Bennett, M. M. & Smith, L. C. Advances in using multitemporal night-time lights satellite imagery to detect, estimate, and monitor socioeconomic dynamics. Remote Sens. Environ. 192, 176–197, https://doi.org/10.1016/j.rse.2017.01.005 (2017).

Several attempts have been made to synthesize consistent NTL time series from the DMSP and VIIRS NTL data. Most of them focus on simulate DMSP NTL data by downgrading VIIRS NTL data. Li et al.11 proposed an inter-calibration model to simulate DMSP NTL data from the VIIRS NTL data to estimate the dynamics of city lights in Syria’s primary human settlement from March 2011 to January 2017. This was achieved by employing a power function for radiometric degradation and a Gaussian low pass filter for spatial degradation. Zheng et al.34 developed a geographically weighted regression model to fit the radiance-calibrated DMSP NTL data and VIIRS NTL data, and generated a DMSP-like NTL dataset for China from 1996 to 2017. Zhao et al.26 conducted a sigmoid model to convert VIIRS NTL data into simulated DMSP NTL data from 1992 to 2018 in Southeast Asia. Li et al.35 generated a harmonized DMSP-like NTL dataset using sigmoid function at global scale from 1992 to 2018. Nechaev et al.25 applied the U-Net Convolutional Neural Network (CNN) and deep residual learning strategy to synthesize an image similar to the DMSP NTL image from the corresponding VIIRS product. This proposed deep learning-based method exhibited high accuracy, generating images visually indistinguishable from real DMSP NTL data. Zhang et al.6 proposed a Night-Time Light convolutional LSTM network to enhance temporal consistency and successfully generated the PANDA-China dataset, which provides annual prolonged artificial NTL data at a 1-km resolution for China from 1984 to 2020.

Xiuxiu Chen: Conceptualization, Methodology, Formal analysis, Software, Validation, Data curation, Visualization, Writing, Editing; Zeyu Wang: Writing, Validation, Visualization; Feng Zhang: Editing, Review, Funding acquisition, Resources, Supervision; Guoqiang Shen: Review, Supervision; Qiuxiao Chen: Review, Supervision.

Li, X., Zhou, Y., Zhao, M. & Zhao, X. A harmonized global nighttime light dataset 1992-2018. Sci. Data 7, 168, https://doi.org/10.1038/s41597-020-0510-y (2020).

Aqua illumination controllermanual pdf

Li, X. & Zhou, Y. Urban mapping using DMSP/OLS stable night-time light: a review. Int. J. Remote Sens. 38, 6030–6046, https://doi.org/10.1080/01431161.2016.1274451 (2017).

where \({\rm{n}}\) is the number of training samples, \({\rm{m}}\) is the number of pixels in the sample. \({{\rm{x}}}_{{\rm{i}},{\rm{j}}}^{{\rm{DMSP}}}\), \({{\rm{x}}}_{{\rm{i}},{\rm{j}}}^{{\rm{NDVI}}}\), \({{\rm{y}}}_{{\rm{i}},{\rm{j}}}^{{\rm{V}}{\rm{I}}{\rm{I}}{\rm{R}}{\rm{S}}}\) is the DMSP NTL value, NDVI value and VIIRS NTL value of the jth pixel value in the ith sample of DMSP NTL, NDVI and VIIRS NTL data, respectively.

Liu, Z., He, C., Zhang, Q., Huang, Q. & Yang, Y. Extracting the dynamics of urban expansion in China using DMSP-OLS nighttime light data from 1992 to 2008. Landsc. Urban Plan. 106, 62–72, https://doi.org/10.1016/j.landurbplan.2012.02.013 (2012).

The version 2 annual global VIIRS NTL composites (Annual VNL V2) is a new consistently processed time series of annual global VIIRS nighttime lights, which is also obtained from the Payne Institute for Public Policy at the Colorado School of Mines (https://eogdata.mines.edu/products/vnl/#annual_v2). Annual VNL V2 is produced from the version 1 monthly cloud-free average radiance grids. Filtering was conducted on Annual VNL V2 to to remove sunlit, moonlit and cloudy pixels. In addition, outlier removal was employed to discard biomass burning pixels and isolate the background using the twelve-month median radiance. The spatial reference is WGS84, with a spatial resolution of 15 arc-seconds (approximately 500 m). The spatial extent covers a range from −180° to 180° longitude and from 65° S to 75° N latitude. Compared to the version 1 annual VIIRS NTL composites (Annual VNL V1), which is only available in 2015 and 2016, the temporal coverage of Annual VNL V2 extends from 2012 to the present. Considering the temporal continuity of annual VIIRS data, Annual VNL V2 was opted in this study, to integrate DMSP data to generate a consistent long-term annual NTL dataset similar to VIIRS. We downloaded the Annual VNL V2 data from 2012 to 2023. The “average_masked” version was utilized, and pixels with value less than 0 were set to 0.

The two existing global extended NTL datasets were utilized for comparison. One was the extended time series of global VIIRS-Like NTL data from 2000 to 2018 provided by Chen et al.41,46 (ChenVNL), while the other was the harmonized global DMSP-like NTL data from 1992 to 2018 provided by Li et al.35,47 (LiDNL). The national GDP statistics from 1992 to 2020 were obtained from the World Bank website (http://www.gadm.org) to aid in assessing the temporal consistency of the generated NTL dataset. The version named “GDP (constant 2015 US$)” was used in this study. The administrative boundary data is sourced from the Global Administrative Areas (GADM) database, which can be accessed at http://www.gadm.org.

The programs used to generate all the results were Python 3.7 and ArcGIS (10.2). The source code and scripts used for training, testing, predicting and validating the NTL data and are available in the open GitHub repository “https://github.com/cxxtribal/NTLSRU-Net”.

