HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via ISI Web of Science (8) Citing Articles via Google Scholar Google Scholar Articles by Zobel, R. W. Search for Related Content PubMed Articles by Zobel, R. W. Agricola Articles by Zobel, R. W. Related Collections Root Development Software

CROP PHYSIOLOGY & METABOLISM

Sensitivity Analysis of Computer-Based Diameter Measurement from Digital Images

USDA-ARS-AFSRC, 1224 Airport Rd., Beaver, WV 25813

^* Corresponding author (rzobel{at}afsrc.ars.usda.gov)

	ABSTRACT

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Root diameters less than 0.5 mm are infrequently documented in the literature, yet they are a critical functional component of the root system. Software capable of measuring root diameters from digital images has been used extensively to measure root length, but little is known of its efficacy in diameter measurements. An Epson 1680 scanner with a transparency unit and a Kodak DCS330 digital camera with a 105-mm Macro lens were each used to image sets of random length wires of four diameters from 0.057 to 1.19 mm and human hair of diameter 0.057 mm over a series of pixel sizes from 0.019 to 0.254 mm (50.5–3.9 p mm^-1). Images were analyzed with the MacRHIZO software package either before or after image sharpening with Adobe Photoshop. Additionally, roots from two experiments were photographed to compare with the wire and hair analyses. Minimum routine mixed object diameter resolution with both this scanner and camera is about 50 µm, though the theoretical minimum is only dependent on lens and scanner technology. With unadjusted images, MacRHIZO pro 3.10b requires a very narrow range of image resolution (7.9–15.7 p mm^-1 with 0.5 to >1-mm diameter objects) for accurate measurement. Higher resolutions cause the software to assume a 1- or 2-pixel thick wire or root is crossing the larger object at points of roughness on the edge of the images. The sharpen filter from Photoshop eliminated this problem at all resolutions tested with scanner and photographic images. Images treated with the maximum sharpening from Photoshop had practical image resolution ranges of 4 to 50 p mm^-1. With sharpening, images of mixtures of roots or other objects with diameters between 50 µm and >1 mm can be accurately analyzed with MacRHIZO. Digital camera and high-resolution scanner appear equally effective in producing digital images for routine length and diameter analysis.

Abbreviations: dpi, dots per inch • tiff, tagged image file format • p mm^-1, pixels per mm • LD, length by diameter class

	INTRODUCTION

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
THERE HAS LONG BEEN A NEED for relatively quick and simple root length and root diameter measurement (Böhm, 1979). Newman (1966) codified the line-intersect method of estimating root length, and Tennant (1975) revised Newman's method. Richards et al. (1979) developed a machine which used the line-intersect method to estimate automatically the length of washed root samples. Voorhees et al. (1980) utilized video photographic methods and Pan and Bolton (1991) and Kaspar and Ewing (1997) used scanner technology coupled with computers to refine further the process of root length measurement. These later methods had the capacity to measure root diameter, and a number of commercial systems have been developed on the basis of these methods. Resolution of very fine diameter roots has, however, been inhibited by the relatively large pixel sizes of the charge-coupled devices used in the video imaging process or inherent inaccuracies at higher resolutions with scanners (Bauhus and Messier, 1999).

