Quantitative Estimation of Missing Value Interpolation Methods for Suomi-NPP VIIRS/DNB Nighttime Light Monthly Composite Images

Composited nighttime light (NTL) data are widely used to evaluate the intensity of human activities and estimate urban dynamics. However, due to the effects of stray light, Visible Infrared Imaging Radiometer Suite (VIIRS) monthly composited data are vastly missing in middle-high latitude areas in summer, resulting in obvious application limitations. Interpolation methods for missing values based on the temporal continuity characteristics of pixels in monthly composite images could be a potential solution to address the limitations. In this paper, we employed least-squares line fitting (LSLF), least-squares quadratic polynomial fitting (LSQPF), least-squares cubic polynomial fitting (LSCPF), cubic spline interpolation (Spline), cubic Hermite interpolation (Hermite), cubic Bezier curve interpolation (Bezier), gray forecast model (GFM) and cubic exponential smoothing (Exponent) to simulate the missing values in VIIRS/DNB data. The rationality and accuracy of the simulation results was quantitatively evaluated. Then, we modeled and decomposed the time series to estimate the missing values of multiple months in NTL data. The results showed that 1)in comparison with the vcm data, the simulation results for LSLF were the closest to the vcm data. In the fitting experiment between the simulation results of these algorithms and vcm data, the correlation coefficient of LSLF was the highest in 4 of the 6 low-latitude cities, ranging from 0.896 to 0.976. 2) In comparison with the socioeconomic data regression, the simulation results of the LSLF, Bezier and Exponent methods showed greater correlation coefficients than vcmsl data. 3) The correlation coefficients between the six months image estimated by the time series decomposition method and the vcm data were all greater than 0.8. It can be inferred from the results that data interpolation methods based on temporal continuity characteristics between pixels in the same location could be potential practical approaches for the simulation of missing values in the vcm monthly composited data.