14 and 18 March 2015 at the World Conference on Disaster Risk Reduction held in Sendai, Japan. This document outlines seven clear targets and four priorities for action to prevent new and reduce existing disaster risks (UNISDR Prevention Web 2015):

• (1) Understanding disaster risk,

• (2) Strengthening disaster risk governance to manage disaster risk,

• (3) Investing in disaster reduction for resilience, and

• (4) Enhancing disaster preparedness for effective response and “Build Back Better” in recovery, rehabilitation, and reconstruction.

Sri Lanka has made a solid commitment to implementing the Sendai Framework for Disaster Risk Reduction 2015–2030. The commitment includes:

• (1) Establishing the Roadmap for disaster risk reduction “Safe and Resilient Sri Lanka” developed by JICA (UDR Prevention Web 2005) for the Ministry of Disaster Management of Sri Lanka, and

• (2) Joining the “ISDR-ICL Sendai Partnerships 2015–2025 for global promotion of understanding and reducing landslide disaster risk” (Sassa 2015).

The Partnership was officially adopted at the ICL-IPL Sendai Partnership Conference on 11–15 March 2015 (Fig. 7). Both NBRO and Central Engineering Consultancy Bureau (CECB), who joined the conference and signed the Sendai Partnership, envisaged together with relevant signatories the first plan for this SATREPS project. After thorough discussions, it passed the final selection round on May 16, 2019, that the new 5-year SATREPS project for Sri Lanka concerning “Development of early warning technology of Rain-induced Rapid and Long-travelling Landslides (Project RRLL)” would start in 2020.

Fig. 7

Signing ceremony of ISDR-ICL Sendai Partnerships 2015–2025 (Sassa 2015)

A signing ceremony for the Minutes of Meeting (MM) between NBRO, Sri Lanka, and JICA, Japan, was held on October 15, 2019, at the auditorium of NBRO (Fig. 8). Mr. Satoshi Nakamura, Leader, Detailed Planning Survey Team, JICA, Japan, and Eng. (Dr.) Asiri Karunawardena, Director General, NBRO, Sri Lanka, signed the MM toward implementing Project RRLL. In the same signing ceremony, Collaborative Research Agreement (CRA) between NBRO, Sri Lanka, and ICL was also signed by Eng. (Dr.) Asiri Karunawardena, Director General, NBRO, Prof. Kazuo Konagai, Leader on the Japanese side of Project RRLL, Principal Researcher, Prof. Kyoji Sassa, Secretary-General, and Prof. Kaoru Takara, Executive Director of ICL.

Fig. 8

Signing ceremony on Oct. 15, 2019 for the minutes of meeting (MM) between NBRO, Sri Lanka and JICA, Japan

The following officers signed Record of Discussions of the Project RRLL: Mr. Fusato Tanaka, Chief Representative, JICA Sri Lanka Office, Eng. (Dr.) Asiri Karunawardena, Director General, NBRO, Major General (Retired), Kamal Gunaratne, Secretary, Ministry of Defence and Mr. Ajith Abeysekera, Director General, Department of External Resources, Ministry of Finance. This Record of Discussion is an official agreement between both governments to confirm the implementation of the 5 year Project RRLL starting on February 5, 2020.

Technical Cooperation Agreement for pursuing the Project RRLL over the five years from March 1, 2020, to February 28, 2020, was implemented between JICA and ICL.

JICA and ICL signed the contract for the first year of Project RRLL with the consent of both parties. The first year started on March 1, 2020, and ended on May 31, 2021.

3.2 Implementation Structure

Figure 9 shows the implementation structure of Project RRLL. The project is carried out exclusively by members of the Joint Coordination Committee (JCC). Simultaneously, it is also open through the “Landslide Technical Forum” to anyone interested in it. The JCC members are signatory entities of the Record of Discussion that JICA and the implementing agencies on the Sri Lankan side signed.

Fig. 9

Implementation structure for project RRLL (as of November 2021)

On the Sri Lankan side, we have:

• (1) NBRO as the primary implementing agency and thus the Secretariat of JCC,

• (2) Disaster Management Center (DMC),

• (3) Department of Meteorology (DOM), and

• (4) Department of Irrigation (DOI).

