This document summarizes a presentation about using remote sensing data and analytics to guide tree management decisions for a utility company. It discusses how the company used LiDAR and imagery data to develop models predicting tree presence, height, and risk. It also describes how they assigned criticality scores to transmission lines then combined this with the tree risk predictions to create priority rankings to optimize vegetation management work. Going forward, the company plans to enhance their models with additional data and feedback from ongoing work.
development of diagnostic enzyme assay to detect leuser virus
Panel Taking action on data-driven insights for Vegetation Management
1. Guiding Tree Management Decisions – A data Agnostic
Approach utilizing Remote Sensing
2022 Trees & Utilities
Adam Helminiak – ATC
Patrick Eisenhauer – E Source
9/21/2022
Key Points 1 - With all that could be done, how much should we do?
Key Points 2 – To identify risk trees you need also to be able to identify good tree.
Don’t become paralyzed with information, analysis paralysis, at the end of the day its about removing risky trees and avoiding outages.
We are empowering utilities to become ““Guardians of their Galaxy”
Key Point 3 – Its about creating an approach that is a flexible solution to allow ATC to evolve their program over time.
Key Point 4 – We are looking to optimize, become more efficient, and recognize cost benefits.
Privately owned by the approximately 25 utilities, municipalities and coops with divested assets.
ATC CULTURE: ONE TEAM – connected by purpose, curiosity and positive energy
ATC business imperatives align with our program direction
ATC Business Imperatives
Manage Risk – compliance, forced outages and storm response, safety, public relations
Manage Budget – continue to find efficient and effective ways to manage
Dev People – challenge and provide opportunities to succeed
Explore technology, challenge the status quo, use data to inform decision making
About 10 years ago ATC hired vegetation specific personnel to support the VM program
That was really the start of our VM program journey which has evolved tremendously in the last 10 years
We started like many programs – high reactive/unplanned work; clearing the easy to recognize lines in poor condition
We now have 5 VM Specialists and a Manager of VM who oversee 6 contractors and about 150 contract employees
We’ve cleared about 85-90% of our lines – We do choose to manage a more aggressive clearing strategy using IVM tools to create a compatible ROW
We’ve made significant progress with clearing our ROW’s to specification but completing the right work at the right time is more challenging and requires more than head knowledge to prioritize more holistically at the system level
Diversity in tree types, growth rates, easements, topography vary across the system and
Our next step in program maturation is technology to enhance the program
Most of us have the largest line item in the maintenance budget and when you have close over grown trees everywhere it’s an easy justification but we’ve found as the VM program matures technology has helped
Prioritize work
Fiscal responsibility
Demonstrate we’re completing the right work at the right time – data driven decisions
VM has a lot of data - but we really had limited ways to use this information
Very manual manipulation of the data
Technology can help us turn data into information
How is ATC becoming data driven
Today we’re focusing on our E Source initiative but keep in mind it’s not just a single technology solution and many times these initiatives feed each other
Other examples of technology to enhance the program
PowerBI – good news is we collected data for about 15 years – scheduling, costs, budget breakdowns (planned vs. unplanned) but we needed a tool to turn this data into something useful
LiDAR initiative started last year and expanded to our FAC applicable lines serving uses of vegetation clearances as well as cycle optimization
Modeling and predicting vegetation growth
AGO
Vegetation Management System for work planning and accessible in the field down to the crew level
Survey123, Quick Capture
This initiative with E Source
We are a management company and we need to leverage our personnel along with industry professionals to help find what works best for us
Leverage those working with other utilities – E Source happened to be the experts we leveraged with this specific initiative
The technology needs to be the right fit and how you get there needs to be the right fit
Key Point - Mapping tree presence system wide is critical to providing a decision framework and approach. Explain.
Key Point - Analytical Design Step 1 – Cast points, at 1m spacing, across the right-of-way extent system wide.
Key Point - Analytical Design Step 2 – Attribute points with relevant variables essentially creating a training dataset.
Key Point – Analytical Design Step 3 – Use variables to predict tree presence and height at each point across the system.
End Point – This is a hierarchical modeling approach, central to spatiotemporal modeling with two components (1) Tree Presence (2) Tree Height. Often multiple models are deployed across service territory, with different variables and multiple sources / types of imagery allowing for the best predictions in specific locations. Some sources of imagery might be better for height and some for presence.
e.g. cloud cover and cornfields.
Generate points at finite resolutions (e.g., 1m spacing) across the territory extent.
Each point can be characterized by:
Tree Presence Yes/No
LiDAR derived height
Satellite imagery
Aerial Imagery
Slope, elevation, terrain
Infrastructure conditions
…
Employ statistical methods to predict the LiDAR derived tree presence and heights
Why are we doing this. We don’t always have LiDAR everywhere. And LiDAR is expensive.
Developed an imagery-based model using public and ATC captured LiDAR data to develop it system wide.
