A machine learning approach to understand nitrate leaching in Iowa watersheds
Date
2022-05
Authors
Patel, Ishan Nalinkant
Major Professor
Advisor
Hu, Guiping
Castellano, Michael
MacKenzie, Cameron
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Altmetrics
Abstract
As one of the corn belt states of the US, Iowa has corn and soybean as the main row crops, which are the main source of nitrate leaching. Many agencies such as USEPA focus on the hypoxic zone in the Gulf of Mexico caused by nitrate leaching in the croplands of MARB. Researchers have utilized quantitative methods such as regression, simulation, and qualitative methods to calculate nitrate load. Since machine learning aims to understand the structure of the data and fit that data to models to predict future outcomes, it can be a great way to tackle this problem because it can predict future outcomes and provide additional insights from the data. The time-series dataset used in this study focused on predicting Nitrate yield (kg NO3-N ha-1 cropland) for 29 watersheds of Iowa, for which the data was collected from 2001 to 2018. The objective of this study was to find relationships between the nitrate yield with the independent variables from the dataset, which can explain the trend and help understand future nitrate leaching in the state of Iowa. The same model can identify potential causes and relationships for different datasets from different states. Walk Forward Cross-Validation approach was used for this study, which focuses on solving time-series analysis problems. The RRMSE value of the trained model for the test year 2018 was 23.68%, with an R2 score of 77.06%. The model suggested that the most important features were annual discharge, rain, corn to soybean ratio, and other variables. The Partial Dependency Plots (PDP) explain their relationship with the target variable, nitrate yield. The relationship from PDP shows an underlying aspect of what value ranges contribute to the sudden changes in nitrate yield and how this finding can help policymakers and environmental agencies understand the problem further.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
thesis