| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Dep. of Crop and Soil Sci., The Univ. of Georgia, 1109 Experiment St., Griffin, GA 30223-1797
* Corresponding author (kingram{at}griffin.peachnet.edu)
Received for publication February 22, 2000.
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
Abbreviations: RMS, Root Measurement System
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
While researchers have developed many methods to observe roots, none is without shortcomings. Minirhizotrons have several advantages over other methods for observing roots. Minirhizotron observations are nondestructive, thereby allowing repeated observations of roots to measure root elongation, branching, and turnover as well as root distribution through the soil profile (Cheng et al., 1991; Merrill, 1992; Steele et al., 1997). If tubes are properly installed, soil disturbance is minimal, and results correlate highly with data from soil cores (Bland and Dugas, 1988; Box et al., 1989; Franco and Arbisqueta, 1997). In addition to providing direct observation of roots growing in soils, minirhizotrons also allow observation of other underground plant organs such as rhizomes, peanut (Arachis hypogaea L.) pods, and legume nodules and other soil organisms such as insects, worms, and fungi (Lussenhop et al., 1991).
The greatest disadvantage of minirhizotron systems has been the tedious, time-consuming process of translating qualitative information from observations to quantitative data (Hendrick and Pregitzer, 1996). Until recently, most root images have been collected on video, film, or overlays traced with a wax pencil. Root length may be estimated by tracing roots (Beyrouty et al., 1988; Cheng et al., 1991) or by counting root intersections across random transects (Box and Ramseur, 1993; Pietola and Smucker, 1995; Tennant, 1976). Some researchers count roots in the field without storing images for analysis (Franco and Arbisqueta, 1997; Merrill, 1992). One tool that may speed image analysis is automatic computer analysis. Automatic image analysis works well for clean roots (Dowdy et al., 1998; Kimura et al., 1999; Murphy and Smucker, 1995), but background noise makes it much more difficult for roots growing in soils (Bakic et al., 1996).
Recently, there have been great advances in quality and reductions in costs for digital image capture through scanners, digital cameras, or video image capture. Pateña and Ingram (2000) described an inexpensive system for capturing and analyzing digital images collected with a minirhizotron camera. This method used Mouse-O-Meter freeware (Hotware, 29a, rue de Mersch, L-8293 Keispelt, Luxembourg) to convert cursor movement to distance traveled. The authors estimated root length by tracing along roots in an image. Accuracy of this method depends greatly on hand–eye coordination of the operator and length of roots in an image. By this method, long roots are much more difficult to trace accurately than short roots that can be traced while the operator rests his or her hand in a single position.
The objective of this note is to present and assess new software that allows rapid, accurate measurement of root length, diameter, and numbers from digital images.
|
MATERIALS AND METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
The principle of operation is based on the fact that each open computer window has a scale that locates cursor coordinates. In RMS, an open image file fills the 640- by 480-pixel image window so that each cursor position in an image is represented by a coordinate pair. When an operator clicks the mouse button to mark one end of a root segment, RMS records the cursor coordinates. As the operator moves the mouse, a straight colored line appears between the previous click point and the cursor location, allowing the operator to see whether the line remains centered over a segment of root. When the operator clicks a second point along a root, RMS records the coordinates of that point and calculates the distance between points. Once a second point is clicked, the line color changes so that the operator knows which root segments have already been measured. Using calibration factors developed as described below, RMS converts end points of a segment from coordinate units to lengths of objects.
As the user clicks sequential points along a root in an image, RMS estimates root length from the series of segments. By counting the number of times a user ends one root and begins another, RMS also records numbers of roots in an image. To measure root diameter, RMS uses a circular cursor, the diameter of which can be increased or decreased using the + or - keys to match the diameter of each root segment. As the operator changes cursor diameter, RMS uses the new diameter to compute surface area and volume of the segment. Although surface area and volume calculations are based on measurements of diameter to one-pixel resolution, length data are stored in 0.1-mm-diam. classes. (The next version of RMS allows users to set diameter class sizes.) In addition to direct measures of root length and diameter and direct root counting, RMS also calculates root volume and surface area, based on the assumption that root segments are cylinders, and root length density assuming that an image at a planar soil surface represents thin soil layer. The default value of the soil layer for computing root length density is 3 mm (Sanders and Brown, 1978), but users may set this value according to their particular condition and calibration data. For each image, RMS creates an ASCII output file to allow subsequent statistical analysis.
