High-Level Project Summary
The BIFROST system consists of a perennial, constant and real-time monitoring network, composed of two parts, essentially: a sensor module and a real-time AI-using monitoring system, utilising neural networks, data analysis and registering, geolocation, geoprocessing and georeferencing. Its main objective is to predict the minimization of damage in environmental tragedies such as landslides and avalanches. Constant and continuous monitoring of data from risk areas not only allows for rapid evacuation in emergencies, but also drives the process of implementing public policies in the region regarding soil preservation and irregular housing.
Link to Project "Demo"
Link to Final Project
Detailed Project Description
Identifying risk with science + communities – The BIFROST Project
Life in society demands constant interaction between individuals and their surroundings, implicating a series of interdependence relations in which many times the aforementioned environment comes to be an active or reflexive agent, influencing social development directly.
In rural communities, wherein this interaction is especially tight, these relations are also resized, making themselves singularly impactful.
Considering landslide events, like the one that took place in 2011 in the rural areas of Teresópolis and Petrópolis, state of Rio de Janeiro, Brazil — at the time considered the biggest climate catastrophe of the country —, it becomes possible to observe the immense proportions of financial damage regarding impairment of infrastructure and the service sector and many other losses to which countless victims were subjected stemming from this type of natural disaster.
According to the World Health Organisation (WHO), landslides can be present all around the globe and are more widespread than any other geological event in the world. Following that line, data provided by the United Nations Office for Disaster Risk Reduction (UNDRR) has registered more than 4 million people affected globally and more than 18 thousand deaths due to landslides between the years 1998 and 2017.
In the case of the 2011 Rio de Janeiro disaster, repercussions were so many that it led the president of the Mineral Resources Department of Rio de Janeiro (DRM-RJ) to mention the occasion had modified the criteria for risk area classification in the country, which began to consider rural areas as also being vulnerable to risk.
On the report of Amarílis Busch and Sônia Amorim’s “A tragédia da região serrana do Rio de Janeiro em 2011: procurando respostas” — “The 2011’s tragedy of Rio de Janeiro’s highland region: searching for answers”—, available at the Brazil National School for Public Administration (ENAP) Repository, in 2010, a year before the tragedy, Brazil’s National Secretary for Protection and Civil Defense (SEDEC) had forwarded a document to the United Nations (UN), in which the Brazilian government described limitation of financial resources and operational capacity, with difficulties concerning monitoring and data dissemination about the vulnerability of its territory. Upon the incident, an authority of the planning area also highlighted the role of previous mobilisation and enablement of community leaders for the execution of contingency plans, which didn’t occur.
Presently, there is a hardship in integrating local data collection to large-scale mapping spatial intake, the most utilised resource at the moment for determining landslide risk, hindering the implementation of risk engagement policies at a local level.
As provided by the Committee on Earth Observation Satellites (CEOS), even in places where there is an amplitude of information about landslides, the greatest part of landslides caused by geophysical or meteorological event triggers goes undetected.
Arises then the need for determining with bigger precision the probability of landslides in rural areas through efficient, scalable, replicable and financially viable methodology, capable of integrating scientific data from national entities to earth monitoring via satellite and promoting dynamic and precise mapping of risk areas, in order to allow the correct planning of local infrastructure with an increased accessibility for community participation in that process.
The BIFROST system consists of a perennial, constant and real-time monitoring network, composed of two parts, essentially: a sensor module and a real-time AI-using monitoring system, utilising neural networks, data analysis and registering, geolocation, geoprocessing and georeferencing based on data provided by Worldview from NASA's EOSDIS and other agencies.
The numerous sensors of the individual probes assess data and transmit it to a central processing unit. The modules will remain permanently connected to their server via radio or WiFi signal, depending on local conditions. In case of signal interruption, or some sensor failure, the server receives an alert immediately.
The equipment is powered by lithium batteries, and recharged automatically through its own solar panel in a totally autonomous manner.
It’s also worth mentioning the low implementation and operation costs and the lesser complexity of data analysis when compared to other solutions available on the market, such as the sole use of satellite imaging or general barometric reading. Because it is both completely modular and built on the Arduíno platform (combined with transmitter modules from the Raspberry Pi platform if necessary for the coverage of bigger areas), installation is mostly effortless, sufficing to simply keep gadgets inside the range of their transmitters. Height and positioning make little to no-difference for transmission and the system sensors can be adapted according to specific terrains, risks to be monitored, or even the type of data to be collected, making it easily standardized and replicable in different environments. Costs, especially hardware maintenance, are also significantly diminished in comparison to other current monitoring systems.
The utilisation of monitoring is not restricted to protection zones or isolated environments, it can also be applied to rural, or even urban environments, assessing a variety of indicators which include immediate soil background, humidity counts, rain precipitation, temperature index, infrasonic sounds, etc. Should one of the modules detect anomalies such as expressive temperature or humidity changes, the server will immediately fire simultaneous alerts to all of the agents involved in the adopted control strategy, no intermediaries. That way, with more precise measurements, public policies for the bettering of the population’s living conditions also become more efficient and dynamic.
The many different types of data get to their own servers, undergoing machine-learning and an Extract, Transform, Load (ETL) process. Afterwards, they are stored in data warehouses, whence the Online Analytical Processing (OLAP) cubes are generated for data-mining and viewing. In turn, processed data will serve as blocks for the construction of a vast information library, which can be used in the prediction of landslides, as well as a way of assessing info about the performance of the system itself and the actuation of the various agents involved in the intelligent network’s operation.
