Comparing methods for estimating leaf area index by multi-angular remote sensing in winter wheat

Li He,Wei Feng,Tiancai Guo,Xingxu Ren,Yangyang Wang,Haiyan Zhang,Beicheng Liu,Wandai Liu

doi:10.1038/s41598-020-70951-w

Abstract

The reflectance of wheat’s canopy exhibits angular sensitivity, which can influence the accuracy of different methods for its leaf area index (LAI) estimation through multi-angular remote sensing. The primary objective of this study was to assess and compare the ability of various methods for LAI estimation from 13 view zenith angles (VZAs). The four methods included: (1) common hyper-spectral vegetation indices (VIs), (2) optimal two-band combination VIs (i.e., VIs: normalized difference index, simple ratio index, and difference vegetation index), (3) back-propagation neural network (BPNN), and (4) partial least squares regression (PLSR). Our results demonstrated that the red-edge plays a key role in estimating LAI, in that the traditional VIs, optimal two-band VIs, and PLSR including the red-edge band all showed satisfactory performance, with coefficient of determination (R2) > 0.72 in the nadir direction. However, the estimation accuracy of LAI was not positively related with band number, and BPNN gave unsatisfactory results under a larger viewing angle, with R2 ≤ 0.60 for extreme angles. The predictive ability of all four methods declined with an increasing VZA, with reliable LAI estimation near the nadir direction. Importantly, by comparing the four methods, PLSR emerged as superior in both its estimation accuracy and angular insensitivity, with R2 = 0.83 in the nadir direction and ≥ 0.65 for extreme angles. For this reason, we highly recommend it be used with multi-angular remote sensing data, especially in agricultural applications.

Highlights

The reflectance of wheat’s canopy exhibits angular sensitivity, which can influence the accuracy of different methods for its leaf area index (LAI) estimation through multi-angular remote sensing
The TCARI/OSAVI produced the largest difference between extreme angles, with 8.28% at − 60° versus 73.52% at + 60°, whereas EVI-1 had the smallest difference: 33.21% in − 60° viewing angle and 36.09% in + 60° viewing angle
(810, 680), DDn, and DD performed best among the 2, 3, and 4-band indices, respectively, with corresponding R2-values of 0.73–0.75 and root mean square error (RMSE)-values of 0.97–1.01 in the nadir direction. These three models provided no advantage under extreme viewing angles (R2-values ≤ 0.58, RMSE-values ≥ 1.57). These results indicated that the ability of vegetation indices (VIs) to monitor LAI is affected by viewing angle

Summary

Introduction

The reflectance of wheat’s canopy exhibits angular sensitivity, which can influence the accuracy of different methods for its leaf area index (LAI) estimation through multi-angular remote sensing. By comparing the four methods, PLSR emerged as superior in both its estimation accuracy and angular insensitivity, with R2 = 0.83 in the nadir direction and ≥ 0.65 for extreme angles. For this reason, we highly recommend it be used with multi-angular remote sensing data, especially in agricultural applications. VIs with two or three bands are common remote sensing parameters, but they may be limited for exploiting the abundant information conveyed in narrow spectral bands of hyperspectral remote sensing data

Objectives

Results

Discussion

Conclusion