(1) NBRO, (2) DMC, and (3) DOM are currently under the purview of the Ministry of Internal Security, Home Affairs, and Disaster Management, while (4) DOI is a department of the Ministry of Irrigation. Just when the project officially started, the COVID-19 pandemic started spreading worldwide. To further promote our activities under this harsh condition, we got new JCC members since the 1st JCC meeting held in April 2021. They are:

• (5) Central Engineering Consultancy Bureau (CECB) and

• (6) Two individual professors from three major universities in Sri Lanka.

On the Japanese side, which the Japan Science and Technology Agency (JST) financially supports, we have:

• (7) The headquarters of ICL as the entity leading the project,

• (8) Tokyo Institute of Technology (TIT),

• (9) Disaster Prevention Research Institute (DPRI) of Kyoto University,

• (10) Forestry and Forest Product Research Institute (FFPRI), and

• (11) Kochi University.

In addition to these entities and organizations, we have individual researchers joining the Project as official members from external entities.

4 Pilot Study Sites

Geologically, the island of Sri Lanka is an extension of Peninsular India and forms part of the Indian Shield, one of the oldest and most stable parts of the earth’s crust. Previous studies have suggested that the larger part of the landslide-prone areas is covered with thick weathered gneiss metamorphosed during Precambrian Era. The tropical climate has favored deep weathering of these metamorphic rocks reaching tens of meters in these mountains with dense tropical vegetation drapes. Two pilot study sites, Aranayake and Athwelthota, are selected as representatives of two major types of RRLL (Fig. 10).

Fig. 10

Pilot study sites: (Left) Alanayake landslide in 2016, and (Right) Athwelthota landslide in 2017 (credit NBRO)

4.1 Aranayake Landslide Area

Aranayake landslide was triggered on May 17, 2016, by exceptional rainfall associated with a slow-moving tropical cyclone. The fluidized landslide mass from the relative elevation of about 600 m ran over an approximately 2 km distance claiming the lives of 125 people. This landslide is unique because it is much more prominent in size and runout distance than the others. Though this type of landslide rarely occurs, a large RRLL can cause a big disaster. This landslide mass ran across two local communities, Elagipitiya and Debathgama Pallebage, with about 1500 and 1100 inhabitants, respectively. Summing up populations of the neighboring communities with similar risks of this type of RRLL expected, the number of beneficiaries of this project will be at least several thousand.

4.2 Athwelthota Landslide Area

This landslide, which occurred in the Athwelthota area, Baduraliya District, on May 26, 2017, destroyed nine houses, killed nine people, and stopped traffic on a national highway. Each landslide of the Athwelthota type will not cause a surprisingly large disaster, but the number of landslides of this type can be huge, causing extensive losses of human lives and properties.

During the heavy rain of 2017, 37 RRLLs reportedly took place all at once, claiming the lives of 262 people. There remains unstable soil masses above the two pilot study sites perching in and around the tops of the exposed bare-earth slopes. We will install necessary equipment on/in these soil masses to measure the causal factors of landslides and creeping deformations. These pieces of equipment include pore pressure sensors, inclinometers, borehole extensometers, etc. We also monitor the movements of these soil masses with Persistent Scatterer Interferometric Synthetic Aperture Radar (PSInSAR). PSInSAR is a remote sensing technique that uses radar signals from a satellite to measure the ground displacement accurately. The method thus allows us to track the motion of each scatterer structure placed on the ground. These measurements will help develop infiltration models for the weathered gneiss in these areas.

