Modeling and Prediction of NPP-VIIRS Nighttime Light Imagery Based on Spatiotemporal Statistical Method

Abstract

The cloud-free monthly composite of the global nighttime light (NTL) data derived from the Suomi National Polar orbiting Partnership with the Visible Infrared Imaging Radiometer Suite (VIIRS) day/night band (DNB) has gained popularity for detecting anthropogenic and socioeconomic activities. However, the monthly VIIRS DNB composite suffers from a data missing problem induced by continuous cloud cover. The full potential of the VIIRS DNB time series is consequently hindered by low-quality and missing observations. This article proposes a spatiotemporal statistical method (STSM) to predict the VIIRS DNB imagery in severe absence of valid observations’ situation. The polynomial with the harmonic model was applied to describe the long-term trends and seasonal cycles in time series. A spatial marginal semivariogram was established to quantify the data dependence in space; we then used spatial interpolation to correct the predicted results from temporal curve fitting. The final predicted values were validated with the actual values based on cross-validation. The results suggest that the STSM is suitable for predicting with a high coefficient of determination ( $R^{2} = 0.922$ ) and a relatively low root-mean-square error (RMSE = 3.40 nW/cm2/sr). We extended the proposed method to forecast future imagery for a five-month period, the performance of which was more stable, with the highest $R^{2}$ /RMSE (0.158 ± 0.010), compared with two other methods. Therefore, the STSM is effective and stable for modeling and predicting the VIIRS DNB monthly composite and will help address the data missing issue.