Longitudinal Effects of Earthquake-Induced Landslide Susceptibility in Papua New Guinea

Earthquake-induced landslides (EQIL) account for 4–5% of all landslides worldwide. The most active earthquake hotspots in the terrestrial environment are susceptible to EQIL. Major earthquakes that have triggered EQIL include: 2008 Wenchuan, 2018 Porgera, 2015 Gorkha and 2010 Haiti . These events have caused more than 25000–30000 landslides, in certain cases more than 100000 landslides, and they have been responsible for more than 100000 deaths and property damage worth billions of dollars (Jesse et al., 2020). The events further trigger cascading hazards such as landslide dams for more than two decades (Fan et al., 2019). Current literature in modelling these repercussions of EQIL has evolved to include the temporal effects of such events in the aftermath of large earthquakes (Sridharan et al., 2024; Dahal et al., 2024).

Papua New Guinea (PNG) is one of the many seismically active regions in the world. The Indo-Australian boundary is a major plate boundary that runs through PNG. This fault zone experienced a major earthquake in 2018 near Porgera that is reported to have triggered more than 10,000 landslides in the region (Tanyas et al., 2022). This work explores the prolonged effects of the earthquake in the region. We use the automatically mapped landslide inventory by Bhuyan et al. (2022) to train and validate our model (Bhuyan et al., 2022). To capture the changes caused by the earthquake, we use a longitudinal GAM (Hastie and Tibshirani, 1990) that estimates the variation in log odds with periodic changes in climatic and seismic inputs. Terrain attributes modelled as static covariates also contribute to the changes observed in the terrain.

Our results show that the model performs well with respect to accuracy measures AUC-ROC, Brier score, and the R2 statistic of susceptibility estimates. We observe that the effect of the seismic activity remains for a short period of a few years after the earthquake. We present the longitudinal susceptibility prediction maps for the PNG at a slope unit level for future reference.

 

References:

Hakan Tanyaş, Kevin Hill, Luke Mahoney, Islam Fadel, Luigi Lombardo, The world's second-largest, recorded landslide event: Lessons learnt from the landslides triggered during and after the 2018 Mw 7.5 Papua New Guinea earthquake, Engineering Geology, Volume 297, 2022, 106504, ISSN 0013-7952

Sridharan, A., Gutjahr, G., and Gopalan, S., “Markov–Switching Spatio–Temporal Generalized Additive Model for Landslide Susceptibility,” Environ. Model. Softw., vol. 173, no. August, p. 105892, Feb. 2024 

Ashok Dahal, Luigi Lombardo, Towards physics-informed neural networks for landslide prediction, Engineering Geology, Volume 344, 2025, 107852, ISSN 0013-7952,

 

Bhuyan, K., Tanyaş, H., Nava, L. et al. “Generating multi-temporal landslide inventories through a general deep transfer learning strategy using HR EO data”. Sci Rep 13, 162, 2023

 

Hastie, T. J.; Tibshirani, R. J. (1990). Generalized Additive Models. Chapman & Hall/CRC.