The raw data of this dataset comes from various remote sensing images and observation data, including Landsat OLI, Sentinel-1 GRD, Sentinel-2 MSI images, as well as high-precision images obtained by drones. The spatial resolution of the 8th band data of Landsat OLI images is 15 meters, mainly obtained from the open platform of the United States Geological Survey (USGS). The selected images have undergone systematic geometric correction (C2L1 product) to ensure their suitability for direct glacier movement velocity calculation. The time selection of images is mainly based on the image pairs at the beginning and end of each month, minimizing cloud and fog interference. If necessary, the time interval can be extended to obtain high-quality image pairs.

The Sentinel-1 GRD image is a radar image, using IW mode ground distance detection data (GRD) with a resolution of 5 × 20 meters, mainly downloaded from the European Space Agency's Copernicus data center. These images overcome the common problem of cloud and fog interference in optical remote sensing in high-altitude areas, providing important support for the construction of time series of glacier surface movement velocity. Due to the revisit period of Sentinel-1 satellite being 12 days, this study generated monthly scale image pairs through time baseline setting (24 or 36 days) to ensure the temporal and complete nature of the data.

The eighth band of Sentinel-2 MSI image is the near-infrared band with a spatial resolution of 10 meters, obtained from the European Space Agency's Copernicus program. The criteria for selecting images are similar to Landsat, focusing on low cloud cover images per month to maximize data quality. The drone images were obtained using the DJI M300 RTK drone equipped with the Ruibo M6 Pros measurement camera, covering some of the glacier end areas in the study area, with a navigation area of approximately 30 square kilometers. Drone images are used to calculate fusion weights and verify the accuracy of fusion results, with extremely high spatial resolution and geometric precision.

At the same time, the study also referred to the glacier distribution dataset in the Gangrigab area and the RGI 7.0 global glacier catalog as a reference for the study area, providing support for regional screening of image data and interpretation of results.

This dataset adopts multi-source data fusion technology to generate high-precision and high spatiotemporal resolution glacier surface velocity data. Data processing consists of five main steps: image preprocessing, motion speed calculation, fusion weight calculation, data fusion, and accuracy verification.

Image preprocessing includes geometric correction, image registration, and noise removal. For Landsat and Sentinel-2 images, COSI-CORR software was used to extract glacier movement velocity, and frequency domain cross-correlation algorithm was adopted. The search window was set to 32 × 32 pixels, with step sizes of 2 pixels and 3 pixels, respectively, and a correlation coefficient threshold of 0.95. For Sentinel-1 images, the offset tracking function in SNAP software is used to set the matching window to 128 × 128 pixels and generate a velocity map with a spatial resolution of 30 meters. The motion speed extraction of drone images is performed using the ImGRAFT toolbox, and the results are obtained through image registration and feature tracking algorithms, and resampled to a resolution of 30 meters.

The fusion weight calculation is based on weighted least squares method. Select Landsat, Sentinel-1, and Sentinel-2 image velocity data from the same period, fit the drone image velocity results, and calculate the weight coefficients (WL8, WS1, WS2) for each data source. Then, perform weighted average fusion on the velocity data of all images. To address the issue of pixel null values, a mask function is introduced to adjust weight allocation and ensure the stability of the fusion process.

After the data fusion is completed, the residual motion velocity of the ice free area is used as the basis for error estimation, and the uncertainty of the data is quantified by calculating its mean and standard deviation. In addition, the accuracy of the fused data is evaluated by subtracting the motion speed results calculated from the drone images, ultimately verifying the reliability of the dataset.

Data quality assessment runs through the entire production process of this dataset to ensure its reliability and applicability. The error of fusion data is estimated by the residual motion velocity of the ice free area. The results indicate that the residual velocity error in the ice free area is 0.0098 m/d, and the error values are normally distributed. This error is significantly lower than the actual movement speed of glaciers, indicating that the dataset has high accuracy.

In order to further verify the data quality, this study will compare and analyze the fusion results with the glacier surface motion velocity calculated from unmanned aerial vehicle images. The mean and standard deviation of the speed difference between the fusion result and the drone image show that the deviation of the fusion data is significantly reduced, and the standard deviation tends to stabilize, indicating that the fusion process effectively reduces noise. Compared with a single data source, the fused data performs well in smoothing and outlier control.

In addition, this dataset adopts strict quality control processes, including geometric accuracy correction of images, optimization of velocity extraction algorithms, and weight calculation methods for multi-source data, ensuring the consistency and reliability of the data. The fusion data not only successfully captured the spatial differences in glacier surface movement in the Gangrigab area, but also provided a continuous time series, providing a solid data foundation for subsequent research. Overall, this dataset has achieved a high level of accuracy, reliability, and practicality.