Using the DSSAT model to simulate wheat yield and soil organic carbon under a wheat-maize cropping system in the North China Plain

Abstract

Crop modelling can facilitate researchers' ability to understand and interpret experimental results, and to diagnose yield gaps. In this paper, the Decision Support Systems for Agrotechnology Transfer 4.6 (DSSAT) model together with the CENTURT soil model were employed to investigate the effect of low nitrogen (N) input on wheat (Triticum aestivum L.) yield, grain N concentration and soil organic carbon (SOC) in a long-term experiment (19 years) under a wheat-maize (Zea mays L.) rotation at Changping, Beijing China. There were two treatments including N0 (zero N application) and N150 (150 kg ha -1 yr -1 N) applied annually as urea before wheat and maize planting, with phosphorus (P) and potassium (K) basal fertilizers applied as 75 kg P 2 O 5 ha -1 and 37.5 kg K 2 O ha -1 , respectively. The DSSAT-CENTURY models were able to satisfactorily simulate measured wheat grain yield and grain N concentration at N0, but could not simulate these parameters at N150, or SOC in either N treatment. Model simulation and field measurement showed that N application (N150) increased wheat yield compared to no N application (N0). The results indicated that inorganic fertilizer application at the rates used did not maintain crop yield and SOC levels. It is suggested that if the DSSAT is calibrated carefully, it can be a useful tool for assessing and predicting wheat yield, grain N concentration, and SOC trends under wheat-maize cropping systems.

Nitrogen in potato rotations with cover crops: Field trial and simulations using DSSAT

“…Even though simulations of some treatments were less successful (e.g., Li et al., 2015; Liu et al., 2017), the results from these studies do not suggest that the DSSAT‐CENTURY model systematically underpredicts N mineralization from SOM. As we did not make any changes to the parameters that govern the SOM turnover in the model, it seems unlikely that the model strongly underpredicted N mineralization in our trial.…”

“…Second, a comparison of N mineralization in undisturbed soil cores incubated revealed that the proportion of total N mineralized in soils sampled from the Tulelake basin was on average lower than in alluvial soils from the Central Valley in California (Miller et al., 2019). Finally, SOM and N turnover have been simulated with the DSSAT‐CENTURY module for several short‐ and long‐term trials with different fertilization treatments (Li, Yang, Drury, & Hoogenboom, 2015; Liu et al., 2017; Tojo Soler et al., 2011; Yang, Yang, Dou, Yang, & Hoogenboom, 2013). In general, the agreement between modeled and measured values has been found to be good or satisfactory.…”

Multi-Model Ensemble Depth Adaptive Deep Neural Network for Crop Yield Prediction

“…To measure wheat production and soil organic carbon under a wheat-maize cropping method in northern China, a Decision Support System for Agro Transfer 4.6 (DSSAT) model [10] was implemented. In this model, various factors related to the yield were investigated by using DSSAT-CENTURT soil model.…”

“…The update rule is given by using adaptive loss function and applying OGD. It is given as follows: (10) In (10), is calculated through backpropagation from error derivatives of . It is noted that the summation starts at since the shallower classifiers do not depend on for making crop yield prediction.…”

“…It adaptively selects the optimal network depth automatically online. It is a shallower model which tends to converge faster than deeper model utilizing a hedging strategy would inferior weights of deeper classifiers to a very small value, which would affect the update in (10). It results in deep classifiers having slow learning.…”