Dynamics of soil water and nitrate within the vadose zone simulated by the WHCNS model calibrated based on deep learning

Abstract

Excessive application of water and nitrogen fertilizer during farmland management practices leads to serious groundwater nitrate pollution. A field experiment was conducted during the spring maize growth period in Alxa, Inner Mongolia, China. The soil Water Heat Carbon Nitrogen Simulator (WHCNS) model was calibrated by the ensemble smoother method (ES(DL)) based on the field experiment. Then dynamics of soil water content, soil water percolation flux, soil nitrate concentration and soil nitrate leaching flux were simulated with the calibrated WHCNS model under different water and nitrogen fertilization treatments within the vadose zone. The main conclusions are as follows: ES(DL) is feasible for the calibration of the WHCNS model. RMSE_Maximum a posteriori (RMSE_MAP) of soil water content are 0.0301 cm3 cm−3 and 0.0302 cm3 cm−3 for model calibration and model validation, respectively. RMSE_MAP of soil nitrate concentration are 5.49 mg N kg−1 and 3.86 mg N kg−1 for model calibration and model validation, respectively. Both average soil nitrate concentration and average nitrate leaching flux increase with increasing irrigation amount under the same nitrogen fertilization level for the depths of 1.2 m and 1.8 m. However, average soil nitrate concentration decreased and leaching flux increased with increasing irrigation amount under the same nitrogen fertilization level for the depths of 10 m and 20 m. The study of nitrate dynamics in the deep vadose zone contributes to better understand the procedure of nitrate pollution in groundwater caused by excessive water and nitrogen application.

Keywords

• Deep learning
• Nitrate leaching
• WHCNS model
• Deep vadose zone