A Functional-structural Model Of Cotton And Prediction Of Cotton Yield In Xinjiang

Functional-structural plant model and machine learning that have become matured to indispensable tools to advance the plant science are playing a vital role not only in understanding the plant perception of,recognition of and responses to environmental stimuli but also in predicting crop yield driven by key factors.As shaping ideal phenotype and improving harvest index are two main approaches to maximize cotton yield,it is necessary to investigate the mechanism of transport and allocation of carbohydrates in cotton plants via functional-structural plant(FSP)modelling.The rapid advancement of information and communications technology is meanwhile creating data explosion in the agricultural sector.The increasing frequency of extreme weather events has already threatened cotton production in Xinjiang.Machine learning provides a powerful tool to predict cotton yield and investigate the major meteorological factors that impact cotton yield.However the plausibility of using machine learning to predict cotton yield has been rarely reported.The study developed a FSP model for cotton and selected the optimal machine learning algorithm to predict cotton yield by integrating literature data,meteorological observations and field measurements.The major conclusions are listed as follows.1.Insertion angle of branches were decreased to avoid shade in cotton.Insertion angle,branch length and leaf area per branch progressively decreased upwardly toward the main stem apex at low densities while they performed inversely at high densities to compete for more light.Vegetative branches at lower main stem ranks were found to abort due to the deficiency of carbohydrates.Mepiquat chloride resulted in a compact plant structure for cotton via shortening intemode and reducing leaf area.The effect of mepiquat chloride on plant structure of cotton was significant in wet year but was not dry year,indicating the interaction between mepiquat chloride and climate.2.As the above-ground dry mass increased while the reproduction allocation was stable at lower densities and decreased at higher densities,seed yield increased gradually to a maximum at 7.5 plants m-2 and then decreased subsequently with increasing density.Mepiquat chloride significantly reduced above-ground dry mass by inhibiting organ expansion,but it can improve the seed yield by increasing the reproductive allocation.The effect of mepiquat chloride on above-ground dry mass and reproductive allocation were not significantly in the experiment under dry condition.3.The heterogeneity of carbohydrate supply for phytomers in a cohort was supported indirectly by the varying possibility of setting matured boll.The local-pool approach was presented on the basis of distribution of carbohydrates from C-labelling results in cotton.CottonXL,a functional-structural plant model for cotton based on a local-pool approach,reproduced light capture,leaf photosynthesis,carbohydrate transport between phytomers,leaf expansion,dry mass allocation,light distribution within canopy and yield formation etc.4.Machine learning was reliable in predicting climate-driven cotton yield in Xinjiang with nRMSE remaining below 20%.The top three algorithms in terms of accuracy were support vector regression with a kernel function of radial,neural network and multiple linear regression.The major meteorological factors impacting cotton yield in Xinjiang has been identified using neural network.From a meteorological perspective,major factors were thermal time,precipitation and sunshine duration in a descending order of importance.From a time perspective,July and August during which flowers and bolls were produced successively were the most important period while June was less important.Extreme heat and frost have been proven to be typical meteorological disasters for cotton production in Xinjiang,suggesting consistency with results from machine learning.