Transactions of the ASABE
Vol. 51(6): 1937-1950
? 2008 American Society of Agricultural and Biological Engineers ISSN 0001-23511937
AUTOMATED IDENTIFICATION OF TILE
LINES FROM REMOTELY SENSED DATA
B. S. Naz, L. C. Bowling
ABSTRACT. Although subsurface drainage provides many agronomic and environmental benefits, extensive subsurface
drainage systems have important implications for surface water quality and hydrology. Due to limited information on
subsurface drainage extent, it is difficult to understand the hydrology of intensively tile‐drained watersheds. In order to
address this problem, a methodology was developed to use image processing techniques for automated detection of tile drains
from multiple dates of aerial photography at the Agronomy Center for Research and Education (ACRE), West Lafayette,
Indiana. A stepwise approach was adopted to first identify potential tile‐drained fields from the GIS‐based analysis of land
use class, soil drainage class, and surface slope using decision tree classification. Based on preliminary classification of
potential tile‐drained area from the decision tree classifier, a combination of image processing techniques such as directional
edge enhancement filtering, density slice classification, Hough transformation, and automatic vectorization were used to
identify individual tile lines from images of 1976, 1998, and 2002. Accuracy assessment of the predicted tile line maps (Hough
transformed and untransformed) was accomplished by comparing the locations of predicted tile lines with the known tile lines
mapped through manual digitization from historic design diagrams using both a confusion matrix approach and drainage
density. Forty‐eight percent of tile lines were correctly predicted for the Hough transformed map and 58% for the
untransformed map based on the producer accuracy. Similarly, 73% of non‐tile area was correctly predicted for Hough
transformed and 68% for untransformed lines. Based on drainage density calculation, 60% of tile lines were predicted from
the aerial image of 1976 and 50% from the aerial image of 2002 for both techniques, while 72% of tile lines were predicted
from the aerial image of 1998 for untransformed and 50% for Hough transformed lines. The Hough transformation provided
the best results in producing a map without discontinuity between lines. The overall performance of the image processing
techniques used in this study shows that these techniques can be successfully applied to identify tile lines from aerial
photographs over a large area.
Keywords. Aerial photography, Decision tree, Edge detection, Hough transformation, Tile lines.
n the Midwestern U.S., the main purpose of subsurface
drainage systems is to control waterlogging in agricul‐
tural fields by installing a series of drainage pipes below
the soil surface. Traditionally, subsurface drainage
pipes were made from ceramic clay tiles; later, clay tiles were
replaced with corrugated polyvinyl chloride (PVC) tubes,
generally still referred to as tile lines. Tile lines can be
installed at different spacings and depths depending on physi‐
cal properties of the soil, crop type, and climate condition
(Varner et al., 2002). Typically, subsurface drainage pipes are
placed at a depth of 70 to 130 cm and at a spacing of 6 to 24?m
(Zucker and Brown, 1998). To reduce cost, earlier systems
were more likely to target specific wet spots in the fields.
However, with improved mechanization, more uniform pat‐
terns of installation have increased. Lateral lines are usually
installed on higher elevations and connected with the main
tile lines at low elevations in the field. This network of sub‐
surface drainage empties either into an open ditch or stream.
Submitted for review in May 2007 as manuscript number SW 7025;
approved for publication by the Soil & Water Division of ASABE in
The authors are Bibi S. Naz, Graduate Student, and Laura C. Bowling,
Assistant Professor; Department of Agronomy, Purdue University, West
Lafayette, Indiana. Corresponding author: Bibi S. Naz, 915 W. State
Street, West Lafayette, IN 47907‐2054; phone: 765‐496‐9522; fax: 765‐
496‐2926; e‐mail: firstname.lastname@example.org.
Although subsurface drainage improves field conditions
by removing excess water from poorly drained soils, there are
also concerns about the potential impacts of these systems on
watershed hydrology and water quality. Subsurface drainage
increases the infiltration of precipitation, decreases annual
evaporation, and lowers the ambient water table (Hill, 1976;
Irwin and Whitely, 1983; Moore and Larson, 1980). These
field‐scale changes to the drainage system have the potential
to produce changes to streamflow response at large scales.
