Our third article has been just published. This time we assessed groundwater level time series in the whole Baltics and proposed a new approach for a more objective assessment of gap-filling techniques. This has been a wonderful collaboration between the University of Latvia, Faculty of Geography and Earth Science partner, and Ezra Haaf from The Chalmers University of Technology from Sweden.
You can access the article here: https://doi.org/10.1016/j.jhydrol.2023.129424
Groundwater levels are a fundamental part of every hydrogeological study
Did you know that automatic groundwater level recordings in the Baltics started quite recently – 2005 for Lithuania and ~2011 for Estonia and Latvia. With the automatization the frequency of groundwater level measurements expanded from a few times per year to a few times per day, giving the possibility to assess groundwater heads’ seasonality and daily variations. Yet, automatic loggers tend to malfunction, thus creating gaps and missing values that are not welcome for further analysis.
Gaps can be found in most time series, but there is a solution
A variety of methods exist how to infill gaps, including our tested machine learning approach. But the greatest challenge is to assess the imputation performance objectively. Here we present a new approach how to mimic gap patterns characteristic of the data set. The gap patterns are then introduced as artificial gaps to assess the imputation performance.
New challenges await
We discovered that the missForest imputation method can effectively infill a variety of gaps, even the challenging ones at the beginning and end of time series and around extremes, just using the rest of the level data set. However, we also discovered that unseen or untypical extremes are complex to infill – for example severe drought events in 2017, and 2018 and human-influenced times series, e.g., by water abstraction. Read more about EU-WATERRES project here: https://eu-waterres.eu/