Impact
Statement

PDF version
for printing

© 2003 Plant Management Network.
Accepted for publication 2 June 2003. Published 2 July 2003.

Evaluation of Digital Image Acquisition Methods for Determining Soybean Root Characteristics

Loretta M. Ortiz-Ribbing, Senior Research Specialist, and Darin M. Eastburn, Associate Professor, Department of Crop Sciences, University of Illinois, Urbana 61801

Corresponding author: Darin Eastburn. eastburn@uiuc.edu

Ortiz-Ribbing, L. M., and Eastburn, D. M. 2003. Evaluation of digital image acquisition methods for determining soybean root characteristics. Online. Crop Management doi:10.1094/CM-2003-0702-01-RS.

Abstract

Advances in root research have occurred with the use of computer digital image analysis. The WinRhizo system has the capability of analyzing images acquired from a flatbed scanner, digital or video camera, and image files. The objectives of this study were to compare the use of a digital camera with that of a flatbed scanner as the method of acquiring root system images, and to determine the usefulness of the WinRhizo software for quantifying root characteristics of soybean [Glycine max (L.) Merr.]. Digital camera and scanner images of string segments and soybean roots from the field and greenhouse were analyzed using the WinRhizo software. Root surface area, volume and average diameter estimates from digital camera images were up to 3, 17, and 6 times greater, respectively, than those estimated from a scanned image of the same root. Immersing roots in water prior to scanning significantly increased estimated values for root length, surface area and volume by 13, 20, and 33 percent, respectively. Digital image analysis of root systems depends on the quality of the root image, and on the preparation of the root system for photographing or scanning. Computer digital image analysis can be used effectively to analyze root systems if special precautions are taken to avoid certain errors.

Introduction

Traditionally, measuring root systems was a labor-intensive, tedious task (2,4,8). However, state-of-the-art digital imaging systems now have the potential to make root analysis less time-consuming (12,17). In addition, analysis of digital images can yield highly accurate root measurements (13,14), and estimates have been shown to correlate well with those obtained through the traditional line-intercept method (12).

For these reasons, characterization of root systems using computer digital imaging software in combination with scanners, digital cameras, and video cameras, has increased in recent years (4). One such system, WinRhizo (Ver. 5.0A; Regent Instruments, Quebec, Canada), has been used for a number of studies involving root analysis. For example, Bouma et al. (3) used WinRhizo in a sensitivity analysis of scanning protocols for estimating root length and average diameter of fine grass root systems. The WinRhizo system has also been used to quickly and accurately determine characteristics of pea (Pisum sativum L.) (11) and muskmelon (Cucumis melo L., var. cantalupensis) (1) root systems from scanned images. In these studies, root responses were related to disease infection by two root rot pathogens. However, in both of these studies, the WinRhizo system was used with a scanner for acquiring images, and root systems were not immersed in water during the scanning process. A high-resolution scanner may be cost-prohibitive for some applications, and root images obtained by other methods, such as a digital camera, may need to be evaluated using the WinRhizo software. In these instances, the accuracy of the WinRhizo system to estimate root characteristics from lower resolution digital images must be evaluated. Scanning root systems immersed in water allows fine roots to spread more easily and decreases root overlap. Although according to Regent, Inc. WinRhizo compensates for root overlap, precision of the estimates should increase as root overlap decreases.

The objectives of this test were to compare the use of a digital camera with that of a flatbed scanner as the method of acquiring root system images, and to compare the accuracy of these two methods using the WinRhizo software program for measuring characteristics of soybean root systems. In addition, a comparison was made between estimates of root characteristics obtained from scanned roots immersed in water and those spread out on the scanner tray without water.

Growing Soybean Plants and Root Systems

Soybean root systems were collected from research plots at the University of Illinois Crop Sciences Research and Education Center in Urbana, Illinois, during the 2001 and 2002 growing seasons. Seeds were planted on 2 May 2001 and 31 May 2002 in 30-inch and 15-inch wide rows, respectively. Soybean plants were collected on 6 June 2001 and 27 June 2002 from 288 and 64 random locations in the field, respectively. In addition, one set of ten root systems also was collected from older plants on 8 August 2002. Soil and roots were sampled to a depth of 10 to 12 inches from a circle (approximately 8-inch diameter) around each plant. To remove soil from the roots, the intact root system and soil were soaked in a 5% Calgon (sodium hexametaphosphate) solution to disperse soil adhering to roots (10) prior to washing on a root screen. Root volumes were determined using the water displacement method, and the roots were kept moist and cold (43°F) until the roots were scanned and root characteristics determined using the WinRhizo system. Scanning occurred within 2 to 14 days after washing.

