EVI data and EVImax trends

The vegetation index data sets from the MOD13A1 version 6 product were used in this study to match the spatial resolution of the land cover data (500 m). These data sets include two primary vegetation layers: the normalized difference vegetation index (NDVI) and the EVI. Due to the greater sensitivity of the EVI relative to that of the NDVI in the high-biomass regions, our results were based on the use of the EVI. For each pixel, the annual maximum EVI (EVI_max) was generated from all the acquisitions over a 16-day period.

We used a Mann–Kendall test, which is a nonparametric test, to detect monotonic trends in the time-series data and evaluate the trends of EVI_max from 2001 to 2018. For a comprehensive and systematic analysis of global urban greenness change, we used the BUA in 2018 to detect urban greening change. Then, if a pixel changed to an urban area in 2002 or later, the greenness change in the pixel was calculated. Notably, if the expended BUA was developed from cropland and natural vegetation land, the greenness of the expended BUA would possibly decrease. In fact, we calculated the BUA where greenness showed a significant increasing trend (P < 0.05). Using this method, we could exclude those pixels that had changed from rural vegetation to BUA (urban expansion).

In this study, the spatial resolution of the MOD13A1 data was 500 m; thus, we wrote a Python program to calculate the Mann–Kendall trend of EVI_max from 2001 to 2018 for each pixel in a raster file. Then, the Geospatial Data Abstraction Library (https://gdal.org) was used to mosaic all the raster files (212 files) of the trends into a global raster map of the EVI_max trend. Finally, we extracted the EVI_max trend for each urban pixel within the urban extent (BUA in 2018) and calculated the ratio (R_greening) of the area of the pixels with a significant increasing EVI_max trend (P < 0.05) to the entire BUA of the corresponding urban extent.