Prediction of Areal Soybean Lodging Using a Main Stem Elongation Model and a Soil-Adjusted Vegetation Index That Accounts for the Ratio of Vegetation Cover

Abstract

In soybean, lodging is sometimes caused by strong winds and rains, resulting in a decrease in yield and quality. Technical measures against lodging include “pinching”, in which the main stem is pruned when excessive growth is expected. However, there can be a decrease in yield when pinching is undertaken when the risk of lodging is relatively low. Therefore, it is important that pinching is performed after the future risk of lodging has been determined. The lodging angle at the full maturity stage (R8) can be explained using a multiple regression model with main stem elongation from the sixth leaf stage (V6) to the blooming stage (R1) and main stem length at the full seed stage (R6) as the explanatory variables. The objective of this study was to develop an areal lodging prediction method by combining a main stem elongation model with areal main stem length estimation using UAV remote sensing. The main stem elongation model from emergence to R1 was a logistic regression formula with the temperature and daylight hours functions f (Ti, Di) as the explanatory variables. The main stem elongation model from R1 to the peak main stem length was a linear regression formula with the main stem length of R1 as the explanatory variable. The model that synthesized these two regression formulas were used as the main stem elongation model from emergence to R8. The accuracy of the main stem elongation model was tested on the test data, and the average RMSE was 5.3. For the areal main stem length estimation by UAV remote sensing, we proposed a soil-adjusted vegetation index (SAVIvc) that takes vegetation cover into account. SAVIvc was more accurate in estimating the main stem length than the previously reported vegetation index (R2 = 0.78, p < 0.001). The main stem length estimated by the main stem elongation model combined with SAVIvc was substituted into a multiple regression model of lodging angle to test the accuracy of the areal lodging prediction method. The method was able to predict lodging angles with an accuracy of RMSE = 8.8. These results suggest that the risk of lodging can be estimated in an areal manner prior to pinching, even though the actual occurrence is affected by wind.