To obtain reliable samples of small root systems, soybean plants were grown in the greenhouse in white, polypropylene 32-ounce containers (pots), measuring 5.5 inches tall, with an upper lip measuring 4.25 inches in diameter tapering down to 3.25 inches in diameter at the bottom. Four holes were drilled in the bottom of each pot for drainage. Pots were filled with a sand:soil (2:1) mix. Two soybean seeds were placed in the center of each pot and seeds were covered with soil. Pots were randomly placed on a greenhouse bench under 1000-watt, high-pressure, sodium vapor lights providing 16 hours of daylight. Temperature was set at 77 ± 3°F (25 ± 3°C) during the day, and 72 ± 3°F (22 ± 3°C) for the 8-hour dark period. Plants were watered as necessary to maintain adequate soil moisture. Plants were thinned to one per pot after emergence. Soybean roots were collected and washed approximately 30 days after planting, and root volumes were determined using the water displacement method. Root systems were then scanned and root characteristics determined through digital analysis.

Acquisition Methods and WinRhizo Analysis

Two methods were used to acquire digital images of string segments and soybean root systems. In the first method, a Sony 10X Digital Mavica camera, model MVC-FD7, was used to capture string segment or root system images. Root systems and string segments were photographed at a camera resolution of 640 × 480 pixels. The string segments or root systems were placed on a camera stand on clear glass and photographed against a black background, and the total image surface area for each image was recorded. Light was supplied by two 150-watt reflector flood light bulbs spaced 39 inches apart, at approximately a 165° angle to the camera stand. Additional lighting caused problems with camera shadows. The upper vegetative portion of the soybean plant was removed above the uppermost lateral root. Forceps and a dissecting needle were used to gently spread apart overlapping roots. The black background around the images was edited in Corel Photo-Paint8 (Corel Corporation, Ontario, Canada) to remove camera, string, and root shadows, as well as dirt and dust specks.

In the second method, string segment or root system images were generated using a digital flatbed scanner (Regent Instruments LA1600, Epson model EU-22) according to the manufacturer instructions. The string segments or root samples were placed in clear plastic trays on the scanner bed. In most experiments water was added to the trays prior to scanning, as per the manufacturers instructions. The scanner was equipped with a “tpu” (transmitted) light source in the lid, so light came from above the roots during the scanning process. Root systems were scanned at 400 dpi with a pixel size of 0.0025 inches (0.063 mm).

Digital images and scanned images were analyzed using WinRhizo (Ver. 5.0A; Regent Instruments, Quebec, Canada) root analysis software on a Dell Dimension 4100, Pentium III, 996 Mhz computer, with 256 Mb ram (Dell Computer Corporation, Round Rock, TX). This method calculated morphological root measurements based on Tennant’s (16) statistical line intersect method. Using the automatic Tennant method, all root images were analyzed for root length, volume, surface area, and average diameter. According to manufacturers directions, root images obtained with the digital camera were analyzed after designating color classes for the root system and background, and after defining image surface area from recorded values for each image.

Experiment 1: Calibration with String and Root Volume

Calibration of other digital analysis systems has been done using segments of string, thread, or wire (8,9,12,14,15). In this study, three pieces of string with individual lengths of 7.91, 7.95 and 7.99 inches (20.1, 20.2 and 20.3 cm) were used to calibrate the WinRhizo system and to compare the two methods of image acquisition. The mathematically calculated surface area of the string was determined (Table 1) using the formula for the total surface area of a cylinder. The volume of the three pieces was calculated mathematically as the volume of a cylinder, and the average diameter was determined with an electronic caliper. For calibration purposes, these three pieces of string were photographed with a digital camera in ten different positions, and analyzed according to the previously described methods. The same three pieces of string were scanned ten times, without immersing in water, arranging the string on the scanner bed in different positions for each scanning.

Table 1. Comparison of a flatbed scanner and digital camera as the method of acquiring digital images of string for dimensional analysis.

* Actual = Mathematical calculations of length, total surface area of a cylinder, volume of a cylinder, and average diameter for the three pieces of string.

† The average of the percent difference was calculated between the mathematically derived treatment means for string characteristics and those estimated using a scanner or digital camera images with the WinRhizo system.

‡ Estimated values within rows followed by the same letter were not significantly different in a paired t test (P < 0.001).

To determine the accuracy of the WinRhizo system to estimate root volume, root volumes estimated by the WinRhizo system were compared with root volumes for the same root systems measured using a water displacement method (7). The root volumes of small root systems, from the field and greenhouse, were first measured using water displacement, and this measurement was used as the standard or reference value. These roots were then placed in clear plastic trays, immersed in water and scanned, and root volume was estimated from the resulting image using the WinRhizo system.

