Corn Yield Prediction With Ensemble CNN-DNN

We investigate the predictive performance of two novel CNN-DNN machine learning ensemble models in predicting county-level corn yields across US Corn Belt (12 states). The developed data set is a combination management, environment, and historical from 1980-2019. Two scenarios for creation are considered: homogenous heterogeneous ensembles. In ensembles, base all same, but they generated with bagging procedure to ensure exhibit certain level diversity. Heterogenous ensembles created different which share same architecture have levels depth. Three types methods were used create several either scenarios: Basic Ensemble Method (BEM), Generalized (GEM), stacked generalized Results indicated that both designed (heterogenous homogenous) outperform five individual ML (linear regression, LASSO, random forest, XGBoost, LightGBM). Furthermore, by introducing improvements over provide most accurate yield predictions states. This model could make 2019 root mean square error 866 kg/ha, equivalent 8.5% relative square, successfully explain about 77% spatio-temporal variation grain yields. significant power this can be leveraged designing reliable tool prediction will, turn, assist agronomic decision-makers.