The temporal trends of global total NTL intensity from 1992 to 2023. (a) raw DMSP data and VIIRS data, (b) three artificial NTL products, namely SVNL, ChenVNL and LiDNL.

The datasets utilized in this study consist of two categories, as shown in Table 1. The first category is utilized for generating the SVNL dataset, encompassing the stable DMSP NTL product, the annual VIIRS NTL product and the Landsat NDVI time series. The second category is ancillary datasets that help to access the accuracy and quality of the generated NTL data, including the existing global extended NTL datasets, Gross Domestic Product (GDP), and administrative boundary data.

The profiles of SVNL data, real VIIRS data, DMSP data, and ChenVNL data in Beijing and Los Angeles are shown in Fig. 8. It is evident that the fluctuation of SVNL data (blue solid line) and ChenVNL data (green solid line) agree well with that of real VIIRS data (orange solid line). The DMSP data reaches a maximum value of 63 and remains constant due to saturation in urban core areas, thereby limiting its ability to capture spatial differences in lighting within the urban. Conversely, both SVNL and ChenVNL exhibit overall trends consistent with those of real VIIRS data, effectively capturing urban physical spatial structures. However, some segments of the SVNL profile in Los Angeles underestimate while parts of the ChenVNL profile in Beijing overestimate when compared with real VIIRS data. The underestimation of SVNL primarily occurs in urban fringe areas, where certain pixels exhibit high NTL intensity while surrounding pixels show relatively lower NTL intensity. In these areas, the CNN operations may result in the reduction of NTL intensity for central pixels by their neighboring pixels.

We would like to thank the Earth Observation Group for providing the original DMSP NTL data and VIIRS NTL data. This work was supported by the National Key Research and Development Program of China under Grants 2019YFE0127400.

Aqua illumination controllermanual

Henderson, M., Yeh, E. T., Gong, P., Elvidge, C. & Baugh, K. Validation of urban boundaries derived from global night-time satellite imagery. Int. J. Remote Sens. 24, 595–609, https://doi.org/10.1080/01431160304982 (2003).

Shi, K. et al. Evaluating the ability of NPP-VIIRS nighttime light data to estimate the gross domestic product and the electric power consumption of China at multiple scales: A comparison with DMSP-OLS data. Remote Sens. 6, 1705–1724, https://doi.org/10.3390/RS6021705 (2014).

Liu, H., Ma, L. & Xu, L. Estimating spatiotemporal dynamics of county-level fossil fuel consumption based on integrated nighttime light data. J. Clean Prod. 278, 123427, https://doi.org/10.1016/j.jclepro.2020.123427 (2021).

Chen, X., Zhang, F., Du, Z. & Liu, R. An unsupervised urban extent extraction method from NPP-VIIRS nighttime light data. Remote Sens. 12, 3810, https://doi.org/10.3390/rs12223810 (2020).

Croft, T. A. Nighttime images of the earth from space. Sci. Am. 239, 86–98, https://doi.org/10.1038/SCIENTIFICAMERICAN0778-86 (1978).

There is an apparent interannual inconsistency in DMSP NTL data due to lacking on-board calibration. The DN values obtained from different satellites or different years at the same location are not directly comparable. Extensive research have been conducted to address this issue. The calibrated DMSP method proposed by Elvidge et al.48 is widely adopted due to its easy implementation and robust performance. However, it is associated with a significant drawback as it leads to a considerable reduction in the range of total NTL DN values compared to the initial pattern. In order to minimize modifications of the original DN values, Li et al.49 developed a stepwise calibration scheme for NTL data derived from different satellites and years. In this study, Li’s method was employed to ensure consistency in the DMSP NTL time series. In some cases, there are two images from different satellites for the same year, e.g., F10-1994 and F12-1994. The annual result was derived from averaging the calibrated images of that year50.

The trained model was utilized to apply the DMSP NTL data and Landsat NDVI data from 1992–2013, in order to generate simulated VIIRS NTL data for the corresponding time period. To balance GPU memory and computational efficiency during prediction, the global NDVI and DMSP NTL datasets were segmented into sub-images. Due to the incomplete context of edge pixels in the convolution process of the U-Net model, directly filling in missing pixel values with 0 for the boundary information will result in lower prediction accuracy for edge pixels compared to internal pixels (Fig. 3b). Accordingly, each sub-image input should have an overlap of width \({f}_{{over}}\) and then the reconstructed simulated VIIRS should be clipped with a border width of \({f}_{{clip}}\), retaining only the predicted results within the sub-image (Fig. 3a). The value of \({f}_{{over}}\) should be greater than or equal to \({f}_{{clip}}\) to ensure that the clipped super-resolved images can be seamlessly merged into a global image.

The quality of the generated SVNL dataset was evaluated from three perspectives: accuracy, spatial pattern, and temporal trend. First, we compared the histograms of real VIIRS and simulated VIIRS NTL intensity in two overlaid years of 2012 and 2013, and analyzed the scatter plots of the two NTL data at multi scales. Subsequently, a visual comparison and profile analysis were conducted to compare the spatial pattern of real and simulated NTL intensity. Furthermore, the temporal trend consistency of SVNL data was validated through qualitative analysis and correlation analysis with social economic data. The ChenVNL data, the only existing extended VIIRS-like time series, was utilized as a benchmark for measuring the performance of SVNL data in these assessments.

Li, X., Zhou, Y., Zhao, M. & Zhao, X. Harmonization of DMSP and VIIRS nighttime light data from 1992-2018 at the global scale. figshare https://doi.org/10.6084/m9.figshare.9828827.v2 (2020).