Experiment 2: Comparing Acquisition Methods

Soybean root systems were used to evaluate the two image acquisition methods. Root systems, from fifteen 4-wk-old greenhouse-grown soybean plants (small root systems) were photographed with a digital camera as described previously. These same roots were then scanned using the flatbed scanner. In order to determine whether immersing roots in water affected the accuracy of calculating root properties, roots were first scanned in the tray without water, and then water was added to the tray and the roots were rescanned. The resulting images (digital camera, scanned without water, and scanned with water) were then analyzed to determine root surface area, length, average diameter, and volume, according to the previously described methods. This experiment was conducted twice, and allowed a comparison between acquisition methods (digital camera vs. scanner) and between scanning methods for small root systems (water vs. no water in the scanner tray). Another trial comparing acquisition methods for root images was conducted using ten 3-month-old soybean root systems from the field (large root systems). These roots were grown, as previously described, in the field in 2002. However, this trial was not repeated.

Statistical Analysis

Simple regression analysis of data was conducted using the General Linear Models procedure in SAS (SAS version, 8.01, SAS Institute, Cary, NC). A paired-comparisons t-test with Proc Means in SAS (alpha level of 0.05) was used to compare the methods. Soybean plants grown in the greenhouse were organized using completely randomized designs. The experiment comparing the accuracy of image acquisition methods had fifteen samples and was repeated twice. The greenhouse experiment for root volume had 240 samples and was conducted three times. The field experiment for root volume had 352 samples and was conducted twice.

Results and Conclusions

Experiment 1: Calibration with string and root volume. Using a scanner to acquire images produced significantly (P < 0.001) more accurate estimates of length, surface area, volume, and average diameter of the string segments than digital camera images (Table 1). The scanned images had slightly overestimated surface area values compared to the actual surface area of the string. The surface area, volume and average diameter of the string segments, estimated from digital camera images, were more than 100 times greater than the actual values. However, string length was significantly underestimated from digital camera images.

Root volumes obtained from scanned root images were lower than those obtained by direct water displacement of the root system (Fig. 1). Estimates of root volume determined using water displacement for soybean root systems from the greenhouse experiments were approximately 2 times greater than root volumes determined through analysis of scanned images (r² = 0.82). For field-grown soybean roots, root volumes determined by water displacement were 3.6 times greater than root volumes determined through analysis of scanned images, and the r² value dropped to 0.57.

Fig. 1. Comparison of root volume measurements determined using water displacement and WinRhizo image analysis methods for soybean plants grown in the greenhouse (top) and field (bottom). Greenhouse data represents the combination of results from three separate trials. Field-grown soybeans were collected in June 2001 and 2002.

Experiment 2: Comparing acquisition methods. Large differences existed for all root characteristic estimates between the two methods of acquiring root images (scanner vs. digital camera) from greenhouse-grown plants (Table 2). Data from the two greenhouse trials were not combined because a significant interaction existed between the trials and the methods of acquisition. However, similar trends were observed with the root images as with the string images. Larger values for root surface area, volume, and average diameter were obtained when estimated from a digital camera image as compared to a scanned image, while root length was less when estimated from a digital camera image. Root surface area estimates of field-grown, large (older) root systems and greenhouse-grown, small (younger) root systems were significantly different when a digital camera was used as a means of acquiring images compared to a flatbed scanner. Scanned images of the smaller root systems, resulted in significantly longer total root length estimates for the same roots when compared to estimates from digital camera images. However, root length estimates of the larger root systems were not significantly different when comparing images acquired with a scanner to those obtained using a digital camera. Values for root volume and average diameter were significantly different between the scanner and camera for both large and small root systems. Smaller differences between scanned root images and digital camera images were seen for all root characteristics when root systems were larger and had fewer fine roots. However, the comparisons between large and small root systems were based on the results of only one trial.

Table 2. Comparison of root characteristic estimates for the same root system obtained using either a flatbed scanner or a digital camera for acquiring the root image for analysis.

* Means from the two greenhouse trials were significantly different and a significant trial by method interaction was observed.

† The average percent difference was calculated from treatment means as the difference between the root characteristic values estimated from scanner images and those estimated from the digital camera images as: ((Camera- Scanned) / Scanned) × 100.

‡ For each trial, means within a row followed by the same letter were not significantly different in a paired t test (P < 0.001).

