Digital tools for a biomass prediction from a plant-growth model. Application to a weed control in wheat crop
Résumé
Weed control is essential for crop quality and yield. Usually, the main factors affecting
decision-making of farmers in weeding are based on the presence of certain weed
varieties and their density (plants/m2). However, the weed management at the scale of
the crop cycle refers to other competition indicators, more reliable, such as plant
biomass or leaf area index (LAI). Classically, plant biomass measurements are obtained
by a destructive method of sampling, time consuming and it turns out to be unworkable
in practice for farmers. This project proposes to study an innovative method of biomass
evaluation which is based on the acquisition of plot images to determine the leaf area
index at young stage. An ecophysiologial model (the crop-growth model Azodyn) was
indirectly fed with visible (RGB) images, considered as input parameters, to predict the
temporal evolution of the plant biomass until a date close to the acquisition date. First,
the model is tested on wheat crop where projected leaf area index (PLAI) of both weed
and wheat is determined from image processing. The PLAI is compared to classical
measurements (leaf area index and aerial biomass) and a good correlation between the
aerial biomass estimated with images and classical biomass measurements is obtained.
Second, biomass prediction of the global weed community is estimated from the model
using as input parameter initial aerial biomass derived from PLAI-aerial biomass
correlation.
This preliminary study is a part of a larger project which aims to develop a decision
support tool to determine the deadline weeding intervention in a wheat crop taking into
account the weed harmfulness on wheat crop through a biomass prediction.
Results are encouraging for next studies of this project. Nevertheless, weed detection
and biomass prediction can be increased looking at individually for each weed species.