5 Preliminary Study of Aranayake RRLL

Natural disasters such as landslides can repeatedly occur in and around the same area, and we often find traces of past tragic events remaining on the terrain. We can also see early signs of future events in the topographical features. Figure 11 shows polygons with possible landslide blocks and rock creep areas image-interpreted from the digital elevation model of Aranayake, the 2016 RRLL (Polygon A) disaster-hit area. In this figure, a clear depression B does not take the appearance of a river valley (such as E) formed by vertical erosional downcutting. This depression might be the source area of a large past RRLL, and the debris mass from this area B might have deposited on area C. Meanwhile, an unstable soil mass (D) remains immediately behind the scar of the bare earth exposed in the 2016 RRLL. Although an RRLL has not occurred yet, the creeping movement may help debris gradually accumulate to reach its stability limit. Given the image interpretation mentioned above and assuming that the inherent soil parameters differ little from those at the 2016 Aranayake landslide zone, LS-RAPID simulations were conducted to examine the following:

Fig. 11

Possible landslide blocks and rock creep areas image-interpreted from the digital elevation model of Aranayake (Tan et al. 2020)

• (1) If the LS-RAPID simulation would reproduce the 2016 RRLL event (A),

• (2) If the past RRLL (B) likely happened, and

• (3) How far out the future potential RRLL (D) can run out.

LS-RAPID is an integrated model that simulates landslide masses’ initiation and downward movements (Sassa et al. 2010).

We took two soil samples (S-1, S-2) for testing from the Aranayake RRLL source area as marked in Fig. 12b. These soil samples were weathered matters of granitic gneiss with grain size distribution curves shown in Fig. 12c. We then conducted a series of ring shear tests using the ICL-2 ring shear apparatus (Khang et al. 2018). These tests allow us to obtain mechanical parameters and the critical displacements, namely DU at which strength reduction starts and DL at which strength converges on its residual value. These parameters are all necessary for the simulations. Undrained tests were then performed under the monotonic stress control, with the initial normal stress σ and its increment $\Delta \sigma$ set at 500 kPa and 1 kPa/s, respectively; the initial shear stress $\tau$ and its increment $\Delta \tau$ set at 200 kPa and 1 kPa/s. The normal stress was firstly increased to the predetermined value (500 kPa) in a drained condition to avoid the generation of excess pore water pressure. The drain valve was then shut to realize the undrained condition before shearing. The applied initial normal stress was determined, taking into account the landslide mass’s average thickness (25 m). The initial shear stress was finally calculated from this normal stress and the slope angle.

Fig. 12

a Location of the landslide region b Satellite picture of the Aranayake landslide and the sampling site. c Grain size distribution of the sample (Dang et al. 2019)

When the effective stress path reaches the failure line, it decreases due to pore pressure buildup along the failure line until the mobilized shear stress reaches the steady-state value. This behavior reflects the sliding-surface liquefaction associated with grain crushing and the collapse of grain fabrics (Sassa et al. 2004, 2010, 2015). As plotted in Fig. 13c, when undrained shearing begins, the stress path in the drained equilibrium with the initial normal stress of 500 kPa (indicated in Fig. 13c) goes obliquely upward, reflecting the porewater pressure buildup. After it reaches the failure line (39°), it decreases along the failure line until it reaches the steady-state shear resistance of 30 kPa. The friction angle holds its peak value (${\varphi }_{p}$) in the pre-failure state until the shear displacement DL is reached (point of failure), then shear strength reduces as the shear displacement develops from DL to DU. DL and DU values observed in this test are marked in Fig. 13b.

Fig. 13

Result of the undrained stress control test on S-2 using ICL-2 apparatus. a Time-series curves. b Stress-displacement curves. c Stress path. Result of S-1 is not employed as increasing the pore pressure failed to trigger the landslide. The reasons are available in Dang et al. (2019)

