Addressing critical environmental data gaps via low-cost, real-time, cellular-based environmental monitoring
Abstract:Models in the environmental sciences are repositories in a sense of the current state of understanding of critical processes. However, as our understanding of these processes (and their accompanying models) become more granular, the data requirements to parameterize them become more limiting. In addition, as these models become more useful, they are often pressed into service for decision support, meaning that they cannot accept the data latency typical of most environmental observations. Finally, the vast majority of environmental data is generated at highly-instrumented, infrastructure-rich “mega sites” in the US/Europe, while many of the most pressing environmental issues are in rural locales and in the developing world. Cellular-based environmental sensing is a promising means to provide granular data in real time from remote locales to improve model-based forecasting using data assimilation. Applications we are working on include drought forecasting and food security; forest and crop responses to weather and climate change; and rural water usage.
Over the past two years, we have developed a suite of integrated hardware, firmware, and backend APIs that accommodates an unlimited variety of sensors, and propagates these data onto the internet over mobile networks. Scientific data holds a unique role for demanding well-characterized information on sensor error and our design attempts to balance error reduction with low costs. The result is a deployment system that undercuts competing commercial products by as much as 90%, allowing more ubiquitous deployment with lower risks associated with sensor loss. Enclosure design and power management are critical ingredients for remote deployments under variable environmental conditions. Sensors push data onto cloud storage and make this data available via public API’s via a backend server that accommodates additional metadata essential for interpreting observations, particularly their measurement errors. The data these pods collect can expand weather monitoring, but more crucially can monitor otherwise unobserved biological (including human) responses to environmental drivers. These data in turn can be assimilated into models, as a means to contextualize and distill these noisy observations into actionable knowledge.