pyLEnM: Machine learning and analytics toolkit for long term water quality monitoring using a remote sensing network

Abstract:

Recent technological advances – in situ groundwater sensors, geophysics, drone/satellite-based remote sensing, reactive transport modeling, and AI – has a great potential to establish the new paradigm of long-term monitoring with improved effectiveness and robustness at groundwater contaminated sites. In particular, in situ sensor networks can be a powerful alternative to conventional groundwater sampling and laboratory analysis; particularly for monitoring controlling and master variables that are often leading indicators of changes prior to plume movement. There are still challenges to effectively establish such networks; where to place sensors, which in situ measurable parameters are most informative, how long monitoring network is expected to continue, how to estimate contaminant concentrations or mobility in real-time based on in situ data.

In this work, we develop a suite of machine learning algorithms to support the sensor network deployment as well as to analyze monitoring datasets effectively. In particular, we focus on extracting critical information from historical/existing monitoring datasets, by analyzing multiple time-series data of groundwater contamination and groundwater quality parameters (such as pH, specific conductance, water table, redox potential). Our algorithms analyze and visualize the correlations between different analytes as well as identify key parameters that control contaminant concentrations and plume mobilities. In parallel, regression models are developed to predict when the contaminant concentrations are expected to reach below the regulatory standard. In addition, we also use principal component analysis, and clustering analysis to group existing wells locations that have similar groundwater dynamics so that we can more effectively select among existing wells for installing new sensors.