Abstract
Temporary streams are streams that do not always contain flowing water. They are abundant and important for water, solute and sediment transport, and are unique habitats. However, they are underrepresented in traditional hydrological monitoring systems. Therefore, new approaches, such as citizen science, have been developed to monitor the flow states of temporary streams. Equipped with smartphones, citizen scientists can make visual flow state observations with the CrowdWater app. Since 2017, they can observe temporary streams by uploading a picture of a stream and choosing one of six flow state classes: dry streambed, damp/wet streambed, isolated pools, standing water, trickling, and flowing. We carried out surveys to assess the agreement among the observations from different people and the aspects they consider when determining the flow state. The pen and paper questionnaires at eight streams were held during 23 field days in summer 2022 (Paper I). The overall agreement on a flow state between participants was 66% and the agreement with the expert’s opinion was 56%. Only 17% of the participants chose a flow state that was more than one class away from the one chosen by the majority. The agreement for the distinction between flowing and not flowing streams was 92%. Participants’ opinions were sometimes inconsistent with their answers to questions detailing aspects of the flow states, particularly when they selected the flow state classes dry or wet/damp streambed. I recommend adding guidelines and assessing this inconsistency for participants who are more experienced in observing temporary streams. Citizen scientists choose the stream site and observation time in the CrowdWater project themselves. This leads to temporal and spatial irregularities in the dataset. Many observations were collected at Zuriberg, Switzerland, thanks to the installation of information boards in this area. We determined the hierarchical structure of stream wetting and drying (Paper II) for this area by ordering the observed stream sites in probabilistic space instead of physical space. More specifically, the streams were ordered in a directed graph from the most to least persistent. For this analysis, we merged the six flow state classes into two: flowing and not flowing. By applying the derived hierarchy, flow states were simulated for most stream sites on observation days even if only a few sites were observed on a specific day. This led to three times as many flow state data points. The accuracy of the derived flow states from the hierarchy compared to the observations was high (99%). By correlating the fraction of flowing streams with climate data, we obtained daily information on the flow states, increasing the total number of flow states from 4,360 observations to 35,569 simulated values. We also combined expert observations of flow states with the CrowdWater dataset and found that these hierarchies did not contradict each other for the same sites and period. From this, we recommend that experts observe different sites than the citizen scientists and on different days to increase the number of flow states that can be derived. To improve the generation of the hierarchy, experts should focus on observing stream sites that could not be placed in the hierarchy with certainty based on the citizen science data alone. We also tested the usefulness of temporary stream observations in combination with discharge data for the calibration of a catchment scale hydrological model (Paper III). For this, we used a dataset of flow states derived by experts in France. Flow states were, similar to those in the CrowdWater app, derived visually and thus the dataset could possibly be compared to a dataset collected by citizen scientists. We applied the flow state data in combination with discharge data for multi-criteria model calibration for 92 catchments. The changes in the flow states were compared to the groundwater level changes simulated by the hydrological model. Using temporary stream data for the calibration did not necessarily improve discharge simulations but could for some catchments improve the uncertainty of parameter K2, which influences low-flow predictions. It is unclear for which kind of catchments the temporary stream data were beneficial. In theory, visual flow state observations of temporary streams provide a great opportunity for multi-criteria model calibration as they are much easier to collect than e.g., groundwater level data. Thus, the value of visual flow state observations for model calibration should be further tested for different kinds of models and study sites. Furthermore, this analysis should be repeated with actual citizen science data of temporary streams. In conclusion, this thesis shows that visual flow state observations of temporary streams by citizen scientists are reliable and useful for further analyses. I, therefore, recommend continued data collection with the help of citizen scientists.
Scheller, Mirjam