Master’s project by Franziska Schwarzenbach

Value of water level class data collected by citizen scientists for hydrological model calibration

My name is Franziska and I work as a community manager for the CrowdWater project. I am also writing my Masterโ€™s thesis in the CrowdWater project. My thesis work is based on hydrological modelling. I want to find out more about the value of the water level class data collected in the virtual staff gauge category of the CrowdWater app for hydrological model calibration. In my thesis, I focus on catchments that fulfill the following two criteria:

  1. Water level class observations are available for the catchment: either observations collected with the CrowdWater app or at the pen-and-paper stations that Barbara and Simon installed.
  2. Streamflow data are available from an official Swiss or Austrian measurement station that is located reasonably close to the CrowdWater spot.

The map below shows the locations of the eleven catchments used in this study.

Data from the Hydrological Atlas of Switzerland (Swiss catchments), Open Data Austria (Austrian catchments), Swisstopo (outlines of Switzerland and Liechtenstein), GADM (outline of Austria).

For each catchment, I have a certain number of water level class observations collected by citizen scientists. I combine these observations with a limited number of streamflow measurements, regularly spread during the year. With these two data types, I obtain 24 different scenarios of data availability, based on all the different combinations of

  • 0%, 25%, 50%, 75% and 100% of the available citizen science data, and
  • 0, 1, 3, 6, 12 streamflow measurements per year.

I calibrated a bucket-type catchment runoff model, the HBV model, for each catchment using the 24 different data availability scenarios. With the obtained parameter sets, I simulated the streamflow for each catchment and compared the resulting hydrograph (time series of streamflow) to the hydrograph that was measured at the official measurement stations. This comparison makes it possible to determine how good the model is or model performance. The higher the value of the model performance, the better is the agreement between the simulated and the measured streamflow time series.

The graph below contains an exemplary result for the Alp in Einsiedeln. One can see that with more data, the model performance tends to get better: the squares are getting darker (indicating a higher value) when moving from the lower left to the upper right corner. Currently, I am trying to find out why the approach works better for some catchments (such as the Alp) than for other catchments.

I will do similar model calibrations again using the quality-controlled citizen science data from the CrowdWater game. This will only be possible for some of the catchments because the data collected at pen-and-paper stations do not contain photos and are thus not part of the CrowdWater game. It will be interesting to see whether the model performance will be better when the citizen science data have a higher resolution and are checked by the crowd.

I thank everyone for contributing to the CrowdWater project and making my thesis possible. I love working with the app and game data! If you have questions or comments regarding my thesis, donโ€™t hesitate to send me an e-mail at