CASM: Global Soil Moisture from Remote Sensing and AI

CASM is a global soil moisture dataset created using a neural network to extend the high-quality data from NASA's SMAP satellite mission back to 2002. It provides estimates for the top soil layer on a 25 km EASE-2 grid with a 3-day resolution from 2002 to 2020. The machine learning approach focused on predicting residuals from the seasonal cycle, aiming to capture extremes, and shows a mean correlation of 0.97 with SMAP data where they overlap.

Use Cases

• Analyzing long-term soil moisture trends and interannual variability based on the 2002-2020 time series.
• Validating hydrological or climate models using the globally consistent, 25 km resolution grid.
• Studying extreme soil moisture events (droughts, floods) using the model trained on seasonal residuals.
• Fusing multi-satellite microwave data (SMOS, AMSR-E/2) to create a homogeneous climate data record.

Strengths

• Exhibits excellent spatial and temporal homogeneity, with a mean correlation of 0.97 against the high-quality SMAP target data.
• Covers a long temporal range (2002-2020), extending SMAP-like data 13 years prior to its launch.
• Validation includes comparison against 367 global in-situ soil moisture monitoring sites.

Limitations

• Metadata on exact file size, row count, and specific column names is unavailable across all sources.
• The dataset's last update date is not specified, which may affect knowledge of its current state.
• Represents moisture only in the top soil layer, which may not capture deeper root zone dynamics.

Provenance

Source

Created by Olya Skulovich at Columbia University, based on NASA SMAP, ESA SMOS, and JAXA AMSR-E/2 satellite data.

Collection Method

Generated using a neural network trained on brightness temperature data from earlier satellite platforms to predict SMAP-quality soil moisture, with a focus on modeling residuals from the seasonal cycle.

Time Range

2002 to 2020

Geography

Global coverage on a 25 km EASE-2 grid.