The 2016 Aranayake RRLL was reproduced in an LS-RAPID numerical simulation to examine its initiation and motion processes. Figure 14 shows the downslope movement of the Aranayake landslide mass frame-by-frame. The landslide mass shows an early sign of being detached at the center of the source area (stage 1), at the elapsed time (from the beginning of simulation) of 40 h 33 m, and the pore pressure ratio (${r}_{u}$) increases to 0.20. The whole landslide mass then starts sliding at 66 h 06 m (r_u = 0.28) (stage 2). The simulated landslide initiation time coincides roughly with the actual time, proving the reliability of the simulation. The landslide flow then splits into two runout paths until they join each other again, where a remarkable volume of debris stagnates (stage 3). Finally, the landslide flow stops at 66 h 08 m (stage 4). Given the satisfactory agreement between the simulated and the observed processes of the Aranayake mass movement, the same geotechnical parameters were used for LS-RAPID simulations to discuss the possible neighboring soil wasting processes. Figures 15 and 16, respectively, show how the soil mass, which could have been detached in the past, moved downslope, and how the unstable soil mass, which still perches atop of the Alanayake mountain slope just behind the scar of the 2016 RRLL, can move downslope (Tan et al. 2020). For the future RRLL case, shallow and deep potential sliding surfaces were assumed to develop through the unstable soil mass. With no information about precipitations for the past and future RRLLs, the pore pressure ratio ${r}_{u}$ was assumed to increase at a steady rate of 0.01/s. These simulations show the potential distal ends when a jumble of wood and soil comes down as the leading wave of a possible debris flow, thus, bringing to light areas of special warning.

Fig. 14

Reproduction of the 2016 Aranayake RRLL on LS-RAPID. The A₁–A₂ profile is shown on the right. (Dang et al. 2019)

Fig. 15

Runout of the past possible RRLL (Tan et al. 2020)

Fig. 16

Runout of the future potential RRLL reproduced on LS-RAPID. a For a deep-seated RRLL and b for shallow-seated RRLL (Tan et al. 2020)

6 Technologies to Develop

This project has the following three groups, G1, G2, and G3:

G1 works as a hub for this joint research and integrates unique technologies developed at the two pilot sites by Groups 2 and 3. Through this activity, Sri Lankan scientists and researchers enhance their ability to cope with the RRLL disaster mitigation.

G2 develops technologies for (1) 24 h in-advance prediction of heavy rainfalls and (2) assessing groundwater pressure build-up, initiation of an RRLL, and its flowing dynamics.

G3 strengthens RRLL risk communication protocol, developing an augmented reality system for sharing predicted risk information and providing public education to develop the capacities of the communities.

As said before, essential technologies that the three groups mentioned above will develop are for:

• (1) Precise weather forecasts in mountainous regions,

• (2) Prediction of groundwater pressure build-up, identification of RRLLs locations and their moving areas, and

• (3) Effective risk communication and public education.

Details for each technology follow.

6.1 Precise Weather Forecast in Mountain Regions

The southwest region of Sri Lanka, where the south-western monsoon brings heavy rain between May to September, is our target region for the precise weather forecast. However, the technology should apply to wherever we need it. Looking ahead to worldwide applications in the future, we use MSSG as our generic platform for the precise weather forecast from this perspective. The developers of this MSSG join the research activity of G2 to help develop the rainfall prediction system suitable for Sri Lanka.

MSSG, standing for Multi-Scale Simulator for the Geo-Environment, is an atmosphere–ocean coupled non-hydrostatic model aimed at seamless simulations from global to local scales. MSSG consists of atmospheric (MSSG-A) and ocean (MSSG-O) components (e.g., Takahashi et al. 2007, 2013; Onishi and Takahashi 2012). MSSG adopts a conventional latitude–longitude grid system for regional simulations and the Yin-Yang grid system (Kageyama and Sato 2004; Baba et al. 2010, Fig. 17a), consisting of two overlapping latitude–longitude grids, thus avoiding the polar singularity problem, for global simulations.

Fig. 17

MSSG is designed to be applicable to a global scale, b meso scales and c up to urban scales. The Yin-Yang grid system, which consists of two overlapping latitude–longitude grids indicated in blue and red, was adopted for global simulations

MSSG can consider the precise topographic effect and the boundary-layer turbulence that affects the cumulonimbus clouds development, (Seifert and Onishi 2016) particularly over slopes against the wind. Thus, it is suitable for better one-day-ahead predictions of heavy rainfalls in the mountains.

Given the initial condition of weather variables such as winds, temperatures, atmospheric pressures, etc., three days before the Aranayake Landslide of May 2016 in Sri Lanka, MSSG simulated cumulative rain falls at Aranayake. Figure 18 shows the rainfall accumulation at Aranayake (80.4546E and 7.1476N) from 00:00 UTC May 14, 2016. The figure shows three curves:

Fig. 18

a Computational nesting domains for the Aranayake rainfall simulation. b Rainfall accumulation from 06:00 UTC on 15 May, 2016

• (1) For the simulation considering the turbulence enhancement (Case-T),

• (2) For the simulation not considering the turbulence enhancement (Case- NoT), and

• (3) The rain gauge record near Aranayake.

Though the reproduced timings of heavy rainfall events are off from the recorded ones, they successfully produced the total rainfall at the end. The Case-T shows a more significant amount of rain than the Case-NoT, which indicates the turbulence enhancement impacted the orographic rain.

One more successful example is in Western Japan, which suffered vast devastation by the successive heavy downpours in late June through mid-July 2018. Figure 19 shows the 6-h rainfall accumulation over Hiroshima. MSSG successfully reproduced the line-shaped torrential rainfall area, with a length of approximately 70 km in the southwest to the northeast direction and a width of roughly 20 km.

Fig. 19

a The computational domain. b The horizontal distribution of the 6-h (11 pm–5 am) rainfall accumulation over Hiroshima. c A three-dimensional visualization of the cloud distribution over Hiroshima at 00:10 am viewed from the southeast direction. The dashed line corresponds to the solid line in (b)

These simulations on MSSG usually require substantial computer resources. For example, the simulation shown in Fig. 19 was conducted on a supercomputer at JAMSTEC with the upper-level computation nest covering a 200 km by 200 km area of the island and the lower-level nest at the resolution of 500 m by 500 m.

The challenge in the one-day-ahead weather prediction in Sri Lanka is that we need to run the MSSG on an affordable workstation. To make it possible, a technique to construct rationally new data points from a low-resolution simulation on the workstation is being developed by applying machine learning technology with good references obtained from simulations on the supercomputer (Onishi et al. 2019).

6.2 Predicting Groundwater Pressure Build-Up

The process of groundwater pressure buildup can trigger an RRLL, and thus it is vital to study the process rationally. For this purpose, researchers at the Disaster Prevention Research Institute (DPRI), Kyoto University, conduct geotechnical centrifuge tests, which are helpful for scale modeling of any large-scale nonlinear problem for which gravity is a primary driving force. With a radius of 2.5 m and a maximum payload of 24 g ton, the centrifuge at DPRI can produce 200 g’s of centrifugal acceleration at its effective radius (Fig. 20).

Fig. 20

Geotechnical centrifuge at DPRI, Kyoto University (DPRI Geotechnical Centrifuge Center 2010)

Studying rainfall-induced slope failures via centrifuge modeling has a long history, and nozzle system is one of the commonly used rainfalls simulating options (Take et al. 2004; Ling et al. 2009; Bhattacharjee and Viswanadham 2018). In this experiment, steady and constant downpours are realized in a rigid soil container (600 × 300 × 140 mm inside) as water flows from 18 misting spray nozzles attached at the top (Xu et al. 2021).

Of particular note is that the soils from mountain slopes in Sri Lanka are weathered fragments of Pre-Cambrian gneiss rock with fine contents ranging from 30 to 40%. In the light of the current circumstances of the COVID-19 pandemic, not the soils from Sri Lanka but weathered granite soil, which is particularly dominant in western Japan and often known as Masado, is used in the experiments. The particle size distribution of this soil says this is well-graded sand with silt. The required amount of the dry soil was moistened with the predetermined amount of water for the specific water content and kept in sealed bags for 24 h to homogenize the soil in terms of water content. The soil was then compacted using the wet-tamping method in layers of 20 mm thickness. During this compaction, Pore Pressure Transducers (PPTs) were placed at predetermined locations shown in Fig. 21. Finally, the soil was cut and shaped to provide the required geometry.

