Model parameters and a priori ranges.

In the context of participatory monitoring projects in hydrology, the collection of water level data by laypersons is used as a simple and cost-effective alternative to automatic water level sensors, especially in poorly gauged catchments in remote areas of countries in the Global South. Such data can be used for the development of hydrological models to support water resources management. However, a common problem with participatory monitoring approaches is the irregularity of data collection and its decreasing frequency over time. Determining the amount and timing of data collection required for satisfactory model calibration is critical. To investigate this further, we examined daily water levels from a four-year project in western Kenya. We set up scenarios that represented datasets of different lengths and seasonal starting points for measurements. These scenarios were then used to calibrate a simple rainfall-runoff model. The data were supplemented with satellite data on evapotranspiration to improve the simulated water balance. The model runs were filtered using a water balance filter, and the Kling-Gupta-Efficiency (KGE) was used to compare model efficiencies. While a single month of water level data collected during the rainy seasons was sufficient to achieve good model performance (KGE ≥ 0.75), several months of data were required for simulations starting in the dry seasons. Similar results were found for the validation period, with lower overall model performance (KGE ≥ 0.6). Water level data collected during high flows generally led to an improvement in model performance compared to data collected during low flows. However, after a certain threshold, more water level data did not lead to further substantial model improvement. Based on the analysis of different datasets, this study demonstrated that short-term participatory monitoring programs that collect water level data during the wet season have the potential to provide sufficient input to calibrate a hydrological model.