Identification Of Unpaved Roads in a Regional
Road Network Using Remote Sensing
Colin N. Brooks, David B. Dean, Richard J. Dobson, Christopher Roussi,
Justin F. Carter, Andrea J. VanderWoude, Tim Colling, and David M. Banach
Abstract
An accurate inventory of unpaved road network length and
condition within a county, state, or region is important for
efficient use of resources to manage and maintain this critical
transportation asset. Object-based classification techniques
provide a cost-effective way to identify unpaved roads within
a local agency’s road network when the road type (i.e., paved
versus unpaved) attribute is missing. We present a Trimble
eCognition® algorithm using four band optical aerial imagery
and object-based classification to classify roads as paved or
unpaved. The ruleset evaluates relationships between bands
and allows separation and segmentation of unpaved roads
from other pavement classes. The algorithm is applied to
unincorporated areas of a six county region in Southeastern
Michigan. Tree shadows on roads and the spectral similarity
of road construction materials pose challenges to classifica-
tion accuracy. An accuracy assessment of the classification
indicated that the algorithm works well with overall classifi-
cation accuracy between 82 and 94 percent.
Introduction
According to the Federal Highway Administration (
FHWA
),
in 2012 there were 2.3 million kilometers of unpaved road in
the United States, accounting for almost 1/3 of the total in our
national transportation infrastructure (USDOT/OST-R, 2015).
Local governments and transportation agencies are responsi-
ble for a large part of this unpaved infrastructure. These agen-
cies need to be able to cost-effectively assess the condition
of their unpaved infrastructure on a periodic basis in order
to effectively manage and optimize their unpaved roads and
maintenance resource allocation. However, many transporta-
tion agencies do not have data on the type of road, whether
paved or unpaved, as an attribute of their existing inventories.
Understanding the location and length of this unpaved
road inventory is a necessary first step in effectively managing
this transportation asset before assessing condition.
Paved roads are characterized by either a bituminous,
mixed bituminous, brick, block, composite, or cement con-
crete cover with a surface base thickness of at least 2.5 cm but
typically 18 cm or more (
FHWA
, 2004). In contrast, an unpaved
road has no “hard” surfacing; they consist of a compacted
aggregate or have no added surfacing. In this paper and in
general use, the former are referred to as gravel roads and the
considered a gravel road; 15 cm to 25 cm is desirable for areas
of high traffic (Walker
et al
., 2002).
These unpaved local roads play an important role in con-
necting farmers to markets, school buses to school children,
and residents to their homes. Michigan Tech Research
Institute (
MTRI
) a research center of Michigan Technologi-
cal University, has developed an unpaved road assessment
system (Dobson
et al
., 2014) that is practical, economical,
and effective using remote sensing from an unmanned aerial
vehicle (
UAV
) as part of the “Characterization of Unpaved
Road Conditions through the Use of Remote Sensing” project
funded by the United States Department of Transportation
Office of the Assistant Secretary for Research and Technology
project (RITARS-11-H-MTU1). The methodology described in
this analysis provides the location of unpaved roads within
a road network, a significant mission planning input for the
larger Unpaved Roads project mentioned above, and a method
of maintaining an inventory of unpaved roads that is useful to
road management agencies.
For the “Characterization of Unpaved Road Condition
Through the Use of Remote Sensing” project, the location of
the unpaved roads to be evaluated is an important part of the
project mission planning system. Identification of unpaved
roads within a county road network from aerial or satellite
imagery builds from methods developed in an earlier proj-
ect that calculated the location and length of unpaved roads
as part of the Transportation Applications of Restricted Use
Technology (
TARUT
) study (Brooks
et al
., 2007;
).
That study used visible-to-near infrared ratios derived from
1 meter multispectral aerial imagery and 60 cm Digital Globe
QuickBird multispectral imagery to map road surface type,
including unpaved roads. The
TARUT
project team was able
to map road surface types with 86 percent accuracy. It was
anticipated that using 4 band 30 cm per pixel imagery, clas-
sification accuracy of at least 90 percent should be possible
with a project goal of reaching 95 percent.
Figure 1 is a sample of 30 cm ground sampling distance
(
GSD
) aerial imagery provided by project partner Southeast
Michigan Council of Governments (
SEMCOG
), where the differ-
ences between natural aggregate road (A), crushed limestone
road (B), and a paved macadam road (C) are all visible. The
project team expected using the higher resolution four band
(R, G, B, IR) aerial imagery would improve classification ac-
curacy. The output from this road surface type analysis is a
GIS
layer that identifies and attributes unpaved roads within
the road network.
Colin N. Brooks, David B. Dean, Richard J. Dobson,
Christopher Roussi, and David M. Banach are with the
Michigan Tech Research Institute, 3600 Green Court, Suite
100, Ann Arbor, MI 48105 (
).
Justin F. Carter is with Resource Data, Inc., 1450 S. Eagle
Flight Way, Suite 150, Boise, ID, and formerly with the
Michigan Tech Research Institute.
Andrea J. VanderWoude is with the NOAA Great Lakes
Environmental Research Lab, 4840 S. State Road, Ann Arbor,
MI 48108.
Tim Colling is with the Michigan Technological University,
1400 Townsend Dr., Houghton, MI 49931
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 5, May 2017, pp. 377–383.
0099-1112/17/377–383
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.5.377
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
April 2017
377
