Supplementary data for: Assessing spaceborne snow reflectance uncertainties using UAV hyperspectral measurements

This dataset contains the supplementary data supporting the article "Assessing spaceborne snow reflectance uncertainties using UAV hyperspectral measurements" (Fonseca-Gallardo et al.). It includes the processed data behind the figures and tables, the CNN classification outputs, and the raster products used to compare UAV hyperspectral reflectance with Landsat 8/9 OLI/OLI-2 Level-2 surface reflectance. UAV hyperspectral data were collected with Resonon Pika L (400–1000 nm) and Pika IR-L (925–1700 nm) imagers at three snow-covered sites: CARC, Montana, USA (prairie; 2024-02-16; Landsat 8), Onion Park, Montana, USA (alpine; 2024-04-20; Landsat 8), and Sodankylä, Finland (taiga; 2024-04-03; Landsat 9). The data cubes were georectified at 30 x 30 cm resolution. The CARC and Onion Park products are in UTM zone 12N, and the Sodankylä products are in UTM zone 35N. Reflectance values are unitless (0–1). CONTENTS Section 3.1 Reflectance characterization of surface types - Sec3.1_<site>_<sensor>_HSI.csv: per-band statistics of hyperspectral reflectance across the full sensor range, for each data cube and CNN surface class (Snow, Vegetation, Shadow). Columns: mean, std, min, max, Q25, Q50, Q75. - Sec3.1_<site>_RSR_LANDSAT<8|9>.csv: the same statistics for the hyperspectral-derived OLI/OLI-2 bands (B1–B6). These were computed with the Landsat relative spectral response (RSR) and with simple band averaging (AVG).- Sec3.1_Classified_rasters.zip: CNN-classified rasters (GeoTIFF) for each data cube, plus the matching classified polygons (GeoPackage), organized by site and sensor. - Sec3.1_CNN_performance.zip: for each site and sensor, the trained CNN model (cnn_model.h5, Keras), the PCA model used for spectral reduction (pca_model.pkl), the confusion matrix (PNG), and precision/F1 metrics (CSV). Section 3.2 Comparison with Landsat reflectance - Sec3.2_<site>_<sensor>_reflectance_difference_LANDSAT<8|9>.csv: per-Landsat-cell comparison (by grid ID and band). Each row holds the hyperspectral statistics, the Landsat SR value, the difference (Landsat minus hyperspectral), RMSE, propagated error, the pixel type (snow / mixed / none), and the simulation type (RSR / AVG). - Sec3.2_<site>_<sensor>_Reflectance_Summary.csv: site-level mean differences and propagated uncertainties by band, pixel type and simulation type. - Sec3.2_Multispectral_difference_rasters.zip: GeoTIFF rasters of the per-pixel reflectance difference within each Landsat cell, by site, sensor, band and simulation type (RSR / AVG). DATA AVAILABILITY The raw and georectified hyperspectral data cubes (Pika L and Pika IR-L) are not included here. They are available from the authors on reasonable request. Requests should be addressed to the corresponding author ([e-mail]). RELATED RESOURCES The code used to process and analyze these data is archived on Zenodo at https://doi.org/10.5281/zenodo.23025601. The development version is maintained at https://github.com/duiliofg/Uncertainties-in-modeled-snow-reflectance