Despite this fact, very little effort has been expended to inves‐
tigate the consequences of subsurface drainage on stream‐
flow response at watershed scales. The analysis of hydro-
logical effects of subsurface drainage systems is complicated
by limited information on the locations of historic tile drains.
Subsurface drainage tile lines that have been installed 50 or
more years ago are still working fully or partially. In most
cases, information on these old tile lines has been lost or does
not exist. A growing emphasis on the water quality impacts
of subsurface tile drainage has also increased the interest of
estimating tile‐drained landscapes at the national level. It is
not possible to identify the tile‐drained areas on such a large
scale using traditional methods such as manual tile‐probing
techniques. This leads to the need of developing a computer‐
based decision support system that not only uses spatially ori‐
ented data but also a mechanism for applying criteria related
to subsurface tile drainage systems.
1938TRANSACTIONS OF THE ASABE
Recent studies have shown that remote sensing can be an
effective tool to accurately map tile lines (Verma et al., 1996;
Varner et al., 2002; Northcott et al., 2000). The key factors
that can affect the reflectance of soil in the visible and near‐
infrared (NIR) portion of the electromagnetic spectrum are
soil moisture, soil texture, organic matter, and tillage prac‐
tice. The amount of moisture held in the soil surface layer is
a function of soil texture. A finer soil texture tends to have
high moisture content in the presence of precipitation, result‐
ing in less incident energy reflectance from the soil surface;
therefore, wet soils appear darker than dry soils in imagery
(Lillesand et al., 2004). Soil immediately above functioning
drains tends to dry faster than surrounding areas after a rain
event, and these differences in soil moisture due to tile drains
have been used to delineate tile lines in fields (Verma et al.,
1996). Using color infrared aerial photographs and GIS anal‐
ysis, Verma et al. (1996) concluded that the best time for tak‐
ing imagery that could be used for tile delineation is 2 to
3?days after 2.54 cm of rain on the field sites. Similarly, or‐
ganic matter and tillage also affect soil reflectance. The
greater the amount of organic matter in the surface horizon,
the greater the amount of soil moisture and the lower the soil
reflectivity (Jensen, 2000). Recent tillage changes the sur‐
face soil moisture, roughness, and the amount of residue left
after cultivation (Morris et al., 2004). These factors can affect
the soil reflectance differently in different wavelengths of the
electromagnetic spectrum and thus lead to large spatial vari‐
ability in surface reflectance patterns.
Farmers evaluate many factors such as soil texture, soil
drainage, soil potential for surface runoff, and slope of land
while installing tile lines. Based on these factors, a decision
support system can be developed to locate the areas where tile
lines may be installed at a larger scale. One such method is
decision tree classification. Studies have been conducted to
evaluate the effectiveness of the decision tree classifier rela‐
tive to conventional classification approaches such as maxi‐
mum likelihood classification and unsupervised clustering
algorithms or to newer methods such as neural network or
fuzzy logic classification (Friedl and Brodley, 1997; Pal and
Mather, 2003). Maximum likelihood classification relies on
statistical assumptions (i.e., distribution of the input data
should follow the normal distribution) and often gives poor
classification accuracy on real‐world problems due to its sta‐
tistical data requirements as the number of attributes in‐
creases. Neural network and fuzzy logic are more
sophisticated in the handling of complex relations among
classes and do not require knowledge of class distribution
(Friedl and Brodley, 1997). The main drawback of these
methods, however, is the complexity and the time involved
in choosing and setting training data sets. In contrast, a deci‐
sion tree classifier is more efficient in terms of its speed and
flexibility to handle both continuous and categorical input
data. The classification accuracy, however, is fully depen‐
dent on the design of the decision tree and the selected input
While a decision tree classifier may be an effective tool in
quantifying potential tile‐drained landscapes, the estimated
area may be used as a guideline in producing more detailed
maps of individual tile lines using image processing tech‐
niques such as edge detection filters. A variety of linear fea‐
ture extraction methods have been explored in the past few
decades (Zlotnick and Carnine, 1993; Manolakis et al., 2001;
Ding et al., 2006). One approach is semiautomatic feature ex‐
traction, in which a starting point and direction are given to
initialize the detection of linear features (Lacoste et al.,
2005). This approach is considered to be a good compromise
between accuracy and digitization time. Another possibility
is to consider a fully automatic feature extraction approach.
