Whilst aboveground plant development and productivity can usually be easily observed and quantified, at least on the plot scale, quantification of roots still pose significant challenges. None of the methods available can quantify root biomass or turnover reliably. Furthermore, most methods are destructive, and there are few methods available to observe roots as they grow.

X-ray CT was first developed for medical applications, and was later used for a variety of industrial applications. The development scanners that could scan a variety of size classes, including small ones (micro-tomo- graphy) made it useful also for geological applications, including soils [1] . X-ray CT has had a variety of useful applications in studies of soil physics (e.g. [2] -[6] ), but studies of biological properties have turned out to be difficult as organic matter has a similar X-ray attenuation as air [7] .

Although imaging roots using X-ray CT can be challenging because roots are difficult to distinguish from pores, progress has been made [8] . The power of the technology lies in that the method is non-destructive, so that roots can be observed repeatedly over time as they grow. Some progress has been made in segmenting soil from pore space [9] -[11] . Segmentation of roots has also been attempted, but has so far succeeded for very young roots or small parts of roots [12] [13] .

In 3D scans of roots in soil the roots can be segmented out and volume and surface area can be calculated using dedicated software. However, it is not known how accurate these measures are. The purpose of this paper is to compare surface and volume calculated by software on 3D scans to weight the length of roots measured by traditional destructive techniques on roots washed out from the samples after scanning.

It was expected that weight and volume of roots would be strongly correlated, and that testing if the volume measured in the 3D image correlates with the weight should be a good test of the calculation from the 3D image. Root length and surface area were also expected to be correlated, but this correlation was expected to be weaker. Overall there was little correlation between measurements done on the roots and the parameters calculated from the 3D image (Table 1; Figure 1). The only strong and significant correlation was between measured weight and calculated volume for the varieties without root hairs showed (Table 1; Figure 1). Tillage treatment did not affect the relationship between measured and calculated parameters, but variety (with or without root hairs) did (Figure 1). How well roots can be quantified from 3D images therefore depends on the type or structure of the roots. Hirano et al. [16] found that detection frequency of roots below 1 cm in diameter was poor using root penetrating radar. Also in this study it was noted that very thin roots were not picked up by the software even when they were seen by the experimenter. Large roots or high resolution would help, but that means that with current technology only a small volume of soil can be examined. Any use of 3D imagery in following root development should start with a calibration like this for the type of roots and resolution to be used. Using a homogenous soil may also help, but the results here suggest that the quality of the roots are more important. It is not known why the roots without root hairs were easier to quantify. Although root hair are too small to be seen on a 3D image, they are also lost in root washing, so they would also not be included in the weight measurements. It is possible that varieties without root hairs compensate by being thicker or in other ways more distinct, and therefore easier for the software to follow.

Larger roots can be reliably quantified in 3D X-ray images, but the quantification is less reliable for smaller roots. However, reliability also depends on root structure in a way that is not fully understood, and any study of in situ root growth using X-ray tomography should start with a calibration of reliability for the roots and resolution to be used.

The author wishes to thank Dr. Tim George at James Hutton Institute in Dundee, UK for support at the field site VGstudioMAX support team in Heidelberg, Germany for support on the use of the software. This work was supported by the University of Abertay Dundee, UK.