Snow variability in southernmost Patagonia by a combination of spectral fusion and mixture analysis with machine learning approach using the Google Engine platform
Abstract
Seasonal snow coverage is a fundamental component of both the global energy budget and the water cycle, and its properties are particularly affected by climate change. Several methods based on satellite data products are available to estimate these properties each one with its pro and cons. This work presents an estimation of the variability of the snow cover extension and its albedo signature using a remote sensing and machine learning approach, using MODIS MOD09 data source at 250 and 500 meters of resolution and climate information over the Google Engine platform. Four main steps are included: (1) increase the spatial resolution with a spectral fusion technique at 250 meters (for all bands used); (2) reconstruction of snow extent variability and albedo assessment with subpixel detection using the endmembers extraction and spectral unmixing (linear and nonlinear mixture) for the period 2000-2018; (3) statistical downscaling of ERA5 Reanalysis data to improve meteorological representation including topographic effects; and (4) correlating the snow properties behaviour with the local climate variability. All these processes were applied over 3 sites with different characteristics: Brunswick Peninsula, Laguna Blanca and Sierra Prat, all located in southernmost Patagonia. The validation process using independent data was performed in 2 steps: (1) the Normalized Difference Snow Index (NDSI) algorithm with Sentinel-2A MSI data was used to evaluate the snow extent reconstruction; (2) record of AWS data, located near the sites, were used to evaluate statistical downscaling. The results show no significant trend between the inter-annual variability of snow properties for the period 2000-2018. However, a shortening of the snow season and warmer winters could be identified, generating changes in the precipitation fraction (solid to liquid). The climate data were used to model the snow extent variability until 1979 found a significant decreasing trend of snow extent. We conclude that this is an adequate approach for estimating the snow properties and climate behaviour in southernmost Patagonia at a high spatial and temporal resolution for the last few decades.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.C24A..04A
- Keywords:
-
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICS;
- 0758 Remote sensing;
- CRYOSPHERE;
- 0794 Instruments and techniques;
- CRYOSPHERE