1. Title, Journal and Authors
Title: Development of early-warning protocol for predicting chlorophyll-a concentration using machine learning models in freshwater and estuarine reservoirs, Korea
Journal: Science of the Total Environment
Authors: YongeunParka, KyungHwaChob, JihwanParka, SungMinChac , JoonHaKima,*
2. Summary
Purpose
-Selection of model for prediction of chlorophyll-a concentration in Juam/Yeongsan reservoirs(JAR, YSR).
-Evaluating model-specific functions based on model performance in terms of statistical evaluation and sensitivity analysis of various input variables.
Methods
An artificial neural network (ANN) and a support vector machine (SVM) were used to compare the predicted and actual values of the Chl-a concentration.
For each model, the correlation coefficient between Chl-a and environmental factors was compared with the previous study results.
Conclusion
Both models reproduced the temporal change in Chl-a concentration within the tolerance range.
Both models use a stochastic error minimization approach, but ANN had limitations in that empirical risk minimization (ERM) was considered only to minimize
training errors. As a result, the two models showed similar performance during the training step, but the SVM showed higher predictive accuracy during the verification step.
The Williams-Kloot test showed that the SVM model performed better than the ANN model in terms of slope and significance.
The structural risk minimization principle (SVM) was more effective than the empirical risk minimization principle (ANN).
In JAR, the factor that had the greatest influence on Chl-a was in the two models, but the second factor that affected Chl-a was'’ in SVM and 'light' in ANN. From previous research results, SVM is considered to be a more reasonable model.
The factors that have the greatest influence on Chl-a in YSR were '' in SVM and'light' in ANN. When the Spearman rank correlation value is applied, nitrogen shows a higher value than light, so SVM is considered to be a more rational model.
Phosphorus appeared to be a more important factor in JAR, a lake where phosphorus was restricted, than in YSR.
In the case of prediction using SVM, the highest accuracy was obtained when the sampling interval was 7 days, and the accuracy was low when the sampling interval was 8 days.
3. Originality and Creativity
4. Application to research
5. Contact
Yeong Gyu Gu / Intern student
Environmental Systems Engineering Lab.
School of Earth Sciences and Environmental EngineeringGwangju Institute of Science and Technology
1 Oryong-dong Buk-gu Gwangju, 500-712, Korea
Phone : +82-10-6589-6653
E-mail : kududrb1@naver.com