Back to search

IKTPLUSS-IKT og digital innovasjon

Distributed Arctic Observatory (DAO): A Cyber-Physical System for Ubiquitous Data and Services Covering the Arctic Tundra

Alternative title: Distribuert Arktisk Observatorium (DAO): Et Cyber-Fysisk System for Allestedsnærværende Data og Tjenester om den Arktiske Tundra

Awarded: NOK 16.1 mill.

The arctic tundra is most sensitive to climate change. The change is quantified by observations on the tundra of animals, vegetation, wind, humidity, CO2, temperature, and more. However, much less than 1% of the arctic tundra is observed today. To get more data, observations must be done in-situ on the tundra, more frequently, at more places, over larger areas, and both above and below snow. The observations must allow for identification of plants, birds and small animals. The data must also be made available when and where needed. It is necessary to deploy a set of nodes (small computers with sensors) to the arctic tundra to do all of this. A node must operate for a long time doing observations, and taking care of, process and report the data. However, on the arctic tundra the common case is that the necessary resources (humans, energy, back-haul data networks) are very limited or not available. We research how a distributed system of observation nodes can be done and will behave under these challenging conditions. A node must rely on a single battery charge because energy from solar, wind and humans are not practical or not allowed for regulatory reasons. Consequently, a node must be very frugal with energy to operate for many months and years. A node typically sleeps most of the time, but wakes up for a few seconds or minutes to do observations, processing and reporting, and then they go back to sleeping. To report data, a node must use one or several wireless back-haul and local area network technologies. If a back-haul network is not directly reachable, the node uses the local area networks looking for a another node having a back-haul network. If this fails, the node must wait for a visit by a mobile data mule node, say, a UAV or a reindeer, to carry data to and from the node. The research-in-progress Edge Connectivity System, CoLoRa, is based around an existing long-range low bandwidth network technology (LoRa), but modifies it to make it appropriate for nodes on the arctic tundra. Nodes must be able to be updated during their operational life. The Update Distribution System first finds a node with a back-haul network and sends an update to it. This node then takes over and uses a local area network, like CoLoRa, to spread the update to the whole neighborhood. Then the Update Install System takes over and installs the update at each node. Both systems are delay and sleep tolerant. Nodes must safe-keep data, as well as making it available to end-users. The research-in-progress Edge Data Abstraction System replicates data between nodes, and provides nodes and end-users with well-defined ways to access the data. Nodes need to process data locally at the nodes to react to events happening around them, and to prepare data for reporting. The research-in-progress Edge Analytics System does this. We research how to do energy-efficient on-node deep learning as well as image size compression. We document that even compressing to 10% of original size only has insignificant impact on the ability of a deep learning computation to determine the species of animals in the image. Reporting compressed images reduces the energy usage and network traffic. Prototype nodes are physical or virtual. Physical prototypes comprises several technologies including the small Raspberry Pi computer, the Sleepy Pi, LoPy and FiPy microcontrollers, and the Intel Movidius neural compute stick. Networks include WIFI, LTE, LoRa, and NB-IoT. Programming languages and libraries include Python, Rust, Go, and Tensorflow. Physical prototype nodes are deployed either to the arctic tundra, locally around the Tromsø area, and in the laboratory. Ten observation nodes nodes were deployed to the Varanger peninsula. They were embedded inside boxes with wild-life cameras being part of multi-year climate experiments. These camera traps are placed onto the tundra in summer, and spends the winter covered by snow and ice taking pictures of rodents entering the traps. The observation nodes presently measure CO2, temperature, time, and the number of images taken by the existing wild-life cameras. The nodes typically wake up from deep sleep every 30 minutes, do observations, compute the next wakeup time, and go back to sleep again. The nodes report the data every morning. The nodes went through several updates to take care of failures, including issues with reporting of the data, corrupt file systems, and too large energy consumption. Virtual prototypes are processes run on a PC with a GPU and on a high-performance computer with thousands of cores. For the virtual prototypes, simulations are done to explore how they behave in a range of situations.

This interdisciplinary project will for the first time provide for in-situ observations and ubiquitous data and services covering the arctic tundra that scales with the size of the observed area, the resolution of the observations, and the volume and freshness of data. This is a direct response to the Climate-ecological Observatory for Arctic Tundra (COAT) science plan stating that the circumpolar arctic tundra is the earth's terrestrial biome most challenged by climate change, but that there presently are too few observations of the arctic tundra. Therefore, there is a high demand for establishing scientifically robust observation systems to enable timely detection, documentation and understanding of climate impacts. The arctic tundra is a demanding region with severe weather, low temperatures, limited network services and energy, and often being physically inaccessible. This project advances the state of the art for cyber-physical systems being exposed to such extreme conditions. The Distributed Arctic Observatory is a novel next-generation scalable, energy sensitive, configurable, and robust observation system enabling many in-situ observations at high resolutions and at many locations throughout the arctic tundra, and with services making the data available and explorable by researchers and the public. There are many challenges facing such a system. The in-situ observation units must be made autonomous so they continue operation despite network limitations, faults, failures, and malware. Observation units have to use their limited resources in an energy sensitive way. Especially the analytics processing to find interesting objects in the observed data requires increased energy efficiency. To be useful in practise the system must be adaptable to new needs, and provide for access to data and for practical analytics and visualizations.

Publications from Cristin

No publications found

No publications found

Funding scheme:

IKTPLUSS-IKT og digital innovasjon