Gauging Through the Crowd: A Crowd‐Sourcing Approach to Urban Rainfall Measurement and Storm Water Modeling Implications

Abstract

Accurate rainfall measurement at high spatial and temporal resolutions is critical for the modeling and management of urban storm water. In this study, we conduct computer simulation experiments to test the potential of a crowd‐sourcing approach, where smartphones, surveillance cameras, and other devices act as precipitation sensors, as an alternative to the traditional approach of using rain gauges to monitor urban rainfall. The crowd‐sourcing approach is promising as it has the potential to provide high‐density measurements, albeit with relatively large individual errors. We explore the potential of this approach for urban rainfall monitoring and the subsequent implications for storm water modeling through a series of simulation experiments involving synthetically generated crowd‐sourced rainfall data and a storm water model. The results show that even under conservative assumptions, crowd‐sourced rainfall data lead to more accurate modeling of storm water flows as compared to rain gauge data. We observe the relative superiority of the crowd‐sourcing approach to vary depending on crowd participation rate, measurement accuracy, drainage area, choice of performance statistic, and crowd‐sourced observation type. A possible reason for our findings is the differences between the error structures of crowd‐sourced and rain gauge rainfall fields resulting from the differences between the errors and densities of the raw measurement data underlying the two field types.

Fast Bayesian Regression Kriging Method for Real‐Time Merging of Radar, Rain Gauge, and Crowdsourced Rainfall Data

“…Figure and Table strongly suggest integrating crowdsourced observations to the radar‐rain gauge data merging process to have huge potential for producing results leading to both more accurate rainfall estimation and better storm water forecasting though only if using an appropriate merging method such the FBRK or KED method. The results give support to the potential importance of merging crowdsourced rainfall data with traditional data for maximum utility, especially when placed alongside earlier results by Yang and Ng (), who found crowdsourced rainfall observations (alone), when compared with rain gauge observations (alone), to lead to less accurate rainfall estimation though more accurate storm water modeling due to the former being better able to capture the true rainfall distribution at the storm center but less so elsewhere. In other words, merging has the potential to combine the best of the different observation sources to yield more accurate rainfall estimates at all rained areas, whether near or away from the storm center, and with that improved storm water forecasting.…”

“…To generate radar observations, we upscale the TF data from above to a 500 m × 500 m × 5 min resolution and then add systematic (bias) and random (noise) errors to the upscaled results to mimic real conditions (Sinclair & Pegram, ). To generate rain gauge observations, as in Yang and Ng (), we presume a random uniform distribution of rain gauges throughout the study region with a 0.08/km 2 density (as derived from real‐world high‐density rainfall monitoring networks, Looper & Vieux, ; Vieux & Vieux, ), and the rain gauge observations as continuous in time with negligible measurement errors. See Figure for a representative realization of the rain gauge locations.…”

“…We further assume each crowdsourced observation, whether scattered or fixed point, to contain a certain measurement error, E , as randomly assigned following the normal distribution below (Clark et al, ; Yang & Ng, ): where I true is the ground truth rainfall intensity, and α and β are the coefficients of variation and bias, respectively, of E . α defines the stochasticity of an observation, while β its tendency to underestimate or overestimate I true .…”

“…We repeat this process for every time step from the start to the end of the simulation period. Lastly, we evaluate the rainfall estimation accuracies of the merged data products in terms of the error statistics REAA I and RMSE I , and their skill as input to the SWMM model of the Chollas Creek watershed from Yang and Ng (, section ) according to the error statistics ARE P , ARE V , RMSE F , and RMSE F10 (which we calculate based on flows at the watershed outlet as predicted by the SWMM model of the watershed). The error statistics are as defined and described in section , and are calculated by comparing the merged rainfall data and the subsequent storm water modeling results against the ground truth TF data and TF‐derived storm water flows.…”

Supporting participative pre‐flood risk reduction in a UNESCO biosphere

“…Current monitoring systems cannot provide sufficiently high‐quality data for accurate storm water modelling; this problem is particularly acute in developing countries. Yang and Ng () demonstrated that utilising crowd‐sourced data would lead to a more accurate modelling of storm water flow than if just using gauge data; though the VGI data was simulated in this instance, the results suggest a potential for use in urban scenarios. Rosser, Leibovici, and Jackson () showed how a combination of social media, remote sensing and terrain models could be harnessed to derive a Bayesian model for estimating flood inundation.…”