Estimation of Winter Wheat Tiller Number Based on Optimization of Gradient Vegetation Characteristics

Abstract

Tiller are an important biological characteristic of wheat, a primary food crop. Accurate estimation of tiller number can help monitor wheat growth and is important in forecasting wheat yield. However, because of leaf cover and other factors, it is difficult to estimate tiller number and the accuracy of estimates based on vegetation indices is low. In this study, a gradual change feature was introduced to optimize traditional prediction models of wheat tiller number. Accuracy improved in optimized models, and model R2 values for three varieties of winter wheat were 0.7044, 0.7060, and 0.7357. The optimized models improved predictions of tiller number in whole wheat fields. Thus, compared with the traditional linear model, the addition of a gradual change feature greatly improved the accuracy of model predictions of wheat tiller number.

Estimation of wheat tiller density using remote sensing data and machine learning methods

“…The results obtained in this way are often biased (relative error > 20%) ( Scotford and Miller, 2004 ), and the generalizability of the method is poor due to the limited applicability of the chosen vegetation index. For example, the NDVI does not fully reflect the wheat tiller density in situations where the leaf area index is particularly high or low or where the amount of cover is high; the sensitivity of the RVI decreases significantly when the vegetation cover is below 50% and thus does not fully reflect the number of wheat tillers ( Wu et al., 2022 ). To address these problems, in this study, multiple vegetation indices were used to complement each other to meet the accuracy requirements of precision agriculture.…”

“…Changes in the canopy density also cause changes in the vegetation cover, which means that the tiller density can be estimated from the vegetation cover: this is consistent with the observation of a strong correlation between the vegetation cover and plant density in wheat. ( Wang et al., 2020 ; Wu et al., 2022 ).…”

“…This can be severely affected by wind and shading between wheat leaves and cause the tiller number to be underestimated ( Fang et al., 2020 ). Spectral characterization models, in contrast, establish a regression between the tiller density and vegetation indices (VIs) to estimate the tiller density ( Flowers et al., 2001 ; Flowers et al., 2003 ; Scotford and Miller, 2004 ; Phillips et al., 2004 ; Wu et al., 2011 ; Wu et al., 2022 ). Most regression models use linear, a few use non-linear ones such as exponential regression.…”

Contrasted reaction norms of wheat yield in pure vs mixed stands explained by tillering plasticities and shade avoidance

“…From our results, a challenge would consist in phenotyping at high throughput the maximum number of tillers at stage BBCH30, possibly at several densities. In pure stands a UAV-based strategy seems feasible (Wu et al ., 2022), but transposing it in mixture at large scale will require the development of a precision sower. Another strategy is to develop functional-structural plant models with an explicit tillering process (Blanc et al ., 2021), allowing an in-silico exploration of mixture behavior after calibration on data as the ones we produced here.…”

“…From our results, a challenge would consist in phenotyping at high throughput the maximum number of tillers at stage BBCH30, possibly at several densities. In pure stands a UAV-based strategy seems feasible (Wu et al, 2022), but transposing it for a large number of mixtures in multi-environment trials will require the development of a precision sower, i.e. a programmable sower able to sow individual seeds from various cultivars according to a pre-specified layout, like what we have done by hand, if possible at a speed comparable to that of conventional sowers.…”