This study investigates whether coupling crop modeling and machine learning (ML) improves corn yield predictions in the US Corn Belt. The main objectives are to explore whether a hybrid approach (crop modeling + ML) would result in better predictions, investigate which combinations of hybrid models provide the most accurate predictions, and determine the features from the crop modeling that are most effective to be integrated with ML for corn yield prediction. Five ML models (linear regression, LASSO, LightGBM, random forest, and XGBoost) and six ensemble models have been designed to address the research question. The results suggest that adding simulation crop model variables (APSIM) as input features to ML models can decrease yield prediction root mean squared error (RMSE) from 7 to 20%. Furthermore, we investigated partial inclusion of APSIM features in the ML prediction models and we found soil moisture related APSIM variables are most influential on the ML predictions followed by crop-related and phenology-related variables. Finally, based on feature importance measure, it has been observed that simulated APSIM average drought stress and average water table depth during the growing season are the most important APSIM inputs to ML. This result indicates that weather information alone is not sufficient and ML models need more hydrological inputs to make improved yield predictions.

We used the Agricultural Production Systems sIMulator (APSIM) to predict and explain maize and soybean yields, phenology, and soil water and nitrogen (N) dynamics during the growing season in Iowa, USA. Historical, current and forecasted weather data were used to drive simulations, which were released in public four weeks after planting. In this paper, we (1) describe the methodology used to perform forecasts;(2) evaluate model prediction accuracy against data collected from 10 locations over four years; and (3) identify inputs that are key in forecasting yields and soil N dynamics. We found that the predicted median yield at planting was a very good indicator of end-of-season yields (relative root mean square error [RRMSE] of ∼20%). For reference, the prediction at maturity, when all the weather was known, had a RRMSE of 14%. The good prediction at planting time was explained by the existence of shallow water tables, which decreased model sensitivity to unknown summer precipitation by 50-64%. Model initial conditions and management information accounted for Abbreviations: APSIM, Agricultural Production Systems sIMulator; RRMSE, relative root mean square error.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. one-fourth of the variation in maize yield. End of season model evaluations indicated that the model simulated well crop phenology (R 2 = 0.88), root depth (R 2 = 0.83), biomass production (R 2 = 0.93), grain yield (R 2 = 0.90), plant N uptake (R 2 = 0.87), soil moisture (R 2 = 0.42), soil temperature (R 2 = 0.93), soil nitrate (R 2 = 0.77), and water table depth (R 2 = 0.41). We concluded that model set-up by the user (e.g. inclusion of water table), initial conditions, and early season measurements are very important for accurate predictions of soil water, N and crop yields in this environment. Neil Huth from CSIRO for their support with the APSIM model, Iowa State University students () for assistance with data collection and managing the field experiments. We also thank the APSIM Initiative for making the software publicly available and for ensuring software quality. ORCIDSotirios V. Archontoulis https://orcid.org/0000-0001-7595-8107 Mark A. Licht https://orcid.org/0000-0001-6640-7856 Kendall R. Lamkey

Shallow water table (WT) influences crop growth and production in many major agricultural regions across the globe. We enhanced the APSIM-soybean model to accurately simulate root depth in fields with shallow water tables. We used data from a controlled experiment (Rhizotron facilities) that included root depth observations for nine WT treatments to develop and calibrate the new model. Analysis indicated that unconstrained root growth occurs until volumetric soil moisture approaches 0.03 mm/mm below saturation. Below that threshold, root growth linearly decreases to zero at saturation. Inclusion of this factor into the model increased accuracy of root depth simulations from R2 of 0.65 to 0.97 and reduced root mean square error from 45 to 9 cm. Validation of root depth simulations using independent field data from Iowa, USA (years 2016, 2017, 2018) confirmed the model. We also found that the inhibition of root growth in response to shallow WT substantially impacted the vertical distribution of the roots in both measurements and simulations. Overall, this work enhances the capability of APSIM in simulating production and environmental aspects of cropping systems, especially in regions with shallow water tables typical of the Corn Belt, USA.

No abstract

We developed a biochar model within the Agricultural Production Systems sIMulator (APSIM) software that integrates biochar knowledge and enables simulation of biochar effects within cropping systems. The model has algorithms that mechanistically connect biochar to soil organic carbon (SOC), soil water, bulk density (BD), pH, cation exchange capacity, and organic and mineral nitrogen. Soil moisture (SW)-temperature-nitrogen limitations on the rate of biochar decomposition were included as well as biochar-induced priming effect on SOC mineralization. The model has 10 parameters that capture the diversity of biochar types, 15 parameters that address biochar-soil interactions and 4 constants. The range of values and their sensitivity is reported. The biochar model was connected to APSIM's maize and wheat crop models to investigate long-term (30 years) biochar effects on US maize and Australia wheat in various soils. Results from this sensitivity analysis showed that the effect of biochar was the largest in a sandy soil (Australian wheat) and the smallest in clay loam soil (US maize). On average across cropping systems and soils the order of sensitivity and the magnitude of the response of biochar to various soil-plant processes was (from high to low): SOC (11% to 86%) > N 2 O emissions (À10% to 43% 43%) > plant available water content (0.6% to 12.9%) > BD (À6.5% to À1.7%) > pH (À0.8% to 6.3%) > net N mineralization (À19% to 10%) > CO 2 emissions (À2.0% to 4.3%) > water filled pore space (À3.7% to 3.4%) > grain yield (À3.3% to 1.8%) > biomass (À1.6% to 1.4%). Our analysis showed that biochar has a larger impact on environmental outcomes rather than agricultural production. The mechanistic model has the potential to optimize biochar application strategies to enhance environmental and agronomic outcomes but more work is needed to fill knowledge gaps identified in this work.