Many of the fully automatic feature extraction algorithms use
edge detector operators. The edge detection algorithms such
as gray‐level gradient, Sobel, Laplacian, Prewit, Robert,
Kirsch, and Rosenfeld threshold operators are designed to
process every pixel on the original image using the gray‐level
variation around the edge area and make decisions about
edge existence using first‐ or second‐order derivatives (Sun,
2003). The major drawback of these techniques is their sensi‐
tivity to noise, particularly for high‐resolution images, which
leads to low accuracy for edge detection algorithms. To re‐
duce this sensitivity to noise, some studies propose the use of
a combination of smoothing and edge enhancing operators
(Lacoste et al., 2005).
Varner et al. (2002) demonstrated the applicability of
semi‐automatic feature extraction in combination with edge
enhancing techniques for the mapping of tile lines in Cham‐
paign and Ford Counties, Illinois. They used multiple image
processing techniques to identify the most effective method
for mapping drainage tiles using data from airborne multi‐
spectral and hyperspectral sensors. Based on their accuracy
assessment of image‐based tile line map, they suggested that
directional edge detect enhancement was the most effective
technique in identifying tile lines in both tilled and no‐tilled
fields. This study also suggested that satellite panchromatic
data could be a possible source of information for tile map‐
ping. These studies focused on identification of tile lines in
individual fields, so limited work has been done to identify
tile lines at larger scales.
The purpose of the present study is to develop a semi‐
automated methodology to predict potentially tile‐drained
area using GIS‐based analysis of land cover, soil, and topog‐
raphy data and to identify individual tile lines from aerial im‐
ages in poorly drained landscapes. The specific objectives of
the study were: (1) to estimate potential tile‐drained area
from spatial data sets using a decision tree classifier, (2) to
identify individual tile lines from digital aerial photographs
using image processing techniques, and (3) to assess the ap‐
propriateness of these techniques for the purpose of auto‐
mated tile line mapping from remotely sensed data.
As shown in figure 1, three nested study sites were se‐
lected : (1) the northwestern part of Tippecanoe County, Indi‐
ana (i.e., Area 1) was selected for the preliminary classifi-
cation of potentially tile‐drained area using a decision tree
classifier; (2) Area 2, which is centered on the Agronomy
Center for Research and Education (ACRE), West Lafayette,
Indiana, was selected for classification of individual tile
lines; and (3) ACRE (Area 3) was chosen for validating the
image‐based predicted tile line map where historic informa‐
tion on tile line locations is available.
ACRE was established in 1949 as a field research station
for crops and soils. The major soil types are Drummer silty
clay loam, Chalmers silty clay loam, and Raub silt loam. The
Drummer series are deep, poorly drained, moderately perme‐
able soils on till plains with 0% to 2% slope. The Chalmers
series consists of very deep, poorly drained soils that formed
in loess and silty till with slopes ranging from 0% to 2%. The
1949 Vol. 51(6): 1937-1950
detection filter, a buffer was applied to predicted tile lines in
order to capture the region of most probable tile lines. How‐
ever, creating a vector buffer may reduce the user accuracy
of the predicted non‐tile area for Final Map 1 and Final Map?2
because a greater number of pixels classified as tile decreases
the non‐tile area and results in lower user accuracy for the
non‐tile class based on pixel‐by‐pixel accuracy assessment.
A higher level of accuracy might be obtained using an edge‐
detector such as a Laplacian of Gaussian (LOG) filter that is
able to locate edges in noisy image as close as possible to its
true position (Pratt, 2001).
The Hough transformation was investigated to extract
straight connected lines from classified images in order to
overcome the problem of disconnected pixels as a result of
image classification. Comparison of final maps created from
Hough transformed and untransformed images shows that
Hough transformation did well in terms of connected lines.
Many of the tile lines derived from Hough transformation are
fully connected except in a few fields, which also resulted in
higher percentages of producer accuracy of the non‐tile class
for the Hough transformed map. The vector map created from
untransformed images did an extremely poor job in terms of
connected lines, with none of the tile lines fully connected.
