Adaptive Learning Techniques for Landslide Forecasting and the Validation in a Real World Deployment Open image in new window
A forecasting algorithm using Support Vector Regression (SVR) used to forecast potential landslides in Munnar region of Western Ghats, India (10.0892 N, 77.0597 E) is presented in this paper. Forecasting for the possibility of landslide is accomplished by forecasting the pore-water pressure (PWP) 24 h ahead of time, at different locations and across soil layers under the ground at varying depths, and computing Factor of Safety (FoS) of the slope. It is done by learning from the real-time sensor data gathered from Amrita University’s Wireless Sensor Network (WSN) system deployed in Western Ghats for monitoring and early warning of landslides. We use two variations of SVR, SVR-Historic and SVR-Adaptive. SVR-Historic algorithm is trained with the data from July 2011 to December 2015 and tested for the period from January to November 2016. SVR-Adaptive algorithm is adaptively trained from July-2011 onwards and tested for the period from January to November 2016. PWP and the computed FoS from both the algorithms are compared with the actual PWP and FoS data and the Mean Square Error (MSE) for the SVR-Historic model is found to be 48.726 and 0.002 whereas the MSE for SVR-Adaptive model is found to be 12.438 and 0.0007 respectively. The PWP and the computed FoS from both the algorithms are tested for correlation using Pearson’s correlation test, with 95% confidence interval and the coefficients for PWP is found to be 0.804 and 0.959 respectively with p-value of 2.2e−16, whereas for FoS it is 0.802 and 0.955 with p-value of 2.2e−16. The confidence intervals for PWP and FoS from both the models is 0.763 to 0.839 and 0.950 to 0.969 respectively. Among the two forecasting models, SVR-Adaptive model performs better with a low MSE of 12.438 and 0.0007 in forecasting PWP and the computed FoS values respectively and correlates with the real-time data ~95% of the times. Application of this forecasting algorithm in real-world can thus provide 24 h extra time for early warning which is a boon for government and public to prepare for landslides after early warnings.
KeywordsLearning techniques Support vector regression Forecasting methods Early warning system
The authors would like to express their immense gratitude to Satguru Sri Mata Amritanandamayi Devi, the chancellor of Amrita University for her constant support and guidance in all the research activities. This work is partly funded by Ministry of Earth Sciences (MoES), Government of India under the project titled “Advancing Integrated Wireless Sensor Networks for Real-time monitoring and detection of Disasters” and partly funded by Amrita University. We wish to express our gratitude to Prof Balaji Hariharan and Ramesh Guntha for their valuable suggestions.
- Brocca L, Ponziani F, Melone F, Moramarco T, Berni N, Wagner W (2012) Improving landslide movement forecasting using ASCAT-derived soil moisture data. In: EGU general assembly conference abstracts, vol 14, p 2307Google Scholar
- Loew S, Gschwind S, Gischig V, Keller-Signer A, Valenti G (2015). Monitoring and early warning of the 2012 Preonzo catastrophic rockslope failure. Landslides, 1–14Google Scholar
- Lukose Kuriakose S, Sankar G, Muraleedharan C (2010) Landslide fatalities in the Western Ghats of Kerala, India. In: EGU general assembly conference abstracts, vol 12, p 8645Google Scholar
- Ramesh MV (2014) U.S. Patent No. 8,692,668. U.S. Patent and Trademark Office, Washington, DCGoogle Scholar
- Russell SJ, Norvig P, Canny JF, Malik JM, Edwards DD (2003) Artificial intelligence: a modern approach, vol 2. Prentice hall, Upper Saddle RiverGoogle Scholar
- Soman KP, Loganathan R, Ajay V (2009) Machine learning with SVM and other kernel methods. PHI Learning Pvt. Ltd, 477. ISBN:978-81-203-3435-9Google Scholar