The potential of crowdsourced personal weather stations for hydrological forecasting in a Dutch lowland catchment

Accurate and real-time available rainfall data are indispensable for flood forecasting and warning. Crowdsourced personal weather stations have a high spatiotemporal resolution (in our case 9 km² and 5 min) and are available in near-real-time, but are prone to errors. In this study, we (1) assessed the accuracy of rainfall observations from personal weather stations in a Dutch lowland catchment (Oude IJssel, 1210 km²) and (2) used these PWS data as input to a rainfall-runoff model (WALRUS) to assess their potential for discharge forecasting.

 

The catchment-averaged rainfall depths measured by personal weather stations slightly overestimated the reference with a bias of only 0.03 mm, which is much lower than the underestimation of the real-time available (unadjusted) radar product (-0.16 mm). Quality control of PWS did not reduce the bias, but time series varied less and correlated better with the reference. For individual stations, quality control reduced the bias with 11% while retaining 85% of the data.

 

Discharge simulations using quality-controlled personal weather stations (NSE=0.98, using simulations with gauge-adjusted radar rainfall data as reference) were better than before quality control (NSE = 0.95) and much better than the real-time available (unadjusted) radar product (NSE=0.70).

 

To conclude, rainfall data from personal weather stations are suitable for real-time hydrological applications, especially after quality control.