Purpose Digital camera color images have been widely and effectively used for assessing soil organic matter (SOM), but they are influenced by both subjective and objective factors. Therefore, we investigated the optimized conditions and models for a colorimeter to rapidly and accurately assess SOM content. Materials and methods This study utilized a colorimeter with three different measure hole sizes (3, 6, and 11 mm) to rapidly estimate SOM content in arable black soils in northeast China, by analyzing the RGB, CIE-Lab, CIE-Luv, and HSI color component features. Multiple linear regression (MLR), radial basis function (RBF) and multi-layer perceptron (MLP) neural network models were employed for predicting SOM within these color spaces. Results and discussion The results revealed a significant negative correlation between all color components reliably obtained using the 11-mm measure hole and SOM contents (P < 0.001), with correlation coefficients (r) ranging from -0.53 to -0.74, except for the H color component. The MLR, RBF and MLP models achieved Rt2 and Rv2 of 0.57–0.58, 0.53–0.57, and 0.54–0.60, with RMSEt and RMSEv of 1.14%–1.18%, 1.22%–1.27%, and 1.20%–1.24% respectively. While no significant differences in performance were observed among these models (P > 0.05), performance improved significantly after soil classification. The neural networks, particularly the MLP algorithm, trained with RGB color components exhibited increased Rt2 and Rv2 for Hapli–Udic Isohumosols, Typic Dark–Aquic Cambosols, Mollic Bori–Udic Cambosols, Pachi–Ustic Isohumosols, and Typic Calci–Ustic Isohumosols, reaching 0.60, 0.73, 0.58, 0.80, and 0.66, as well as 0.62, 0.77, 0.63, 0.80, and 0.78, respectively. Concurrently, the RMSEt and RMSEv values decreased to 0.99%, 0.87%, 1.17%, 0.48%, and 0.63%, as well as 0.97%, 0.83%, 1.06%, 0.53%, and 0.57%, respectively. Conclusions We recommend using a colorimeter, particularly with MLP neural network modeling following soil type determination, for large–scale applications in assessing SOM content. This method offers a rapid and alternative approach for determining SOM in northeast China, facilitating in-situ and profile soil investigation, as well as enabling quick assessments of cultivated land quality grades. Graphical Abstract

Swift evaluation of black soil organic matter content in northeast China using a colorimeter

期刊论文
发表日期：

2025年1月31日

查看全文

Feng Zhang,
Yadan Wang,
Yiyi Deng,
Wenyou Hu,
Decheng Li,
Dongshen Yu,
Shunhua Yang,
Fengqin Chi,
Chao Zhang,
Yingde Xu,
Jun Jiang,
Renkou Xu

摘要

Purpose

Digital camera color images have been widely and effectively used for assessing soil organic matter (SOM), but they are influenced by both subjective and objective factors. Therefore, we investigated the optimized conditions and models for a colorimeter to rapidly and accurately assess SOM content.

Materials and methods

This study utilized a colorimeter with three different measure hole sizes (3, 6, and 11 mm) to rapidly estimate SOM content in arable black soils in northeast China, by analyzing the RGB, CIE-L*a*b*, CIE-L*u*v*, and HSI color component features. Multiple linear regression (MLR), radial basis function (RBF) and multi-layer perceptron (MLP) neural network models were employed for predicting SOM within these color spaces.

Results and discussion

The results revealed a significant negative correlation between all color components reliably obtained using the 11-mm measure hole and SOM contents (P < 0.001), with correlation coefficients (r) ranging from -0.53 to -0.74, except for the H color component. The MLR, RBF and MLP models achieved R_t² and R_v² of 0.57–0.58, 0.53–0.57, and 0.54–0.60, with RMSE_t and RMSE_v of 1.14%–1.18%, 1.22%–1.27%, and 1.20%–1.24% respectively. While no significant differences in performance were observed among these models (P > 0.05), performance improved significantly after soil classification. The neural networks, particularly the MLP algorithm, trained with RGB color components exhibited increased R_t² and R_v² for Hapli–Udic Isohumosols, Typic Dark–Aquic Cambosols, Mollic Bori–Udic Cambosols, Pachi–Ustic Isohumosols, and Typic Calci–Ustic Isohumosols, reaching 0.60, 0.73, 0.58, 0.80, and 0.66, as well as 0.62, 0.77, 0.63, 0.80, and 0.78, respectively. Concurrently, the RMSE_t and RMSE_v values decreased to 0.99%, 0.87%, 1.17%, 0.48%, and 0.63%, as well as 0.97%, 0.83%, 1.06%, 0.53%, and 0.57%, respectively.

Conclusions

We recommend using a colorimeter, particularly with MLP neural network modeling following soil type determination, for large–scale applications in assessing SOM content. This method offers a rapid and alternative approach for determining SOM in northeast China, facilitating in-situ and profile soil investigation, as well as enabling quick assessments of cultivated land quality grades.

Graphical Abstract