Pilot: How can cheaply available sensor technologies support our climate change resilience?

Collaboration between K2 Dronotics and Luminant Data, April 2018

Thin Ice: Mapping Safe Arctic Alaskan Ice Roads

Original Pilot Proposal, March 2018

2018 has brought a year of unprecedented low sea ice extent in the Arctic, along with thinner ice. Thin ice can cause ice roads to be dangerous or even unusable. A team of two Alaskan-run companies specializing in UAV surveying and statistical modeling propose a low-cost scientific investigation capturing multi-sensor data that correlates with ice safety in the Chukchi Sea in April 2018. Our goal is to map meaningful metrics of ice safety with survey data acquired via UAV-mounted sensors, supplementing indigenous knowledge to aid the safe navigation of ice roads.

The Need

Throughout long winters, ice is utilized as transportation infrastructure for people, goods, and animals in Arctic Alaska. Decreases in thickness and extent of ice across the Arctic cause significant constraint of local travel  routes, subsistence living, and community connections. As 2018 ice extent in the Bering Sea and Chukchi Sea reach es all-time lows, the toll is more significant than ever -- isolating communities, disrupting supply routes, failure of structures, and even causing deaths. Whether ice is safe to traverse on foot or vehicle is determined by thickness among other factors [1] . Thickness is critical to know, yet arduous to measure.

Arctic sea ice reached its annual maximum extent on March 17, according to the National Snow and Ice Data Center, with cover at 5.59 million square miles. It's the second-lowest cover on record and just 23,000 square miles more extensive than the record low, which happened last year. The four lowest maximum extents on record have happened recently, in 2015, 2016, 2017 and now, 2018.

Local residents across the Arctic have been dealing with hazardous traveling conditions for weeks now. Some have had to cut short trips to camp for fear the early overflow would make it too dangerous to wait. Others have postponed traveling altogether, not wanting to risk getting stranded by the warm conditions.

- Outlook shows no break in the heat , The Arctic Sounder, March 30, 2018

Overall ice extent is extremely low, approximately ½ to ⅓ of km^2 of prior year coverage

https://twitter.com/alaskawx/status/972901359279267841

Quality and Availability of Information
Alaskan communities typically use word of mouth and communication technologies to disseminate information about the conditions and safety of travel routes. In Kotzebue, KOTZ radio is a critical communication channel for weather and climate conditions that impact when and where people decide to travel. Weather information is sourced from government weather stations, forecasting, and local anecdote. Reporting includes typical weather measures (temperature, wind, precipitation, barometric pressure, etc), but does not present quantitative local reports on the safety of travel routes. Information about ice safety, if available, originates from anecdotes or occasional physical measurements.

Climate change, data, visibility
While scientists quantitatively observe significant changes in ice extent, governmental funding for science and meteorology is declining. For many years, coastal Alaska has suffered the consequences of global climate change, particularly in the form of increased storms, precipitation changes, permafrost melt, water current changes, erosion, and animal habitat elimination. At minimum, this impacts 7.8% of Alaskans (57,664 people) who live on the Bering or Chukchi Seas. Having the longest coastline of any state in the US, Alaska is particularly connected to the ocean and its changes, but is far less visible due to remoteness. In the last year, critical deployments of weather telemetry supporting weather predictions have been suspended in many parts of Alaska . With less of this data, climate conditions along the Alaskan coastline become less visible to scientists and the public.

Better ice monitoring with inexpensive higher quality data
In recent years, scientists monitoring climate change and arctic ice have utilized sensor systems and data processes that are expensive, offline, decentralized, and intermittent. This research is not specifically focused on supporting community safety, and thus has also not focused on the localized scale. This project team specializes in technology that improves upon legacy approaches in terms of efficiency, fidelity, quality, and cost (93.7% cheaper). These technologies present a new opportunity to utilize high quality data to predict the safety of local travel routes from as-needed surveying, open source data modeling, and accessible hardware.

Technology in Arctic Alaska

We imagine a  future where communities will be able to map safe travel routes automatically and as needed, from data collected with publicly-owned hardware, leveraging access to internet-based maps and predictive technologies. As broadband reaches even the most remote  Alaskan Arctic coastline , connectivity has exposed the opportunity to provide high quality, on-demand weather information and specialized tools over the internet. Our opportunity is to extend this even further with  new technologies, building hyper-localized predictive measures of weather.