LiDAR is expensive to procure, do you really need it everywhere?
Key Points – Ecologically Specific
Key Points – Using best available data
Key Point – Fairly northern, had to be careful of imagery sources used during spring capture. Snow can impact model performance.
Key Point - Utilizing best available data across the system.
Candidate variables. We narrow these down via analytical works to get at the key model drivers.
Best
Not all Lidar is the same.
If I was standing in front of a tree example.
SO, how well are the models performing.
Not perfect. Reminder that LiDAR is not perfect either.
Goodness of fit metric for continuous variables.
Goodness of fit metric for yes/no is AUC.
Perform cross validations – Train Test
Key Point - 35Million Points
We have used generalized model risk.
Don’t have enough outages to train on like we do on distribution.
Data Capture - So, we setup a sampling design.
Prediction – ACU < 0.80
This is how our outage risk analytics work.
Crews -
Identify Red Canopy
Captured Why
Sampling can be performed and would improve model performance.
Risk Tree Predictions were estimated using point process models (32 million points) across ATC’s entire Transmission System (10,025 miles)
Geo-spatial datafiles at multiple resolutions (point, span, circuit) are delivered to support ATC’s vegetation management and mitigation planning (as a risk prioritization tool)
As you can see, there are tons of points generated. In this example, we display points with predicted heights (blue points are short, red points are tall)
We need to take this immense amount of data and aggregate it to deliver actionable tree analytics
One potential spatial aggregation method is tree canopy polygons containing the heights of detected trees
In this example, white polygons represent short trees and red polygons represent tall trees
Another potential spatial aggregation is at the span level, or each section of conductor from pole to pole.
Here is an example spatial aggregation of the total number of trees per span
Blue lines contain the least number of trees, red lines contain the most number of trees
We can scale this up further and spatially aggregate tree presence and height at the segment level as well, or even broader spatial resolutions
In this example, Blue segment sections have the least number of trees, and red segments contain the most trees
Because we
Key Point - Imagery provides a great advantage due cost and speed to acquisition / delivery
Key Point - Thinking beyond tree presence and heights. IE Risk and criticality.
Key Point – Nested approach provides operational power and each level of the organization.
Expert Opinion based scoring system
Score was fit 0-1
How do we tie this
Tree Risk – What is the probability of a tree caused outage.
Criticality – Given that we are going to have an outage, what is the impact
Opener Slide Number 2 - Tell the story to set the stage.
1 - High tree outage risk warrants an investigation.
2 - High Line Criticality warrants an investigation.
3 – Bad Apples
This can be done at multiple resolutions including, polygons, spans, circuits,
Key Point - Imagery provides a great advantage due cost and speed to acquisition / delivery
Key Point - Thinking beyond tree presence and heights. IE Risk and criticality.
Key Point – Nested approach provides operational power and each level of the organization.
Imagery-based and LiDAR-based estimates of tree conditions were highly correlated.
Imagery provides an advantage due to reduced cost and speed to acquisition / delivery.
Think beyond:
A single data source. e.g., LiDAR
Trees alone e.g., Risk and criticality.
Used a nested / scaled approach to provide decision support at each level of the organization.
Initially a challenge but this initiative provided opportunities to improve
Forced ATC to look at appropriate data sources – line load, sensitive customers, etc
Data sharing permissions and security
In some cases there were multiple data sources which provided an opportunity corporately to identify appropriate data sources
We were missing conductor location and had to make assumptions
Ecologically specific tree models were developed using additional data such as time of year – snow, agriculture
It took more time but more data was used to overcome this
Cost
LiDAR can be 30x cost – this won’t be used as a compliance tool!
E Source adaptable – it wasn’t that we had to have X, Y and Z to make it work – they adapted to what we had available or were able to use what we had to use where it fit best
Aligning the data and expectations with the deliverables
You don’t necessarily need LiDAR data to predict tree risk
We all have a lot going on within our routine work and yes it took work, but having the right partner can help minimize the work
Prioritization used to validate our annual work plan
Right work at the right time
AGO map is a visual integrated with our AGO map tools for field use
Our industry is changing rapidly with generation sources
We’re seeing a significant amount of solar and wind
Line criticality adjusts as system conditions changes – such as new generation added
This helps tie VM objectives with system operation objectives
We anticipate this data be used during VM patrols and work planning
We can’t 360 degree inspect every tree across 10K miles but we can use the tool to help narrow those areas of higher risk
We still have some work to do
Supplement LiDAR with the models as it becomes available
Ground truthing will help refine the accuracy of the models which was not in scope with this phase of the project
Update models as data changes – work, outages, line risk
Closing:
There isn’t one single solution, and it needs to fit your program needs
Don’t be afraid to start this journey because you don’t feel you have the right information, data or resources
Many of these solutions area flexible, adaptable
Start your technology journey by leveraging the experience in the industry – but start that journey