기상 및 토양환경 정보를 이용한 노지 채소의 토양수분 변화 및 예측 모델 개발

Abstract

Soil moisture is an essential element for plant growth and is biochemically affected by weather and soil environments. Farmers have difficulties in managing soil moisture, and existing soil moisture prediction models require a lot of environmental information. Therefore, we intend to develop a model that predicts soil moisture per hour using only limited environmental information. First, the relationship between soil moisture and the environment is identified through correlation analysis. Next, a statistical soil moisture prediction model was developed through a nonlinear regression model (NLS) using a sigmoid function. Finally, in order to confirm whether the proposed model is suitable as a soil moisture prediction model, it was compared with the machine learning algorithms of support vector regression (SVR), random forest (RF), and extreme gradient boosting (XGB), which are widely used in previous studies. Through root mean square error (RMSE) and mean absolute error (MAE), it was found that the NLS model was superior in field cultivation and underestimated the maximum soil moisture in paddy cultivation compared to other models. Since the proposed model is easy to apply in various soil environments, it is expected to help as a useful soil moisture management system for farmers.