This indicates that Hough transformation will reduce the
amount of manual work that will be required to make a tile
line map with fully connected lines, which can be used in oth‐
er applications such as hydrological model simulations. The
Hough transformation, however, also transforms noise in the
original image to lines in diagonal directions. Selection of a
threshold for the inverse transformation may remove lines
from the resultant image that may represent tile lines, which
also resulted in lower producer accuracy of the tile line class
for the Hough transformed map. Reduction of noise and
masking of known linear features may avoid this problem. In
addition, the computationally intensive Hough transform is
very time consuming. For larger images, it takes significant
computer resources in terms of time and computer memory
to run the transformation.
Few tile lines were identified from the 2002 image for the
southern part of ACRE and for the fields where tile lines were
installed in 1988 and 1997. Similarly, the tile lines installed
in 1988 were not identifiable on the 1998 image. While
comparing the observed tile lines installed prior to 1976 with
tile lines detected from the 2002 and 1998 images, it was
found that only 6% of tile lines were identified from the 2002
image and 3% from the 1998 image. In contrast, 71% of these
tile lines were identified in the 1976 image. There was a small
amount of rainfall before taking the aerial photographs of
1998 and 2002, but it was not sufficient to maximize the soil
moisture differentials immediately above functioning drains
(as indicated by previous studies, the amount of rain event
should be at least 25 mm two or three days before taking the
aerial photograph). The rainfall totals prior to the 1976 image
are unknown, but surface reflectance differences are more
pronounced in this image. It is possible that in the case when
the rainfall is not sufficient, using an image that is taken soon
after the installation of tile lines may be more desirable. Dis‐
turbance of the top soil during installation of tile lines makes
the locations of the tile more distinct on the images relative
to other features. In this case, the aerial images from multiple
years based on tile installation time are required to obtain the
best possible tile line map.
The effect on the prediction of individual tile lines caused
by tile drainage performance variations was not investigated
in this study. The individual tiles were identified using aerial
images that were taken in different years and near the time of
the tile line installation, so presumably the majority of the
tiles were functioning well. A study by the Institute for
Technology Development (ITD, 2002) showed that color in‐
frared imagery can identify areas of poorly draining tiles such
as broken or clogged tiles, but recent color infrared images
taken at the right time and weather condition suitable for tile
lines delineation were not available for the study area.
Although the spatial accuracy was not as good as needed,
future work may be focused on using image‐based tile line
maps as guidelines to identify approximate tile line locations,
which could be very useful for farmers who need a general
idea of the drainage system in their fields and for hydrologic
modelers who need an estimate of drainage density in order
to predict drain flow at watershed scale. The problem of ac‐
quiring the right images at the right time complicates the is‐
sue of creating an image‐based tile line map, particularly for
a large area. However, selecting images from multiple pre‐
vious years that were taken two to three days after a signifi‐
cant rainfall event or soon after the tile line installation with
no residue cover can increase the probability of identifying
tile lines from remotely sensed data.
Beaty, J. 2005. Personal communication. Superintendent,
Agronomy Center for Research and Education, Purdue
University, West Lafayette, Ind.
Baird, F., and F. Baret. 1997. Crop residue estimation using
multiband reflectance. Remote Sensing Environ. 59(3): 530‐536.
Basu, M. 2002. Gaussian‐based edge‐detection methods: A survey.
IEEE Trans. Systems, Man, and Cybernetics C 32(3): 252‐260.
Daughtry, C. S. T. 2001. Discriminating crop residues from soil by
shortwave infrared reflectance. Agron. J. 93(1): 125‐131.
Ding, X., W. Kang, J. Cui, and L. Ao. 2006. Automatic extraction
of road network from aerial images. In Proc. IEEE 1st Intl.
Symp. Systems and Control in Aerospace and Astronautics,
220‐223. Piscataway, N.J.: IEEE.
ESRI. 2004. ArcGIS Desktop Help. Redlands, Cal.: Environmental
Systems Research Institute.
Franzmeier, D. P., W. D. Hosteter, and R. E. Roeske. 2001.
Drainage and wet soil management. Extension Bulletin AY‐300.
West Lafayette, Ind.: Purdue University.
Friedl, M. A., and C. E. Brodley. 1997. Decision tree classification
of land cover from remotely sensed data. Remote Sensing
Environ. 61(3): 399‐409.
Hill, A. R. 1976. Environmental impacts of agricultural land
drainage. J. Environ. Mgmt. 4(3): 251‐274.