Software that automatically calculates root length and diameter from a scanner or from saved digital images, using a thinning algorithm similar to that of Lebowitz (1988), is commercially available (RHIZO—Regent Instruments, Quebec City, Quebec, Canada, http://www.regent.qc.ca; verified 9 Oct. 2002)¹. RHIZO is normally combined with a high-resolution scanner (nominal resolution <64 pixels mm^-1 [p mm^-1]), and can provide relatively rapid root length and diameter measurements from mixed populations of roots. Bauhus and Messier (1999) analyzed fine roots (0.5- to 2-mm diam) of Picea glauca (Moench) Voss with their scanner set at 11.8 p mm^-1 (0.085-mm pixel width) and RHIZO set for diameter class sizes of 0.5 mm. Their research suggests that this is the maximum effective resolution for RHIZO with roots of this diameter distribution. These settings were chosen to obviate overestimation of root length that is often observed when the resolution of the scanner is >11.8 p mm^-1 and when the diameter class size is smaller than the minimum diameter to be analyzed, in which case data show root lengths for diameters of roots that are smaller than any roots in the image. This type of software has the potential to revolutionize the task of routine root diameter measurement if a method can be found that deals with mixed populations of roots having a minimum root diameter of 0.05 mm or less. Bouma et al. (2000) and Costa et al. (2001) suggest that staining of fine roots (<0.3-mm diam.) can help in measuring these nearly transparent roots. Their research was based on measuring root length, rather than diameter. It is difficult to measure root diameter accurately when the root is 0.3 mm in diameter and the pixel width is 0.085 mm, because if the accuracy of the system is ±one pixel, this represents a 28% shift per pixel. In addition to staining for contrast, a method to improve the accuracy of measurement of roots >1 mm at scanner resolutions above 30 p mm^-1 is needed.

Böhm (1979) classified roots on the basis of diameter ranges. The minimum diameter range of his classification is "very fine roots" (<0.5-mm diam). Our data from corn (Zea mays L.) and the data of others with barley (Hordeum vulgare L.), demonstrate a large population of roots less than 0.5-mm diam which are normally short and determinate in growth habit (Cahn et al., 1989). This observation is supported by the data of Varney and McCully (1991), which showed a large population of first order lateral roots of corn that were in the very fine category and were 30 mm long and determinate. These very fine determinate roots are generally thought to be the active sites of nutrient uptake. Wright et al. (1999) discuss the roots of soybean [Glycine max (L.) Merr.] and two common weeds of North Carolina. They found that with field grown Amaranthus palmeri S. Wats., nearly 68% of the root length is made up of roots less than 0.1-mm diam, while in Senna obtusifolia (L.) Irwin & Barneby, 84% of the roots are less than 0.25-mm diam, and in soybean, 45% of the roots were less than 0.25-mm diam. There is little literature describing roots this fine. Miller (1981) measured corn seedling roots from hydroponic culture and found that they were in the range of 0.09- to 0.125-mm diam. Later, Varney et al. (1991) measured the length and diameter of first order laterals of young corn plants, and found that these roots varied from about 0.1- to 0.8-mm diam with an average length of about 30 mm. The Varney et al. (1991) data show that the majority of the first order laterals of corn are less than 0.5-mm diam.

A further observation by Varney et al. (1991) is that small and very fine roots can be classified into a small number of diameter classes. They suggest 0.2-, 0.35-, and 0.53-mm diam class mid points for their corn root data. Unfortunately, little or no equivalent data is available for other species, including pasture and range plant species. The Wright et al. (1999) data were not collected in a fashion that would allow an equivalent type of assessment with their species. Published frequency distributions of roots smaller than 1 mm in diameter often demonstrate frequency peaks at about the same diameter classes as in the Varney et al. (1991) data [e.g., with apple roots (Wells and Eissenstat, 2001)]. Routine root length and diameter measurements of roots in the fine and very fine categories are needed.

If a system for routine washing of roots can be established that retains roots in the 50 µm and above range, it will be possible to evaluate total rooting patterns in pastures and range lands, and to determine the impact of grazing on rooting and root longevity. A variation on the system described by Bauhus and Messier (1999), which uses a digital camera in place of a scanner is discussed here. The initial concept was that the camera was more portable and had multiple uses. If it produced comparable results, a digital camera would be a good alternative for field-based experiments. There appears to be no literature on the accuracy of the RHIZO software in analyzing root diameters (Bauhus and Messier, 1999, focused on root length). Therefore, a sensitivity analysis was conducted on scanner and camera systems utilizing the RHIZO software to analyze diameters.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Equipment
An EPSON Expression 1680 scanner (63 p mm^-1 maximum direct resolution), with a positive film transparency unit, attached to a power Macintosh G4 computer was used to produce scanned images. Scanner software was the Epson software package provided with the scanner. Images were saved to disk as tagged image file format (tiff) files and later copied to a compact disc for permanent storage and reference. Objects to be scanned were placed directly on the scanner surface. For this series of tests, the scanned area was approximately 4 x 10³ mm² (63 mm on a side). This size was selected to meet the maximum image memory size limitation of 30 MBytes at 50.5 p mm^-1, for this version of the analysis software. Scanning parameters in addition to resolution were positive transparency and 24 bit color. Color imaging allows visual assessment of the images for root color, such as secondary thickening, root death, and pathology. All images were loaded into Adobe Photoshop (Adobe Systems Inc., San Jose, CA) and saved as standard tiff files.

