Open Collections

Mee, Emily

University of British Columbia

Data-driven methods are popular in hydrogeology research but have been limited in practice due to data sparsity and implementation challenges. There is increasing opportunity for data-driven analysis as mining operations and other major development projects collect large volumes of hydrogeological data for site characterization and monitoring. To illustrate the utility of machine learning in a practical context, we use the self organizing map (SOM) method to extract patterns from groundwater head time series for 85 piezometers at the Red Chris Mine in northern Canada. The SOM results reveal spatiotemporal trends that were not readily apparent from previous site characterization studies. SOM results are strongly influenced by the choice of map size, but there is no overall preferred method for map size selection. We selected various map sizes using common error measures. A small map was optimal to visualize general trends in the aquifer system and larger maps distinguished local anomalies like borehole repair. The efficacy of SOM analysis to distinguish aquifer stresses and heterogeneity is not well understood. We applied SOM clustering to simulated head time series from models with known properties to evaluate its efficacy. The clustering results reveal spatial variation associated with aquifer recharge and aquitard discontinuities. SOM analysis did not clearly distinguish variation in aquifer properties due to the overlapping pattern of head drawdown due to groundwater pumping. SOM results should be tested and supplemented with other data visualization and analysis methods. With awareness of its strength and limitations, SOM analysis can enhance aquifer system characterization.

Text

2024-10-03

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/89332

Geological Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/