Aerial view of Kotzebue, March 2018

Photo by Pong Decrito of Anchorage


Proposal

In order to enable  safe navigation of changing ice-based travel routes, we propose an investigation of ice safety at the local scale. Using UAV (unmanned aerial vehicle) hardware, sensor technologies, spatial data modeling software, and data analysis methodologies, the team will map and collect data of ice roads originating in Kotzebue, Alaska with a variety of sensors. Using direct and indirect  observation, the team’s goal is to  determine if non-contact methods can produce data that will effectively map areas of unsafe ice.

This team will collect data along travel routes surrounding Kotzebue, Alaska over 3 days, process resulting spatial data, and finally publish and present findings to stakeholders and the public.

Hypothesis: It is possible to map meaningful metrics of ice safety with survey data acquired via UAV-mounted sensors.

Investigation Process

  1. Interview  locals and travelers   to identify sites impacted by ice safety changes (Qualitative)
  2. Capture data  of potential correlative measures with ice safety (Surveying)
  3. Correlate predictive measure with target to determine how non-contact data predicts safety (Statistical Analysis / Learning)
  4. Build travel safety map from data (Mapping and Geostatistical Analysis)
  5. Journalistically present and publicize impacts of changing ice safety on less-visible rural Alaskan communities

Technical Strategy

Approaches to validating the Hypothesis

Data Strategy : Correlate ice thickness measurements (physically collected “target” data) with inexpensive UAV survey measurements (no-contact “predictor” data) to determine whether ice safety can be established inexpensively by portable aerial vehicles.

Strategy : Using multiple different sensor systems, we will investigate the correlation of UAV sensor data with samples of ice thickness. This will result in a learned relationship between the predictor and target variables. This in turn determines acceptance of the hypothesis.

where the learning problem is defined as:

y = P(a)

y  is the direct measurement of ice thickness

a  is the predictor variable from UAV based-sensors

P  is the learned relationship between y and a

1. [Requirement]

Target Variable:  Deterministic measure of ice thickness

Strategy : Take samples benchmarking/calibrating thickness (base of correlation)

2. [Requirement]

Predictor Variables : Data likely to correlate with target data

Strategy : Collect data from sensors that correlates with thickness of ice, such as presence of overflow, water, cracks, heat, density of snow or ice

3. [Methodology]
Statistical Analysis, aka “Data Science”

Strategy:  Correlate target and predictor variables to determine relationship

4. [Methodology]

Visualize trends

Strategy : Geospatially map collected data, correlations, and photography

Advantages: New Technology and Cost

Electronic technologies are known to decrease in cost as production costs decrease, scale increases, standards emerge, and platforms become commoditized. For this particular project, the change in surveying cost from only 8 years ago is extreme. UAV surveying has never been more financially accessible and convenient. These costs are expected to decrease even further as a result of software development specifically supporting UAV surveying data analysis.

In short, UAVs and open source software generally provide a 93.7%  reduction in cost of surveying technology. This difference is even greater in Arctic Alaska, where fuel and aircraft costs are always higher than average.

New Technologies

Traditional Technologies

Advantages of New Technology

Cost Change

“UAV” Unmanned Aerial Vehicle

Owned

Helicopter, Plane Rental

Inexpensive (comparatively)

Closer to ground (~10 ft)
Relatively small (luggage)

No risk of human harm

$1k/hr [2]  → cost of UAV ($500-$5,000)

GPR Sensors (2018)

GPR Sensors (2010)

Cheaper (Moore’s Law)

Lighter

Faster

$30k [3]  → $400 [4]

Open source statistical software

Proprietary statistical software

Open source

Widely used and tested

Transparent source code

$2k [5]  → $0

Open source geospatial software

Proprietary geospatial software

Open source

Widely used and tested

Transparent source code

$2,500 [6]  → $0

Cloud data storage (2018)

Cloud data storage (2010)

Cheaper

Diversified and new products

(Amazon S3)

$0.110/GB/mo [7]  → $0.023/GB/mo

Cloud hosting & compute (2018)

Cloud hosting &  compute (2010)

