Abstract
A in situ hyperspectral dataset containing multiple growth stages over multiple growing seasons was used to build paddy rice leaf area index (LAI) estimation models with a special focus on the effects of paddy rice growth stage development. The univariate regression method applied to the vegetation index (VI), the traditional multivariate calibration method of partial least squares regression (PLSR), and modern machine learning methods such as support vector regression (SVR), random forests (RF), and artificial neural networks (ANN) based on the original and first-derivative hyperspectral data were evaluated in this study for paddy rice LAI estimation. All the models were built on the whole growing season and on each separate vegetative, reproductive and ripening growth stage of paddy rice separately. To ensure a fair comparison, the models of the whole growing season were also validated on data for each separate growth stage of the standalone validation dataset. Moreover, the optimal band pairs for calculating narrowband difference vegetative index (DVI), normalized difference vegetation index (NDVI) and simple ratio vegetation index (SR) were determined for the whole growing season and for each separate growth stage separately. The results showed that for both the whole growing season and for each single growth stage, the red-edge and near-infrared band pairs are optimal for formulating the narrowband DVI, NDVI and SR. Among the four multivariate calibration methods, SVR and RF yielded more accurate results than the other two methods. The SVR and RF models built on first-derivative spectra provided more accurate results than the corresponding models on the original spectra for both whole growing season models and separate growth stage models. Comparing the prediction accuracy based on the whole growing season revealed that the RF and SVR models showed an advantage over the VI models. However, comparing the prediction accuracy based on each growth stage separately showed that the VI models provided more accurate results for the vegetative growth stages. The SVR and RF models provided more accurate results for the ripening growth stage. However, the whole growing season RF model on first-derivative spectra could provide reasonable accuracy for each single growth stage.
Highlights
Leaf area index (LAI), which is defined as half of the all-sided green leaf area per unit ground area [1], is a key biophysical parameter that reflects biochemical and physiological processes of plants
The optimal band pairs to formulate narrowband difference vegetative index (DVI), normalized difference vegetation index (NDVI) and simple ratio vegetation index (SR) were determined within each single growth stage and for the whole growing season of paddy rice in this study
The univariate regression method on vegetation indices (VIs), traditional multivariate calibration method partial least squares regression (PLSR) and modern machine learning methods such as support vector regression (SVR), random forests (RF), and artificial neural networks (ANN) based on the original and first-derivative hyperspectral data were evaluated in this study for paddy rice LAI estimation with a special focus on the effects of growth stage development
Summary
Leaf area index (LAI), which is defined as half of the all-sided green leaf area per unit ground area [1], is a key biophysical parameter that reflects biochemical and physiological processes of plants. There are two main approaches to build LAI estimation models from remote sensing data; the empirical statistical approach and the radiative transform model (RTM) approach [8] The former approach includes univariate regression models built on a vegetation index (VI) and multivariate-calibration-based models using the full reflectance spectrum [14,15,16]. These multivariate calibration techniques include the partial least squares regression (PLSR) methods and modern machine learning methods such as support vector regression (SVR), random forests (RF), and artificial neural networks (ANN). The accuracy of the RTM inversion results are highly reliant on the realism of the RTM simulation and appropriate RTM parameter initialization [8,17,18]
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
