Extracting subpixel vegetation NDVI time series for evaluating the mixed pixel effect on GPP estimation in urban areas

The normalized difference vegetation index (NDVI) is the most widely used vegetation index for monitoring vegetation vigor and cover. As NDVI time series are usually derived at coarse or medium spatial resolutions, pixel size often represents a mixture of vegetated and non-vegetated surfaces. In heterogeneous urban areas, mixed pixels impede the accurate estimation of gross primary productivity (GPP). To address the mixed pixel effect on NDVI time series and GPP estimation, we proposed a framework to extract subpixel vegetation NDVI (NDVIvege) from Landsat OLI images in urban areas, using endmember fractions, mixed NDVI (NDVImix), and NDVI of non-vegetation endmembers. Results demonstrated that the NDVIvege extracted by this framework agreed well with the true NDVIvege cross seasons and vegetation fractions, with R-2 ranging from 0.74 to 0.82 and RMSE ranging from 0.03 to 0.04. The NDVIvege time series was applied to evaluate vegetation GPP in Wuhan, China. The total annual GPP estimated with NDVIvege was 28-35% higher than the total annual GPP estimated with NDVImix, implying uncertainty in the GPP estimations caused by mixed pixels. This study showed the potential of the proposed framework to resolve NDVIvege for characterizing vegetation dynamics in heterogeneous areas.

Automated summary

The normalized difference vegetation index (NDVI) is widely used to monitor vegetation vigor and cover. In heterogeneous urban areas, mixed pixels affect the accurate estimation of gross primary productivity (GPP). This study proposed a framework to extract subpixel vegetation NDVI (NDVIvege) and showed its potential for characterizing vegetation dynamics in heterogeneous areas.