Improved generality of wheat green LAI models through mitigation of the effect of leaf chlorophyll content variation with red edge vegetation indices

The retrieval of wheat green leaf area index (LAI_G) from satellite imagery is critical for monitoring crop growth and assessing food security. Numerous vegetation indices (VIs) derived from spectral reflectance have been widely used to estimate LAI_G. In particular, red edge VIs can mitigate the confounding effect of multiple factors, such as the soil background and leaf inclination angle variation, and typically are highly correlated with LAI_G. However, their relationship to LAI_G tends to be affected by variations in leaf chlorophyll content (LCC), because the position of the red edge of vegetation spectra shifts with changes in LCC. This issue directly limits the operational use of VI-LAI_G models, especially those employing red-edge bands. Therefore, to reduce the sensitivity of VI-LAI_G relationships to LCC variation, this study proposed an innovative approach, called the Difference Combination between Spectral Indices (DCSI). Using synthetic data simulated with the PROSAIL radiative transfer model, we tested the dependence of the algebraic difference between common VIs on LCC. The results show that many combinations of VIs are insensitive to LCC variation. The newly developed DCSI combination between the Sentinel-2 red edge position (S2REP) and B6-red edge band (RE2) (i.e., DCSI(S2REP&RE2)), produces the most accurate LAI_G model when LCC varies. We also modified the constant of this DCSI combination, to develop the Sentinel-2 modified red edge position (S2MREP) for LAI_G retrievals. In comparison to traditional VI-LAI_G models, the S2MREP-LAI_G model has higher accuracy, with $R_{\mathit{cal}}^2$$R_{\mathit{cal}}^2$ of 0.76 in calibration, and in validation $R_{\mathit{val}}^2$$R_{\mathit{val}}^2$ of 0.72 and $\mathit{RRMSE}$$\mathit{RRMSE}$ of 23.61 %. In addition, the S2MREP-LAI_G model ($\mathit{RRMSE}$$\mathit{RRMSE}$=28.64 %) also outperforms the existing Sentinel-2 LAI product ($\mathit{RRMSE}$$\mathit{RRMSE}$=38.20 %) in the retrieval of wheat LAI_G. In summary, the proposed DCSI approach and S2MREP effectively mitigate the impact of LCC variations on LAI_G retrievals, thus facilitating the large-scale retrieval of LAI_G and the spatial mapping of wheat LAI_G.