Abstract

Optical remote sensing (RS) with robust algorithms is needed for accurate assessment of crop canopy features. Despite intensive studies on algorithms, their performance using RS needs to be improved. We evaluated five different algorithms (partial-least-squares regression (PLSR), support vector regression (SVR), random forest regression (RFR), locally-weighted-PLSR (PLSRLW) and PLSR with feature selection (PLSRFS)) for rapid assessment of leaf area index (LAI) and canopy water content (CWC) for rice canopies using canopy reflectance spectra over visible to short-wave infrared region. Two pooled datasets of LAI (600) and CWC (480) were collected from two replicated field experiments during 2014–15 and 2015–16 rice growing season. The performance of each algorithm was evaluated using coefficient of determination (R2). Results showed that PLSRLW performed more accurately than other algorithms with R2 values 0.77 and 0.66 for LAI and CWC, respectively. We also used a bootstrapping approach to generate a kernel density estimator of root mean squared error values for each model. The results suggested that the improvement in prediction accuracy of LAI and CWC can be achieved if a suitable algorithm is selected by assigning higher weights to calibration samples, which has similar canopy structure as the test sample. Subsetting of the canopy spectral data results large error values in test dataset, therefore the use of entire season canopy spectral data should be used for model calibration.

Full Text
Published version (Free)

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 call