In order to find the key-variables for the process, all of the platform data needs to undergo a transformation procedure: applying a data-mining technique, the Partial Approximative Set Theory (PAST), the system acts in the identification of significant variables.
The identified variables are then forwarded to the autoconfigured convolutional artificial neural network model. The Multi-Layer Perceptron (MLP) network has been applied successfully to the solution of many complex problems by way of its supervised training, which uses the backpropagation error algorithm, based on learning through auto-correction. The incoming signals are propagated, layer upon layer, until the outgoing vector is obtained in the last layer.
The architecture of an MLP network consists in the topological arrangement of Neuron-Processing Units (NPUs), the respective weight values associated with their connections. These synaptic weights are adjusted so that the real network response is approximated to the desired response. Due to the complexity of the situation, there arises the need for a mono-objective optimisation by means of a mono-objective function or a multiobjective function; the problem of mono-objective optimisation having been solved by the Multi-Particle Collision Algorithm (MPCA), with the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) used for multiobjective optimisation.
Finally, BIFROST’s graphic interface exhibits input from each of the sensory probes (or sensory modules) in real-time. By applying data, the system can predict if landslides are in fact occurring or not in designated areas. Simultaneously, data is constantly transferred to a server, which in turn processes and cross-references captured information with other data banks, such as satellite monitoring, or regional meteorological records. Furthermore, convolutional neural networks are specialised in the identification and detection of objects in imaging, such as provided by these external data banks. By means of cross-referencing with government reports, for example, regional demographic info is obtained, allowing the system to quantify potential human losses and predict the impact of environmental catastrophes, including their many financial, personal, natural and other implications.
Ideally, this procedure will serve to delimitate local public policies for soil management in risk areas, additionally allowing the confrontation of quantitative and qualitative data concerning eventual human interventions in said environment, such as earthings or dewaterings, making it possible for local communities and governments to view, interpret and include info, whilst also being able to analyse decision-making results.
Integration with local communities is essential, and one of the biggest challenges for nowadays most used monitoring and prevention models. In the proposed model, such communication is stimulated: communities belonging to monitored areas have access not only to the collected material via phone application, but they are also alerted in real-time in case of imminent risk, being able to register potential landslide signs or other hazardous situations, as well as confirm or dismiss false alerts via the same platform.
The possibility of constructing a broad historical record of environmental conditions is essential to action-planning and decision-making, so that the integration with local communities becomes not only a mere strategy for fighting the direct effects of landslides, but also a tool for environmental education.
BIFROST is quite a suitable tool, easily read and of quick response in the proper assessment of landslides.
The BIFROST system, as well as its code, can be made openly available for public consultation as a whole through a GitHub repository. It is also possible to provide an Application Programming Interface (API) which enables state agents and researchers to access data generated by the system, including its efficiency details, giving room for upgrades.
Space Agency Data
We were inspired by data provided by Worldview from NASA EOSDIS and other agencies such as the European Space Agency and Jaxa Global Rainfall Watch, which in our project are compared with data obtained on land, thus arriving at much more accurate and reliable results.
Hackathon Journey
It was a unique experience to interact with several students and professionals from various fields of science, in favor of a better and safer society, and a more balanced environment.
We chose this challenge because our country is regularly affected by natural disasters of this type, and we want to contribute to improving this scenario.
Through these two days of challenge, we learned to realize that everything is connected, just as the "Bifrost Bridge" from mythology connects the nine worlds.
Also, we would like to thank NASA and all agencies (and all your collaborators), that made their data public for this challenge.
References
Busch, Amarílis, e Sônia Amorim. “A tragédia da região serrana do Rio de Janeiro em 2011: procurando respostas.” ENAP Casoteca de Gestão Pública, 2011.
Busch, Amarílis, e Sônia Naves David Amorim. “A tragédia da região serrana do Rio de Janeiro em 2011: procurando respostas.” Escola Nacional de Administração Pública (ENAP), 2011.
Comittee on Earth Observation Satellites. “Landslides.” Comittee On Earth Observation Satellites. s.d. https://ceos.org/ourwork/workinggroups/disasters/landslide-pilot/ (acess in October 3rd 2021).
Copernicus Open Access Hub . s.d. https://scihub.copernicus.eu/ (acess in October 3rd 2021).
Jaxa Global Rainfall Watch. s.d. https://sharaku.eorc.jaxa.jp/GSMaP/index.htm (acess in October 3rd 2021).
Landslides . 2021. https://ceos.org/ourwork/workinggroups/disasters/landslide-pilot/ (acess in October 3rd 2021).
Landslides. s.d. https://www.who.int/westernpacific/health-topics/landslides (acess in October 3rd 2021).
World Health Organization. “Landslides.” World Health Organization. s.d. https://www.who.int/health-topics/landslides#tab=tab_1 (acess in October 3rd 2021).
Worldview. s.d. https://worldview.earthdata.nasa.gov (acess in October 3rd 2021).
Tags
#bifrost #bridge #NineRealms #earth #sea #sky #wind #landslide #allconnected #mythology
Global Judging
This project has been submitted for consideration during the Judging process.