Spatial modeling of snow avalanche susceptibility using hybrid and ensemble machine learning techniques


AKAY H.

CATENA, cilt.206, 2021 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 206
  • Basım Tarihi: 2021
  • Doi Numarası: 10.1016/j.catena.2021.105524
  • Dergi Adı: CATENA
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aquatic Science & Fisheries Abstracts (ASFA), BIOSIS, CAB Abstracts, Environment Index, Veterinary Science Database, DIALNET, Civil Engineering Abstracts
  • Anahtar Kelimeler: Boosting ensemble models, Machine learning models, Random subspace ensemble model, Snow avalanche susceptibility, Uzungol Basin, LOGISTIC-REGRESSION, DECISION-MAKING, SLOPE FLOWS, REGION, HAZARD, TURKEY
  • Gazi Üniversitesi Adresli: Evet

Özet

In this study, snow avalanche susceptibility maps of Uzungol Basin, which is a special environmental protected zone located in Eastern Black Sea Basin of Turkeywere conducted. For this, the belief function values of snow avalanche conditioning factors extracted from snow avalanche inventory data exploring geographical information system software were used. Classes of snow avalanche conditioning factors based on J48, random tree (RT), random forest (RF), and AdaBoost M1, Bagging, Real AdaBoost, and Random Subspace ensemble models of J48, RT, and RF models were implemented. Then snow avalanche susceptibility maps of each model were generated based on the combination rule of Dempster-Shafer theory. The model predictions were validated using a receiver operating characteristics curve, seed cell area index, and statistical tests. AdaBoost M1 ensemble of RT model was found to be the best performing results. However, ensemble models did not improve the prediction results unexpectedly. It can be expected that findings of this study will lead to the implementation of dynamic snow avalanche modeling studies on avalanche paths highlighted in the prepared maps, planning of land-use change projections, controlling snow avalanche movement, and mitigating the risks of loss of life and possible snow avalanches.