Spatiotemporal effects of atmospheric correction on fractional vegetation coverage retrieval in coastal wetlands based on the pixel dichotomy model

Coastal wetlands are highly dynamic and ecologically vulnerable ecosystems, making accurate vegetation monitoring essential for ecological conservation and restoration. Fractional Vegetation Coverage (FVC) serves as a crucial indicator for assessing these environments. However, atmospheric scattering and absorption significantly interfere with satellite signals, complicating accurate assessments. To clarify the spatiotemporal differences in FVC of coastal wetlands before and after atmospheric correction, this study conducted a comprehensive analysis based on the Google Earth Engine (GEE) cloud computing platform. We systematically analyzed the FVC data of the Yellow River Delta coastal wetlands from 2021 to 2023 from the perspective of seasonal variations. The FVC was retrieved based on the Pixel Dichotomy Model using the Normalized Difference Vegetation Index (NDVI). Specifically, NDVI serves as the measurement basis, where its low values reflect bare soil information and high values represent pure vegetation. By extracting these two representative values as a baseline, the Pixel Dichotomy Model assumes that the information within a mixed pixel is composed of these two categories, therefore, the proportion of the vegetation component directly reflects the fractional vegetation cover. The results showed that although there was a highly significant statistical difference (p<0.001) between the FVC before atmospheric correction (TOA_FVC) and the FVC after atmospheric correction (SR_FVC), their dynamic variation trends remained highly consistent with the natural growth cycle of vegetation. Quantitatively, SR_FVC was generally higher than TOA_FVC. During the seasons when SR_FVC exceeded TOA_FVC, the difference was 2.40 percentage points on average, displaying a broader spatial coverage specifically within the middle-to-high level FVC regions. In terms of specific land cover classifications, including those dominated by herbaceous and shrub vegetation, as well as composite land cover types where vegetation and non-vegetation elements coexist, the SR_FVC levels were generally higher. Both of these quantitative and categorical differences were found to be particularly obvious during the summer and autumn seasons, corresponding to the peak periods of vegetation growth. Furthermore, spatial centroid analysis revealed that the migration magnitude of the SR_FVC centroid was significantly greater than that of TOA_FVC. Specifically, there was an obvious difference in the displacement trajectory during the winter-spring period. This indicates that the atmospheric correction process more clearly revealed the seasonal spatial heterogeneity of coastal wetland vegetation. In addition, the anomalously high values of SR_FVC detected in winter and spring reflected a specific methodological limitation: when NDVI is used to quantitatively characterize vegetation information, it is profoundly difficult to effectively suppress complex surface background signals in low-coverage areas. Ultimately, these conclusions not only provide a critically new perspective for understanding the profound impact of atmospheric correction on FVC reshaping, but also offer robust data support and a solid theoretical basis for improving the long-term vegetation monitoring accuracy of coastal wetland ecosystems.