Precipitation and temperature are the two main characteristics of weather and climate, they also play an important role on the outcome of crop yield. In this study the goal is try to predict crop yield for corn, wheat and soybeans using meteorological data using machine learning predictive models like Lasso, Decision Tree Regressor and Random Forest Regressor. A feature analyses will also be conducted to understand the importance each meteorological parameter have on the outcome of the model. It will be interesting to verify if the feature importance will be the same for each crop or not. Once we can quantify the importance of weather parameters listed below for crop yield we can analyze the impact of climate change on food security.
The dataset used is the result of a long-term agroecological research project in Beltsville, Maryland, the Farming Systems Project (FSP). The data available for analyses span 20 years, 1996-2016.
The meteorological data was measured daily on the same site where crops were cultivated and covers the period of January 1st, 1996 to December 31st, 2016.
Deep Neural Networks (DNN) was used to predict crop yield in the study conducted by Khaky and Wang [1] in response to the 2018 Syngenta Crop Challenge. DNN outperformed Lasso, Shallow Neural Networks (SNN) and Regression Tree. They also found that environmental factors (weather) had greater effect on crop yield than genotype and soil composition. Crane-Droesch [2] also used DNN to predict crop yield and climate change impact assessment in agriculture. In his study he included growing degree days (GDD) as a feature in his model which ended up having a great impact on crop yield prediction.
The best result for corn is for Decision Tree Regressor customized model with for 15 weeks. The best result for soybean is for Random Forest Regressor model for 14 weeks and for wheat the best model was Lasso Regression for 31 weeks.
Overall performance of all the models was rated as good (74% to 83%) but it could be better if more data were available to train the models.
The best crop yield model for each crop was unique.
Relative humidity is more important than precipitation for crop prediction. The temperature play an important role on crop prediction.
Climate change will have an impact on crop yield. The increase in temperature is often accompanied by drought and flooding conditions. For a future study, a sensitivity analyses can be performed to understand the climate anomalies on yield prediction.
REFERENCES
Khaki, Saaed, and Lizhi Wang. “Crop Yield Prediction Using Deep Neural Networks.” Frontiers, Frontiers, 26 Apr. 2019. Retrieved February 25, 2020 from https://www.frontiersin.org/articles/10.3389/fpls.2019.00621/full.
Crane-Droesch, Andrew. “Machine learning methods for crop yield prediction and climate change impact assessment in agriculture.” Environmental Research Letters, Vol 13, Num 11, 26 Oct. 2018. Retrieved March 1, 2020 from https://iopscience.iop.org/article/10.1088/1748-9326/aae159
Explanation of Growing Degree Days, Midwestern Regional Climate Center. Retrieved March 1, 2020 from mrcc.illinois.edu/gismaps/info/gddinfo.htm.
Scikit-learn (2019). Sklearn.preprocessing.RobustScaler. Retrieved May 12, 2019 from
https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.RobustScaler.html
5. StatisticsHowTo (2019). Lasso Regression: Simple Definition. Retrieved May 12, 2019 from https://www.statisticshowto.datasciencecentral.com/lasso-regression/
6. Scikit-learn (2019). Sklearn.ensemble.RandomForestRegressor. Retrieved May 12, 2019 from https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestRegressor.html
7. Tf.keras.optimizers.RMSprop TensorFlow Core v2.2.0. (n.d.). Retrieved May 09, 2020, from https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/RMSprop
8. 3.3. Metrics and scoring: Quantifying the quality of predictions. (n.d.). Retrieved May 08, 2020, from https://scikit-learn.org/stable/modules/model_evaluation.html
9. Coefficient of determination. (2020, April 29). Retrieved May 08, 2020, from https://en.wikipedia.org/wiki/Coefficient_of_determination