MINIMUM VARIANCE ESTIMATION OF YIELD PARAMETERS OF RUBBER TREE WITH KALMAN FILTER

Abstract

Although growth and yield data are available in rubber plantations in Nigeria for aggregate rubber production planning, existing models poorly estimate the yield per rubber tree for the incoming year. Kalman filter, a flexible statistical estimator, is used to combine the inexact prediction of the rubber production with an equally inexact rubber yield, tree girth, tapping height, stimulation and tapping system measurements to obtain an optimal estimate of one year ahead rubber production. Six rubber clones-GT1, PB260, PB217, PB28/59, PB324 and RRIM703 were studied using 12-year data, generated from permanent experimental plots. Stochastic autoregressive model was fitted to the data to identify optimal management strategy that accounts for risk due to seasonality. STAMP, an OxMetric modular software system for time series analysis, was used to estimate the yield parameters. Our results show that significant test of actual yield to model forecast is less than 1.96.  Hence, the null hypothesis that the actual yield is within the forecasted value is accepted at 5% significant level. Based on the impulse response function of the lead equations, the long-run elasticity of yield was estimated to be highest for PB324 (2211gm/tree) and lowest for RRIM703 (1053gm/tree). PB260 is the best short term clone with the highest dynamic multiplier of 0.59. More important, the estimator minimized the variance of estimation errors from 55% of plantation prevision to 10%. It is our opinion that Kalman filter is a robust estimator of the biotechnical dynamics of rubber exploitation system