A digital camera (Kodak DCS 330 with a 105-mm Macro lens and an imaging screen 18.1/13.5 mm with 2008/1504 pixels) was attached to a photographic stand, and resolution controlled by varying camera height. When necessary, images were cropped to 30 MBytes with Adobe Photoshop. Camera height was adjusted to obtain resolutions over the range normally expected for root measurements. Wires and hair to be imaged were placed directly on the light box surface, and roots were contained in a 31- by 20-cm rectangular Pyrex dish with 200 mL distilled water. Photographs were normally taken at ASA 200, manual focus, f13, 1/23 s. Photographs were acquired from the camera with Photoshop and Kodak proprietary software (Kodak DCS Acquire Module 5.9.3) and saved to disk as Kodak proprietary format tiff files to maintain a minimum file size. Acquisition was either via a firewire connection between a Macintosh G3 computer and the camera, or by direct downloading from the camera's PC memory card to the computer. Images were later copied to a compact disc for permanent storage and later reference.

Camera height, the distance from an object to the lens, required for a specific resolution can be estimated by multiplying the imaging surface pixel width (mm p^-1) by the focal length of the lens (mm), and dividing the product by the desired resolution (mm p^-1), and adding the focal length to the result. The actual focal length of the lens changes with distance to the object to focus on an object. A rough, though reasonable, estimate of focal length can be obtained by measuring the distance from the lens surface to the imaging surface with the camera focused at different distances. For instance, the minimum actual focal length of the lens used here is 107 mm (calculated from calibration images produced with the lens set to focus at infinity i.e., >1.5 m), while the actual calculated focal length was 111 mm at 1 m, 120 mm at 0.5 m, and 150 mm at 0.33 m. Because scanner resolution is calibrated in dots per inch (dpi), Table 1 provides a translation table for camera height and dpi to pixels per mm (p mm^-1), which is an accepted standard for digital resolution. For convenience, Table 1 also includes the translation from resolution to pixel size (mm on a side, assuming square pixels). Table 2 provides a general reference table for conversion of dpi to p mm^-1 to mm p^-1.

	SOFTWARE

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

For every pixel in the thinned image, the distance to the edge of the root (in pixels) is determined and a diameter calculated and adjusted for angle from horizontal. By means of this estimated diameter, a pixel length adjusted for root angle is then added to the appropriate diameter class within which the estimated diameter falls. For example, if a 0.947-mm wire is imaged at 23.7 p mm^-1, then pixel size is 0.042 mm, and the wire is 22.4 pixels in diameter. If the calculated diameter at a given pixel along the length is 0.947 mm (22.4 pixels), then the 22-pixel diameter class will be incremented by the calculated pixel length, but if the measured diameter is 0.95 mm, the diameter in pixels would be 22.5, and the 23-pixel diameter class will be incremented.

Diameter classes are represented in the saved data table as a diameter range, for instance the 22-pixel diameter class would be represented as 0.907 < x 0.949 mm, and the 23 pixel diameter class as 0.949 < x 0.992 mm. The impact of this process is briefly discussed in the results, relative to Fig. 1. The sum of the lengths from all diameter classes then equals the length measured from the thinned line. In this document the diameter class ranges are represented in graphs by their midpoint.

