Advanced machine learning techniques have been used in remote sensing (RS) applications such as crop mapping and yield prediction, but remain under-utilized for tracking crop progress. In this study, we demonstrate the use of agronomic knowledge of crop growth drivers in a Long Short-Term Memory-based, domain-guided neural network (DgNN) for in-season crop progress estimation. The DgNN uses a branched structure and attention to separate independent crop growth drivers and captures their varying importance throughout the growing season. The DgNN is implemented for corn, using RS data in Iowa, U.S., for the period 2003–2019, with United States Department of Agriculture (USDA) crop progress reports used as ground truth. State-wide DgNN performance shows significant improvement over sequential and dense-only NN structures, and a widely-used Hidden Markov Model method. The DgNN had a 4.0% higher Nash-Sutcliffe efficiency over all growth stages and 39% more weeks with highest cosine similarity than the next best NN during test years. The DgNN and Sequential NN were more robust during periods of abnormal crop progress, though estimating the Silking–Grainfill transition was difficult for all methods. Finally, Uniform Manifold Approximation and Projection visualizations of layer activations showed how LSTM-based NNs separate crop growth time-series differently from a dense-only structure. Results from this study exhibit both the viability of NNs in crop growth stage estimation (CGSE) and the benefits of using domain knowledge. The DgNN methodology presented here can be extended to provide near-real time CGSE of other crops.

Accurate in-season crop phenology estimation (CPE) using remote sensing (RS)based machine-learning methods is challenging because of limited ground-truth data.In this study, a biophysical crop model was used to guide neural network (NN)-based, in-season CPE. Using the Decision Support System for Agrotechnology Transfer (DSSAT), we conducted uncalibrated simulations for corn (Zea mays L.) across Iowa and Illinois in the U.S. Midwest with in-season weather and historical information for planting and harvest. We investigated guiding the NN CPE method with growth stage (GSTD) and water stress factor (WSF) outputs from these simulations. Results show that guided NNs are able to estimate onset and progression of phenological stages more accurately than an unguided baseline and a crop model-only method. GSTD guidance improved CPE during seasons when progress deviated from a regional average because of temperature but was detrimental during seasons of delayed planting and harvest. WSF guidance improved CPE during seasons when planting and harvest were delayed by heavy rainfall but performed less well during grainfill and mature stages. Neural network-based CPE guided by both GSTD and WSF provided the most accurate estimates for pre-emergence, emerged, silking, and grainfill stages as well as lower RMSE for the median stage transition date than reported in three full-season CPE studies. An accurate RS method for estimating planting could link DSSAT simulations to the current planting window and improve upon these results. This model-guided approach can be extended to other crops and regions to unlock in-season crop risk assessments that are directly linked to crop phenology.

Accurate, in-season crop progress estimation (CPE) using remote sensing-based (RS) machine learning methods is challenging due to ground-truth data. In this study, a biophysical crop model was used to guide neural network-based (NN) in-season CPE. Using the Decision Support System for Agrotechnology Transfer (DSSAT), we conducted uncalibrated simulations for corn across Iowa and Illinois in the US Midwest with in-season weather and historical information for planting and harvest. We investigated guiding the NN CPE method with growth stage (GSTD) and water stress factor (WSF) outputs from these simulations. Results show that guided NNs are able to estimate onset and progression of growth stages more accurately than an unguided baseline and a crop model-only ridge regression. GSTD guidance improved CPE during seasons when progress deviates from a regional average due to temperature, but were detrimental during seasons of delayed planting and harvest. WSF guidance improved CPE during seasons when planting and harvest were delayed by heavy rainfall, but performed less well during grainfill and mature stages. NN-based CPE guided by both GSTD and WSF provided the most accurate estimates for pre-emergence, emerged, silking, and grainfill stages compared to other methods. For silking, the most vulnerable growth stage to drought. Since the DSSAT simulations in this study are based upon historical planting information, the effectiveness of guidance was decreased by delayed in-season planting. An accurate RS method for estimating time of planting could link simulations to the current planting window and improve upon these results.