Chen, Z. et al. A new approach for detecting urban centers and their spatial structure with nighttime light remote sensing. IEEE Trans. Geosci. Remote Sensing 55, 6305–6319, https://doi.org/10.1109/TGRS.2017.2725917 (2017).

Chen, Z. et al. An extended time-series (2000–2018) of global NPP-VIIRS-like nighttime light data from a cross-sensor calibration. Earth Syst. Sci. Data 13, 889–906, https://doi.org/10.5194/essd-13-889-2021 (2021).

A linear regression model was conducted to compare GDP and total NTL intensity at global and national levels (Fig. 11). The SVNL data has a similar trend with the GDP from 1992 to 2023, with an R2 of 0.88 in global (Fig. 11a1). As shown in Fig. 11a2, most countries exhibit an R2 exceeding 0.5, with China, India, and Afghanistan demonstrating high values above 0.8. However, Japan, the United Kingdom, and Canada display relatively low R2 below 0.3. This may be due to the fact that the urbanization level in these developed countries had already reached a high level in the 1990s, and the increase in NTL intensity over the past three decades has not been significant. A correlation analysis of ChenVNL and GDP time series was conducted as a baseline for comparison (Fig. 11b). The R2 for SVNL and ChenVNL are close and similar both at global and national levels. These findings indicated the relatively high reliability of the SVNL data.

Zhao, M. Applications of satellite remote sensing of nighttime light observations: Advances, challenges, and perspectives. Remote Sens. 11, 1971, https://doi.org/10.3390/rs11171971 (2019).

Levin, N. et al. Remote sensing of night lights: a review and an outlook for the future. Remote Sens. Environ. 237, 111443, https://doi.org/10.1016/j.rse.2019.111443 (2020).

Yu, B. et al. Poverty evaluation using NPP-VIIRS nighttime light composite data at the county level in China. IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 8, 1217–1229, https://doi.org/10.1109/JSTARS.2015.2399416 (2015).

Zheng, Q., Seto, K. C., Zhou, Y., You, S. & Weng, Q. Nighttime light remote sensing for urban applications: Progress, challenges, and prospects. ISPRS-J. Photogramm. Remote Sens. 202, 125–141, https://doi.org/10.1016/j.isprsjprs.2023.05.028 (2023).

Previous studies have demonstrated that VIIRS NTL data outperforms DMSP NTL data in estimating socioeconomic indicators and identify urban spatial structure36,37,38,39,40. It is a formidable challenge to simulate VIIRS NTL data using DMSP NTL data, as it involves taking fewer pixels and a limited range of bright values to reconstruct more detailed pixels and a broader range of radiance values. Chen et al.41 constructed a global annual VIIRS-like NTL extended time series from 2000 to 2018 for the first time. They used vegetation index and auto-encoder neural network model to obtain additional information and achieved the transformation of DMSP NTL data into VIIRS NTL data. The vegetation index in Chen et al. was derived from MODIS images, which have been available since 2000.

AI Hydracontroller

Correlation analysis of the generated SVNL (before 2012) with the DMSP data and ChenVNL data. (a1) Scatter plot of total NTL intensity of global countries for SVNL and DMSP in 2005; (a2) Scatter plot of total NTL intensity of global countries for ChenVNL and DMSP in 2005; (b) R2 for y = x linear fit and correlation coefficient of SVNL and ChenVNL (2000-2012).

To assess the spatial consistency between the generated SVNL data and real VIIRS data, we conducted visual comparisons and profile analysis using SVNL data, real VIIRS data, DMSP data, and ChenVNL data. Visual comparisons were performed in eight selected regions around the world: Beijing, Shanghai, Guangdong-Hong Kong-Macao Greater Bay Area (GBA), London, Los Angeles, New York, Tokyo, and Cape Town (Fig. 7). These regions were chosen due to their extensive urban areas, significant spatial variations in NTL intensity, and typicality of their global distribution.

). These Landsat images have undergone geometric, terrain, and radiometric corrections, and the CFMASK algorithm was used to generate masks for clouds, shadows, water, snow, and saturated pixels. Images with minimal cloud cover were selected using the “pixel_qa” band to compose annual data. The maximum value strategy is employed for each pixel to compose the annual image for each year. Compared to the mean or median value strategies, the maximum value strategy greatly reduced the impact of scan line artifacts in Landsat 7 (e.g., ETM+) images since 2003. Missing values for specific years were filled using pixels from neighboring year images. The generated annual Landsat NDVI was resampled to 15 arc-seconds with the spatial reference of WGS84, snapping the image cells to VIIRS NTL image.

Need help? We're available at (507)-304-2579 Email us at sales@fragtasticreef.com or LIVE CHAT Questions? Feedback? powered by Olark live chat software

Although an increasing number of satellite platforms are equipped with nighttime light detection capabilities, only the two NTL datasets have relatively long historical records, which are derived from the Defense Meteorological Satellite Program-Operational Linescan System (DMSP-OLS) and Suomi National Polar-Orbiting Partnership-Visible Infrared Imaging Radiometer Suite (NPP-VIIRS)24,26,27. The DMSP NTL dataset is publicly available from 1992 to 2013, providing the longest historical records, while the VIIRS NTL dataset is available since 201328. Significant inconsistency exist between DMSP NTL data and VIIRS NTL data in terms of spatial resolution, radiometric resolution, and data quality due to variations in sensor design, data collection techniques, and storage methods. The DMSP NTL data experiences interannual inconsistency, saturation, and blooming, due to the lack of onboard calibration, the 6-bit quantization, the accumulation of geographic bias in data synthesis and the coarse spatial resolution26,29,30,31. In contrast, the VIIRS NTL data has a higher spatial resolution, more sensitive low-light detection capabilities, and a broader dynamic range of NTL radiance values, with on-board radiance calibration and few blooming32,33. Hence, the synergistic integration of these two NTL datasets is critical to long-term applications.