View larger version (18K): [in this window] [in a new window]   Fig. 1. Length by diameter class (LD) curves for scanned image diameter estimation of 0.947 mm diameter wire at three resolutions, 15.2 p mm-1 (66 = 0.066 mm width pixels), 27.6 p mm-1 (36 = 0.036 mm width pixels), and 42.1 p mm-1 (24 = 0.024 mm width pixels). Data are the same as those in Fig. 3, sharpened- scan. Scanner setting: color photograph, high-resolution printer output, saved as tiff files. Diameter class size was set at 0.01 mm.

View larger version (20K): [in this window] [in a new window]   Fig. 3. Estimated diameter of 0.947 mm wire. Photographed and scanned images were made at 7 different resolutions and two image adjustments. Camera setting: ASA 200, 1/23 sec, f-16, manual focus. Scanner setting: color photograph, high-resolution printer output, saved as tiff files. Adjustments: original = no image modification; sharpened = image sharpened with Photoshop. See text for parameters.

Measured Objects
We produced images for sensitivity analysis from test samples of wire or hair of different diameters cut to random numbers of random lengths. As measured with an engineering micrometer, average diameter from 10 measurements per wire were 1.19-mm diam, 0.95 m long; 0.947-mm diam, 0.93 m long; 0.627-mm diam, 0.86 m long; and 0.229-mm diam, 1.12 m long; and hair diameter was 0.057 mm, 3 m long. The largest diameter wires were straight, while the others were curved or irregular in shape.

Roots of switchgrass (Panicum virgatum L. cv. Cave-N-Rock), grown in the greenhouse in soil in 4-L pots, were hand washed from the pots after removal of the shoot and mesocotyl. Roots were imaged with the camera system described above and analyzed with MacRHIZO. An independent assessment of root length of these samples was performed with a Comair Root Length Scanner (Hawker De Havilland Victoria Ltd., Port Melbourne, Victoria, Australia) according to the procedures of Clark et al. (1999). Root diameter data was acquired only from the MacRHIZO analysis. Because the imaged area was smaller than the tray in which the roots were placed, the roots were spread uniformly and photographed with three non-overlapping images. Actual imaged area was calculated from the photograph of the grid and the average of the three measured images adjusted to the area of the tray to estimate values for the whole sample.

A second set of root samples consisted of roots of mixed species—fescue (Lolium arundinaceum [Schreb.]), orchardgrass (Dactylis glomerata L.), and clover (Trifolium repens L.)—established at the Morrison Plumley research site, Bragg, WV. Roots were hand washed from soil cores (10-cm diam), photographed, and analyzed with MacRHIZO.

Sensitivity Analysis
For the first test, scanner resolution and camera height (Table 1) were chosen to achieve 17 similar resolutions from 3.9 to 35.4 p mm^-1. Sets of wires of each of the three diameters and hair were placed individually on the scanner or light box in a random orientation and an image taken for each resolution. A set composed of a mix of subsets of wires and hair was also imaged at each resolution. Camera images had extensive light reflection from the polished wires, so sensitivity analyses for length were done only from the scanner images. This test was repeated 11 times in different configurations with results being essentially identical in all repetitions so only one test is reported here.

Adobe Photoshop was used to adjust the images before analysis. Photoshop offers many possible treatments of digital images, a large selection of which we assessed in preliminary trials but do not report here. Of special interest were contrast adjustment and image sharpening. No other image adjustment method approached image sharpening in efficacy for either scanner or camera images. For the data presented here, two treatments are compared: no adjustment (original), and sharpen (Filter: Sharpen: Unsharp mask: amount = 500%, radius = 250.0 pixels; threshold = 1 level). This setting, when used with our color images, provides a very efficient and precise contrast adjustment to create an essentially black and white image with very sharp edges. Variations on these settings will probably be needed for specific imaging situations, but these were the best for the conditions and studies reported here.

Unsharp mask is a traditional method to correct for blurring that may be introduced during photographing or scanning. The process essentially identifies pixels differing in contrast from surrounding pixels and increases the contrast. More information may be found at the following URL: http://adobedoc.kanisasolution.com/Photoshop6/Help.htm; verified 9 Oct. 2002.

