A new sugarcane yield model using the SiPAR model. (30th December 2021)

• Record Type: Journal Article
• Authors: Hu, Shun; Shi, Liangsheng; Zha, Yuanyuan; Huang, Kai
• Is Part Of: Agronomy Journal. Volume 114:Number 1(2022)
• Page Start: 490
• Page End: 507
• Publication Date: 2021-12-30
• Subjects: Agronomy -- Periodicals; 630
• DOI: 10.1002/agj2.20949
• Languages: English
• ISSNs: 0002-1962

Abstract

Abstract: Physical process–based crop yield models are subject to extensive input requirements, and traditional statistical models often lack robustness in a changing environment. The purpose of this study was to develop a new and simple semi‐physical sugarcane ( Saccharum officinarum L.) yield model, called the SiPAR model (intercepted photosynthetically active radiation partitioned to stem), with less data requirement than the process‐based models and strong robustness. The SiPAR model was developed using the normalized difference vegetation index, the leaf area index, and solar radiation data. A 3‐yr field experiment was used to evaluate model performance. The SiPAR model was also compared with three traditional statistical models. The results showed that (a) the SiPAR model obtained the highest accuracy among the compared methods and reproduced the spatial pattern of yield well (RMSE = 5.88–8.65 t ha –1 ; R 2 = .44–.87; normalized RMSE (Ry) = 7.22–11.77%) and (b) the SiPAR model had better spatial and temporal stability than the other three statistical models through cross‐validation because the SiPAR model considered the stem biomass accumulation features of sugarcane by introducing a weighting factor to reflect the stem potential growth rate and using intercepted photosynthetically active radiation to represent the energy available for photosynthesis. The SiPAR model has the potential to be applied at a regional scale because it requires less data and fewer parameters and is robust and highly accurate. Core Ideas: A simple semiphysical yield model is proposed for sugarcane. The exponential form of normalized NDVI is defined as an enhanced weighting factor. The proposed model only has two parameters and requires LAI, NDVI, and SR data.