EPRI Proposal: Monitoring Vegetation Encroachment Along Electric Power Transmission and Delivery Lines

Monitoring Vegetation Encroachment Along Electric Power Transmission and Delivery Lines With Remote Sensing Techniques

A Research Proposal Submitted to: Electric Power Research Institute Palo Alto, California Submitted by: Regents of the University of California Davis, CA 95616

EPRI Vegetation Monitoring and Maintenance Scheduling Algorithm

Principal Investigator:

Susan L. Ustin
Associate Professor of Resource Science
Department of Land, Air, and Water Resources
University of California
Davis, CA 95616
PH: (530) 752-0621, FAX: (530) 752-5262
email: slustin@ucdavis.edu

Introduction:

Power utilities have a continuous job to monitor integrity of the electric power supply system. The extensive network of powerlines, nationally and in California alone must be continuously monitored and maintained to ensure power delivery. These lines are subject to many sporadic natural disasters that can and do disrupt the delivery system, e.g., floods, earthquakes, severe weather, and human interference. Of the natural disasters that are of continuous concern for disaster management and mitigation response, vegetation encroachment on power right-of-way is one of major concern. Vegetation can grow into power lines allowing branches to contact the lines, trees or branches may fall on the lines, possibly breaking lines, or causing them or the transmission poles to collapse or start fires. The California Depart Fire Protection database shows that three of the 20 largest wildfires based on the number of structures (homes and other buildings) lost were caused by powerlines (City of Berkeley, 1923, with a loss of 130 ac. and 584 structures; Laguna, in 1970, with 275,425 ac. and 382 structures lost; and Riverside in 1993, with 25,100 ac. and 107 structures lost). For comparison, the long-term California average wildfire loss is 16.2 ac (http://www.fire.ca.gov). These fires may be large also as evidenced by the CDF list of 20 largest wildfires which includes the Laguna fire, Clampitt (Los Angeles, Co.) at 105,212 ac. (86 structures), and PG&E (Glen Co.) 1965 fire which lost 71,945 ac. Thus, the industry faces considerable liability when powerlines cause fires in addition to their direct losses. Alternatively, if a wildfire starts, dense vegetation under or near the power lines can cause the fire to burn into the lines. Conversely, without any low vegetation ground cover under or near lines located on steep slopes and hillsides may allow heavy winter rains to cause mudflows and landslides also endangering the power delivery system. Therefore it is not just absence of vegetation in a buffer around power lines that is needed but an absence of trees and shrubs with the presence of low growing, low biomass, herbaceous vegetation. The current method used by the power companies is to assign personnel to directly observe the lines (e.g., by walking or driving) to locate sites of vegetation encroachment. This is expensive, time-consuming, and subject to observer bias or a failure to observe trouble spots at the right time. An automated or semiautomatic monitoring method using remote sensing could help to identify potential trouble spots and dispatch crews more efficiently and cost-effectively, thus, saving resources and providing better maintenance service.

Despite the economic costs related to vegetation management around powerlines, an exhaustive survey of scientific publications, exceeding 2000 journal and proceedings entry’s in several citation databases, failed to yield any publications directly addressing the use of airborne photogrammetry or remote sensing to monitor powerlines or of evaluating powerline management and wildfires. There were numerous citations for evaluation of herbicide, mechanical, and other treatments for vegetation removal and suppression under powerlines. While the EPRI database includes relevant studies related to monitoring by remote sensing techniques, these publications have been unavailable to us. Despite this, based on the general ecological and agricultural literature several conclusions can be reached.

In the past aerial photography has been a key component of most natural resource monitoring systems. However, manual interpretation of photos is also labor intensive and slow. The methods employed by visual interpretation of aerial photography have some advantages for detection of vegetation but some clear disadvantages also. First, the spatial resolution is generally high making it possible to readily observe vegetation growing under utility lines. Aerial photography is at significantly higher spatial resolution than existing spaceborne sensors, like Landsat Thematic Mapper (TM) at 30 m pixel resolution and SPOT (Systeme Probatoire d’Observation de la Terre) at 20 m multispectral resolution and 10 m panchromatic band resolution. If utility right-of-ways are generally 75 ft on either side of the utility line, this provides only a two pixel Landsat TM buffer on either side of the powerline for determining vegetation encroachment.

