Preprocessing Methods¶

Bases: PaperImplementation

A specialized filter for VNP46A2 NTL data focusing on primary quality indicator.

Unlike broader screening implementations, this class relies strictly on the instrument's internal quality assessments. It isolates pixels that are labeled as 'High Quality' by the mandatory sensor checks.

Bases: PaperImplementation

Applies a spatial buffer for removing cloud and snow edge pixels.

Implements the cloud and snow masking process for daily NASA's Black Marble NTL data as described in:

Li, T., Zhu, Z., Wang, Z., Román, M. O., Kalb, V., & Zhao, Y. (2022). Continuous monitoring of nighttime light changes based on daily NASA's Black Marble product suite. Remote Sensing of Environment, 282, 113269. https://doi.org/10.1016/j.rse.2022.113269

This class filters out observations affected by clouds, cirrus, and snow/ice surfaces, and dilates the resulting mask to ensure edge pixels are removed.

Bases: PaperImplementation

Filters out low nighttime light radiance values.

Implements a low radiance background filter as described in:

Li, T., Zhu, Z., Wang, Z., Román, M. O., Kalb, V. L., & Zhao, Y. (2022). Continuous monitoring of nighttime light changes based on daily NASA's Black Marble product suite. Remote Sensing of Environment, 282, 113269. https://doi.org/10.1016/j.rse.2022.113269

Bases: PaperImplementation

Implements the four-step high-quality data retrieval process to filter daily Black Marble NTL radiance as described in:

Jia, M., Li, X., Gong, Y., Belabbes, S., & Dell'Oro, L. (2023). Estimating natural disaster loss using improved daily night-time light data. International Journal of Applied Earth Observation and Geoinformation, 120, 103359. https://doi.org/10.1016/j.jag.2023.103359

This class filters out observations affected by solar stray light, clouds, abnormal sensor values, and high lunar illumination.

Bases: PaperImplementation

Implements a rolling modified z-score outlier removal as described in:

Zheng, Q., Jiang, W., Wang, W., Lei, X., & Li, Z. (2021). Daily consistent NOAA-20 VIIRS nighttime light data in China. International Journal of Remote Sensing, 42(22), 8538-8561. https://doi.org/10.1080/01431161.2021.1969057

This method utilizes a rolling window to calculate the median and the Median Absolute Deviation to identify temporal spikes (outliers).

By using a 30-day backward-looking window, the filter remains compatible with walk-forward processing while remaining robust to outliers that typically bias a standard mean-based z-score.

Bases: PaperImplementation

Implements a 2D spatial moving average (mean pooling) to reduce noise and enhance the signal-to-noise ratio in daily Nighttime Light (NTL) radiance.

This spatial filtering technique is detailed in the Black Marble product documentation:

Román, M. O., Wang, Z., Sun, Q., Kalb, V., Miller, S. D., Molthan, A., ... & Enenkel, S. (2018). NASA's Black Marble nighttime lights product suite. Remote Sensing of Environment, 210, 113-143. https://doi.org/10.1016/j.rse.2018.03.017

This class applies a rolling mean across the spatial dimensions (x, y) of the dataset. This is often used to mitigate high-frequency noise, such as atmospheric artifacts or sub-pixel flickering, while preserving the underlying socioeconomic light signals.

CAUTION: While spatial pooling improves stability, it acts as a low-pass filter that reduces spatial resolution. Larger filter sizes will result in 'blurring' of fine-scale urban structures.

Bases: PaperImplementation

Implements the A-Averaging filter to reduce high temporal variation and blooming effects in daily NTL data as described in:

Hu, Y., Zhou, X., Yamazaki, D., & Chen, J. (2024). A self-adjusting method to generate daily consistent nighttime light data for the detection of short-term rapid human activities. Remote Sensing of Environment, 304, 114077. https://doi.org/10.1016/j.rse.2024.114077

The A-Averaging method decomposes NTL radiance into 'fixed' (stable) and 'mismatch' (variable) components. By applying spatial averaging only to the mismatch component, it reduces blooming and temporal noise while preserving the sharp gradients of urban light structures.

CAUTION: The identification of 'Fixed Light' requires a sufficient temporal baseline (ideally one year) to calculate meaningful quantiles.

Bases: PaperImplementation

Implementation of the angular correction method for daily NTL data:

Hu, Y., Zhou, X., Yamazaki, D., & Chen, J. (2024). A self-adjusting method to generate daily consistent nighttime light data for the detection of short-term rapid human activities. Remote Sensing of Environment, 304, 114077. https://doi.org/10.1016/j.rse.2024.114077

Bases: PaperImplementation

Implements a View Zenith Angle (VZA) correction for Nighttime Light radiance following the quadratic regression normalization approach detailed in:

Zheng, Q., Zeng, Y., Zhou, Y., Wang, Z., Mu, T., & Weng, Q. (2025). Nighttime lights reveal substantial spatial heterogeneity and inequality in post-hurricane recovery. Remote Sensing of Environment, 319, 114645. https://doi.org/10.1016/j.rse.2025.114645

This class normalizes pixel-wise radiance to a nadir observation (VZA = 0°) by calculating a multiplicative correction factor 'f' derived from the ratio between predicted nadir radiance and the predicted radiance at the observed angle.

Bases: PaperImplementation

Simple linear interpolation imputation baseline.

This class uses linear interpolation along the time dimension to fill missing values in the data.

Bases: PaperImplementation

Implements the Disturbance Factor Correction (DFC) model as described in:

Yue, H., et al. (2026). Improved Daily Nighttime Light Data as High-Frequency Economic Indicator. Applied Sciences, 16(2), 947. https://doi.org/10.3390/app16020947

It uses a Local Ordinary Least Squares (OLS) regression to decompose daily NTL radiance into three components: physical disturbances, socioeconomic signals (trends/seasonality), and random noise.

The model accounts for: - Lunar Illumination (LI) - Sensor Geometry: Viewing Zenith Angle (VZA) and Viewing Azimuth Angle (VAA) - Anisotropic Interaction: VZA x VAA - Cloud-contamination artifacts - Long-term Day Trend and Seasonal Dummies (Spring, Summer, Autumn, Winter)

LOGIC: The regression acts as a 'control' framework. While all factors are regressed simultaneously to prevent omitted variable bias, only the physical disturbances (Moon, Geometry, Clouds) are subtracted from the raw signal if they are statistically significant (p < 0.05). The socioeconomic signals (trend and seasonality) are preserved in the corrected output.