A Phenology-Informed Fuzzy Neural Network Model for Accurate Forecasting of Yellow Stem Borer (Scirpophaga incertulas) Population Dynamics in Rice (Oryza sativa L.) in the North Coastal Zone of Andhra Pradesh, India
Annepu Jhansi *
Department of Statistics and Computer Applications, Agricultural College, Bapatla, India.
P. Lavanya Kumari
Department of Statistics and Computer Applications, SMGR Agricultural College, Udayagiri, India.
P. Uday Babu
Agricultural Research Station, Ragolu, India.
G. Ramesh
Department of Statistics and Computer Applications, Agricultural College, Rajamahendravaram, India.
*Author to whom correspondence should be addressed.
Abstract
Rice (Oryza sativa L.) is a major staple crop supporting food security in India, but productivity is constrained by insect pests, including the yellow stem borer (Scirpophaga incertulas). Accurate forecasting of pest population dynamics is important for effective pest management. Previous studies have mainly used weather-based statistical and machine-learning approaches; however, crop phenological stages and previous-week pest incidence may also influence current pest status. These factors were therefore incorporated in the present study.
Weekly light-trap observations of yellow stem borer (YSB) during the Kharif seasons from 2011 to 2023 at the Agricultural Research Station, Ragolu, Andhra Pradesh, were used. Integer-valued Generalised Autoregressive Conditional Heteroscedastic models with exogenous variables (INGARCHX), Artificial Neural Network models with exogenous variables (ANNX), Support Vector Regression models with exogenous variables (SVRX), Extreme Learning Machine models with exogenous variables (ELMX) and Fuzzy Neural Network models with exogenous variables (FNNX) were developed and evaluated.
Among the evaluated models, the FNN model incorporating crop phenological stages produced the lowest training error, with an MSE of 35.038 and an RMSE of 5.919. The results indicate that crop phenology influenced YSB dynamics. The milk-to-maturity stage recorded the highest single pest incidence (80) and the highest mean infestation (19.2), indicating peak pest pressure. The FNN model was effective for forecasting YSB population dynamics and may support timely pest management decisions under the study conditions.
Keywords: Rice, Oryza sativa L., `, yellow stem borer, Scirpophaga incertulas, fuzzy neural network, FNNX, INGARCHX, pest forecasting, crop phenology, machine learning, population dynamics