Smart Agriculture: A Climate-Driven Approach to Modelling and Forecasting Fall Armyworm Populations in Maize Using Machine Learning Algorithms

Kalisetty Vani Sree^1*Upendhar Sudharshanam¹NAGESH KUMAR MALLELA VENKATA¹Rajashekhar Mandla¹Akula Sreenivas¹Mallaiah Bedika¹Bhadru Dharavath¹Sreelatha Dogga¹Ramakrishna Babu A¹Chandra Sekhar Javaji²

• ¹Professor Jayashankar Telangana State Agricultural University, Hyderabad, India
• ²ICAR - Indian Institute of Maize Research, Ludhiana, India

The fall armyworm (Spodoptera frugiperda) poses a significant threat to global maize production owing to its rapid life cycle, extensive host range, and strong dispersal capabilities. We developed a forecasting system for fall armyworm outbreaks over one week using weekly pheromone trap counts (2019–2023) from the Maize Research Centre in Rajendranagar (Hyderabad), combined with weather data such as air temperature, relative humidity, and rainfall. Three modelling approaches, INGARCHX, SVRX and ANNX, were evaluated based on performance metrics: Integer Valued GARCH with Exogenous Variables (INGARCHX), Support Vector Regression with climate inputs (SVRX), and Artificial Neural Network with climate inputs (ANNX). During the training phase, the ANNX model delivered the best performance, recording a mean square error of 0.42 and a root mean square error of 0.65. These results outperformed the SVRX model, which produced a mean square error of 7.29 and a root mean square error of 2.70, and also exceeded the INGARCHX model, showing a mean square error of 2.91 and a root mean square error of 1.70. During testing, the ANNX model consistently outperformed the alternatives, yielding a mean squared error of 25.13 and a root mean squared error of 5.01. SVRX recorded scores of 34.07 and 5.84, while INGARCHX showed 48.90 and 6.99, respectively. Diebold–Mariano tests verified that ANNX's edge over SVRX and INGARCHX is statistically significant at the 5%. By integrating climate variables, this neural network is a dependable early-warning system that predicts fall armyworm population surges with roughly 80% accuracy, one week ahead. This timely and geographically targeted forecasting allows for precise pest-control actions, minimizing maize yield losses and advancing sustainable agricultural strategies.