For the second test, wire of 0.947-mm diam with a flat black coating was placed on the scanner or light box in a random fashion and imaged at 12.5 to 50.5 p mm^-1 with both the scanner and camera. Settings were as follows: camera setting—ASA 200, 1/23 s, f-16, manual focus; scanner setting—color photograph, high-resolution printer output, saved as tiff files. The frame size varied among camera images and not all of the wire length was included in every frame. As a result, these images were used only to demonstrate the efficacy of the two imaging methods for estimating diameters. Again, images were analyzed with and without sharpening to demonstrate the impact of sharpening on the final result. Repeated three times with identical results, only the last run is reported.

For the third test, wire pieces of 947 µm with a flat black coating were placed on the light box in a random fashion and imaged, with the camera, at 23.1 p mm^-1. Camera settings were ASA 200, 1/23 sec, manual focus. Images were made at 12 different f-stops around f11. Images were analyzed both with and without sharpening. Repeated twice with nearly identical results, so only the last run is reported.

For the fourth test, wire of 0.947-mm diam and 930.5-mm length with a flat black coating was placed on the scanner and oriented in random fashion with from 0 to 40 wire crossings, 0 to 3 wire touches, and 0 to 2 wire crosses at a touch. Scanner settings were 23.7 p mm^-1, color photograph, high-resolution printer output, saved as tiff files, and adjusted with Photoshop before analysis. This test was to assess the ability of RHIZO to adjust for crossing and touching objects.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Image Sharpening
Digital scanner and photograph images without additional treatment with Photoshop gave good length measurements when set at the appropriate resolution for the diameter (see Bauhus and Messier, 1999). For simplicity, data from the 0.947-mm diam wire are not presented. Data from camera images of hair were identical to those of the scanner, so camera data are also omitted. These procedures tended to over estimate length when the resolution became finer than a broad optimum for each diameter (Fig. 2). For the wires, overestimation of length at high resolutions resulted from spurious skeleton lines drawn, more or less perpendicular to the single pixel skeleton of the primary object from one or two pixel diameter irregularities on the surface of the imaged object. These length overestimates are then associated with the smallest diameter classes in the resulting data set. Bauhus and Messier (1999) found a similar overestimation of length when the resolution pixel size is smaller than 15% of the diameter of the imaged object. On the other hand, with the 0.057-mm diam hair, pixel size at resolutions below 18.9 p mm^-1 is greater than the diameter of a hair, which led to underestimation of length. The appropriate scanner resolution for roots 1-mm diam and larger is 11.8 p mm^-1 or less when the images are not adjusted with Photoshop. Calculating from the data presented here, analysis of original images is limited to resolutions that have pixel sizes greater than 15% of the actual diameter of the sample objects, up to the diameter of the object itself. This precludes accurate measurement of root length, for example, when the roots are in a mixed population with roots larger than 1 mm and smaller than 0.150-mm diam. If Photoshop is used to sharpen the image before analysis, this difficulty disappears, as long as the pixel size is smaller than the diameter of the smallest object (Fig. 2). Analysis of mixed objects after sharpening is limited only by the lowest resolution of the scanner (0.016 mm) and camera (0.009 mm).

View larger version (24K): [in this window] [in a new window]   Fig. 2. Scanner estimation of length of two wires (1.190 mm diameter, 0.95 m long; and 0.229 mm diameter, 1.12 m long) and hair (0.057 mm diameter and approx. 3 m long) with and without image sharpening. Images are from scans at 17 different resolutions. Scanner setting: Color photograph, high-resolution printer output, saved as tiff files. Adjustments: original = no image adjustment; sharpened = image sharpened with Photoshop. See text for parameters.

