A CNN-RNN Framework for Crop Yield Prediction

Abstract

Crop yield prediction is extremely challenging due to its dependence on multiple factors such as crop genotype, environmental factors, management practices, and their interactions. This paper presents a deep learning framework using convolutional neural networks (CNNs) and recurrent neural networks (RNNs) for crop yield prediction based on environmental data and management practices. The proposed CNN-RNN model, along with other popular methods such as random forest (RF), deep fully connected neural networks (DFNN), and LASSO, was used to forecast corn and soybean yield across the entire Corn Belt (including 13 states) in the United States for years 2016, 2017, and 2018 using historical data. The new model achieved a root-mean-square-error (RMSE) 9% and 8% of their respective average yields, substantially outperforming all other methods that were tested. The CNN-RNN has three salient features that make it a potentially useful method for other crop yield prediction studies. (1) The CNN-RNN model was designed to capture the time dependencies of environmental factors and the genetic improvement of seeds over time without having their genotype information. (2) The model demonstrated the capability to generalize the yield prediction to untested environments without significant drop in the prediction accuracy. (3) Coupled with the backpropagation method, the model could reveal the extent to which weather conditions, accuracy of weather predictions, soil conditions, and management practices were able to explain the variation in the crop yields.

Keywords: convolutional neural networks; crop yield prediction; deep learning; feature selection; recurrent neural networks.

Figure 1

The left and right plots show the average corn yield and the average soybean yield across Corn Belt from 1980 to 2018, respectively. The unit of yield is bushels per acre.

Figure 2

The unrolled modeling structure of the proposed CNN-RNN model. Input variables W_t, S_t, ${\overline{Y}}_t$, ${\widehat{\overline{Y}}}_t$, and M_t denote the weather, soil, average yield, predicted average yield, and management data at time step t, respectively, and k denotes the length of time dependencies. At the test phase, ${\overline{Y}}_{t-1}$ was used as an estimator for ${\widehat{\overline{Y}}}_t$ in this paper.

Figure 3

The left and right maps show the absolute prediction errors of the year 2018 for corn and soybean yield predictions, respectively. The counties in black color inside the cornbelt indicate that ground truth yields were not available for these counties for year 2018. The unit of error is bushels per acre.

Figure 4

Bar plot of estimated effects of six weather components on corn measured for 52 weeks of each year, starting from January. The vertical axes were normalized across all weather components to make the effects comparable.

Figure 5

Bar plot of estimated effects of six weather components on soybean measured for 52 weeks of each year, starting from January. The vertical axes were normalized across all weather components to make the effects comparable.

Figure 6

Bar plot of estimated effects of soil conditions measured at different depths of soil and soil conditions measured at the soil surface on corn. The vertical axes were normalized across soil conditions to make the effects comparable. Separate normalizations were done for soil conditions measured at different depths of soil and soil conditions measured at the soil surface. The NCCPIcorn and NCCPIA stand for national commodity crop productivity index for corn and national commodity crop productivity index for all crops, respectively.

Figure 7

Bar plot of estimated effects of soil conditions measured at different depths of soil and soil conditions measured at the soil surface on soybean. The vertical axes were normalized across soil conditions to make the effects comparable. Separate normalizations were done for soil conditions measured at different depths of soil and soil conditions measured at the soil surface. The NCCPIcorn and NCCPIA stand for national commodity crop productivity index for corn and national commodity crop productivity index for all crops, respectively.

Figure 8

Bar plot of estimated effects of planting date on the crop yield. The left and right plots are for corn and soybean, respectively. The vertical axes were normalized across planting dates variables to make the effects comparable.

Figure 9

Plots of predicted state average yield and RMSE based on the predicted weather data for the state of Iowa. The predicted weather data was updated every week with its corresponding ground truth 2018 weather data and prediction results were obtained for each week. The units of yield and RMSE are bushels per acre.