Digital Soil Mapping of Soil Organic Matter with Deep Learning Algorithms

Digital soil mapping has emerged as a new method to describe the spatial distribution of soils economically and efficiently. In this study, lightweight organic matter (SOM) based on deep residual network, which we call LSM-ResNet, is proposed make accurate predictions with background covariates. ResNet not only integrates information around observed environmental covariates, but also reduces problems such loss, undermines integrity prediction uncertainty. To train model, rectified linear units, mean squared error, adaptive momentum estimation were used activation function, loss/cost optimizer, respectively. The was tested Landsat5, meteorological data from WorldClim, 1602 sampling points set Xinxiang, China. performance LSM-ResNet compared traditional machine learning algorithm, random forest (RF) training (80%) test (20%) created both models. results showed that (RMSE = 6.40, R2 0.51) model outperformed RF in roots square error (RMSE) coefficient determination (R2), accuracy significantly improved 6.81, 0.46). trained for SOM district plain terrain maps can be deemed an reflection variability distribution.