Accurate rainfall measurement at high spatial and temporal resolutions is critical for both research and urban management. With the recent advances in technology, there emerges a new crowd-sourcing approach, where common citizens use smartphones, surveillance cameras, and other devices as sensors to measure rainfall intensity. The crowd-sourcing approach is promising as it has the potential to provide high-density measurements, but it also associates with relatively large individual errors and relatively unreliable participants. In this thesis we develop a framework for modeling the crowd-sourcing approach in rainfall monitoring, and use this modeling framework to tap into the following issues: (i) the proof of concept of the crowd-sourcing approach, (ii) the integration of cro...[ Read more ]

Accurate rainfall measurement at high spatial and temporal resolutions is critical for both research and urban management. With the recent advances in technology, there emerges a new crowd-sourcing approach, where common citizens use smartphones, surveillance cameras, and other devices as sensors to measure rainfall intensity. The crowd-sourcing approach is promising as it has the potential to provide high-density measurements, but it also associates with relatively large individual errors and relatively unreliable participants. In this thesis we develop a framework for modeling the crowd-sourcing approach in rainfall monitoring, and use this modeling framework to tap into the following issues: (i) the proof of concept of the crowd-sourcing approach, (ii) the integration of crowdsourced data with the existing radar and rain gauge data, and (iii) the optimal management of crowd-sourcing participants.

The main body of this thesis consists of three parts. The first part develops a new rainfall data crowd-sourcing model, which involves two data types: (i) scattered observations made by individuals and (ii) fixed-point observations made using fixed sensors. Using this model, we explore the potential of the crowd-sourcing approach for urban rainfall monitoring and the subsequent implications for stormwater modeling through a series of synthetic but realistic simulation experiments. The results show that even under conservative assumptions, crowdsourced rainfall data lead to more accurate modeling of stormwater flows as compared to rain gauge data. We also 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 crowdsourced observation type.

The second part investigates the merging of crowdsourced rainfall data with traditional radar and rain gauge data to maximize their utility. For this purpose, we develop a tailored Fast Bayesian Regression Kriging (FBRK) method combining regression kriging and Laplace approximation in a Bayesian framework. A strength of this method lies is its ability to capture the differences in the errors of the three data types. Another lies in its fast yet reasonably accurate approximation of the Bayesian posterior, making it suitable for use in real-time. We conduct computer simulations to evaluate the FBRK method alongside three other merging methods. In the simulations, we compare the accuracies of their resulting rainfall estimates, as well as the skill of those estimates as input to a stormwater flow forecasting model. In both aspects, we observe the FBRK method to lead to more exact results and truer representations of the associated uncertainties. We observe also, however, the performance of the FBRK method to be sensitive to the choice of the Bayesian prior under certain conditions. Finally, we find merging crowdsourced data with traditional data leads to more accurate estimation of the ground truth rainfall field and subsequently, more accurate flow forecasts (though only when an adequate merging method, such as the FBRK method is used), and the results to be fairly robust to bias in the input crowdsourced data.

The third part proposes and tests the idea of incentive allocation (IA) among crowd-sourcing participants, where IA refers to the allocation of quantitatively measurable and limited rewards. We modify our model in the first part to develop an integrated model for this purpose. The new model consists of a real-time IA component for the manager, an agent-based model to simulate interactions between the manager and participants, and a rainfall simulation model to evaluate the performances of different IA policies. In the model, we generate six different IA policies, and each follows a specific design principle and associates with a level of administrative cost (AC). These policies and the integrated model is tested through a theoretical but realistic case study. Results suggest the performance with each policy to be positively affected by the incentive budget, the level of environmental education and awareness, and the spatial uniformity of participants, and the impact of budgets is less significant with the increase of spatial uniformity of participants. The participant density weighted maximum participation policy (MDPP) which considers the spatial distribution of participants performs better than all other policies under almost all investigated conditions. However, MDPP associates with a high level of AC, which makes the choice of optimal IA policy a trade-off between performance and AC.

To sum up, this thesis provides direct answers to concerns regarding the feasibility, data processing, and participant’s reliability of the crowd-sourcing approach for urban rainfall monitoring. It also offers more quantitative understandings regarding this new approach, and the models developed here could be used as a framework for future studies on the application of the crowd-sourcing approach in other research areas.