f-Stop Adjustment
The greatest differences between the camera and scanner are that scanners have a fixed focal length and a fixed light intensity. With camera systems, the f-stop must be adjusted to provide the correct amount of light at a given distance and lighting condition. Image sharpness and contrast are both affected by this parameter. Figure 4a demonstrates that length estimation is accurate over a range of f-stops, with f-stop values >6 being acceptable analyses from sharpened images and f-stop values > 8 being acceptable for analyses from original images. On the other hand, diameter estimation requires that the f-stop be set within one stop of the optimum value (Fig. 4b). The optimum f-stop value for sharpened images was 13 with values from 11 to 16 being acceptable. Analyses from original images, however, only approached acceptable levels at an f-stop of 27. Back and side lighting of a light box can confuse the exposure meter on a digital camera. Therefore, for precision, a curve, like that of Fig. 4b—sharpened, should be made for the operating conditions of each imaging session. Once a system is established and environmental conditions established routine imaging with the camera could be done without generating a new f-stop curve for each session.

View larger version (22K): [in this window] [in a new window]   Fig. 4. Camera f-stop analysis of length (a) and diameter (b) of 0.947 mm wire 0.74 m long cut into random length pieces. Camera settings: ASA 200, 1/23 sec., manual focus with a resolution of 23.1 p mm-1. Each image was analyzed both with and without sharpening in Photoshop. Horizontal line represents the 0.947 mm point. Vertical line represents 1 pixel width (0.043 mm).

View larger version (18K): [in this window] [in a new window]   Fig. 5. Scanner estimation of average length and diameter of 947 µm wire 930.5 mm long with 0 to 40 non-parallel wire crossovers and combinations of 40 crossings with 0 to 3 overlaps (40t to 40t3—an overlap is a parallel touch or crossover) and up to two crossovers at an overlap (x or xx). Scanner setting: 23.7 p mm-1, color photograph, high-resolution printer output, saved as tiff files. Adjustments: original = no image adjustment; sharpened = image sharpened with Photoshop (see text for parameters). Pixel width = 0.042 mm.

As pixel size gets smaller, the curve includes more data points, but has a narrower spread. Data in Fig. 1 have 5 data points significantly greater than zero for 0.066-mm pixels, 4 data points for 0.036-mm pixels, and 5 for 0.024-mm pixels, giving curve widths of 0.198, 0.108, and 0.096 mm, respectively. Because the software produces some background noise, especially at diameter classes larger than a curve estimating a diameter, values less than 5% of the highest data point of a curve can be assumed to be 0. With a conservative assessment that a curve is commonly represented by 5 to 6 data points, pixel size needs to be less than one-third the diameter of the smallest object being imaged, assuming that the middle data point is at the midpoint for the object diameter and there are two to three peaks on either side. A resolution at 25% of the diameter of the smallest object is a good choice. This result suggests that for this scanner system the effective lower limit of analyzed diameters is about 0.05 mm with a RHIZO diameter class size of 0.01 mm using 65 p mm^-1 resolution (1660 dpi). This analysis will result in a measurement of 0.05 mm ± 0.045 mm for the LD curve of a 0.05 mm diameter object. Newer versions of RHIZO allow smaller class sizes and accordingly smaller threshold diameter sizes, and presumably scanner technology will soon allow greater resolutions.

When the LD plot of a mixture of wires is superimposed on the combination of LD plots of the component wires, individual diameters in the mixture are easily identified (Fig. 6). Thus estimation of diameters of a mixture of objects with discrete diameters will result in an LD plot that is a set of curves that represent the individual elements in the mixture. If the mixture is a set of objects with a smooth or uniform distribution of diameters, that is, if the difference in diameter between adjacent sized objects is less than one pixel, the LD plot should be smooth.

View larger version (26K): [in this window] [in a new window]   Fig. 6. Length by diameter class (LD) curves for four wire diameters and a mixture of the four. Scanned at 23.7 p mm-1—same data set as Fig. 2, sharpened. Scanner setting: Color photograph, high-resolution printer output, saved as tiff files. Diameter class size set to 0.042 mm.

View larger version (18K): [in this window] [in a new window]   Fig. 7. Log scale plots of root length by measured diameter class (LD plot). Switchgrass photographed at 19.2 p mm-1 with the diameter class size set at 0.052 mm; Pasture photographed at 14.3 p mm-1, with diameter class size set tot 0.070 mm. Camera settings: ASA200, 1/23 sec, f-13, manual focus.

