Forecasting China’s GDP at the pixel level using nighttime lights time series and population images

China’s rapid economic development greatly affected not only the global economy but also the entire environment of the Earth. Forecasting China’s economic growth has become a popular and essential issue but at present, such forecasts are nearly all conducted at the national scale. In this study, we use nighttime light images and the gridded Landscan population dataset to disaggregate gross domestic product (GDP) reported at the province scale on a per pixel level for 2000–2013. Using the disaggregated GDP time series data and the statistical tool of Holt–Winters smoothing, we predict changes of GDP at each 1 km × 1 km grid area from 2014 to 2020 and then aggregate the pixel-level GDP to forecast economic growth in 23 major urban agglomerations of China. We elaborate and demonstrate that lit population (brightness of nighttime lights × population) is a better indicator than brightness of nighttime lights to estimate and disaggregate GDP. We also show that our forecast GDP has high agreement with the National Bureau of Statistics of China’s demographic data and the International Monetary Fund’s predictions. Finally, we display uncertainties and analyze potential errors of this disaggregation and forecast method.