Significant differences in values of root system characteristics were observed for the same root system when it was immersed in water and scanned compared to when it was scanned without adding water to the tray (Table 3). There was no significant interaction between trial and the use of water during scanning, so data from both trials were combined. Average root diameter of greenhouse-grown plants was the only estimated value that was not significantly affected by the presence or absence of water in the scanner tray. Estimates of root length, surface area, and volume were significantly greater when root systems were immersed in water during the scanning process.

Table 3. Comparison of root characteristic estimates for the same root systems of greenhouse-grown soybean plants scanned with or without the use of water.

* Data are the combined results of two trials since there were no significant interactions between trial and method.

† The average percent difference was calculated from treatment means as the difference between the root characteristic values estimated from images scanned with water in the scanner tray and those scanned without water in the scanner tray as: ((no water - water) / water) × 100.

‡ Means by row followed by the same letter are not significantly different (P < 0.0001).

Considerations when using digital imaging. Computer image analysis of digital images of root systems obtained using a camera or flatbed scanner was an effective method for estimating root characteristics. However, there were problems that reduced the accuracy estimates obtained by image analysis. The biggest source of error in our research was the digital image itself. The accuracy of digital image analysis of root systems was only as good as the initial image of the roots. In addition, the quality of the image depended on the preparation of the root system prior to photographing or scanning. Root systems need to be thoroughly cleaned to remove soil particles and root debris. Root debris was a problem for photographing or scanning root systems because, in both methods, minute amounts of soil and debris could appear in the images and affect the results.

Large differences were observed between root characteristic estimates for the same root from scanned and camera images. This inconsistency could have occurred because the positions of the root system in the two digital images were different. Regardless of how carefully roots were placed, individual roots would be crossed differently and laying in different orientations on the camera stand and in the scanner tray. We found that scanning roots was more convenient than photographing roots, and that scanning produced a better image.

The digital camera used could only photograph pictures at resolutions of 640 × 480 pixels. Fine roots that could not be distinguished from the background were not measured using this system; therefore, root length was underestimated. In addition, when color scales where designated within the WinRhizo system so that it could distinguish between root colors and background colors, dark colored areas on the root caused by damage, dirt, and normal discoloration were not always distinguished from the background. Hence these areas were not included in the total root length. Video and digital cameras often produce low-resolution images, and this is a major problem when using them with computer imaging systems (4). Commins et al. (5) found results of image analysis of roots in resin impregnated blocks depended largely on camera resolution and contrast and focus settings on the camera.

Another problem associated with the use of the digital camera was that it was nearly impossible to photograph the string or roots at a proper exposure without some string, root, or camera shadow appearing in the image. In order to avoid shadows, string and roots had to be photographed at a lower exposure with less lighting. For roots, this resulted in a greater portion of the taproot or smaller roots not being distinguished from the background. Murphy and Smucker (12) found that the fineness of the roots caused detection problems for the camera. Ottman and Timm (14) experienced similar problems with lighting.

With digital camera images, smaller roots were not easily discernable, and some roots were grouped, giving the appearance of thicker individual roots. In addition, because of low resolution of the digital camera, edges of individual roots were not smooth and appeared larger. Some root shadows could not be removed even with the image-editing program Corel Photo-Paint8. As a result, when digital root images were analyzed their root surface areas were overestimated, because the shadow made individual roots appear wider. We feel this probably resulted in the significant trial by method interaction because the soybean plants in Trial 2 had healthy, more fibrous root systems with a large amount of fine roots. Plants in Trial 1 had been inoculated with a root pathogen and the root systems were smaller and less fibrous. Ottman and Timm (14) found the opposite. They found lower root surface area estimates when photographed roots were analyzed with a computer. These factors would explain why estimates of root surface area, root volume, and average diameter were greater when camera images were analyzed. Therefore, it is advisable not to combine data obtained by different methods of acquiring images unless calibration between methods has shown similar results.

Care needs to be taken in order prevent technical errors when using computer digital image analysis systems. When calibrated with string, the WinRhizo system slightly overestimated the actual values of length, surface area, volume, and average diameter. According to manufacturers instructions, the accuracy of this system increases as root overlap decreases, but it was almost impossible to spread out the root system to completely avoid overlap even when the roots were immersed in water. With the WinRhizo system, root volume was underestimated when compared to volumes determined using water displacement. It is uncertain why this occurred. One explanation could be that root volumes determined by water displacement occurred immediately after sampling roots, when root systems were fresh. Even though stored properly, the root systems could have lost enough moisture to account for a difference in volume, over the 2-to-14-day period it took to scan roots. The volume and other root parameters might have increased if the root systems were stained. Staining has been suggested as a means of sample preparation for scanner-based image analysis, particularly for species with a large amount of fine roots (3,6). However, staining has been shown not to enhance image analysis of soybean roots (6).