Time Required for Image Analysis
We assessed the speed and accuracy of operating RMS by three methods. First, we asked 10 different people to use RMS to analyze an image of a 24 mm s-shaped line drawn on a piece of paper and captured with a minirhizotron camera (Model BTC-100X, Bartz Technol. Corp., Santa Barbara, CA). To draw a 24-mm line, we first cut a straight piece of wire 24 mm long and then bent the wire into an s-shape and drew a line along the side of the wire with a felt pen. The camera was set to the lowest magnification (about 10x) so that the total area viewed in each image was approximately 19.3 by 13.8 mm. We calibrated RMS before operators analyzed images, and all operators used the same computer for the analysis. Each operator measured the line on the image 10 times, all with the same orientation. Each operator was given the same set of minimal instructions. We assured them that the analysis was neither a race nor a test but requested that they be as accurate as possible. Operators could begin from either end of a line. They were told that they could click as many or as few times along the curve as they wanted but that they should keep the red line stretching from last click to the next between the borders of the image line. Finally, after finishing a trace of each root, operators were to click the right mouse button, select the end root option from the menu, and begin the next measurement. For each operator, we collected data on total time for analysis. For each image measurement, we recorded numbers of line segments and total measured length of the line.
In the first test, the line was of uniform diameter along its length, so this test did not allow us to assess the effect of adjusting diameter on speed of operation. A second test was done with 10 different operators having a range of previous experience with RMS. The operators measured both lengths and diameters of pseudo roots for each of 10 images made from bent wires of different lengths and diameters and were instructed on how to increase and decrease circular cursor dimension to match the wire (pseudo root) diameter. Other operator instructions and data collection procedures were similar to those of the first test.
For our third method of assessment, we logged the time required for image analysis from field data sets used in the calibration test described below. A single technician measured length of tall fescue (Festuca arundinacea Schreber) roots from images collected at different depths from 32 minirhizotron tubes. In each sample date, a total of 920 images were collected. Numbers of images analyzed each day and time spent in that analysis were recorded.
Calibration
We printed a 20- by 40-mm grid with 1- by 1-mm hatches using a laser printer. We used digital calipers to assure correct dimensions. The grid was wrapped around the outside of a clear acrylic tube of the same diameter as our minirhizotron tubes. We took an image of this grid with the minirhizotron camera before sampling each of 32 minirhizotron tubes installed in field plots of tall fescue.
To calibrate RMS, we first open the image file with a grid of known dimensions. During calibration, RMS uses a cross-hair cursor to aid in aligning the cursor with an intersection of the grid. Then, by first clicking on a grid intersection on the left side and then on the right side of the grid, a line segment is drawn from left to right along a horizontal line of known length, 18 mm in our case. Similarly, a line is drawn from top to bottom, 13 mm long in our case. These grid dimensions compare with a total image size of 19.3 by 13.8 mm. After the user enters the actual lengths of these calibration lines, RMS multiplies the horizontal calibration factor by difference between x-coordinate values of sequential points and vertical calibration factor by differences between y-coordinate end point values and applies the Pythagorean equation to convert distance in screen coordinate points to length of an object in an image.
We conducted analysis of variance for horizontal and vertical calibration factors for four sample dates to test the repeatability of measurements. Some grid images were discarded or lost, so this analysis is based on a total of 104 calibration images, all measured by a single technician.
|
RESULTS AND DISCUSSION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
0.1 mm for all but Operator 10. Average length for all operators (23.3 mm) underestimated the nominal length, but we believe this difference arose from making the image rather than measuring it. If we consider the first six measurements as practice and limit our analysis to the final four measurements of each operator, average length increased to 23.7 mm. Thus, it takes very little practice for operators to become proficient in use of RMS.
|
Length of pseudo roots measured by 10 operators having a wide range of experience did not differ significantly (Table 2). Though operators differed significantly in time required to analyze images and number of segments used to trace roots (Table 2), there was no significant correlation among time for analysis, number of segments used, or total length. Operators with more experience using RMS analyzed images in about half the time of the slowest inexperienced operators. Experience with RMS was probably not the only factor involved; basic familiarity with computers and mouse applications may also explain some of the differences in speed among operators. In addition, when analyzing images of roots growing in soil, it will be important for operators to have training to discern what is and is not a root and whether a brown root is dead or alive. There is no simple rule for relating root color to function or viability. We have observed white branch roots growing from apparently dead and decaying brown nodal roots of tall fescue.
|
|
Roots in soil do not normally grow in an s-shape pattern, nor do they have uniform diameters along their length as did the wires used in the second test. Within the 19.3- by 13.8-mm observation area of our minirhizotron images, most roots are nearly straight. There is great variation in numbers of roots in different minirhizotron images depending on soil depth and time after planting. Using images of tall fescue roots under field conditions, we found that one technician could analyze up to 38 images h-1 early in the season when roots were few and many images had no roots at all. Later, as root length increased to an average of 34 mm image-1, with some images having >200 mm of roots, the same technician could analyze only 24 images h-1 on average.