Illingworth, J., and J. Kittler. 1998. A survey of the Hough
transform. J. Computer Vision, Graphics, and Image Processing
Irwin, R. W., and H. R. Whiteley. 1983. Effects of land drainage on
stream flow. Canadian Water Resources J. 8(2): 88‐103.
ITD. 2002. Ag20/20 2002 Soybean and Corn Project: Delineation
of field tile locations using remotely sensed imagery. Stennis
Space Center, Miss.: Institute for Technology Development.
Available at: www.iftd.org/pdf/prj_pdf.5682.
2002.1022093820.pdf. Accessed 31 July 2008.
Jensen, J. R. 1996. Introductory Digital Image Processing: A
Remote Sensing Perspective. 2nd ed. Upper Saddle River, N.J.:
1950TRANSACTIONS OF THE ASABE
Jensen, J. R. 2000. Remote Sensing of the Environment: An Earth
Resource Perspective. 2nd ed. Englewood Cliffs, N.J.: Prentice
Lacoste, C., X. Descombes, and J. Zerubia. 2005. Point processes
for unsupervised line network extraction in remote sensing.
IEEE Trans. Pattern Analysis and Machine Intelligence 27(10):
Lillesand, T. M., R. W. Kiefer, and J. W. Chapman. 2004. Remote
Sensing and Image Interpretation. 5th ed. New York, N.Y.: John
Wiley and Sons.
Manolakis, D., C. Siracusa, and G. Shaw. 2001. Hyperspectral
subpixel target detection using the linear mixing model. IEEE
Trans. Geosci. and Remote Sensing 39(7): 1392‐1409.
Moore, I. D., and C. L. Larson. 1980. Hydrologic impact of
draining small depressional watersheds. J. Irrig. and Drainage
Morris, D. K., C. J. Johannsen, S. M. Brouder, and G. C. Steinhardt.
2004. Remote sensing. In Encyclopedia of Soils in the
Environment, III: 385‐392. D. Hillel, ed. Amsterdam, The
NASS. 2002. National Agricultural Statistics Database.
Washington, D.C.: USDA National Agricultural Statistics
Service. Available at: www.nass.usda.gov. Accessed 15
Northcott, W. J., A. K. Verma, and R. A. Cooke. 2000. Mapping
subsurface drainage systems using remote sensing and GIS.
ASAE Paper 002113. St. Joseph, Mich.: ASAE.
Pal, M., and P. M. Mather. 2003. An assessment of the effectiveness
of decision tree methods for land cover classification. Remote
Sensing Environ. 86(4): 554‐565.
Pratt, W. K. 2001. Digital Image Processing. 3rd ed. New York,
N.Y.: John Wiley and Sons.
Streck, N. A., D. Rundquist, and J. Connot. 2002. Estimating
residual wheat dry matter from remote sensing measurements.
Photogrammetric Eng. Remote Sensing 68(11): 1193‐1201.
Sun, T. L. 2003. A detection algorithm for road feature extraction
using EO‐1 hyperspectral images. In Proc. IEEE 37th Intl.
Carnahan Conf. on Security Technology, 87‐95. Piscataway,
USDA. 2004. Soil Survey Geographic Database (SSURGO).
Washington, D.C.: USDA Natural Resources Conservation
Service. Available at: www.ncgc.nrcs.usda.gov. Accessed 6
Varner, B. L., T. Gress, K. Copenhaver, and S. White. 2002. The
effectiveness and economic feasibility of image based
agricultural tile maps. Urbana, Ill.: University of Illinois,
Beckman Institute for Advanced Science and Technology.
Verma, A. K., R. A. Cooke, and L. Wendte. 1996. Mapping
subsurface drainage systems with color infrared aerial
photographs. In Proc. 32nd AWRA Annual Conf. and Symp. on
GIS and Water Resources. Middleburg, Va.: American Water
Zlotnick, A., and P. D. Carnine. 1993. Note: Finding road seeds in
aerial images. CVGIP: Image Understanding 57(2): 243‐260.
Zucker, L. A., and L. C. Brown, eds. 1998. Agricultural drainage:
Water quality impacts and subsurface drainage studies in the
Midwest. Extension Bulletin 871. Columbus, Ohio: Ohio State