Estimating canopy parameters of winter wheat at different stages using hyperspectral data combined with soil variables

ABSTRACT

Remote sensing technology has many advantages in real-time estimating of crop growth. Crop growing estimation primarily establishes the direct statistical model between spectral data and crop biophysical variables. However, the model accuracy rarely improves with the increase in the number of independent variables owing to the multicollinearity; hence, the addition of a data set different from spectral variables may solve this problem. Considering the strong correlation between crop growth and soil environment, this study aims to investigate whether and how the addition of important soil variables can improve estimation model accuracy. This study collected LAI and Chlorophyll substitution index (SPAD value) of wheat canopy and canopy spectral data under different soil environments to quantify the correspondent relationship. Important spectral parameters (IPs) and important soil variables (ISVs) were selected by least absolute shrinkage and selection operator (LASSO) to establish linear and nonlinear models for wheat growth estimation and the effect of multiple soil variables in enhancing wheat growth estimation was tested. The results indicated LASSO can effectively reduce feature dimensionality for wheat growth estimation with maintaining model accuracy; the extra ISVs can improve the model accuracy due to the high collinearity of spectral parameters. The optimal models of wheat LAI estimation (R² = 0.83, RMSE = 0.500) and SPAD estimation (R² = 0.75, RMSE = 1.835) were constructed based on orthogonal partial least squares analysis (OPLS) by IPs and IPs+ISVs, respectively. Finally, we discussed the applicability of spectral parameters and soil variables. This research combines remote sensing features of crops with crucial growth variables to obtain an efficient and mechanical crop growth estimation.

KEYWORDS:

• Hyperspectral remote sensing
• wheat growth
• leaf area index
• SPAD
• soil environment
• modelling

3. Highlights

• To select the important variables from the independent variables with multicollinearity to simplify the model through LASSO

• To estimate wheat growth based on hyperspectral data combined with soil variables

• To evaluate the model’s accuracy, stability and simplicity, and determine the optimal estimation of the wheat growth.

Acknowledgements

We thank the staff of ecological park for their efforts on experimental site. We are grateful to the academic editors and anonymous reviewers for their valuable opinions and comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was supported by Jiangsu Province University Innovation Team Project [NO. (2019)1468]. This research was also funded by the Postgraduate Research & Practice Innovation Program of Jiangsu Province [NO. KYCX22_2589], and the Graduate Innovation Program of China University of Mining and Technology [NO.2022WLKXJ106].