Other factors that may slow measurements include background noise in images that may be difficult to distinguish from roots, frequent changes of root diameter, and operator fatigue. Ability to distinguish between roots and other soil features depends mostly on experience. At the same time, one of the strengths of RMS compared with automatic image analysis is that it uses the superior capacity of the human to make such distinctions. Measuring roots of similar size in sequence minimizes the time spent increasing and decreasing the cursor circle to match root diameter. Though it is possible for an operator to measure roots with RMS for a full 8-h day, we find that shorter work periods of no more than 4 h per day keep the operator fresh and less prone to error.
Accuracy and Repeatability
For 104 calibration grid measurements, we found an average of 71.49 ± 0.06 points mm-1 in the horizontal direction and 71.14 ± 0.05 points mm-1 in the vertical direction. The difference between horizontal and vertical calibration factors may result from image distortion, which is why it is important to have both vertical and horizontal calibration factors.
Variance in calibration factors within sample dates was very small, with all standard deviation values <0.84% of means. This error could arise from slight differences in distance from the camera lens to the grid when collecting calibration images or from slight differences in where the operator positions the cross-hair cursor during calibration.
To assure data quality, we recommend recalibration as often as is practically possible. At a minimum, we recommend recalibration at the beginning and end of every sample day and each time before and after the camera zoom setting is changed. A good record of calibration factors will help operators evaluate their experimental methods and may identify problems in minirhizotron hardware before they become serious.
Differences between Root Measurement System and Other Methods
There are several other methods that researchers have used to analyze images from minirhizotrons, including root counts, estimating root length by Tennant's (1976) line-intercept method, using a linear probe (tracing wheel) to trace roots on a photograph or video screen, and automatic image analysis. Whether in situ or from images, roots may be counted in less than half the time required to trace roots using RMS. If roots are counted in situ, researchers can save additional time by avoiding the image storage step. The disadvantage of merely counting roots is that the relationships are not constant among numbers of roots and other more physiologically meaningful root measurements such as length, surface area, and volume. Root length in images may be analyzed by the line-intercept method faster than they can be traced by RMS, and using a linear probe takes about the same amount of time as RMS. The disadvantages of the line-intercept and linear probe methods are that neither includes measurement of root diameter and estimation of characteristics that can be computed from length and diameter. Finally, although there are several automatic image analysis systems that work very well for cleaned roots against a uniform background (Kimura et al., 1999; Pietola and Smucker, 1995), none has proven successful in analyzing roots against a soil background.
Because RMS measures root length and diameter and counts roots, its outputs can be compared directly with results from other experiments that have used these other methods. By using the human eye and brain to distinguish roots from other image features, RMS is less subject to error from background noise than are automated systems. Relying on human judgment also allows us to analyze images with very modest computer requirements. Of course, operators do need to learn how to distinguish between roots that are dead and alive and how to handle roots that cross and roots that grow away from and then return to the observation surface.
RooTracker (Duke Univ., Durham, NC) is a software system similar to RMS in many ways. Although a version of RooTracker has recently been released in a beta testing version for Microsoft operating systems, our experience is that it takes much longer to learn to operate and because root length and diameter are measured in separate operations, RooTracker takes as much as twice the time to analyze images as RMS.
Though RMS was designed to analyze root images collected by minirhizotrons, it could be used to measure roots in any digital image, whether from a rhizotron, soil pit, or anywhere else. The only constraint is that either the image must be of known dimensions or a calibration standard must be included in the image. With its speed of operation, accuracy, and ease of learning, RMS makes it possible to analyze sufficient numbers of minirhizotron images to allow detection of treatment effects despite the large variability of most root observations.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Crop Science | Vadose Zone Journal | |||
| Journal of Natural Resources and Life Sciences Education |
Soil Science Society of America Journal | ||||
| Journal of Plant Registrations | Journal of Environmental Quality |
The Plant Genome | |||
