Weather-based forecasting model for the presence of bagworm (metisa plana walker) in oil palm plantation using regression analysis and artificial neural network

Metisa plana is one of the top leaf-eating insect pests in oil palm plantation. A moderate level of infestation could significantly reduce yield for over two years and causes a severe economic loss to the industry. The devastating losses that Metisa plana bring about is one of the reasons the execution of control method should be properly planned. Most of the conventional control method does not associate Metisa plana with weather parameters. Therefore, it is important to elucidate the relationship between these two prior to development of an early warning system so that the pest can be controlled efficiently. Hence, the objectives of this study were i) to examine the population density of bagworm under field condition and weather parameters, ii) to investigate the most dominant weather parameters at different time-lags that influence changes in bagworm population density, and iii) to develop a prediction model for bagworm population density by using regression models and artificial neural network (ANN). This study was conducted in Estate Sungai Mengah owned by Tabung Haji Plantation located in Muadzam Shah, Pahang from July 2016 to June 2017. Two fields were selected: Block 16 and Block 21, and these fields have severe and mild bagworm infestations, respectively. Bagworm censuses were done by identifying Metisa plana’s larval stage 1 (L1) to 7 (L7) from 13 random palms by cutting off frond number 17. The larval stages were then recorded and summed up biweekly. A Davis Vantage Pro 2 weather station was installed in each block to acquire weather data i.e., temperature, rainfall, relative humidity, solar radiation, wind speed, wind direction and heat index. The weather data were then averaged or summed up biweekly to produce mean temperature (MT), total rainfall (RF), mean relative humidity (RH), mean solar radiation (SR), mean wind speed (WS), and mean heat index (HI). The timelags used in the analysis consisted of lag two weeks (T2), four weeks (T4), six weeks (T6), eight weeks (T8), ten weeks (T10) and twelve weeks (T12). The relationship between bagworm and weather parameters were analysed using Shapiro-Wilk’s test, Spearman’s Rank correlation, multiple linear regression (MLR) and ANN. For the ANN, two models were developed particularly i.e., ANN based on correlation analysis and feature selection. The results showed that bagworm population in Block 16 was higher because the field was significantly hotter, less humid and received more solar radiation than Block 21. Bagworms were negatively correlated with mean temperature, mean heat index, and mean wind speed while positively correlated with total rainfall and mean relative humidity. Most of the interactions between bagworm and weather parameters occurred frequently at time-lag 2 weeks in Block 16 and time-lag 12 weeks in Block 21. The results showed that highest R2 values were obtained through ANN-Correlation ranging from 0.329 to 0.989, followed by ANNFeature selection ranging from 0.266 to 0.995, and multiple linear regression ranging from 0.000 to 0.798. The best models were obtained through ANNCorrelation method i.e., for L1 larval stages utilizing mean temperature, mean relative humidity, mean wind speed, and mean heat index at time-lag 2 and 4, mean temperature, mean relative humidity, and mean heat index at time-lag 6, and mean solar radiation at time-lag 12 with 99.58% accuracy. This was followed by the L2 larval stage model utilizing mean temperature, total rainfall, mean relative humidity, mean wind speed, and mean heat index at time-lag 2, mean temperature, mean wind speed, and mean heat index at time-lag 4, mean temperature, total rainfall, mean wind speed and mean heat index at time-lag 6, and mean solar radiation at both time-lag 8 and 12 with 99.91% accuracy. A query performed using both models suggested that the favourable weather condition for Metisa plana under field condition was 20 to 24°C mean temperature, 15 to 20 mean heat index and 138 to 210 Wm-2 mean solar radiation. Prediction of Metisa plana’s L1 and L2 larval stages could be achieved with high accuracy using ANN by incorporating weather parameters and time-lag analysis.

Text FP 2022 1 IR.pdf Download (1MB)

Actions (login required)

View Item