Comparison of two methods for estimating parameters of harvest index increase during seed growth

Abstract

The linearity of harvest index (HI) increase has provided a simple means to analyze and predict seed growth and yield in experimental and simulation studies. When using the linear HI concept, it is necessary to estimate three parameters: (1) the period from anthesis to the onset of linear increase in HI (LAG), (2) the time of cessation of linear increase in HI (HIM), and (3) the rate of linear increase in HI with time (dHI/d t). Thus, an appropriate method is needed for accurate estimation of the parameters. In this paper, we compare the capability of two methods for this purpose. In the first, conventional method (M1), a simple, linear regression model ( y= a+ bx) is used to describe HI increase versus time using data contained within the main seed growth period. dHI/d t is taken as b and LAG as − a/ b. In this method, physiological maturity is considered as HIM. The second method (M2) is based on applying a non-linear, segmented regression model. The segmented model consists of two intersecting lines, a sloping line ( y= a+ bx) for the linear increase in HI ( b=dHI/d t) and LAG (− a/ b) and a horizontal line ( y= a+ bx 0) which determines HIM ( x 0) and consequently maximum HI (HI max). Data sets of HI versus time for 13 spring wheat ( Triticum aestivum L.) cultivars grown under rainfed conditions were used. The segmented model described HI increase with time well and its performance was comparable to quadratic, cubic and logistic models that have been used for this purpose. R 2 values for the segmented model were between 0.93 and 0.98. M1 and M2 provided considerably different estimates of LAG, HIM and dHI/d t. Comparison of measured maximum HI with those calculated using parameter estimates of the conventional method (M1) showed that this method is not accurate (RMSE=5.4%). The deficiency in M1 in estimating HI-related parameters may be a source of error in grain yield prediction based on the linear HI increase concept. M2, however, was significantly better (RMSE=0.3%) and thus is proposed as the preferred method for future use.