ESA CCI SM FREEZE/THAW Long-term Climate Data Record of surface conditions from merged multi-satellite observations

This dataset was produced with funding from the European Space Agency (ESA) Climate Change Initiative (CCI) Plus Soil Moisture Project (CCN 4 to ESRIN Contract No: 4000126684/19/I-NB ”ESA CCI+ Phase 2 New R&D on CCI ECVS Soil Moisture”).  Project website: https://climate.esa.int/en/projects/soil-moisture/

This dataset contains information on the Surface Soil Moisture (SM) state derived from satellite observations in the microwave domain.

The operational (ACTIVE, PASSIVE, COMBINED) ESA CCI SM products are available at https://catalogue.ceda.ac.uk/uuid/c256fcfeef24460ca6eb14bf0fe09572/ 

Abstract

Understanding whether the soil surface is frozen or thawed is crucial for interpreting satellite-based soil moisture measurements and for many Earth system applications. The physical state of water in the soil strongly affects its dielectric properties, which in turn determine how satellites sense moisture content. Current ESA CCI Soil Moisture products exclude data when the surface is likely frozen, as reliable retrievals are not possible under such conditions. Yet, the freeze/thaw state itself carries valuable environmental information: it reflects the changing energy and water exchange between land and atmosphere, shapes seasonal hydrological cycles, and influences agriculture, ecosystems, and climate feedbacks across much of the Northern Hemisphere.

This dataset provides global estimates of the soil moisture freeze/thaw state for the period from 11-1978 to 12-2025 derived from PASSIVE (radiometer) and ACTIVE (scatterometer) satellite observations within the ESA CCI Soil Moisture framework. These sensors, operating in the K- and C-band frequency range, are sensitive to surface temperature, enabling the detection of frozen versus thawed conditions at daily temporal and ~25 km spatial sampling. Data from L-band missions (e.g., SMAP, SMOS) are not included, resulting in a total number of 17 satellites. Only nighttime observations are considered for the passive sensors.

The classification algorithm, described in Van der Vliet et al. (2020), was originally developed to flag frozen conditions in passive soil moisture retrievals and has since evolved into a dedicated data product. It applies a decision-tree approach using multi-frequency satellite measurements to classify the surface state for each sensor. Similarly, Naeimi et al. (2012) have developed an algorithm based on ASCAT backscatter for freeze/thaw classification in C-band scatterometer retrievals. Individual classifications are then merged into a single spatiotemporal record using a maximum-likelihood estimator based on the sensitivities (probability that the sensor classifies the soil as frozen given that it is truly frozen) and specificities (probability that the sensor classifies the soil as thawed given that it is truly thawed) of the individual sensors, as described in Jaffe et al. (2015). These are calculated per pixel via Categorical Triple Collocation Analysis (McColl et al., 2016, Wang et al., 2026) using triplets of a passive sensor, an active sensor and the GLDAS-Noah land surface model. In areas where this algorithm fails, sensitivities are estimated via a global linear regression model for each sensor based on the climatological (long-term mean) fraction of frozen soil of the GLDAS-Noah land surface model. Specificities, which vary very little geographically, are estimated via the global mean of the available datapoints. 

The current product achieves an estimated accuracy of 83% (up from 75% in the last version) against in situ surface temperature observations and 92% compared to ERA5 reanalysis data.

Summary

• Daily binary (true/false) freeze/thaw surface soil moisture state classification dataset (~25 km spatial sampling) for the period November 1978 to December 2025.
• Based on a satellite brightness temperature (K-band) classification algorithm (Van der Vliet et al., 2020) from 12 satellite radiometers and a satellite backscatter (C-band) classification algorithm (Naeimi et al., 2012).
• A pixel is classified as "frozen" if the combined classifications of all sensors providing valid measurements are more likely to occur when the true state of the soil is frozen rather than thawed (maximum-likelihood estimation).
•  Approximately 83% agreement with in situ surface temperature measurements (Dorigo et al., 2021) and 92% with ERA5-Land reanalysis temperature fields (Muñoz-Sabater et al., 2021)

Programmatic (bulk) download

You can use command-line tools such as wget or curl to download (and extract) data for multiple years. The following command will download and extract the complete data set to the local directory ~/Download on Linux or macOS systems.

#!/bin/bash

# Set download directory
DOWNLOAD_DIR=~/Downloads

base_url="https://researchdata.tuwien.at/records/awtmr-a9d08/files"

# Loop through years 1978 to 2024 and download & extract data
for year in {1978..2025}; do
    echo "Downloading $year.zip..."
    wget -q -P "$DOWNLOAD_DIR" "$base_url/$year.zip"
    unzip -o "$DOWNLOAD_DIR/$year.zip" -d $DOWNLOAD_DIR
    rm "$DOWNLOAD_DIR/$year.zip"
done

Data details

Filename template

The dataset provides global daily estimates for the 1978-2025 period at 0.25° (~25 km) horizontal grid resolution. Daily images are grouped by year (YYYY), each subdirectory containing one netCDF image file for a specific day (DD) and month (MM) of that year in a 2-dimensional (longitude, latitude) grid system (CRS: WGS84). The file name follows the convention:

ESACCI-SOILMOISTURE-L3S-FT-YYYYMMDD000000-fv10.1.nc

Data Variables

Each netCDF file contains 3 coordinate variables

• lon: longitude (WGS84), [-180,180] degree W/E
• lat: latitude (WGS84), [-90,90] degree N/S
• time: datetime, encoded as "number of days since 1970-01-01 00:00:00 UTC"

 and the following data variables

• ft: (int) Soil moisture freeze-thaw state binary indicator (0=not frozen, 1=frozen, -1=missing data)
• ft_agreement (float): Classification agreement between available sensors. 1 means that the frozen/unfrozen classification was the same for all merged sensors. The number decreases as the classification results between available satellites contradict.
• sensor_count (int): Total number of merged sensors
• sensor_count_frozen (int): Total number of measuring sensors that detected frozen soils
• mode: (int) Indicator for satellite orbit(s) used in the retrieval (1=ascending, 2=descending, 3=both, 0=missing data)
• sensor: (int) Indicator for satellite sensor(s) used in the retrieval. For more details, see netcdf attributes.

Additional information for each variable is given in the netCDF attributes.

Version Changelog

Changes in v10.1 (this version): 

• This version changes the merging algorithm and replaced the unanimity rule with a maximum-likelihood estimator (Jaffe et al. (2015)). Sensitivities and specificities of the individual sensors are estimated via Categorical Triple Collocation (McColl et al., 2016, Wang et al., 2026). Only nighttime observations are considered for the passive sensors. Covers the period from 11-1978 to 12-2025.

Changes in v9.2 (first released version):

• This version applies the classification algorithms described by Van der Vliet et al. (2020) and Naeimi et al. (2012) to 17 sensors and a unanimous merging approach. Covers the period from 11-1978 to 12-2024.

Software to open netCDF files

These data can be read by any software that supports Climate and Forecast (CF) conform metadata standards for netCDF files, such as:

• Xarray (Python)
• netCDF4 (Python)
• esa_cci_sm (Python)
• Similar tools exist for other programming languages (Matlab, R, etc.)
• Software packages and GIS tools can open netCDF files, e.g. CDO, NCO, QGIS, ArcGIS
• You can also use the GUI software Panoply to view the contents of each file

Related Records

This record and all related records are part of the ESA CCI Soil Moisture science data records community.