Modeling pesticide leaching from golf courses using artificial neural networks

Abstract

The objective of this work was to develop a computer model that accurately predicted pesticide leaching of pesticides applied to turfgrass areas. After much investigation, the number of inputs used to train the Artificial Neural Networks (ANN) was reduced to pesticide solubility, pesticide soikwater partitioning coefficient (Koc), time after application, and the irrigation application practice. For comparison reasons, 1st and 2nd order polynomial regression models were developed. An artificial neural network is a form of artificial intelligence enabling the program to learn relationships instead of the relationships being defined by the programmer. The ANN proved to be a feasible modeling technique for pesticide leaching. The ANN predictions for the test cases had much less error than the 1st or 2nd order regression equations (sum of the squared error between measured and predicted values were 17.4, 528.4, and 522.3, respectively). An interactive World Wide Web (www) site has been developed where this artificial neural network can be accessed (http://www.eece.ksu.edu/∼starret/KTURF/). The www site is called KTURF and is accessible through the Internet. Used as an assessment tool, KTURF can help to reduce pesticide leaching by allowing users to experiment with different pesticide/irrigation schemes. They can thus optimize their practices to reduce the likelihood of pesticide leaching beyond the rootzone.

Notes

Corresponding author (e‐mail address: [email protected]).