A stream computing approach for live environmental models using a spatial data infrastructure with a waterlogging model case study

Abstract

Traditional environmental model simulations often use archived data as inputs. Recent advancement of Sensor Web technologies in Spatial Data Infrastructures (SDIs) allows real-time observations to be fed into models to generate “live” models. A key challenge is how to efficiently process observation streams in models, which is particularly important in time-critical cases like disaster management. This paper presents an observation stream computing model for live modelling, which couples Sensor Web and models in stream computing environment to provide timely decision-support information. Observation Streams are proposed as information models to deal with observation stream processing. The approach shows how MapReduce and Apache Spark stream processing can be leveraged to support coupling of observation streams and models. The approach is applied in a disaster management case, where in-situ observation streams are processed to compute the waterlogging information in near real time. The results illustrate applicability and effectiveness of the approach.