Cheaper

Diversified and new products

(EC2 m2.4xlarge)

$2.00/hr [8]  → $0.24/hr

Estimate A: Traditional Technologies

Estimate B: Assume rental of GPR unit in 2010

A $38,551

B $14,551

Estimate: New Technologies

$910

Cost difference (not inclusive of human labor)

Example: 4-hour land survey, 20 hours compute, 100 GB data

Estimation Method (sheet)

2.4 - 6.3%

of 2010 costs

93.7%  cost reduction

Results

Beyond a validation of the hypothesis, the following will result from the project:

Kotzebue airspace, March 2018

Project Timeline

March / April 2018

Phase

Dates

Outcome

1a. Calibration and Data Testing

2 days

Calibrate new sensor system. Survey ice near Anchorage, river and ocean. De-risks issues in onsite. (UAV)

1b. Site pre-identification

5 days

Identify potential sites for surveying with indigenous residents and climatologists based on travel routes and safety concerns. (Journalism)

Some locations contingent on FAA approval. (UAV)

2. (Onsite) Surveying and Interviews

3 days

April 14-17

Onsite conversations with locals, hand-mapping routes (Journalism)

Survey identified sites (UAV)

3. Data Processing

2 weeks

Process and produce maps of collected data (Data processing)

Statistical processing, correlation, metric-based validation (Data Science)

4. Journalism

3+ weeks

Write narrative and conclusions for a) lay audience and b) scientific/tech development report (Journalism)

5. Presentation

4/18+

Publish, present findings

Project Team

Luminant Data is an Artificial Intelligence Consultancy specializing in rapid feasibility testing and tech adoption life cycles for custom-built machine learning products. Founder: Clare Corthell clare@luminantdata.com  

K2 Dronotics is an Alaskan drone UAV service company specializing in engineering inspections, custom payloads, fleet pilot training, GIS, and analysis. Founders: Ben Kellie, Nick Kellie nick@k2dronotics.com

Funding

This project is uniquely accessible to the teams given their expertise in new low-cost, non-proprietary technologies, Luminant Data in statistical modeling, K2 Dronotics in UAV Surveying. The team can deliver the project at approximately 15-20x less expense than other groups, consultants, and technologies.

Minimum budget (travel + materials): $3,000

Target budget (travel, materials, nominal hours): $10,000

Future (Winter)

If the team validates the hypothesis that it is possible to estimate ice safety with survey data acquired via UAV-mounted sensors, we would look to utilize this data in the most impactful way possible: warn travelers of specific areas of dangerous ice and provide alternative routes in a timely and consumable manner.

Scientists and communities confirm the trend of a year over year decrease in ice extent, which places a greater reliance on routine surveying and mapping of safe transit routes for communities across the Arctic, not just in Alaska. We can close that gap at a low cost and provide communities with tools to safely rely on critical natural infrastructure such as ice roads.

Future Potential Hypotheses

[9]


Appendix I: Ice Thickness Safety Advisories

Source: MN DNR

Climate Change

Public Safety and Infrastructure

Scientific Data Collection

Technology for Alaska

an alaskan tech collaboration


[1] Eicken, Hajo & Mahoney, Andrew. (2015) Sea Ice: Hazards, Risks, and Implications for Disasters. 381-401.

[2]  See https://www.quora.com/How-much-does-a-helicopter-charter-cost  

[3]  See https://trenchlesstechnology.com/ground-penetrating-radar/  

[4]   See PiCoR $400 https://www.xethru.com/blog/posts/picor-ice  

[5]  Example: Mathematica $1890 https://www.wolfram.com/mathematica/pricing/industry-individuals.php  

[6]   Industry Standard: ESRII Arc-gis http://www.cloudpointgeo.com/blog/blog/2014/5/9/a-summary-of-arcgis-platform-and-products  

[7]  AWS S3, 2010 https://aws.amazon.com/blogs/aws/what-can-i-say-another-amazon-s3-price-reduction/  

[8]  AWS EC2, 2010 https://aws.amazon.com/blogs/aws/amazon-ec2-price-reduction/  

[9]   Source: InsideClimate News 2/17/18 https://insideclimatenews.org/news/17022018/arctic-sea-ice-record-low-extent-alaska-bering-hunting-whales