Spatial variations of soil carbon, nitrogen and their stable isotopic compositions under diverse vegetation types in an urban park: application of laboratory-based hyperspectral reflectance spectroscopy

Abstract:

Urban soil is a heterogeneous mixture of various parent materials and significantly affected by anthropogenic activities. In this study, we first investigated the spatial variations of soil organic carbon (SOC), total nitrogen (TN), δ¹³C and δ¹⁵N values in urban soils across different vegetation types in an urban park. We took hyperspectral reflectance for each soil sample using ASD-FieldSpec after drying and removing stones (> 2 mm diameter), then used partial least squares (PLS) regression to establish the predictive models for SOC, TN, δ¹³C and δ¹⁵N values in urban soils. A total of 102 samples were collected and the SOC, TN, δ¹³C and δ¹⁵N values of topsoil (0-20 cm) were measured under five vegetation types, including mixed forests, deciduous forest, needle-leaf forest, urban lawn and waterfront plant. The SOC and TN stocks varied between 0.33 kg m^-2 to 12.51 kg m^-2 and 0.02 kg m^-2 to 0.78 kg m^-2, respectively. The δ¹³C and δ¹⁵N data varied between -30.18‰ to -17.17‰ and -3.39‰ to 3.20‰, respectively. The average SOC, TN, δ¹³C and δ¹⁵N data showed a clear vegetation-dependent pattern. The PLS model achieved acceptable results with coefficient of determination (R_c²) and root mean square error (RMSE_c) of calibration set for SOC (R_c² = 0.67; RMSE_c = 2.2%), TN (R_c² = 0.87; RMSE_c = 0.2%), δ¹³C (R_c² = 0.98; RMSE_c = 5.0‰) and δ¹⁵N (R_c² = 0.97; RMSE_c = 2.2‰), respectively. The leave-one-out cross-validation (LOOCV) procedure confirmed the robust performance of PLS model. The results indicated that 1) the SOC, TN, δ¹³C and δ¹⁵N can be estimated with reasonable accuracy across heterogeneous urban landscapes based solely on the hyperspectral reflectance spectroscopy, and 2) the strategy has the potential of upscaling soil properties toward city-level assessments of soil health.