The strategy of input patches for global NTL prediction and dataset generation. (a) over-lapped sub-images cropping, and (b) edge padding with 0 value during prediction.

Yu, B. et al. Urban built-up area extraction from Log-Transformed NPP-VIIRS nighttime light composite data. IEEE Geosci. Remote Sens. Lett. 15, 1279–1283, https://doi.org/10.1109/LGRS.2018.2830797 (2018).

Liu, X. et al. High-spatiotemporal-resolution mapping of global urban change from 1985 to 2015. Nat. Sustain. 3, 564–570, https://doi.org/10.1038/s41893-020-0521-x (2020).

Chen, X. Wang, Z. & Zhang, F. A history reconstructed time series (1992-2011) of annual global NPP-VIIRS-like nighttime light data through a super-resolution U-Net model. figshare https://doi.org/10.6084/m9.figshare.22262545.v8 (2024).

Liu, X. et al. High-resolution multi-temporal mapping of global urban land using Landsat images based on the Google Earth Engine Platform. Remote Sens. Environ. 209, 227–239, https://doi.org/10.1016/j.rse.2018.02.055 (2018).

The version 4 DMSP annual stable NTL time series is the most commonly applied DMSP NTL product, which is available from the Payne Institute for Public Policy at the Colorado School of Mines (https://eogdata.mines.edu/products/dmsp/#v4_dmsp_download). The stable DMSP NTL product has undergone a filtering process to eliminate transient light and background noise, while retaining stable light sources from continuous illumination sources. The spatial reference is WGS84, with a spatial resolution of 30 arc-seconds (approximately 1 km). The spatial extent covers a range from -180° to 180° longitude and from 65° S to 75°N latitude. We downloaded the stable DMSP NTL data from 1992 to 2013. The data from 2012 and 2013 were used to train the NTLSRU-Net calibration model, while the remaining data was employed to reconstruct the simulated VIIRS NTL data from 1992 to 2011.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Wu, K. & Wang, X. Aligning pixel values of DMSP and VIIRS nighttime light images to evaluate urban dynamic. Remote Sens. 11, 1463, https://doi.org/10.3390/rs11121463 (2019).

The scatter plots comparing the simulated VIIRS data with real VIIRS NTL data at multi scales (unit: \({\rm{nW}}{{\rm{cm}}}^{-2}{{\rm{sr}}}^{-1}\)). The solid line denotes the 1:1 line. The scatter plots for our data and real VIIRS NTL data in 2012 are shown in a1-a4, those for 2013 are shown in b1-b4, and the comparison between ChenVNL and real VIIRS NTL data in 2012 is shown in c1-c4.

Histogram comparison of the generated SVNL data (yellow outline) and real VIIRS NTL data (blue fill) for pixel values at global test spots by ranges: a1, a2 and a3 show sub-histograms for pixel value ranges of 0-10, 10-60, and above 60 in 2012, respectively; b1, b2 and b3 display the corresponding ranges for 2013, respectively.

Several modifications were made to the original U-Net network for the super-resolution task of DMSP NTL data. One modification involved removing certain network layers in order to preserve image details. To address the issue of fine-grained information loss in NTL super-resolution tasks, all pooling layers from the original U-Net were eliminated. Although pooling operations enhance learning efficiency of CNNs based on local correlation51, they tend to result in loss of fine-grained information in this specific task. Additionally, batch normalization layers52 following convolutional operations were removed from the expansive path. ReLU53 was employed as the activation function between convolutional blocks except for the final 1 × 1 convolution operation in the output layer. Another modification was about the expansive path. The up-sampling operations were conducted using transposed convolution network54, and features obtained from skip connections were compressed through a 1 × 1 convolution for feature channel reduction before being fed into two consecutive convolutional layers for feature fusion. This approach effectively reduced model parameters while enhancing computational efficiency. The last modification pertained to feature padding, where a zero-value pixel was appended to the periphery of the feature map after each 3 × 3 convolutional operation in order to preserve the input image size. Given that the background pixels dominate the NTL data and bright pixels constitute only a minor fraction, incorporating padding in convolutional operations has little effect on network learning.

AI LEDcontrollerMK II

The spatial dynamics of NTL intensity growth in SVNL data from 1992 to 2023 is shown in Fig. 10. The lit areas predominantly concentrated in regions such as the United States, China, Europe, Southeast Asia and the Middle East (Fig. 10a). As illustrated in Fig. 10b2, there is no significant change in NTL intensity during the 1992–2002 period, higher NTL intensity clustered in the region from 20 to 45°N, while lower values were observed in the Southern Hemisphere. The region between 10°N and 20°N experienced a significant increase in NTL intensity during the period of 2002–2012, whereas the rise occurred between 35°N and 45°N during the period of 2012–2022. Additionally, for the region spanning from 20°N to 35°N, there was a notable increase in NTL intensity after 2002. In the longitudinal direction (Fig. 10b1), one of the NTL intensity peaks within the Western Hemisphere region was mostly located in the United States (from 70° to 100° W). Three peaks in the Eastern Hemisphere region experienced substantial growth from 2002 to 2022, which are located within the ranges of 40° to 60°E, 70° to 80°E, and 100°–130°E. This result aligns with the global economic development trend. The spatio-temporal evolution of NTL intensity in the Beijing-Tianjin-Hebei region (BTH), GBA, and Los Angeles over the past 30 years as depicted in Fig. 10c1-c3, reveals a consistent temporal trend of continuous spatial expansion for high luminance pixels from urban cores to fringe areas for SVNL time series.