For small soybean root systems, we found it was more important to float the roots in a tray of water during scanning. Immersing roots in water allowed individual roots of small root systems to spread apart more easily. Therefore, scanning roots in water would pick up a greater number of individual roots, accounting for significantly larger estimates for all root characteristics except average root diameter when compared to scanning roots without water. We recommend that small soybean root systems having many flexible, fine roots be immersed in water for scanning. This procedure is not feasible for large, rigid soybean root systems that grow in 3 dimensions.

Computer digital image analysis can be used effectively to aid in the study of root systems. The method used to acquire the digital image and the quality of the image is important and will make a difference in the estimated values for root characteristics. Similar to the results of Kraft and Boge (11) and Biernacki and Bruton (1), we found that computer image analysis made it easy to measure root system characteristics. In our study, the use of computer image analysis was easier and provided more information than the conventional water displacement method for measuring root volume. Others have found that estimates of root length obtained through computer image analysis were as accurate and easier to generate than estimates obtained using the traditional line-intercept method (11,12).

Acknowledgements

Funding was provided by the Illinois Soybean Program Operating Board and by the UIUC Campus Research Board. We want to thank Dr. Scott Isard for the use of his digital camera and for his time reviewing this manuscript. We also thank Matthew Curtis for his help in the lab and greenhouse.

Literature Cited

1. Biernacki, M., and Bruton, B. D. 2001. Quantitative response of Cucumis melo inoculated with root rot pathogens. Plant Dis. 85:65-70.

2. Böhm, W. 1979. Methods of studying root systems, Pages 64-71 in: Ecological studies: Analysis and synthesis, Vol. 33. W. D. Billings, F. Golley, O. L. Lange, J. S. Olson, eds. Springer-Verlag, New York.

3. Bouma, T. J., Nielsen, K. L., and Koutstaal, B. 2001. Sample preparation and scanning protocol for computersied analysis of root length and diameter. Plant and Soil. 218:185-196.

4. Box, J. E., Jr. 1996. Modern methods for root investigations. Pages 193-237 in: Plant Roots: The Hidden Half. Y. Waisel, A. Eshel, and U. Kafkafi, eds. Marcel Dekker, New York.

5. Commins, P. J., McBratney, A. B., and Koppi, A. J. 1991. Development of a technique for the measurement of root geometry in soil using resin-impregnated blocks and image analysis. J. Soil Sci. 42:237-250.

6. Costa, C., Dwyer, L. M., Hamel, C., Muamba, D. F., Wang, X. L., Nantais, L., and Smith, D. L. 2001. Root contrast enhancement for measurement with optical scanner-based image analysis. Can. J. Bot. 79:23-29.

7. Curran, M., James, P., and Allaway, W. G. 1996. The measurement of gas spaces in the roots of aquatic plants-Archimedes revisited. Aquatic Botany 54:225-261.

8. Dowdy, R. H., Nater, E. A., and Dolan, M. S. 1995. Quantification of the length and diameter of root segments with public domain software. Commu. Soil. Sci. Plant Anal. 26:459-468.

9. Harris, G. A., and Campbell, G. S. 1989. Automated quantification of roots using a simple image analyzer. Agron. J. 81:935-938.

10. Khan, A., Atibalenja, N., and Eastburn, D. M. 2000. Influence of inoculum density of Verticillium dahliae on root discoloration of horseradish. Plant Dis. 84:309-315.

11. Kraft, J. M., and Boge, W. 2000. Root characteristics of pea in relation to compaction and Fusarium root rot. Plant Dis. 85:936-940.

12. Murphy, S. L., and Smucker, A. J. M. 1995. Evaluation of video image analysis and line-intercept methods for measuring root systems of alfalfa and ryegrass. Agron. J. 87:865-868.

13. Nilsson, H. E. 1995. Remote sensing and image analysis in plant pathology. Annu. Rev. Phytopathol. 15:489-527.

14. Ottman, M. J., and Timm, H. 1984. Measurement of viable plant roots with the image analyzing computer. Agron. J. 76:1018-1020.

15. Smika, D. E., and Klute, A. 1982. Surface area measurement of corn root systems. Agron J. 74:1091-1093.

16. Tennant, D. 1975, A test of the modified line intersect method for estimating root length. J. Ecol. 63:995-1001.

17. Wright, S. R., Jennette, M. W., Coble, H. D., and Rufty, T. W., Jr. 1999. Root morphology of young Glycine max, Senna obtusifolia, and Amaranthus palmeri. Weed Sci. 47:706-711.