Date of Award


Publication Type

Master Thesis

Degree Name



Earth and Environmental Sciences


Landslides, Machine Learning, Natural Hazards, Prediction, Self Organizing Maps


Chris Houser




Landslides are natural hazards commonly understated in both number of occurrences and cost of economic impacts. The Costa Rican terrain is predominately geologically young and therefore, severely impacted by landslides. It has limited resources and infrastructure and with a large portion of the population being poor, this causes communities to build in hazardous locations and infrastructure that can be easily crippled by landslides. Being able to identify where and when landslides are going to occur is key to mitigating the effects, either by stabilizing the slope or by evacuating communities. Machine learning is one method that has been increasingly used to monitor and predict landslides in recent times. These methods do not have the shortcomings of traditional analytical methods and can be easily adapted for different locations, changing or missing data, and number of factors studied. This research proposes that Self Organizing Maps (SOM) can be used as a versatile and effective method for landslide prediction. The results of this study have shown how SOM can be used for multi scale susceptibility analysis and for prediction with use of precipitation data, by producing significant results identifying high risk areas with a varying number and combination of variables. It has also shown that when precipitation data is used, it can identify high risk locations based on precipitation amounts and static variables (slope, TWI, curvature, NDVI, etc.). At the five-time scales tested, four of the tests produced correlations between increased precipitation and higher landslides risk (6 hour r2 = 0.38, 12 hour r2 = 0.36, 1 day r2 = 0.24, 1 month r2 = 0.33). This study has shown the versatility and effectiveness of SOM by producing significant results, as well as being able to use current weather conditions to produce landslide prediction analysis.