Automatic segmentation and quantitative analysis of soil preferential flow using dye tracer technology

Abstract: Preferential flow is widely considered to be a common phenomenon of water movement in soil. Currently, dye tracer can be one of the most efficient ways to characterize the preferential flow using soil-stained images. However, the general image processing software, such as Photoshop, Image Pro Plus, and Image J, cannot specifically extract the soil-stained images with inconsistent chromaticity and low contrast between dyed and non-dyed areas. A larger error occurs normally in the subsequent quantitative analysis for the preferential flow pathways. This study aimed to propose an automatic segmentation for preferential flow pathways using dyed tracer images and to further improve the accuracy and efficiency of quantification. An image processing was performed on the dyeing images of preferential flow, thereby quantitatively analyzing specific parameters. Firstly, brilliant blue dye was used to stain subsurface flow pathways in soil plots from natural secondary forest and hazelnut shrub forest during simulated rainfall events under dry conditions. The dyed tracer images were converted into the hue-saturation-value (HSV) space for the extraction of hue (H) component, in order to improve the contrast of dyed images and highlight the preferential flow path. Fuzzy C-means based on H component and morphology (HM-FCM) was selected to automatically segment the dyeing area. Morphological opening and closing arithmetics were used to fix under- and over-segmentation in the images. Secondly, mathematical statistics were selected to quantificationally analyze multiple indicators of soil preferential flow in the high-precision graphs of natural secondary forest and hazelnut shrub forest. The specific parameters included total dyeing area ratio, matrix flow depth, preferential flow ratio, and fractal dimension. The proposed segmentation well accurately identified the distribution of preferential flow pathways in forest soil and automatically segmented the dyeing area. Furthermore, multiple indicators were achieved for the subsequent evaluation of preferential flow and topological structure. Specifically, the preferential flow in the natural secondary forest occurred earlier than that in the hazelnut forest, whereas, the development degree of preferential flow in the natural secondary forest soil was higher than that in hazelnut forest soil. The dyeing areas of the two forests were generally concentrated in the soil layer of 0-50 cm, where the dyeing area ratio of hazelnut forest was higher than that of natural secondary forest. The water infiltration behaved mostly the uniform flow with less preferential flow. It was found that HM-FCM effectively segmented the soil dyeing areas of two forests. The segmentation accuracy was 87.9% for the images of natural secondary forest, and the harmonic mean was 90.5%, whereas, the segmentation accuracy was 83.3% for the images of hazelnut shrub forest, and the harmonic mean was 80.3%. There were different development degrees in the priority flow (P<0.05). The proposed automatic segmentation can be widely expected to identify the preferential flow and migration in the underground soil of various woodlands for sustainable forestry.