Integrative remote sensing and machine learning approaches for SOC and TN spatial distribution: Unveiling C:N ratio in Black Soil region

Abstract

The carbon-to-nitrogen ratio (C:N ratio) in soil is a key indicator for assessing soil quality and health. Research and monitoring of this ratio is critical for understanding soil ecosystem functions and agricultural productivity. However, mapping multiple soil properties simultaneously is more challenging than mapping individual attributes. Therefore, this study aimed to develop an approach for jointly mapping soil organic carbon (SOC) and total nitrogen (TN) and to evaluate their spatial C:N ratio. In this study, we used multi-year remote sensing imagery, environmental covariates, and 188 soil samples. Optimal features were selected using the Recursive Feature Elimination (RFE), and the Random Forest model was applied to map the spatial distribution of SOC and TN in a typical black soil region. Finally, we analyzed the C:N ratio in the study area. The results indicated that: (1) Multi-temporal remote sensing imagery significantly enhanced SOC and TN mapping compared to single-temporal imagery. Environmental covariates positively contributed to mapping accuracy, but data redundancy remained; (2) RFE improved mapping accuracy, increasing the R 2 value of SOC by 0.035 and reducing RMSE by 0.28 g/kg, while TN's R 2 value increased by 0.040, and RMSE decreased by 0.02 g/kg; (3) The sensitive features for SOC and TN mapping differed, with the B2 and B3 bands of Sentinel-2 imagery being most sensitive for SOC mapping, while the B12 and B11 bands were most sensitive for TN mapping; (4) The contrast between paddy and dry fields was a key factor influencing the spatial distribution of the C:N ratio in the study area, with the C:N ratio in dry fields being higher than in paddy fields, primarily due to the excessive nitrogen content in paddy fields. In summary, this study presents an effective remote sensing monitoring method for accurately mapping the spatial distribution of SOC and TN in typical black soil region, and enhances understanding of soil health and agricultural ecosystems through C:N ratio analysis. • Multi-temporal imagery significantly improved SOC and TN mapping accuracy, but data redundancy remained. • RFE can effectively remove redundant data and improve mapping accuracy. • TN and SOC mapping had different optimal sensitive bands in Sentinel-2 imagery. • The significant difference in C: N ratio between dry and paddy fields indicates the impact of land use on soil health.

Topics & Concepts