Improving a weed population model using a sequential Monte Carlo method

Abstract

Makowski D, Chauvel B & Munier‐Jolain N (2010). Improving a weed population model using a sequential Monte Carlo method. Weed Research50, 373–382.SummaryField measurements and weed population model predictions have been proposed as the basis for recommendations on the need for chemical or mechanical treatment, but both approaches have some limitations. This study shows how a Sequential Monte Carlo (SMC) method can be used to combine weed count measurements and model predictions, to derive a better estimate of weed population characteristics. SMC was applied to a dynamic model simulating weed densities, seed production and seedbank densities for Alopecurus myosuroides (blackgrass). The benefit resulting from SMC was quantified for several types of weed count data, using experiments carried out in seven plots during 6 years. Compared with the initial model predictions, SMC reduced the root mean squared error (RMSE) by 33.5–81.5%. Compared with the weed densities derived from the weed counts alone, SMC reduced the RMSE by 1.2–10%. SMC should be preferred to the single use of model or of weed count data, because it can improve weed density predictions and because the probability distributions computed by SMC can be used to analyse the uncertainty about the state of the system.