Journal of Environmental Biology

Artificial neural network techniques for predicting severity

of Spodoptera litura (Fabricius) on groundnut

S. Vennila¹*, G. Singh², G. K. Jha², M. S. Rao³, H. Panwar¹ and M. Hegde⁴

¹Agricultural Knowledge Management Unit, National Research Centre for Integrated Pest Management, New Delhi -110 012, India

²Department of Economics, Indian Agricultural Research Institute, New Delhi-110 012, India

³Division of Crop Sciences, Central Research Institute for Dryland Agriculture, Hyderabad- 500 059, India

⁴Department of Entomology, University of Agricultural Sciences, Dharwad-580 005, India

*Corresponding Author E-mail: svennila96@gmail.com  

Key words

Groundnut,

Neural network,

Sensitivity analysis,

Spodoptera litura,

Weather variables

Publication Data

Paper received: 19.09.2015

Revised received: 29.02.2016

Re-revised received: 29.06.2016

Accepted: 13.07.2016

Abstract

Aim: Approaches to modelling pest populations range from simple empirical models to advanced soft computing techniques that have advantages as well as limitations. A comparative analyses of modelling approaches result in selection of betterpest forecast model with a higher prediction accuracy.

Methodology: Artificial neural network (ANN) techniques viz., multi-layer perceptron neural network (MLP-NN) and polynomial neural networks (PNN) were used along with the multiple and polynomial regressions to predict the moth population of tobacco caterpillar Spodoptera litura (Fabricius) in groundnut cropping system. S. litura pheromone trap catch and weather data of twenty five years (1990-2014) for kharif season (26 to 44 standard meteorological weeks (SMW)) was used for predictive modelling. The weekly male moth catches of S. litura (numbers/trap/week) during maximum severity period (34 SMW) was modelled using weather variables viz., maximum and minimum temperature (^?C), rainfall (mm), morning and evening relative humidity (%) lagged by two weeks. The performance of the models was evaluated using coefficient of determination (R²), root mean square error (RMSE) and mean absolute percentage error (MAPE) estimates.

Results: The study clearly demonstrated the superiority of MLP-NN (R²:0.89) over all other models for predicting the peak severity of S. litura. Sensitivity analysis of MLP-NN model indicated that the maximum temperature lagged by two weeks and evening relative humidity of the previous week was two most important factors influencing the peak population of S. litura. Validation also demonstrated the effectiveness of MLP-NN followed by PNN in dealing with non-linear relation between S. litura population and weather variables.    

Interpretation: All model equations developed in the present study can be used to predict S. litura peak (34 SMW) in conjunction with weather of 32 and 33 SMW during kharif season, and in issuing need based advisories for its effective management on groundnut. 

Copyright ? 2017 Triveni Enterprises. All rights reserved. No part of the Journal can be reproduced in any form without prior permission. Responsibility regarding the authenticity of the data, and the acceptability of the conclusions enforced or derived, rest completely with the author(s).