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

	NOTES

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
¹ Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture.

Received for publication January 30, 2002.

	REFERENCES

TOP ABSTRACT INTRODUCTION NOTES MATERIALS AND METHODS SOFTWARE RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 

• Bauhus, J., and C. Messier. 1999. Evaluation of fine root length and diameter measurements obtained using RHIZO image analysis. Agron. J. 91:142–147.[Abstract/Free Full Text]
• Böhm, W. 1979. Methods of studying root systems. Springer-Verlag, New York.
• Bouma, T.J., K.L. Nielsen, and B. Koutstaal. 2000. Sample preparationand scanning protocol for computerized analysis of root length and diameter. Plant Soil 218:185–196.
• Cahn, M.D., R.W. Zobel, and D.R. Bouldin. 1989. Relationship between root elongation rate and diameter and duration of growth of lateral roots of maize. Plant Soil 119:271–279.
• Clark, R.B., R.W. Zobel, and S.K. Zeto. 1999. Effects of mycorrhizal fungus isolates on mineral acquisition by Panicum virgatum in acidic soil. Mycorrhiza 9:167–176.
• Costa, C., L.M. Dwyer, C. Hamel, D.F. Muamba, W.L. Wang, L. Nantais, and D.L. Smith. 2001. Root contrast enhancement for measurement with optical scanner-based image analysis. Can. J. Bot. 79:23–29.
• Kaspar, T.C., and R.P. Ewing. 1997. ROOTEDGE: Software for measuring root length from desktop scanner images. Agron. J. 89:932–940.[Abstract/Free Full Text]
• Lebowitz, R.J. 1988. Digital image analysis measurement of root length and diameter. Env. Exp. Bot. 28:267–273.
• Miller, D.M. 1981. Studies of root function in Zea mays. II. Dimensions of the root system. Can. J. Bot. 59:811–818.
• Newman, E.I. 1966. A method for estimating the total length of root in a sample. J. Appl. Ecol. 3:139–145.
• Pan, W.L., and R.P. Bolton. 1991. Root quantification by edge discrimination using a desktop scanner. Agron. J. 83:1047–1052.[Abstract/Free Full Text]
• Richards, D., F.H. Goubran, W.N. Garwoli, and M.W. Daly. 1979. A machine for determining root length. Plant Soil 52:69–76.
• Smit, A.L., J.F.C.M. Sprangers, P.W. Sablik, and J. Groenwild. 1994. Automated measurement of root length with a three-dimensional high-resolution scanner and image analysis. Plant Soil 158:145–149.
• Tennant, D. 1975. A test of a modified line-intersect method of estimating root length. J. Ecol. 63:995–1001.
• Varney, G.T., M.J. Canny, X.L. Wang, and M.E. McCully. 1991. The branch roots of Zea. I. First order branches, their number, sizes and division into classes. Ann. Bot. (London) 67:357–364.[Abstract/Free Full Text]
• Varney, G.T., and M.E. McCully. 1991. The branch roots of Zea. II. Developmental loss of the apical meristem in field-grown roots. New Phytol. 118:535–546.
• Voorhees, W.B., V.A. Carlson, and E.A. Hallauer. 1980. Root length measurement with a computer-controlled digital scanning microdensitometer. Agron. J. 72:847–851.[Abstract/Free Full Text]
• Wells, C.E., and D.M. Eissenstat. 2001. Marked differences in survivorship among apple roots of different diameters. Ecology 82:882–892.[ISI]
• Wright, S.R., M.W. Jennette, H.D. Coble, and T.W. Rufty. 1999. Root morphology of young Glycine max, Senna obtusifolia, and Amaranthus palmeri. Weed Sci. 47:706–711.

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via ISI Web of Science (8) Citing Articles via Google Scholar Google Scholar Articles by Zobel, R. W. Search for Related Content PubMed Articles by Zobel, R. W. Agricola Articles by Zobel, R. W. Related Collections Root Development Software