머신러닝 모델을 이용한 최적관리기법 적용에 따른 경작지 양분 유출 부하량 예측 (Predicting Nutrient Runoff Load in Cultivated Lands using Best Management Practices with Machine Learning Models)

서지정보

· 발행기관 : 강원대학교 농업생명과학연구원
· 수록지 정보 : 농업생명환경연구 / 36권 / 2호 / 152 ~ 163페이지
· 저자명 : 이서로, 배주현, 이관재, 신형진, 양재의, 김종건, 임경재

초록

Increases in livestock manure and chemical fertilizers have led to excessive nutrient runoff into rivers, degrading crop productivity, soil health, and water quality. Effective nutrient runoff management is crucial for sustainable agriculture and ecosystem protection. This study proposes a new methodology for nutrient management by evaluating the predictability of nutrient runoff loads under various best management practices (BMPs) scenarios using a machine learning model. The Agricultural Policy and Environmental Extender (APEX) model was employed to simulate the effectiveness of BMPs in reducing nutrient runoff across farmlands within riparian and drinking water protection zones. Additionally, the predictability of nutrient runoff reduction was assessed using six machine learning models, with the best model selected based on cross-validation performance. Results indicate that the filter strip scenario within the APEX model was the most effective, achieving a nutrient runoff reduction efficiency of 61.6%. Xtreme Gradient Boosting demonstrated superior predictive performance, with a coefficient of determination (R2) of 0.75, Nash–Sutcliffe efficiency (NSE) of 0.65, total nitrogen (T-N) and an R2 of 0.55 and NSE of 0.48 for total phosphorus. These findings suggest that nutrient runoff reduction can be effectively evaluated using topographical and meteorological data alone. This study provides a methodological framework for predicting nutrient runoff loads and assessing the impacts of BMPs in agricultural watersheds.

영어초록

Increases in livestock manure and chemical fertilizers have led to excessive nutrient runoff into rivers, degrading crop productivity, soil health, and water quality. Effective nutrient runoff management is crucial for sustainable agriculture and ecosystem protection. This study proposes a new methodology for nutrient management by evaluating the predictability of nutrient runoff loads under various best management practices (BMPs) scenarios using a machine learning model. The Agricultural Policy and Environmental Extender (APEX) model was employed to simulate the effectiveness of BMPs in reducing nutrient runoff across farmlands within riparian and drinking water protection zones. Additionally, the predictability of nutrient runoff reduction was assessed using six machine learning models, with the best model selected based on cross-validation performance. Results indicate that the filter strip scenario within the APEX model was the most effective, achieving a nutrient runoff reduction efficiency of 61.6%. Xtreme Gradient Boosting demonstrated superior predictive performance, with a coefficient of determination (R2) of 0.75, Nash–Sutcliffe efficiency (NSE) of 0.65, total nitrogen (T-N) and an R2 of 0.55 and NSE of 0.48 for total phosphorus. These findings suggest that nutrient runoff reduction can be effectively evaluated using topographical and meteorological data alone. This study provides a methodological framework for predicting nutrient runoff loads and assessing the impacts of BMPs in agricultural watersheds.

참고자료

· 없음