European Centre on Geodynamical Risks of High Dams - GHHD


 

1 Alexidze Street
Tbilisi 0171
Georgia

Email: [email protected]
Director: Prof. Tamaz Chelidze

 

Foundation/Legal status

The Centre was founded in December 1995at the Ministry of Environmental Protection of Georgia. Now the GHHD is the independent NGO, registered 18/07/2008 in Tbilisi Regional Centre of tax inspectorat, serial number N 004760.

Centre's host organisation: Institute of Geophysics at Tbilisi State University

 

Structure

President of the Scientific Committee: Dr. Martin Wieland
Director: Prof. T. Chelidze, Institute of Geophysics
Secretary General: Dr. T. Matcharashvili, 
Treasurer: N. Taniashvili

 

Main activities and latest achievements

The Centre is created for development of multinational, multidisciplinary approach to the problems of geodynamical hazards, generated by high dams, including:

  • development and testing of modern methods of multidisciplinary monitoring/diagnostics of local and regional geodynamical processes in the proximity of large dams on the basis of Enguri Arc Dam International Test Area using contemporary approaches of nonlinear dynamics and machine learning;
  • creation of up-to-date early warning systems;
  • mathematical modelling of geodynamical processes at large dams, forecasting of impending geodynamical events (earthquakes, tectonic deformations, landslides) and prognosis of response of large dams to these impacts;
  • monitoring of processes and associated variations in physical properties of foundation rocks and dam material;
  • creation of databases of geodynamical observations on large dams;
  • analysis and generalization (in collaboration with other European centres) of possible geodynamical hazards, creation of scenarios of possible damage and instructions for public education on what to do in case of alarm, during and after the disaster using contemporary approaches of nonlinear dynamics and machine learning;
  • active participation in international, regional and national projects related to major disasters and environmental problems.

 

Activities implemented in the last three years:

2022-2023: Development of new methods of analysis of strains/seismicity in the area of large dams, using machine learning tools. 

Application of new measures of high dams stability using nonlinear dynamics tools. In the present report a new approach, namely, analysis of Tsallis entropy variation in dam strain time series was performed. For statistical and dynamical investigation of dam stability problem, modern methods of nonlinear dynamics analysis of strain/tilt time series were used, namely include detrended fluctuation analysis (DFA), Recurrence Plots (RP) and Recurrence Quantitative Analysis (RQA), algorithmic (Lempel-Ziv) complexity measure (LZC), mutual information (MI). In the present report a new approach, namely, analysis of entropy, namely, Tsallis entropy variation in strain time series was performed.  It has been shown that the probability of occurrence of different microstates is correlated: the distribution of events in such systems obey a power law due to long-range interactions. Tsallis suggested a generalized approach, non-extensive statistical mechanics (NESM), in which interactions among the elements of a system at all lengths are taken into account.

As a result, we performed the EQ forecast for the West Caucasus region, using data on geomagnetic variation, water level in deep wells, earth tides as well as additional predictive seismological parameter W(t) on 3-years long learning/testing period. W(t) characterize the EQ activity in the 5-days intervals before events of M≥ 3.5. Besides, special attention was paid to compensate the imbalance effect leading to overfitting of data. We show that by application of the oversampling approach it is possible to obtain balanced assessments. The confusion matrix was obtained, which show that such statistical measures, as Matthews correlation coefficient and F1 score give good results in forecasting regional events of M≥3.5, namely MCC in the range 0.8±0.012 and F1 score in the range 0.85±0.010. After randomization of the EQ catalog the values of both MCC and F1 score decrease considerably.

2024-2025: Application of machine learning to analysis of landslide/mudflow hazard in mountainous country (example of Georgia. Testing new tilt/acceleration monitoring equipment at Enguri dam).

Georgia, due to its mountainous landscape, climatic conditions, large-scale hazardous geological activity, growth of population, intensive land use, vulnerable infrastructure, and number of large engineering constructions, belongs to one of the hardest-hit regions in the world in relation to landslides. Thus, it is of great importance to study the intensity and spatial distribution of landslide risk to create reliable and cost-effective early warning systems (EWS) for monitoring and predicting mass-movements in potentially dangerous areas.   The first step in assessing landslide risk is the detailed landslide stationary susceptibility mapping, whose results can reveal the spatial distribution of the future landslide likelihood. On the other hand, it is known that intense and prolonged rainfall is the most frequent landslide trigger around the world, with almost 80% of landslides being rainfall-triggered. Thus, research on the intensity and spatial distribution of rainfall-induced landslide risk is of great importance for Georgia.

The results of work are presented in the paper (Chelidze at al, 2022), where the EQ forecast problem for the West Caucasus region is considered, using data on geomagnetic variation, water level in deep wells, earth tides as well as additional predictive seismological parameter W(t) on 3-years long learning/testing period. W(t) characterize the EQ activity in the 5-days intervals before events of M≥ 3.5. Besides, special attention was paid to compensate the imbalance effect leading to overfitting of data. We show that by application of the oversampling approach it is possible to obtain balanced assessments. The confusion matrix was obtained, which show that such statistical measures, as Matthews correlation coefficient and F1 score give good results in forecasting regional events of M≥3.5, namely MCC in the range 0.8±0.012 and F1 score in the range 0.85±0.010. After randomization of the EQ catalog the values of both MCC and F1 score decrease considerably.
 

Publications

Chelidze T, Kiria T, Melikadze G, Jimsheladze T and Kobzev G (2022) Earthquake Forecast as a Machine Learning Problem for Imbalanced Datasets: Example of Georgia, Caucasus. Front. Earth Sci. 10:847808. doi: 10.3389/feart.2022.847808

Chelidze, T., Melikadze, G., Kobzev, G., Jimsheladze, T. and Dovgal, N. (2023) Geophysical Reactions to Remote 2022 Tonga Eruption and to Türkiye Earthquakes in Georgia (Caucasus): Hydrogeology, Geomagnetics and Seismicity. Open Journal of Earthquake Research, 12, 223-237

Chelidze T, Dovgal N., Kiria J, et al. 2024. Long-Term Strain Dynamics of the Fault Crossing the Engury Dam Foundation (Georgia). Bull. Georgian Nat. Acad. Sci. v.18, no. 4.

Telesca L, Tsereteli N., Tugushi, N., Chelidze T. 2025 Spectral investigation of the relationship between seismicity and water level in the Enguri high dam area. Geosciences 2023,