Hyperspectral reflectance-based partial least squares regression models for predicting cotton leaf physiological traits
Data files
Sep 23, 2025 version files 215.93 MB
-
PLSR_Coefficients_cotton.xlsx
215.93 MB
-
README.md
3.39 KB
Abstract
Alterations in the mechanistic drivers of photosynthesis have the potential to improve crop productivity, but their measurement is inherently time-consuming using traditional methods. High-throughput approaches to estimate photosynthesis using hyperspectral reflectance could be developed by leveraging variation in cotton (Gossypium hirsutum L.) leaf traits generated through nitrogen management, synthetic growth regulation strategies, and leaf position within the canopy. Currently, no such models exist for cotton, and interactions among the aforementioned factors are relatively unexplored for cotton leaf traits. This study aimed to (1) evaluate the effects of N application rate, mepiquat chloride (MC) management, and leaf position within the canopy on photosynthesis and its components, and (2) develop and validate hyperspectral reflectance-based partial least squares regression (PLSR) models for predicting cotton leaf physiological traits. N rate and leaf position interacted to affect net photosynthetic rate (AN), electron transport rate, chlorophyll a, and chlorophyll b, while mepiquat chloride influenced only leaf pigments and specific leaf weight (SLW). AN reductions under N deficiency were driven by declines in the maximum rate of Rubisco carboxylation (Vc,max) and ribulose-1,5-bisphosphate regeneration (Jmax), whereas high N had no effect on AN. PLSR models exhibited good to high predictive accuracy, with R² values ranging from 0.62 to 0.87 for most traits, except for SLW. These findings enhance our understanding of the physiological responses to N rate, MC strategy, and leaf position and highlight the potential of hyperspectral reflectance-based PLSR as a high-throughput tool for predicting leaf physiological traits to improve photosynthetic efficiency in cotton.
PLSR Model Coefficients for Cotton Leaf Traits
This folder contains Partial Least Squares Regression (PLSR) model coefficients and jackknife coefficient files for estimating cotton leaf traits from reflectance spectra (500–2400 nm).
File Naming Convention
- [Trait]PLSR_Coefficients[#comp].csv
Standard coefficients for estimating traits using the specified number of PLSR components. - [Trait]_Jackkife_PLSR_Coefficients.csv
Jackknife coefficients used to estimate the uncertainty (standard deviation and confidence intervals) of trait predictions.
Example:
A_PLSR_Coefficients_15comp.csvA_Jackkife_PLSR_Coefficients.csv
Column Descriptions
To assist users unfamiliar with spectral modeling, below is a description of the variables (column headers) in the CSV files:
1. Standard Coefficients Files ([Trait]PLSR_Coefficients[#comp])
- Unnamed: 0
Predictor variable label. This includes the intercept term(Intercept)and spectral band identifiers (e.g.,Wave_500,Wave_501, ...). - [Trait].[#comp].comps
PLSR coefficient value for the specified trait using the given number of PLSR components.- Units: Dimensionless regression weights.
- Interpretation: These coefficients, when multiplied by the predictor values (reflectance), contribute to estimating the trait. Units of the final prediction match the trait being modeled (e.g., chlorophyll content, nitrogen concentration, water).
2. Jackknife Coefficients Files ([Trait]_Jackkife_PLSR_Coefficients)
- Iteration
Index of the jackknife resampling iteration. Each row represents one jackknife sample used to estimate variability in coefficients. - Intercept
Intercept coefficient for that jackknife iteration. - Wave_[λ] (e.g., Wave_500, Wave_501, … Wave_2400)
Coefficients corresponding to reflectance at each wavelength (in nanometers, nm).- Wavelengths cover the 500–2400 nm spectral range.
- Units: Dimensionless regression weights.
- Usage: Apply within the PLSR model to generate predictions; variation across iterations provides uncertainty estimates (e.g., standard deviation, confidence intervals).
Instructions for Users
- Download the dataset
- Obtain the full file:
PLSR_Coefficients_cotton.xlsx(hosted on Dryad under CC0).
- Obtain the full file:
- Prepare the coefficient files
- Open the Excel file and save each worksheet as an individual CSV file.
- Make sure each CSV filename exactly matches the names referenced in the R script (e.g.,
A_PLSR_Coefficients_15comp.csv,A_Jackkife_PLSR_Coefficients.csv).
- Organize your local working environment
- Place all CSV files inside a directory named
data/within your local project folder. - Maintain the directory structure expected by the script (i.e.,
data/plsr_coefficients/).
- Place all CSV files inside a directory named
- Run the R script
- Download the script (
Estimate_Cotton_Leaf_Traits_PLSR.Ror the PDF version for documentation) from Zenodo (MIT License). - Execute the script to generate trait estimates.
- Output files will include predicted trait values with associated uncertainty metrics (standard deviation, 5% and 95% confidence bounds).
- Download the script (
Contact
For questions or technical support, please contact: Ved Parkash