County-scale crop yield prediction by integrating crop simulation with machine learning models.

Crop yield prediction is of great importance for decision making, yet it remains an ongoing scientific challenge. Interactions among different genetic, environmental, and management factors and uncertainty in input values are making crop yield prediction complex. Building upon a previous work in which we coupled crop modeling with machine learning (ML) models to predict maize yields for three US Corn Belt states, here, we expand the concept to the entire US Corn Belt (12 states). More specifically, we built five new ML models and their ensemble models, considering the scenarios with and without crop modeling variables. Additional input values in our models are soil, weather, management, and historical yield data. A unique aspect of our work is the spatial analysis to investigate causes for low or high model prediction errors. Our results indicated that the prediction accuracy increases by coupling crop modeling with machine learning. The ensemble model overperformed the individual ML models, having a relative root mean square error (RRMSE) of about 9% for the test years (2018, 2019, and 2020), which is comparable to previous studies. In addition, analysis of the sources of error revealed that counties and crop reporting districts with low cropland ratios have high RRMSE. Furthermore, we found that soil input data and extreme weather events were responsible for high errors in some regions. The proposed models can be deployed for large-scale prediction at the county level and, contingent upon data availability, can be utilized for field level prediction.

Crop traits enabling yield gains under more frequent extreme climatic events.

Climate change (CC) in central China will change seasonal patterns of agricultural production through increasingly frequent extreme climatic events (ECEs). Breeding climate-resilient wheat (Triticum aestivum L.) genotypes may mitigate adverse effects of ECEs on crop productivity. To reveal crop traits conducive to long-term yield improvement in the target population of environments, we created 8,192 virtual genotypes with contrasting but realistic ranges of phenology, productivity and waterlogging tolerance. Using these virtual genotypes, we conducted a genotype (G) by environment (E) by management (M) factorial analysis (G×E×M) using locations distributed across the entire cereal cropping zone in mid-China. The G×E×M invoked locally-specific sowing dates under future climates that were premised on shared socioeconomic pathways SSP5-8.5, with a time horizon centred on 2080. Across the simulated adaptation landscape, productivity was primarily driven by yield components and phenology (average grain yield increase of 6-69% across sites with optimal combinations of these traits). When incident solar radiation was not limiting carbon assimilation, ideotypes with higher grain yields were characterised by earlier flowering, higher radiation-use efficiency and larger maximum kernel size. At sites with limited solar radiation, crops required longer growing periods to realise genetic yield potential, although higher radiation-use efficiency and larger maximum kernel size were again prospective traits enabling higher rates of yield gains. By 2080, extreme waterlogging stress in some regions of mid-China will impact substantially on productivity, with yield penalties of up to 1,010 kg ha-1. Ideotypes with optimal G×M could mitigate yield penalty caused by waterlogging by up to 15% under future climates. These results help distil promising crop trait by best management practice combinations that enable higher yields and robust adaptation to future climates and more frequent extreme climatic events, including flash flooding and soil waterlogging.

A time-dependent parameter estimation framework for crop modeling.

The performance of crop models in simulating various aspects of the cropping system is sensitive to parameter calibration. Parameter estimation is challenging, especially for time-dependent parameters such as cultivar parameters with 2-3 years of lifespan. Manual calibration of the parameters is time-consuming, requires expertise, and is prone to error. This research develops a new automated framework to estimate time-dependent parameters for crop models using a parallel Bayesian optimization algorithm. This approach integrates the power of optimization and machine learning with prior agronomic knowledge. To test the proposed time-dependent parameter estimation method, we simulated historical yield increase (from 1985 to 2018) in 25 environments in the US Corn Belt with APSIM. Then we compared yield simulation results and nine parameter estimates from our proposed parallel Bayesian framework, with Bayesian optimization and manual calibration. Results indicated that parameters calibrated using the proposed framework achieved an 11.6% reduction in the prediction error over Bayesian optimization and a 52.1% reduction over manual calibration. We also trained nine machine learning models for yield prediction and found that none of them was able to outperform the proposed method in terms of root mean square error and R2. The most significant contribution of the new automated framework for time-dependent parameter estimation is its capability to find close-to-optimal parameters for the crop model. The proposed approach also produced explainable insight into cultivar traits' trends over 34 years (1985-2018).

Coupling machine learning and crop modeling improves crop yield prediction in the US Corn Belt.

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.

Modeling Flood-Induced Stress in Soybeans.

Despite the detrimental impact that excess moisture can have on soybean (Glycine max [L.] Merr) yields, most of today's crop models do not capture soybean's dynamic responses to waterlogged conditions. In light of this, we synthesized literature data and used the APSIM software to enhance the modeling capacity to simulate plant growth, development, and N fixation response to flooding. Literature data included greenhouse and field experiments from across the U.S. that investigated the impact of flood timing and duration on soybean. Five datasets were used for model parameterization of new functions and three datasets were used for testing. Improvements in prediction accuracy were quantified by comparing model performance before and after the implementation of new stage-dependent excess water functions for phenology, photosynthesis and N-fixation. The relative root mean square error (RRMSE) for yield predictions improved by 26% and the RRMSE predictions of biomass improved by 40%. Extensive model testing found that the improved model accurately simulates plant responses to flooding including how these responses change with flood timing and duration. When used to project soybean response to future climate scenarios, the model showed that intense rain events had a greater negative effect on yield than a 25% increase in rainfall distributed over 1 or 3 month(s). These developments advance our ability to understand, predict and, thereby, mitigate yield loss as increases in climatic volatility lead to more frequent and intense flooding events in the future.