Fig. 21

Schematic diagram of the centrifuge model

With time, the downpour from the misting spray nozzles led to pore water pressure (PWP) buildup. As shown in Fig. 22, the buildups of the pore pressure at different locations in the soil mass were reproduced in the analyses adopting a simple elastic constitutive model with the soil–water characteristic curve (SWCC) based on the van Genuchten (VG) model (Van Genuchten 1980). The governing equations for the numerical analysis are available in Uzuoka and Borja (2012). The first step of the numerical simulation was to put the soil mass in the steadily increasing gravitational field up to 50 G. Then, the water started seeping into the ground from the flux boundary. The aerial elements were placed immediately above the slope surface to reproduce the smooth infiltration across the air–soil boundary (Uzuoka et al. 2011). The reproduced pore pressure buildups are in good agreement with those observed, as shown in Fig. 23.

Fig. 22

Numerical model

Fig. 23

Comparison of pwp development of experimental results (Test A) and numerical results

6.3 Effective Risk Communication and Public Education

It will take several hours to make a one-day-in-advance forecast of the occurrence of RRLLs using the computer programs mentioned above (MSSG, LS-RAPID, etc.) on high-performance workstations. The obtained results will then be timely relayed to the last mile as augmented reality dioramas of the predicted rains and RRLLs with a bird’s-eye view of the terrain as its background on tablets’ and PC screens (Fig. 24). The system physically allows for bi-directional communication between transmitter and receiver sides.

Fig. 24

Rendering image of augmented reality dioramas of the predicted rains and locations of RRLLs with a bird’s-eye view of the area as their background on tablet’s screen

However, we need to take the following into account for better implementation strategies:

• (1) In Sri Lanka, districts are the second level administrative divisions under provinces (Statoids 2015). Each district, administered under a District Secretary, is divided into several Divisional Secretary’s (DS) divisions, which are further subdivided into extremely many local communities called “Grama Niladhari (GN)” divisions.

• (2) The number of local GN divisions in RRLL-prone areas is increasing year by year reflecting human interventions.

• (3) Though the Disaster Management Centre (DMC) of the Ministry of Défense has sole authority in terms of the issuance of evacuation alerts and evacuation orders, each Divisional Secretary (DS) can use his/her discretion in taking necessary actions in case of urgency.

• (4) NBRO has been helping their decisions providing information alerting them of landslides.

• (5) In this regard, both NBRO and JICA has started a joint project “Project SABO” (NBRO and JICA 2020), for capacity strengthening on development of non-structural measures for landslide risk reduction. The goals include: (a) strengthening of hazard mapping and risk assessment capacity, and (b) revision of regional-level early warning issuance criteria, etc. Thus, Project RRLL is expected to play a complementary role with Project SABO, providing proactive information that allows people to take precautions against RRLLs (Fig. 25, NBRO and JICA 2020).

  Fig. 25

  

  Rain-induced landslides to be covered by both project SABO and project RRLL (SATREPS)

As said in the “Background story,” NBRO has had local community people actively involved in the Community-Based Landslide Early Warning (CBLEW). The AR image of potential heavy rains and RRLLs displayed one day in advance on tablets and PC screens will undoubtedly help them keep further watchful eyes on the movements of potentially dangerous slopes. In this light, people should know the intrinsic nature of the creeping ground movement, which can be a precursor of the RRLL. The researchers at Kochi University study this nature using model slopes with different slope angles, water contents, etc., built in a steel flume (Fig. 26). Behind their experiments is the empirical finding by Fukuzono (1985). Fukuzono reported a linear relationship between the logarithm of acceleration and the logarithm of velocity for the tertiary (or accelerating) creeping movement of a landslide mass with an increasing creep rate, which leads to a landslide.

