Validation of Sentinel-2 simplified level 2 prototype (SL2P) processor in retrieving leaf chlorophyll concentration over dusty environment

Abstract

Reliable information on leaf chlorophyll concentration (LCC) in mining-impacted regions is critical for vegetation assessment and management. However, the presence of foliar dust (FD) significantly alters canopy reflectance, introducing uncertainty in satellite-based chlorophyll estimation. The present study, possibly for the first time, aims to evaluate the performance of the globally trained Sentinel-2 Simplified Level 2 Prototype Processor (SL2P) against locally calibrated empirical models and in-situ measurements for estimating LCC in FD-affected mining landscape. Furthermore, this study explores the LCC-FD nexus based on in-situ observations. In-situ LCC measurements were collected using a handheld chlorophyll meter across 40 sites over an industrial region in India. Sentinel-2B spectral bands (surface reflectance) and vegetation indices (VIs) were used to develop empirical models, while SL2P-derived LCC estimates were validated against in-situ measurements. The findings of the study revealed a non-linear FD–LCC relationship, indicating distinct vegetation responses across low, intermediate, and high dust loads. Furthermore, the study evidenced that SL2P consistently underestimates LCC under both dusty and non-dusty conditions, exhibiting substantial negative bias (–8.18 to –14.61 µg/cm²) and high uncertainty (RMSE = 10.10–15.09 µg/cm²), indicating limited reliability for LCC retrieval. In contrast, several empirical models demonstrated improved performance, particularly under dusty conditions. Band-based models using the red-edge (RE2), red, and near-infrared (NIR) bands achieved low dispersion (MAD ≈ 2.1–3.2 µg/cm²) and low relative uncertainty (nRMSE ≈ 7–8%). Among VIs, the Transformed Soil-Adjusted Vegetation Index (TSAVI) showed stable performance in dusty environments (MAD ≈ 2.0 µg/cm²; nRMSE ≈ 9%), while Global Environmental Monitoring Index (GEMI) and Modified Chlorophyll Absorption in Reflectance Index (MCARI) exhibited lower transferability across conditions. These results highlight the limited accuracy of globally trained biophysical algorithms across diverse environments, and advocate for locally calibrated, adaptive models for improved LCC estimation accuracy.