The Snow Water Equivalent Synthetic Aperture Radar and Radiometer (SWESARR) is a dual microwave instrument meant to fill in information gaps in the remote sensing data of Snow Water Equivalent (SWE). The aim of this work is to improve and validate SWESARR measurements of SWE for areas with tree canopy using unsupervised machine learning methods. This information is critical to NASA’s SnowEx mission for understanding the spatial and temporal variability of snow. SWE is an integral part of the climate system and affects many other climate-related processes, thus an accurate understanding of how SWE is changing with climate change is crucial for future water resource management. We have made use of a suite of parameters to help in identifying features most important in predicting SWE in areas that have missing satellite data due to vegetation. Our aim is to validate and improve SWESARR measurements using unsupervised machine learning clustering algorithms with the goal of being able to better quantify spatial and temporal changes of SWE due to climate change.Other relevant datasets that have been central to identifying parameters for predicting SWE and data validation have been from the ASO (Airborne Snow Observatory), NSIDC (National Snow and Ice Data Center), and ground snow pit observations taken by the SnowEx field work team. This work is a result of NASA’s SnowEx team based out of Goddard Space Flight Center within the Climate Change Research Initiative at NASA GISS.

The SWESARR instrument produces unreliable results when flying over terrain populated by foliage. As a result, our team used a modified version of the LOF algorithm to detect subtle and extreme disturbances in the SWESARR data. We were only able to work with the VV polarization of the X band and the VH polarization of the Ku band as those were the sets of data where we found a significant correlation between the LOF data and the original data. Currently, we can only use the LOF data to make predictions as to where the foliage might be. But in the future, we aim to overlay the two sets of data to find precise locations and remove the need for ground truth measurements. NASA’s SWESARR airborne instrument collected the relevant data used in this research experiment.

Satellites are particularly well-suited to provide spatially distributed observations of global snow. For hydrological research and applications, Snow Water Equivalent (SWE) is the most important observation, but it is also our biggest gap in snow remote sensing. Currently, no satellite sensor has the ability to measure SWE globally at the accuracy, resolution and frequency needed, because of a number of factors that impact the signals such as forests, mountains, clouds and the snow characteristics themselves. The NASA Climate Change Research Initiative (CCRI) SnowEx team uses machine learning approaches to attempt to address the unanswered questions of snow science. The team has collaborated with other similar NASA wide programs and leveraged skills and resources which led to the formation of a community machine learning (ML) working group.

Mountain snow fields act as natural storage basins in areas that rely on snowmelt, and by understanding the long-term trends of these systems, alongside the spatial component, we can make more informed decisions on their management and conservation. However, current flight intervals make the production of continuous airborne data limited. Our aim is to demonstrate the necessity of more temporal analysis and modeling of Snow water equivalent (SWE) for the management and conservation of water resources from the Tuolumne River Basin, and globally, through Time Series Forecasting and Uni/Multivariate analysis. In association with NASA SnowEx, we aim to create two Long-Short-Term Memory (LSTM) regression models to predict SWE, by itself and how it changes based on Snow Depth and Snow Density. The relevant raster data sets were collected by the ASO (Airborne Snow Observatory), and obtained from the NSIDC (National Snow and Ice Data Center).

During 2020 and 2021, the NASA SnowEx Mission performed a time series with UAVSAR, and L-band InSAR, in the Western U.S. Small field efforts were performed throughout the time series, and in addition, SnowEx carried out an intensive campaign on Grand Mesa, involving five aircraft with seven different airborne instruments, and a large field campaign. Snow properties can vary significantly over distances of 50-200 meters, and therefore rapid techniques for measuring bulk snow properties are valuable for calibration and validation of snow remote sensing efforts. We developed and deployed a ground-based microwave radar from a snowmobile, during the 2020 NASA SnowEx campaign on Grand Mesa. These observations provide information about the spatial distribution of snow depth, snow water equivalent, and stratigraphy, and were performed coincident with many different in-situ and airborne snow remote sensing observations.