A digital camera as an alternative tool for estimating soil salinity and soil surface roughness

• DOI: 10.1016/j.geoderma.2019.01.028
• Publication Date: 2019-01-24T00:50:04Z
• Authors: L. Xu; C.L. Zheng; Z.C. Wang; M.J. Nyongesah
• Keywords: 2. Zero hunger; 0401 agriculture, forestry, and fisheries; 04 agricultural and veterinary sciences; 15. Life on land; 6. Clean water

Abstract Determining soil salt content (SSC) and soil surface roughness (SSR) rapidly and simply is important for various environmental and agricultural applications. Recently, increasing attention has been focused on the color analysis to detect soil properties. Based on the fact that soil attributes affect soil color, this study investigated the feasibility of utilizing digital camera observations as an alternative for SSC and SSR estimation. In this study, field measurements were first conducted in the farmlands and wastelands in the inland river basin of Northwest China, and soil photographs were captured within the random-selected spots that were sampled for SSC and SSR measurements. Then, four color components, red (R), green (G), blue (B) and gray (Gr), were extracted from the raw photograph, and the digital number (DN) value of each pixel in each color component was calculated and counted. Each number of DN value was differently correlated with SSC and SSR. To find out certain DN ranges that are mostly sensitive to the variation of SSC and SSR, the whole DN value range (0–255) was separated equably into several partitions (from 2 to 10, 16, and 32), and the pixel number of each partition were recounted while the portion of every partition was computed for all pixels in each color component. After that, soil properties and these portions of four color components were regressed by partial least squares regression to find the best partitions according to the best determining coefficient (R2) and the ratio of performance to deviation (RPD). To obtain a better fit, we trained the models based on random training dataset for 100 times. Finally, SSC inversion model was built and evaluated with the excellent accuracy (R2 = 0.90 and RPD = 3.11), and SSR model with comparatively satisfactory accuracy (R2 = 0.71 and RPD = 1.87). We conclude that it is viable to apply digital cameras to estimate SSC and SSR. Findings from this study provide a base for research that can maximize the potential of digital cameras in estimating soil attributes under field conditions.