Second, since aerial flight lines can be reflown it is possible to get overlapping photos making canopy height determinations possible. While stereo imagery is possible to obtain from SPOT data present evidence indicates that the vertical resolution is too coarse for adequately determining canopy height from this satellite data. Lastly, aerial flights can be scheduled to obtain data at the optimal contrast between vegetation and surrounding environmental conditions. On the negative side, an argument against the use of photogrammetry is the fact that high resolution requires that many photos be manually analyzed and interpreted with no easy application of automated methods. With 10’s of thousands of miles of utility line in California, this is not a small consideration. Registration of photos to ground coordinates for entry into a GIS is also not trivial since low flying planes have significant yaw, pitch and roll problems that must be rectified. Satellites offer the possibility of obtaining frequent acquisition of image data over sites and to be able to quantitatively compare changes within and between datasets. One of the major advantages is to be able to mathematically analyze data to enhance and isolate vegetation information from other scene dependent sources of variation.

The next generation of high resolution spaceborne and airborne remote sensing systems will have spatial resolutions and multiple spectral bands that have the potential to provide digital information at a resolution better suited for monitoring this problem. Thus, it is timely to begin to develop applications that can be transferred to the commercial sector for improving monitoring assessments and making them more cost efficient and timely. Several commercial aerospace companies will offer one-five meter resolution satellite imagery and other airborne photogrammetry companies will offer sub-meter to several meter digital photographic and image data. Lockheed will launch the first of these satellites, the Space Imaging sensor, in December 1997. While the potential to supply spatial information in a timely manner with these sensors is clear, methods to analyze and deliver the data need to be developed. Optical sensors can detect green plant foliage based on its strong contrast with soils and geologic materials in the visible and near-infrared spectral region. Any analysis of vegetation encroachment in powerline right-of-way would utilize this information. While this information is of secondary importance, it can aid in defining the type of vegetation present (and assist in determination of the potential for vertical growth into the overhead powerlines) and in evaluation of fuel buildup for ignition and wildfire spread. However, the spectral distinction between plant litter and stem material from soils is not as easily separated in remote sensing data as is green (foliage) from soil. Because of community phenological differences, acquisition of satellite or airborne data might require more then one data acquisition to fully evaluate vegetation encroachment. A possible alternative to multiple image analysis methods is to use other spectral information or ancillary information in a GIS. Further, because one issue of concern to the power industry is to separate low stature vegetation, like grasses or herbaceous vegetation, from larger trees or shrubs that presents greater danger to the utility system. Both spatial and spectral band selections need to be evaluated to determine optimal methods that are efficient and cost-effective management tools.

Over the years, a variety of methods for processing remote sensing images have been standardized. These include simple display of false color images, band ratio images, including vegetation indices like the Normalized Difference Vegetation Index (NDVI), multi-band classification algorithms e.g., maximum likelihood methods, and spectral mixture analysis methods. Newer techniques include advanced mixture methods like use of wavelets and neural networks. In some cases, tradeoffs between spatial and spectral resolution can be made, and coarser spatial images may produce the most cost-effective methodology. In other cases the advantages of multiple spectral bands do not outweigh the advantage of high spatial resolution of a red and near-infrared band pair. This research will examine these possibilities and will recommend a strategy for data analysis to give maximum efficient protection of the utility corridor.

Objectives:

1. Immediate goal of the proposed research is to create a methodology for monitoring vegetation encroachment on power transmission and delivery line system using modern information technologies, like digital aerial photography or satellite imaging, geographic information systems (GIS) and location of field sites using global positioning satellite technology (GPS). The study will utilize existing spaceborne and airborne remote sensing data to explore the application of different spectral and spatial scales for monitoring risk for vegetation encroachment.

2. Longer-term goal is to develop near real-time automated or semi-automated monitoring methodologies for "just in time" management to reduce vegetation related natural hazards, like wildfire, erosion, and blown-down trees and limbs.

