Improvement of the DRAINMOD model's performance under time-variable surface storage capacities using neural network models

Abstract

DRINAMOD is a superior model being used globally to assess the impacts of different agricultural drainage layouts and practices on farmlands' water balance. But the model considers a fixed value of surface storage capacity (SSC) for the whole simulation period, which causes the model to predict improper water distribution patterns in farmlands where SSC varies with time due to agricultural practices. The present study presents a novel artificial neural network model (ANNM) that mimics the DRAINMOD model and overcomes such a shortage of considering a fixed SSC value when predicting farmlands' water balance. After training the ANNM by sets of data, which were generated by the DRAINMOD model after being calibrated and validated in rice-wheat rotation farmland at the lower reaches of the Yangtze River basin, China, groundwater tables in the study area were predicted by both models under different scenarios of drainage layouts and management practices. Results show that there were excellent agreements between these predictions, which validates the applicability of the developed ANNM in predicting the water balance in artificial-drained farmlands. The developed ANNM enhances water fate predictions in artificial-drained farmlands characterized by time-variable surface storage capacities, thus resulting in a better assessment of the available water resources alongside high agricultural productivity. Findings from the present study point also to the potential role of neural network models as a tool to enhance the performance of hydrologic models without modifying their source code.

Keywords

• DRAINMOD
• Farmlands’ water balance
• Hydrologic simulations
• Subsurface drainage systems
• Surface storage