Evaluation of the performance of Suomi NPP VIIRS top of canopy vegetation indices over AERONET sites

• Analyzed VIIRS TOC VI retrievals over AERONET sites
• Performance of atmospheric correction is limited by AOT overestimation near clouds.
• Uncertainties in atmospheric correction affect TOC NDVI, but TOC EVI is resistant.
• Cloud and snow masks are effective in screening low quality TOC VIs.
• Monitoring of VIIRS VI over AERONET sites has been automated with a web tool.

The Suomi NPP VIIRS Vegetation Index (VI) Environment Data Record (EDR) includes the Top Of the Atmosphere Normalized Difference Vegetation Index (TOA NDVI) and the Top Of the Canopy Enhanced Vegetation Index (TOC EVI). TOC NDVI will be included in the VI EDR suite for the upcoming JPSS-1 and JPSS-2 missions. Currently, the VIIRS VI suite is undergoing extensive Calibration and Validation (Cal/Val) activities to quantify product performance and to provide guidance for algorithm improvement. In this study we utilized AErosol RObotic NETwork (AERONET) based Surface Reflectance (SR) match-up data sets. Match-ups are pairs of VIIRS SR and SR derived from atmospheric correction of VIIRS TOA reflectances using AERONET ground measurements (aerosol and water vapor parameters) and ancillary data. Atmospheric correction is performed with the Second Simulation of a Satellite Signal in the Solar Spectrum (6S) radiative transfer code, the same code used in the operational VIIRS atmospheric correction. Daily time series January 1, 2013–March 31, 2014 of 101 × 101 pixel window match-up subsets at 375 m VIIRS Imagery resolution, centered at the selected AERONET site locations were used. The overall objective of the study was to characterize the performance of the VIIRS TOC VIs (TOC EVI and TOC NDVI) over AERONET sites as a function of accuracy of inputs of atmospheric correction algorithm (aerosol, water vapor and others) and quality control screening (cloud and snow masks). We performed three types of analyses: (1) analysis of overall performance of VI product over all sites, (2) analysis of time series over select sites representative of vegetation types, and (3) sensitivity analysis of VI to cloud, aerosols and snow contamination. Over the period of time series average Accuracy, Precision and Uncertainty statistics were 0.009, 0.035, and 0.038, respectively, for TOC NDVI and − 0.004, 0.015, and 0.016, respectively, for TOC EVI. Those statistics were derived based on screening to retain only confidently clear, snow free, non-urban pixels. The reason for the substantial difference in the performance of VIs is a different sensitivity of VIs to residual atmospheric contamination. According to our study VIIRS cloud mask performed reasonably. However, Aerosol Optical Thickness (AOT) was substantially overestimated in the vicinity of clouds, which resulted in overcorrection of VIIRS visible channels and ultimately overestimation of TOC NDVI. TOC EVI, on the other hand, remained largely unaffected, as an atmospherically resistant index. TOC EVI was also found to be less sensitive to residual snow contamination. Results of this study indicate the need to develop VI-specific quality control, which effectively screens data based on index sensitivity to atmospheric contamination. Monitoring VIIRS VIs over AERONET sites has been automated with a web-based STAR JPSS VI Monitor.