Sensors & big data in hydrology: complication or opportunity?

Morning: data science meets hydrology

In a vibrant morning programme, speakers from Deltares, the Netherlands Statistics (CBS) and TU Delft talked about a range of exciting new developments that will soon empower hydrologists in harnessing the opportunities of large-scale datasets. Deltares presented a range a recent research projects in which Google Earth Engine was used to map the state of water reservoirs worldwide, centered around the theme of ‘drop detection’. Jelrik Bakker, from Netherlands Statistics, demonstrated what big data can tell about society when properly organized and processed. He also called for a discussion about the possible benefits of an integrated database with all water data of The Netherlands. As a final talk in the morning programme, TU Delft and eScience center presented a new software platform that will soon allow researchers to share open source scripts and data.

Afternoon: machine learning and novel sensoring showcases

So, what can machine learning really do for hydrologists? And what are the pros and cons of the next-generation sensors? Talks in the afternoon programme showed that machine learning can truly add value to hydrologists. Vortech presented new insights about the Volkerak lake, purely derived from data analysis. Erwin Vonk showed how machine learning enables water utilities to better estimate peaks in water demand for different climate change scenarios. Recurring theme during the talks was that we should not forget that machine learning and physical models are only as useful as the quality and quantity of underlying data. Sensors and automated data validation methods can therefore be seen as a key ‘enabling technology’ for data-driven analysis. Presentations of Moisture Matters and the Royal Netherlands Meteorological Institute (KNMI) showed how novel sensoring methods can lead to richer datasets, while Leiden University outlined efforts to integrate satellite observations in a more intelligent way.

At the end of the day, participants discussed whether new data-driven algorithms should be seen as a replacement of existing modelling practices or merely as a new tool in the ever growing toolbox of hydrologists. The consensus of the day was that while the growing supply of data and mainstream machine learning unlocks new possibilities, current physical models and ‘expert judgement’ are simply very hard to beat. It seems that machine learning will thus co-exist with the current hydrological tools and workflows, as opposed to replacing it.