Elvidge, C. D., Baugh, K. E., Zhizhin, M. N. & Hsu, F. Why VIIRS data are superior to DMSP for mapping nighttime lights. Proceedings of The Asia-Pacific Advanced Network 35, 62–69, https://doi.org/10.7125/APAN.35.7 (2013).

Aqua illumination controllerfor sale

The random hierarchical sampling strategy was employed to gather over 20,000 test spots at a global scale. The histograms presented in Fig. 4 depict the distribution of real VIIRS data and predicted data from DMSP for the above test spots. As the dominance of the high-count 0–10 range obscured comparison with other ranges, the histograms for these test spots were divided into sub-histograms by ranges, as shown in Fig. 4. Overall, the histograms derived from the predicted data closely resemble those of real VIIRS data, particularly in midrange pixel values (3–60). However, there are also some differences. The most obvious difference occurs near values close to 0 due to dim light pixels (1–3) in VIIRS data being predominantly background value (0) in DMSP data, which was not captured by the CNN model. In the high range (above 60), the histogram distribution of predicted data is lower than that of real VIIRS data, primarily attributed to the saturation of DMSP data.

He, C., Liu, Z., Tian, J. & Ma, Q. Urban expansion dynamics and natural habitat loss in China: a multiscale landscape perspective. Glob. Change Biol. 20, 2886–2902, https://doi.org/10.1111/gcb.12553 (2014).

Nechaev, D. et al. Cross-sensor nighttime lights image calibration for DMSP/OLS and SNPP/VIIRS with residual U-net. Remote Sens. 13, 5026, https://doi.org/10.3390/rs13245026 (2021).

Ultimately, a global annual time series of simulated VIIRS NTL data spanning from 1992 to 2023 was generated by integrating simulated data from 1992 to 2011 with real VIIRS NTL data from 2012 to 2023.

The global annual time series of simulated VIIRS NTL data from 1992 to 2023 consists of the simulated VIIRS data (1992–2011) derived from DMSP NTL data and the annual VIIRS NTL composition (2012–2023). The data is in the WGS84 coordinate system with a spatial resolution of 15 arc-seconds (~500 m), and is stored in independent TIF format, each representing the NTL data for a specific year, named SRUNet_NPP-VIIRS_V2_Like_Year.tif. The data is freely accessible at https://doi.org/10.6084/m9.figshare.22262545.v855.

Elvidge, C. et al. A fifteen year record of global natural gas flaring derived from satellite data. Energies 2, 595–622, https://doi.org/10.3390/en20300595 (2009).

Dong, C., Loy, C. C., He, K. & Tang, X. Image super-resolution using deep convolutional networks. IEEE Trans. Pattern Anal. Mach. Intell. 38, 295–307, https://doi.org/10.1109/TPAMI.2015.2439281 (2014).

Zheng, Q., Weng, Q. & Wang, K. Developing a new cross-sensor calibration model for DMSP-OLS and Suomi-NPP VIIRS night-light imageries. ISPRS-J. Photogramm. Remote Sens. 153, 36–47, https://doi.org/10.1016/j.isprsjprs.2019.04.019 (2019).

Shi, K. et al. Detecting spatiotemporal dynamics of global electric power consumption using DMSP-OLS nighttime stable light data. Appl. Energy 184, 450–463, https://doi.org/10.1016/j.apenergy.2016.10.032 (2016).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Department of Regional and Urban Planning, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, 310058, China

Ioffe, S. & Szegedy, C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. in Proceedings of Machine Learning Research: 32nd International Conference on Machine Learning, Lille, France, July 7-9, 2015, Proceedings (eds. Bach, F. & Blei, D.), 37, 448-456, https://dl.acm.org/doi/10.5555/3045118.3045167 (ACM, 2015).

Cao, X. et al. A simple self-adjusting model for correcting the blooming effects in DMSP-OLS nighttime light images. Remote Sens. Environ. 224, 401–411, https://doi.org/10.1016/j.rse.2019.02.019 (2019).

Zhang, L. et al. A prolonged artificial nighttime-light dataset of china (1984-2020). Sci. Data 11, 414, https://doi.org/10.1038/s41597-024-03223-1 (2024).

During the prediction process, sub-images with a spatial range of 2° × 2° (\({f}_{{sub}}=480\) pixels) were cropped in the WGS84 geographic coordinate system, with an overlapping width of 0.3° (\({f}_{{over}}\) = 72 pixels) between adjacent sub-images. The output images of each sub-image only retained the central part at 380 × 380 pixels (\({f}_{{clip}}\) = 50 pixels). The “Mosaic To New Raster” tool in ArcGIS 10.2 was used to merge the sub-images with only the central part retained to obtain a global super-resolution NTL image similar to VIIRS.

Aqua illumination controllersetup

Li, X., Li, D., Xu, H. & Wu, C. Intercalibration between DMSP/OLS and VIIRS night-time light images to evaluate city light dynamics of Syria’s major human settlement during Syrian Civil War. Int. J. Remote Sens. 38, 5934–5951, https://doi.org/10.1080/01431161.2017.1331476 (2017).

The global total NTL intensity of SVNL,real VIIRS, raw DMSP, ChenVNL, and LiDNL time series are depicted in Fig. 9. It is evident that the raw DMSP time series exhibit temporal inconsistencies with multiple trajectories among different satellites, and there are significant temporal discontinuities between DMSP and VIIRS data. The total NTL intensity time series of SVNL, ChenVNL, and LiDNL from 1992 to 2012 exhibit similar growth trends and fluctuation characteristics. The LiDNL time series exhibits a noticeable jump during the overlapping period of DMSP and VIIRS data, while the SVNL and ChenVNL time series keep a high temporal consistency.

