Estimation of different soil properties using fast and inexpensive color sensor data

Soil color is one of the obvious characteristics of soil that usually has much to do with other soil properties. The NixTMPro color sensor can detect different amounts of soil color and allow you to check different soil properties. This sensor is less sensitive to the user's environmental and mental conditions than the conventional method of Mansell manual to determine the color of the soil, and it is effortless to use. Therefore, in this study, NixTMPro color sensor was used for fast and cheap estimation of different soil properties. For this purpose, 150 soil samples were collected from the semi-arid study area of Qazvin province. Values related to each soil characteristic were measured in the laboratory. Then, using this color sensor, the spectra of each color system were recorded for each soil sample. Two methods were used for this purpose. In the first method, by examining the correlation between the variables of the color system, an attempt was made to introduce a standard color system with the highest correlation coefficient with all soil properties. In the second method, all variables of different color systems were examined by the method of recursive feature elimination, which selects the features that have the highest accuracy by choosing the most important features. According to the results of both methods and to introduce a standard color system, in this study, CIEL*a*b color system was used to estimate soil properties. Because this system showed the highest degree of correlation with different soil properties. Then, using the random forest algorithm, the values related to each soil feature were estimated. Soil properties include sand, silt, clay, salinity, calcium carbonate (CaCO3), organic matter and soil bulk density. According to the results of random forest forecast, the amount of root mean square error (RMSE), mean error (ME), ratio of performance to interquartile distance (PRIQ), and the value of correlation coefficient (r) for each soil feature were determined. For sand, silt and clay particles, the amount of RMSE was 10.07, 6.28, 7.26%, and the correlation was 0.70, 0.49, and 0.77, respectively. PRIQ statistics for sand particles (2.09), clay (2.37), and calcium carbonate (1.78) are at an excellent and acceptable level. The RMSE values of organic matter, calcium carbonate and soil bulk density were 0.57, 0.02, 0.11%, and the correlation coefficient were 0.55, 0.58, and 0.70, respectively. Based on these results, it can be said that the spectra obtained from the NixTM Pro color sensor can be helpful in the rapid prediction of soil properties.