Presentation Type
Poster
Student
Yes
Abstract
Accurate crop yield predictions can help farmers make adjustments or changes in their farming practices to optimize their harvest. Remote sensing data is an inexpensive approach to collecting massive amounts of data that could be utilized for predicting crop yield. This study employed linear regression and spatial linear models were used to predict soybean yield with data from Landsat 8 OLI. Each model was built using only spectral bands of the satellite, only vegetation indices, and both spectral bands and vegetation indices. All analysis was based on data collected from two fields in South Dakota from the 2019 and 2021 harvest years. The 2019 yield data was used as training data and the 2021 data was used as test data. The models were compared based on their computing time, mean squared error (MSE), mean squared prediction error (MSPE), residual prediction deviation (RPD), and r-squared values. In terms of MSE, RPD, and r-squared for 2019 data, the best model was the Bayesian spatial model built using vegetation indices. For 2021 data, the Bayesian spatial model using bands outperformed the other models in terms of MSPE. It is worth noting that the Bayesian spatial linear model always outperformed the Bayesian linear regression in all criteria except computation time. Although the spatial model built with bands is not the best model in terms of fit for the training data, the difference is negligible. Furthermore, since prediction is the main purpose, the Bayesian spatial regression model based on bands is recommended for future use.
Start Date
2-6-2024 1:00 PM
End Date
2-6-2024 2:00 PM
Included in
Predicting Crop Yield Using Remote Sensing Data
Volstorff A
Accurate crop yield predictions can help farmers make adjustments or changes in their farming practices to optimize their harvest. Remote sensing data is an inexpensive approach to collecting massive amounts of data that could be utilized for predicting crop yield. This study employed linear regression and spatial linear models were used to predict soybean yield with data from Landsat 8 OLI. Each model was built using only spectral bands of the satellite, only vegetation indices, and both spectral bands and vegetation indices. All analysis was based on data collected from two fields in South Dakota from the 2019 and 2021 harvest years. The 2019 yield data was used as training data and the 2021 data was used as test data. The models were compared based on their computing time, mean squared error (MSE), mean squared prediction error (MSPE), residual prediction deviation (RPD), and r-squared values. In terms of MSE, RPD, and r-squared for 2019 data, the best model was the Bayesian spatial model built using vegetation indices. For 2021 data, the Bayesian spatial model using bands outperformed the other models in terms of MSPE. It is worth noting that the Bayesian spatial linear model always outperformed the Bayesian linear regression in all criteria except computation time. Although the spatial model built with bands is not the best model in terms of fit for the training data, the difference is negligible. Furthermore, since prediction is the main purpose, the Bayesian spatial regression model based on bands is recommended for future use.