Jiang, W. et al. Assessing light pollution in China based on nighttime light imagery. Remote Sens. 9, 9020135, https://doi.org/10.3390/rs9020135 (2017).

Gholamalinezhad, H. & Khosravi, H. Pooling methods in deep neural networks, a review. Preprint at https://arxiv.org/abs/2009.07485 (2020).

Zhao, M. et al. Mapping urban dynamics (1992–2018) in Southeast Asia using consistent nighttime light data from DMSP and VIIRS. Remote Sens. Environ. 248, 111980, https://doi.org/10.1016/j.rse.2020.111980 (2020).

Ge, W., Yang, H., Zhu, X., Ma, M. & Yang, Y. Ghost city extraction and rate estimation in china based on NPP-VIIRS night-time light data. ISPRS Int. J. Geo-Inf. 7, 219, https://doi.org/10.3390/ijgi7060219 (2018).

In this study, the reconstruction of the global annual VIIRS NTL data in the early period of 1992–2011 using DMSP NTL data was attempted for the first time, and subsequently a complete and consistent global annual simulated VIIRS NTL dataset (SVNL) spanning from 1992 to 2023 at a resolution of 500 m was generated by integrating it with real VIIRS NTL data. The constructed SVNL dataset provides a novel and valuable resource for long-term studies related to human activities, filling a crucial gap in existing NTL datasets with its full temporal coverage. This facilitates more precise and extensive investigations into urban development, socioeconomic dynamics, and environmental impact assessments, especially in regions with limited official statistical data.

Nighttime light (NTL) data can capture various sources of low-light emissions from earth at night, with the majority of stable low-light emissions originating from artificial lighting in urban areas1,2,3. It has been recognized as a reliable proxy for measuring the scope and intensity of human activity4,5,6, and is increasingly being used in studies such as urbanization monitoring7,8,9,10, socioeconomic estimation11,12,13, and ecological environment assessment14,15. Time series analysis of NTL data is particularly valuable for detecting, estimating, and monitoring dynamics in human activities, such as urban expansion mapping16,17,18, carbon emissions19,20,21, and economic development22,23, especially in regions lacking reliable official statistics. However, the potential of NTL time series has been constrained by limited temporal coverage and substantial discrepancies in existing NTL datasets24,25.

Chen, Z. et al. An extended time-series (2000-2023) of global NPP-VIIRS-like nighttime light data. harvard dataverse https://doi.org/10.7910/DVN/YGIVCD (2020).

Dou, Y., Liu, Z., He, C. & Yue, H. Urban land extraction using VIIRS nighttime light data: An evaluation of three popular methods. Remote Sens. 9, 175, https://doi.org/10.3390/rs9020175 (2017).

Imhoff, M. L., Lawrence, W. T., Stutzer, D. C. & Elvidge, C. D. A technique for using composite DMSP/OLS “City Lights” satellite data to map urban area. Remote Sens. Environ. 61, 361–370, https://doi.org/10.1016/S0034-4257(97)00046-1 (1997).

Elvidge, C. D., Baugh, K. E., Kihn, E. A., Kroehl, H. W. & Davis, E. R. Mapping city lights with nighttime data from the DMSP Operational Linescan System. Photogramm. Eng. Remote Sens. 63, 727–734, https://doi.org/10.1016/S0924-2716(97)00008-7 (1997).

Zheng, Q., Zheng, Q., Weng, Q. & Wang, K. Correcting the pixel blooming effect (PiBE) of DMSP-OLS nighttime light imagery. Remote Sens. Environ. 240, 111707, https://doi.org/10.1016/j.rse.2020.111707 (2020).

Zhao, M. et al. Building a series of consistent night-time light data (1992-2018) in Southeast Asia by integrating DMSP-OLS and NPP-VIIRS. IEEE Trans. Geosci. Remote Sensing 58, 1843–1856, https://doi.org/10.1109/TGRS.2019.2949797 (2020).

Zhang, Q. & Seto, K. C. Mapping urbanization dynamics at regional and global scales using multi-temporal DMSP/OLS nighttime light data. Remote Sens. Environ. 115, 2320–2329, https://doi.org/10.1016/j.rse.2011.04.032 (2011).

Ronneberger, O., Fischer, P. & Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. in Lecture Notes in Computer Science: 18th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), Munich, Germany, October 5-9, 2015, Proceedings (eds. Navab, N., Hornegger, J., Wells, WM. & Frangi, AF.). 9351, 234-241, https://doi.org/10.1007/978-3-319-24574-4_28 (Springer, 2015).

Zhao, N., Samson, E. L. & Currit, N. Nighttime-lights-derived fossil fuel carbon dioxide emission maps and their limitations. Photogramm. Eng. Remote Sens. 81, 935–943, https://doi.org/10.14358/PERS.81.12.935 (2015).

Abrahams, A. S., Oram, C. & Lozano‐Gracia, N. Deblurring DMSP nighttime lights: A new method using Gaussian filters and frequencies of illumination. Remote Sens. Environ. 210, 242–258, https://doi.org/10.1016/j.rse.2018.03.018 (2018).

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

The correlation analysis was performed to assess the reliability of our simulated data for previous years (before 2012) in comparison with DMSP and ChenVNL data, as illustrated in Fig. 6. Given the differences in units and magnitudes between DMSP and VIIRS-like data, we analyzed the data using the year 2005 as an example. The correlation coefficient between the simulated data and DMSP data for total NTL intensity at the national level in 2005 was 0.973, with an R2 value of 0.947 for linear fitting (Fig. 6a1). Additionally, we calculated the correlation between ChenVNL and DMSP as a reference, yielding a correlation coefficient of 0.968 and an R² value of 0.937 (Fig. 6a2). Furthermore, we conducted annual correlation analyses and y = x fitting between the simulated data and ChenVNL from 2000 to 2012, since they were all simulated VIIRS datasets (Fig. 6b). The annual correlation coefficient exceeded 0.99, and the R2 for the y = x linear fit consistently surpassing 0.9. These observations indicated the relative reliability of the simulated data in earlier years.