Approach:

EPRI and its member utility will determine the study site selected. Every attempt will be made to find a location for which image and airborne sensor data already exist or where a contractor can supply airborne data to EPRI. Commercial vendors for digital airborne imagery operate in the San Francisco Bay area and elsewhere. NASA Ames Research Center, Moffett Field, CA maintains a public domain library of all airborne data acquired by NASA in California. Other sites within the state operate public databases and the power utilities also maintain databases. It is likely that the study can be performed using existing data from one or more of these sources. Figure 1 (106k) The study will be done using GIS and image processing methods. The GIS database will be provided by EPRI Utility Company partner and will be from a site in California. The service area chosen will probably be located in the central California coast range or Sierra Nevada region. The analysis will use GIS to determine powerline location within the image datasets, to coregister data to a base map, and to provide a comparison to DEMs and existing vegetation maps. The images will be registered to base maps for topographic comparisons, superposition of power lines, transmission and delivery poles, existing vegetation maps, and other factors of interest to the member utility and EPRI. We recommend starting on a Mokelumne River watershed drainage because of PG&E’s long-standing interest in the area, and because of our existing database on the area, including high resolution DEM, aerial photography and several years of digital satellite imagery, digital soils data, and management history. The advantage of this choice would be to limit the effort that must be put into developing the database and allow all of the funding from this project to directly apply to addressing the spatial and spectral resolution questions relevant to EPRI. Figure 2 (46k)

1. Map locations of vegetation cover in proximity to power lines, provide a map of vegetation type (and height potential) and estimate of current vegetation height. We propose to develop and test image processing methodologies that will explore use of several vegetation indices, based on reflectance differences between red and near-infrared spectral regions, from normalized vegetation indices (NDVI) to those atmospheric/soil adjusted vegetation indices. These indices are indicators of variation in green vegetation. This information will provide a map of vegetation cover variation. For values of low leaf area index (less then 4.0) as are expected for maintained right-of-way locations, canopy vegetation indices are expected to increase linearly or near linear with increasing biomass. Thus, these indices may provide a satisfactory indication of canopy distribution and a quantitative or qualitative estimate of amount. If additional bands are available (e.g., with TM) we will use spectral mixture analysis (SMA) to determine the vegetation density and percentage canopy cover. SMA has been used to estimate sub-pixel mining when the objects of interest vary at scales smaller then the pixel. One advantage of this technique is the estimate of the litter and woody biomass fraction. An additional benefit is the shade/shadow estimate, which has been used to estimate canopy biomass and height under full cover conditions. Thus, SMA may allow development of a hierarchical monitoring procedure, where SMA is used on TM data (with 30 x 30 m pixels) to identify possible problem sites. These sites would be targeted for another examination with a higher resolution sensor (1-5 m range) or field visits. Thus, making an efficient procedure to locate possible remediation sites. Mature canopies are often characterized by increasing gaps and roughness in the canopy and this creates deeper shadows within the canopy. Spectral mixture analysis is well suited to characterize this residual shading and we will determine if this will provide better estimates of canopy height then simple thresholding. Images will be analyzed at full resolution (for the number of spectral bands available) and the analyses degraded spatially to determine the impact on interpretation.

Other image analysis approaches for vegetation mapping and classification can be applied but the specifics depend on the choice of sensors. EPRI can be consulted for approval on the final work plan based on the appropriate methodology for the image data sets available. Depending on the number and placement of spectral bands, vegetation classification techniques and/or texture analysis techniques may provide a basis for inferring canopy height from images. Registration of multiple images acquired at different sun-view angles may allow estimates of height or thresholds where canopy height is indicated as of concern.

2. Create baseline data management map for the selected utility service area. One goal of the project is to develop a methodology for change detection and developing an appropriate baseline with which to compare changes. The approach we have taken is to first establish confidence in the image predictions for vegetation encroachment based on field recognizance of sites selected from the imagery. Subsequently, sites where vegetation condition (i.e., the cover extent and vegetation type) are considered "good" for soil stability and of low stature will be identified in the database. These sites will become targets for determining future vegetation condition goals. The set of these "good condition site indicators" can be expanded over time as different areas along the utility lines move closer to the optimal condition. This type of study can be done with image to image comparisons and the results stored in a GIS database. The update methodology to compare current vegetation distribution against a "control" condition, i.e., that which vegetation distribution presents a low risk for electric power distribution needs to be developed. The optimal goal is to create a reliable "just in time" management plan.

3. Evaluation Criteria. The project will use field methods and photography to gain confidence in the models and to statistically evaluate their robustness. The project will use field sampling and GPS-located ground-based photography, and any high-resolution aerial photography available to evaluate and validate image analysis. Field measurements will also include samples of common materials (plants, soils, and geologic materials) within the region to develop spectral training sets for analysis and interpretation of airborne and satellite imagery.

1998, Center for Spatial Technologies and Remote Sensing (CSTARS)
University of California, Davis