Sparse sensing to dense assessment: Incorporating spatial autocorrelation for assessing flood impacts

A rapid and comprehensive assessment of flood impacts is critical to assist emergency managers in conducting effective relief operations. With advances in information technologies, various types of sensors have been widely used to assess flood impacts promptly as they are capable of providing rapid flood impact information. However, sensor-driven approaches are limited in the provision of a comprehensive impact assessment as sensors are often sparsely distributed. In this research, the authors integrate the sparse flood impact information obtained from sensors and the spatial autocorrelation of flood-impacted areas, in order to achieve a rapid and comprehensive flood impact assessment. To achieve such a purpose, a systematic approach is proposed to (1) extract flood impact information from sparsely distributed sensors; (2) model the spatial autocorrelation of flood-impacted areas based on flood evolution and geography principles; (3) learn the parameters of the spatial autocorrelation model through a gradient descent method; (4) infer the flood impacts of sensor-uncovered areas based on the sparsely sensed impacts and the modeled spatial autocorrelation. To illustrate the proposed approach, we studied flood impacts on Highways in Houston, Texas during Hurricane Harvey. Results show that the spatial autocorrelation model presents a decent generalization capability in inferring the probability of neighboring highway blocks having the same flood impacts. Compared to purely sensor-driven approaches, the proposed approach is capable of greatly extending the coverage of flood impact assessment while maintaining the nearly same accuracy.