Chen, Z. et al. The potential of nighttime light remote sensing data to evaluate the development of digital economy: a case study of China at the city level. Comput. Environ. Urban Syst. 92, 101749, https://doi.org/10.1016/j.compenvurbsys.2021.101749 (2022).

Elvidge, C. D., Baugh, K., Zhizhin, M., Hsu, F. C. & Ghosh, T. VIIRS night-time lights. Int. J. Remote Sens. 38, 5860–5879, https://doi.org/10.1080/01431161.2017.1342050 (2017).

AI Hydra 52

Correlation analysis of SVNL data and Gross Domestic Product (GDP) time series: (a) SVNL, (b) ChenVNL as baseline for comparison. (a1) and (b1) depict scatter plots of NTL and GDP time series at global level, while (a2) and (b2) show the spatial distribution of R2 for NTL and GDP time series at national level.

Glorot, X., Bordes, A. & Bengio, Y. Deep sparse rectifier neural networks. Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics 15, 315-323 (PMLR, 2011).

Shi, K. et al. Modeling spatiotemporal CO2 (carbon dioxide) emission dynamics in China from DMSP-OLS nighttime stable light data using panel data analysis. Appl. Energy 168, 523–533, https://doi.org/10.1016/j.apenergy.2015.11.055 (2016).

Dumoulin, V. & Visin, F. A guide to convolution arithmetic for deep learning. Preprint at https://arxiv.org/abs/1603.07285 (2016).

The NTL images from 2012 and 2013 were utilized for training the NTLSRU-Net model. To build the sample set, firstly, we employed the nearest neighbor algorithm to upsample DMSP NTL images to a resolution of 15 arc seconds and align their pixels with VIIRS NTL images. Then, the pairs of image patches for DMSP, NDVI, and VIIRS data were cropped and selected from the global images using a designed cropping strategy. The strategy aimed to increase urban image patches and decrease the proportion of background pixels in the sample set. Clipping centers were randomly selected from regions with DN values exceeding 55 to obtain urban image patches. These patches had a wide DN dynamic range and abundant contextual information, making a more conducive for learning the mapping function. The sample set comprised 12049 pairs of image patches, with 80% randomly selected for the training set. The remaining samples were divided into test and validation sets at a ratio of 2:1. The loss function adopted in this study was the mean square error function and was then optimized by the Adam algorithm proposed by Kingma and Ba in each deep learning step. The loss function can be formed as Eq. (3).

The annual Landsat NDVI data from 1992 to 2013 was derived from the the available Landsat 4 ETM (LANDSAT/LT04/C02/T1_L2), Landsat 5 ETM (LANDSAT/LT05/C02/T1_L2), Landsat 7 ETM+ (LANDSAT/LE07/C02/T1_L2), and Landsat 8 OLI level-2 surface reflectance (LANDSAT/LC08/C02/T1_L2) products archived on the Google Earth Engine (GEE) platform (https://code.earthengine.google.com). The scripts for preprocessing Landsat time series data in GEE was provided by Liu et al.45 (https://code.earthengine.google.com/1c901129fa8c9d81b292824e8fb4ff1c

Li, X. et al. Potential of NPP-VIIRS nighttime light imagery for modeling the regional economy of China. Remote Sens. 5, 3057–3081, https://doi.org/10.3390/rs5063057 (2013).

Spatial dynamics of NTL intensity growth in SVNL data from 1992 to 2023. (a) global spatial distribution of SVNL data in 2012, (b) the line plots of total NTL intensity for SVNL data by longitude (b1) and latitude (b2) in 1°bins, and (c) enlarged subplots of NTL intensity dynamics from 1992 to 2023 interval 5 years in BJH (c1), GBA (c2), Los Angeles (c3).

Aqua illuminationDirector for sale

Peng, J. et al. Spatiotemporal evolution of urban agglomerations in China during 2000–2012: a nighttime light approach. Landsc. Ecol. 35, 421–434, https://doi.org/10.1007/s10980-019-00956-y (2020).

For the cross-calibration of NTL images from DMSP and VIIRS, we utilize an convolution neural network with U-Net architecture and super-resolution technique. Let \({{\rm{X}}}^{{\rm{DMSP}}}\), \({{\rm{X}}}^{{\rm{NDVI}}}\), \({{\rm{Y}}}^{{\rm{VIIRS}}}\), and \({{\rm{Y}}}^{{\rm{SR}}}\) be the DMSP NTL data, Landsat NDVI data, VIIRS NTL data, and simulated VIIRS NTL data, respectively. Our goal is to learn a mapping function \({\rm{F}}\left(\bullet \right)\) that can forecast the corresponding VIIRS-like NTL data via taking full use of the inputted pairs of DMSP NTL and Landsat NDVI data, as expressed as Eq. (1) and Eq. (2).

Du, X., Shen, L., Wong, S. W., Meng, C. & Yang, Z. Night-time light data based decoupling relationship analysis between economic growth and carbon emission in 289 Chinese cities. Sust. Cities Soc. 73, 103119, https://doi.org/10.1016/j.scs.2021.103119 (2021).

