Intercalibration of Luojia1-01 and Suomi-NPP-VIIRS Monthly Nighttime Light Composite Using a Spatial-Temporal Residuals Correction Random Forest Model

Satellite-derived Nighttime light (NTL) data from the Suomi National Polar-orbiting Partnership (Suomi-NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) and Luojia1-01 (LJ1-01) have been widely used in multiple urban aspects. However, the inconsistency in NTL intensity (the same spatial coverage and month) between the two sensors impedes their integration on finer scales. To alleviate this issue, a novel model was developed to simulate LJ1-01-like NTL in March 2019, considering Guangzhou main city (GZMC) as the study area, Shenzhen, Dongguan, and Huizhou as validation cities to confirm the spatial simulation ability of the model. First, the spatial features originated from VIIRS NTL and geographical data to enhance the prior spatial information for NTL. Second, the temporal features were derived from the VIIRS NTL time series between 2018 and 2020, such as inter-annual and intra-annual trends, to describe the temporal fluctuation of NTL. Third, a spatial-temporal residuals correction random forest (STRCRF) regression model was proposed to generate LJ1-01-like NTL validation with 10-fold cross-validation (CV). It established the nonlinear relationship between LJ1-01 NTL intensity and spatial-temporal features at the pixel level. It generated LJ1-01-like NTL data with a CV-R2of 0.982 and a CV-RMSE of 3469. The validation accuracy with R2ranged from 0.918 to 0.945, and RMSE varied from 4139 to 7298 in Shenzhen, Dongguan, and Huizhou. The evaluations of LJ1-01-like NTL based on the statistic analyses, spatial patterns, and profile analyses indicated the effectiveness of the STRCRF model in improving NTL intensity consistency at the pixel and landscape level. Moreover, the STRCRF model outperformed the previous related studies and improved the CV-R2by 0.4. Our approach offers the potential for simulating LJ1-01-like NTL by taking advantage of the prior spatial and temporal information and the ensemble regression with residuals correction. This method could contribute to a higher quality NTL dataset for mapping human activities and monitoring the urban environment.