Hyperspectral inversion of SPAD in Camellia oleifera based on integrated Taylor-CC and ANN with optimal wavelength selection

Abstract Background Reliable assessment of leaf chlorophyll status can support the monitoring of Camellia oleifera , but point-based SPAD measurements provide limited spatial coverage. Hyperspectral data offer a non-destructive alternative, although the transferability of spectral-SPAD models across phenological stages remains uncertain. This study evaluated a wavelength-selection and modeling framework for canopy SPAD estimation in Camellia oleifera . Methods Canopy hyperspectral reflectance and corresponding SPAD measurements were collected during the spring-shoot (April), fruit-expanding (August), and full-blooming (November) stages. Taylor expansion-based correlation coefficient (Taylor-CC) analysis was used to select core wavelengths across a range of correlation thresholds. Artificial neural network (ANN) models were developed and compared with random forest and support vector regression approaches. Stage-specific and cross-stage predictions were assessed using R2, root mean square error (RMSE), and the ratio of performance to deviation (RPD). Forward-difference analysis was used to characterize the local sensitivity of model predictions to reflectance perturbations. Results Stage-matched models performed better on their corresponding datasets than in cross-stage predictions. The ANN models achieved R 2 values of 0.8350, 0.7984, and 0.7280 for April, August, and November, respectively, with stage-matched RPD values above 1.9. In cross-stage predictions, R 2 remained below 0.5 and RPD below 1.5. Taylor-CC-based wavelength configurations performed favorably relative to full-spectrum inputs in the reported comparisons, although the magnitude of improvement varied across model and stage combinations. The April model showed the lowest forward-difference sensitivity among the evaluated models; this indicates a weaker local response under the specified perturbation procedure, rather than establishing physiological causation or external robustness. Conclusions The results indicate limited transferability of canopy spectral-SPAD relationships among the three sampled datasets and support stage-aware calibration under the conditions studied. April is a candidate monitoring window for the sampled plantation and acquisition protocol, but validation across years, sites, and cultivars is required before recommending a generally optimal monitoring time. Taylor-CC wavelength selection combined with ANN modeling warrants further evaluation using independent field data.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11500533/v1
Primary Topic
Remote Sensing in Agriculture
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Hyperspectral inversion of SPAD in Camellia oleifera based on integrated Taylor-CC and ANN with optimal wavelength selection

Xuehai Tang, Zhibin Sun, Xingzi Kou, Chunyu GE et al.
3 links
Research Square
Remote Sensing in Agriculture
preprint

Hyperspectral inversion of SPAD in Camellia oleifera based on integrated Taylor-CC and ANN with optimal wavelength selection

Xuehai Tang, Zhibin Sun, Xingzi Kou, Chunyu GE, Xinyue Jiang, Yue Zhang
preprint en

Abstract

Abstract Background Reliable assessment of leaf chlorophyll status can support the monitoring of Camellia oleifera , but point-based SPAD measurements provide limited spatial coverage. Hyperspectral data offer a non-destructive alternative, although the transferability of spectral-SPAD models across phenological stages remains uncertain. This study evaluated a wavelength-selection and modeling framework for canopy SPAD estimation in Camellia oleifera . Methods Canopy hyperspectral reflectance and corresponding SPAD measurements were collected during the spring-shoot (April), fruit-expanding (August), and full-blooming (November) stages. Taylor expansion-based correlation coefficient (Taylor-CC) analysis was used to select core wavelengths across a range of correlation thresholds. Artificial neural network (ANN) models were developed and compared with random forest and support vector regression approaches. Stage-specific and cross-stage predictions were assessed using R2, root mean square error (RMSE), and the ratio of performance to deviation (RPD). Forward-difference analysis was used to characterize the local sensitivity of model predictions to reflectance perturbations. Results Stage-matched models performed better on their corresponding datasets than in cross-stage predictions. The ANN models achieved R 2 values of 0.8350, 0.7984, and 0.7280 for April, August, and November, respectively, with stage-matched RPD values above 1.9. In cross-stage predictions, R 2 remained below 0.5 and RPD below 1.5. Taylor-CC-based wavelength configurations performed favorably relative to full-spectrum inputs in the reported comparisons, although the magnitude of improvement varied across model and stage combinations. The April model showed the lowest forward-difference sensitivity among the evaluated models; this indicates a weaker local response under the specified perturbation procedure, rather than establishing physiological causation or external robustness. Conclusions The results indicate limited transferability of canopy spectral-SPAD relationships among the three sampled datasets and support stage-aware calibration under the conditions studied. April is a candidate monitoring window for the sampled plantation and acquisition protocol, but validation across years, sites, and cultivars is required before recommending a generally optimal monitoring time. Taylor-CC wavelength selection combined with ANN modeling warrants further evaluation using independent field data.

Research Square
Anhui Agricultural University (CN), Nanjing Normal University (CN)
Remote Sensing in Agriculture
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon