The vegetation score was determined to follow a letter grade of A, B, C or D based on three-dimensional distances to conductor of 9+ ft (A score), 6-9 ft (B score), 3-6 ft (C score) and 0-3 ft (D score). In other words, the highest-risk segment due to the closest encroachment was a D score based on the risk of vegetation contact of 0-3 feet. The Priority score (P1, P2, P3, P4) is designed to accommodate for the incorporation of the vegetation score (A, B, C or D) combined with priority variables like population, criticality of the area served and even wildfire risk, using a Wildfire Risk model.
Scoring is further enhanced by analytics, which can include things like maintenance data to refine an initial score. For example, a D score, once maintained in the field, can be reverted to an A score in the application. For each of the scores, especially those scores deemed highest risk like D and C vegetation condition scores, a monetary value can be assigned. All this data is hosted, analyzed and displayed with a tool set called IBM Environmental Intelligence Suite.
Early Results
In the first two years of the project, PEC began a well-thought-out plan to make use of high-resolution spectral satellite data for vegetation encroachment scoring. The co-op also began an effort to tag roughly 200,000 tree species across its overall service territory. Not only would targeting areas for high-priority work due to potential encroachments help transition PEC from the cycle-based to condition-based approach to maintenance, but the cooperative also wanted to know where certain species were located to better plan for removals of hazardous trees. This transition allowed PEC to create a program where some of the maintenance on species with timing considerations like oak wilt could be deferred to the most appropriate time of year.
When PEC first introduced the technology and satellite data, the co-op immediately saw the value of it. With satellite maps and AI technology, PEC was able to move away from a cycle-based program to a targeted approach. This allows the co-op to see the entire area and create a plan to proactively prune trees before they create an outage.
Challenges to the Early Technology
One problem with satellite data is the two-dimensional nature of the imagery. This leaves some gaps in accuracy when a high-precision KPI is required to direct millions of dollars of maintenance budget towards condition-based work.
Satellite data makes use of multi-spectral analytics to detect vegetation using Normalized Difference Vegetation Index (NDVI) and other elements and then “estimate” tree heights using AI modeling. Reasonable accuracy can be achieved with tree heights, but that is only half of the equation. To achieve high levels of accuracy, the program must predict not only the height and shape of vegetation, but also must also know exactly where the poles and conductors are located in the X, Y or Z plane to derive the proximity of vegetation to assets. The analytics to measure these encroachments between trees and conductors are executed in the GIS environment, and because satellite data cannot detect poles and conductors with great certainty, especially in overgrowth situations, some error is introduced. Even with highly accurate GIS data, as was the case with PEC, conductor heights were still estimated based on pole heights, which do not capture line sag or pole lean with certainty.
Additionally, satellite data tended to struggle with certain types of directional pruning such as “V” cuts or “L” cuts, when trees were under the lines and existing acceptable clearance was indistinguishable, making the accuracy somewhere between 70-80%. The same condition would be even more true with overhang scenarios.
