A new method for high-precision estimation of soil organic matter using two-dimensional correlation spectroscopy—to support collaborative use of global open soil spectral libraries

Soil organic matter (SOM) plays a crucial role in controlling soil function and quality, mitigating greenhouse gas emissions, and improving the global carbon cycle. However, spectral analysis and scale effects have long been challenging issues in remote sensing estimation of SOM. The collaborative utilization of soil spectral libraries (SSL) provides a solution to scale effects, but the lack of effective spectral variables and the numerous protocols within SSL remain significant obstacles to implementing this technology. To address these issues, this study developed a new SOM estimation method using a local spectral library (470 samples) from the dry farming areas of China. In response to the difficulties in spectral analysis, the response mechanism of SOM in different spectral dimensions was analyzed. The impact of different mathematical transformations on the one-dimensional reflectance spectroscopy (ODRS) was analyzed, and the optimal mathematical transformation method (logarithmic first-order differential, (lgR)') was identified. However, ODRS still faced difficulties in accurately estimating SOM due to signal overlap and low signal strength. Two-dimensional correlation spectroscopy (TDCS) enables comprehensive analysis of interactions between different wavelengths, highlighting hidden information. In this study, an improved TDCS was developed by combining (lgR)' and TDCS. The enhanced TDCS increased the sensitivity and resolution of the spectra, providing a platform for studying interactions between different functional groups in the soil. Several optimization algorithms were used to optimize the best models in different spectral dimensions. The TDCS-(lgR)'-DSI-CARS model, which combines competitive adaptive reweighted sampling (CARS) with TDCS, obtained optimal parameters (RV2, RMSEV, and RPIQ reaching 0.92, 1.51, and 5.75, respectively), achieving precise estimation of SOM. Using mathematical models, simulations were performed for different spectral equipment scenarios, demonstrating that the proposed model was adaptable to the majority of high-spectral devices available on the market, effectively addressing the SSL alignment issue. Finally, the proposed modeling strategy was replicated in different soil types within a large-scale area, achieving excellent estimation accuracy (both RV2 and RPIQ exceeded 0.80 and 3, respectively). Thus, the integration of this research's modeling strategy with SSL effectively responds to the scale effects in SOM remote sensing estimation. Additionally, this study proposed a methodology for constructing the TDCS library, which offers a straightforward process and lower costs for the collaborative use of SSL. The research findings provide theoretical and methodological support for SOM remote sensing mapping, sensor development, and the collaborative utilization of remote sensing data with large-scale SSL.