We plotted the NTL intensity scatters between our simulated VIIRS data and real VIIRS data at multi scales, including pixel level, city level, province level and national level, as shown in Fig. 5. Three metrics, R Squared (R2), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE), were employed to assess the accuracy of the simulated NTL intensity. The scatter plot analysis utilized NTL intensity of the above test spots at pixel level and total NTL radiance intensity at regional levels. The R2 for the simulated data and real VIIRS data in 2012 and 2013 were 0.617 and 0.564 at the pixel level, respectively. It is evident that the scatter dot distributions of the simulated data and real VIIRS data are predominantly concentrated around the y = x line and clustered within the low to mid range of NTL intensity. Furthermore, in 2012, the R2 for the simulated data and real VIIRS data were found to be 0.747 at city level, 0.874 at province level, and 0.964 at national level; while in 2013 these values were observed to be 0.647, 0.841,and 0.943 respectively. The estimation accuracy demonstrates an increasing trend with larger scales. For comparison, we also incorporated ChenVNL data, the only existing artificial VIIRS-like NTL dataset, into the scatter plot analysis as depicted in Fig. 5c1-c4. The R2 for ChenVNL data in 2012 were 0.512, 0.694, 0.860, and 0.966 at pixel level, city level, province level, and national level, respectively. The estimation accuracy of our data was closely to that of ChenVNL. These results suggested that our data exhibits good estimated accuracy across multiple scales in comparison to real VIIRS data.

Consequently, a complete time series of global VIIRS-like data since 1992 has not been available to date. To address this issue, we proposed a Nighttime Light U-Net super-resolution network (NTLSRU-Net) for the cross-sensor calibration between DMSP NTL data and VIIRS NTL data. The NTLSRU-Net calibration model leveraged the U-Net CNN and employed image super-resolution (SR) deep learning techniques to transform DMSP NTL data into VIIRS NTL data. The U-Net architecture has proven effective in recovering spatial details and accurately predicting pixel positions from blurred images42, showing promise in modeling the spatial dependency of NTL data. SR is a technique for reconstructing a high resolution image from a low resolution image43, and the process of transforming DMSP NTL data into VIIRS NTL data follows a similar approach, from low-resolution to high-resolution images. Additionally, Landsat NDVI was incorporated into the calibration process, which was commonly used to reduce the saturation and blooming in DMSP NTL data17,44.

As illustrated in Fig. 2, the proposed NTLSRU-Net was considered as our target mapping function \({\rm{F}}\left(\bullet \right)\), with a contracting path and an expansive path for feature down-sampling and up-sampling. The network consists of 23 convolutional layers and 4 transposed convolutional layers, with 3 × 3 kernels with a stride of 1. The receptive field of each pixel in the final super-resolution output is 37.

Sign up for the Nature Briefing: Anthropocene newsletter — what matters in anthropocene research, free to your inbox weekly.

Li, X. & Zhou, Y. A stepwise calibration of global DMSP/OLS stable nighttime light data (1992–2013). Remote Sens. 9, 637, https://doi.org/10.3390/rs9060637 (2017).

The visual comparison result is illustrated in Fig. 7, showing that the generated SVNL data exhibits a spatial pattern similar to real VIIRS data. It is shown that SVNL effectively addresses the challenges posed by saturation and blooming effects of DMSP data, exhibiting superior efficacy compared to DMSP data in capturing spatial changes and distribution of NTL intensity in urban areas. The spatial pattern of ChenVNL data also agree well with real VIIRS data. However, both SVNL and ChenVNL data have their limitations. ChenVNL tends to overestimate NTL intensity in urban core areas and underestimate it in peri-urban areas (e.g., Beijing, Shanghai, GBA), while SVNL tends to underestimate NTL intensity in urban core areas (e.g., London, New York, Los Angeles). Both SVNL and ChenVNL perform well in Cape Town.

In this study, a four-step framework was proposed for generation of the long-term global simulated VIIRS NTL dataset, as illustrated in Fig. 1. Firstly, the raw stable DMSP NTL data underwent preprocessing to rectify interannual inconsistency. Secondly, a U-Net super-resolution network was employed to model the relationship between the calibrated DMSP and VIIRS NTL data, with the assistance of Landsat NDVI. Thirdly, the derived relationship at the global scale was applied to obtain the simulated VIIRS data from DMSP for the period 1992–2013. Subsequently, a long time series NTL dataset was generated by integrating the simulated VIIRS NTL data (1992–2011) and real VIIRS NTL data (2012–2023). Finally, the comprehensive accuracy and performance validation were conducted on the generated NTL dataset.

Nighttime light (NTL) data is recognized as a reliable proxy for measuring the scope and intensity of human activity, finding wide application in studies such as urbanization monitoring, socioeconomic estimation, and ecological environment assessment. However, the substantial discrepancies and limited temporal coverage of existing NTL datasets have constrained their potential for long-term research applications. To address this, a Nighttime Light U-Net super-resolution network is proposed for the cross-sensor calibration between the Defense Meteorological Satellite Program-Operational Linescan System (DMSP-OLS) NTL data and the Suomi National Polar-Orbiting Partnership-Visible Infrared Imaging Radiometer Suite (NPP-VIIRS) NTL data. This network is applied to generate a continuous and consistent 500-meter global annual simulated VIIRS NTL dataset (SVNL) from 1992 to 2023. Validation results indicate a high confidence in the quality of the SVNL data, demonstrating its superiority in capturing longer NTL dynamics, maintaining higher temporal consistency, and presenting greater spatial detail compared with other NTL datasets. The SVNL could be utilized for prolonged human activities monitoring, and further research on regional or global urbanization.

Chen, X., Wang, Z., Zhang, F. et al. A global annual simulated VIIRS nighttime light dataset from 1992 to 2023. Sci Data 11, 1380 (2024). https://doi.org/10.1038/s41597-024-04228-6