首页|An improved model for estimating soil organic matter content in salt-affected farmlands based on multi-source spectral data coupled with environmental variables
An improved model for estimating soil organic matter content in salt-affected farmlands based on multi-source spectral data coupled with environmental variables
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
Springer Nature
Abstract Purpose Accurate estimation of soil organic matter (SOM) content in a timely manner is crucial for soil management in salt-affected farmlands. The aim of the present study was to develop a machine learning model for SOM estimation in salt-affected farmlands using multi-source remote sensing data coupled with environmental variables.Materials and methods SOM content and ground hyperspectral reflectance (H) were measured in nine representative farmland sites of the Hetao Plain, northern China. Multi-source remote sensing images were acquired by Landsat 9 OLI (L), Sentinel- 2 MSI (S), and Sentinel- 1 SAR. In addition to single bands, 13 spectral indices (SI) were constructed and 11 environmental variables (EV) were introduced for SOM modeling. After variable selection by the gradient boosting machine, random forest models were developed based on different variable combination strategies.Results The SOM contents in the study area were generally low (2.24–23.70 g·kg−1) with moderate spatial heterogeneity. Temperature, precipitation, evapotranspiration, and combined radar polarimetric indices contributed substantially to SOM modeling. The model based on H + L + S + SI + EV showed the best performance, and its R2 value (0.893 in the validation set) was 0.249–0.318 greater than those of the models based on H + SI + EV, S + SI + EV, and L + SI + EV.Conclusion Shapley Additive exPlanations analysis identified mean annual precipitation as the overarching environmental variable influencing SOM estimation. Accurate mapping of SOM distribution across the study area was achieved using the optimal model with Sentinel- 2 images. This study presents a useful tool for rapid monitoring of SOM in salt-affected farmlands over large scales.
NETL
NSTL
Springer Nature