Fig. 26

Slope model prepared in a steel flume (Sasahara 2021)

They have examined three different cases with different slope angles and water contents (Table 1). The three model slopes were prepared with highly weathered granite sand (Fig. 27, Masado in Japanese), and steady artificial rainfalls of 46 mm/h were given to these slopes. These slopes showed different buildups of displacements, as shown in Fig. 28. Displacements also differed from location to location on each slope surface. However, it is surprising that the logarithm of acceleration takes almost the same ascending line as the logarithm of velocity increases, no matter the slope geometry and water content, as shown in Fig. 29. The scatter in the plots, particularly for Case 2, are due to the spontaneous formation of a bulge at the toe part of the slope. Once people find they are somewhere on this unique line, they should take proactive actions to prepare for the potential RRLL disaster. Further studies will be necessary to work out detailed action plans.

Table 1 Three cases examined (Sasahara 2021)

Fig. 27

Grain size distribution of the soil for the model (Sasahara 2021)

Fig. 28

Displacement buildups for three slope cases: the target plate for displacement measurement was placed 85 cm downslope from the upper end of the flume (Sasahara 2021)

Fig. 29

Logarithms of acceleration and logarithms of velocity for three different cases: the target plate for displacement measurement was placed 85 cm downslope from the upper end of the flume (Sasahara 2021)

7 Summary

Influenced by the effects of global climate change and, more seriously, by human-induced interventions in landslide-prone areas, the number of tragic RRLL events in Sri Lanka has been on a general soaring trend with some remarkable spikes in 2003, 2007, 2011, 2014, 2016, and 2017. In this situation, ICL and NBRO have started this project, “Development of early warning technology of rain-induced rapid and long-traveling landslides in Sri Lanka” (Project RRLL) in 2020 after the preceding period of 2019. The project is in line with the activities of the Sri Lanka Comprehensive Disaster Management Programme, output 1.3 “National and community level landslide early warning system is in place.” NBRO has been running one more JICA technical cooperation project, Project SABO, to strengthen the development of non-structural measures for landslide risk reduction. Project RRLL is thus complementary with Project SABO providing extra lead time for emergency responses evacuations, namely, one of the most critical missing pieces of the jigsaw puzzle for landslide-hazard mitigation.

The critical technologies to be developed in Project RRLL include:

• 24 h in-advance prediction of heavy rainfalls in mountains,

• Assessing groundwater pressure build-up, initiation of an RRLL, its flowing dynamics, and

• Effective risk communication and public education.

For the (1) 24 h in-advance prediction of heavy rainfalls in the mountains, we use MSSG (Multi-Scale Simulator for the Geo-Environment) as our generic platform for the precise weather forecast. MSSG can consider the topographic effect and thus the boundary-layer turbulence that affects the cumulonimbus clouds development (Seifert and Onishi 2016), particularly over slopes against the wind. It is, therefore, suitable for better one-day-ahead rainfall predictions, particularly in the mountains. The current challenge at the Tokyo Institute of Technology is to run the MSSG on an affordable workstation. To make this possible, a technique to construct rationally new data points from a low-resolution simulation on the workstation is being developed by applying machine learning technology.

For (2) assessing groundwater pressure buildup, researchers at the Disaster Prevention Research Institute (DPRI), Kyoto University, conduct geotechnical centrifuge tests, which are helpful for scale modeling of any large-scale nonlinear problem for which gravity is a primary driving force. Of particular note is that the soils from mountain slopes in Sri Lanka are weathered fragments of Pre-Cambrian gneiss rock with fine contents ranging from 30 to 40%. Thus, well-graded sand with the inclusion of silt was used in the experiments. The pore pressure buildups at different soil mass locations were successfully reproduced in numerical simulations, thus validating the numerical model.

For (3) effective risk communication, the communication tool to be developed will timely relay the prediction to the last mile as augmented reality dioramas of the potential rains and RRLLs with a terrain image displayed on tablets and PC screens. The AR image of possible heavy rains and RRLLs displayed on screens will undoubtedly help the local people and officers keep further watchful eyes on the movements of potentially dangerous slopes. Researchers at Kochi University have found that the logarithm of acceleration of the creeping landslide mass takes almost the same ascending line as the logarithm of velocity increases, regardless of the slope geometry and water content. Once people find they are somewhere on this unique line, they will take proactive actions to prepare for the potential RRLL disaster. This finding is what people should learn through the educational program to be developed.