A global urban microwave backscatter time series data set for 1993-2020 using ERS, QuikSCAT, and ASCAT data

Abstract

Urban settlements are rapidly growing outward and upward, with consequences for resource use, greenhouse gas emissions, and ecosystem and public health, but rates of change are uneven around the world. Understanding trajectories and predicting consequences of global urban expansion requires quantifying rates of change with consistent, well-calibrated data. Microwave backscatter data provides important information on upward urban growth - essentially the vertical built-up area. We developed a multi-sensor, multi-decadal, gridded (0.05° lat/lon) data set of global urban microwave backscatter, 1993-2020. Comparison of backscatter from two C-band sensors (ERS and ASCAT) and one Ku-band sensor (QuikSCAT) are made at four invariant non-urban sites (~3500 km²) to evaluate instrument stability and multi-decadal pattern. For urban areas, there was a strong linear correlation (overall R² = 0.69) between 2015 ASCAT urban backscatter and a continental-scale gridded product of building volume, across 8450 urban grid cells (0.05° × 0.05°) in Europe, China, and the USA. This urban backscatter data set provides a time series characterizing global urban change over the past three decades.

Fig. 1

Schematic overview of the urban backscatter data sets. Shaded bars represent range of sensor data (see Methods) and solid circles represent summer season means for each year for ERS (1993–2000), QuikSCAT (1999–2009), ASCAT (2007–2020). Urban data domain masks: 0.05° lat/lon gridded backscatter data were masked by open water minimum threshold of 50%, and GHSL (Global Human Settlement Layer) 2014 built-fraction minimum threshold of 20%. ASCAT 2015 urban backscatter is compared to gridded building volume and height data for Europe, China, and the USA.

Fig. 2

Backscatter time series for tropical evergreen forests. Mean Jan-Mar backscatter power return (PR) time series for four ‘invariant’ tropical forest sites (see Table 2 for site locations). For the QuikSCAT data, a simple Ku-band to C-band offset is computed for each site as the mean difference over the 5 overlapping years (1999–2000, 2007–2009) and QuikSCAT + offset is also plotted. Each point includes a standard deviation range for the 121 grid cells (0.05° lat/lon); in all cases this range is smaller than the symbol size. ERS backscatter data were missing for Jan-Mar 1994 at two sites. Linear fits (backscatter vs. year) are computed separately for ERS, QuikSCAT, and ASCAT data; trends (yr⁻¹) are reported below the data points. Note: y-axis scale matches the y-axis scales of Figs. 3–6 below.

Fig. 3

Urban backscatter vs. building volume. Mean summer 2015 ASCAT backscatter power return (PR) vs. smoothed building volume for 0.05° grid cells in 200+ major cities masked by (a) GHSL 2014 Built Fraction >=20%, and (b) GHSL 2014 Built Fraction >=10%. Data disaggregated by region: China (orange), USA (blue), Europe (green). All urban areas are in the northern hemisphere, so mean Jul-Sep backscatter is used. Linear fits for each region are shown in colored lines, linear fit for all data is the black line. Linear fit statistics are reported in Table 3, for both ASCAT PR vs. building volume, as in figures above, and ASCAT PR vs. building height (figures not shown).

Fig. 4

Backscatter time series for large cities. Mean summer backscatter power return (PR) time series for six large cities, using ERS, QuikSCAT, and ASCAT data. Centered on each city, an 11 × 11 grid of 0.05° lat/lon grid cells was selected, masked to remove open water >50%, and built fraction in 2014 (left column) <20%, or (right column) <10%. Mean and standard deviations of remaining grid cells are plotted. For the QuikSCAT data, a simple Ku-band to C-band offset is computed for each site as the mean difference over the 5 overlapping years (1999–2000, 2007–2009) and QuikSCAT + offset is also plotted.

Fig. 5

Backscatter time series for large cities. Mean summer backscatter power return (PR) time series for six large cities, using ERS, QuikSCAT, and ASCAT data. Centered on each city, an 11 × 11 grid of 0.05° lat/lon grid cells was selected, masked to remove open water >50%, and built fraction in 2014 (left column) <20%, or (right column) <10%. Mean and standard deviations of remaining grid cells are plotted. For the QuikSCAT data, a simple Ku-band to C-band offset is computed for each site as the mean difference over the 5 overlapping years (1999–2000, 2007–2009) and QuikSCAT + offset is also plotted.

Fig. 6

Urban backscatter vs. building volume for New York City, Los Angeles, and Washington DC. Mean summer 2015 ASCAT backscatter power return (PR) vs. smoothed building volume, using the GHS BF₂₀₁₄ ≤0.2 mask. Note: panel axis scales match Fig. 3; black (blue) dashed line is linear fit to all regions (USA) data from Fig. 3.