Spatial Estimation of Soil Organic Matter and Total Nitrogen by Fusing Field Vis-NIR Spectroscopy and Multi-spectral Remote Sensing Data

original data and the main code for the study. (1) Remote Sensing Data Processing First, the GaoFen-1 (GF-1) remote sensing data is clipped in ArcGIS to match the extent of the study area. Then, the clipped remote sensing data is processed using the ‘Remote sensing processing.R’ script in R, where spectral indices (such as NDVI, EVI, etc.) and the coefficient of variation (CV) are calculated to provide input data for subsequent modeling analysis. (2) Field Spectral Data Processing The field vis-NIR spectral data is standardized through Direct Standardization (DS) transformation, which is carried out using the 'PLS_Workspace Browse' tool in MATLAB. The DS transformation aims to eliminate environmental influences on the spectral data, thereby improving the comparability and consistency of the data. The transformed spectral data undergoes Principal Component Analysis (PCA) to extract principal components, which are then spatially interpolated in ArcGIS to generate a distribution map of the spectral principal components across the study area. (3) Modeling Analysis After the preprocessing of both remote sensing and field spectral data, modeling analysis is conducted using the ‘modelling.R’ script in R. This script integrates the preprocessed remote sensing data (spectral indices and coefficient of variation) and field spectral data (principal components). Machine learning algorithms,including Partial Least Squares Regression (PLSR) and Random Forest (RF), are employed to construct soil property prediction models and assess the model's performance (by ‘Accuracy. R’ script).