The land surface temperature (LST) CCI project aims to deliver a significant improvement on the capability of current satellite LST data records to meet the strict GCOS requirements for climate applications of LST data.
Accurate knowledge of land surface temperature (LST) plays a key role in describing the physics of land-surface processes at regional and global scales as they combine information on both the surface-atmosphere interactions and energy fluxes within the Earth Climate System. This provides important information across a range of disciplines including monitoring drought, impact on human health, and changes in vegetation.
Phase 1 of the programme of work has achieved some excellent progress:
• Detailed climate user input into the specifications of the LST ECV products, and user assessment of these products to drive LST exploitation in climate science
• Strong buy-in from the climate science community coordinated by the Climate Research Group
• A suite of high quality IR LST ECV Products and MW LST ECV Products for geostationary (GEO) and low earth orbit (LEO) satellites from the ATSRs, MODIS, SLSTR, SEVIRI, GOES, MTSAT and SSM/I
• An improved long-term LST CDR of +20 years from 1995 to 2020 for ATSR-2 through to SLSTR
• A +10 year Merged LST product combining the advantages of both GEO and LEO satellites
• Algorithm, cloud masking and uncertainty consistency across datasets
We present here the approaches taken and the results to realise the full potential of long-term LST data for climate science.

The ESA Land Surface Temperature (LST) Climate Change Initiative (LST_cci, https://climate.esa.int/en/projects/land-surface-temperature/) project aims to provide an accurate view of temperatures across land surfaces globally over the past 25 years. The main LST_cci data records are derived from inverting infrared observations from a variety of satellite sensors. This is the established technique to derive reasonably accurate and spatially resolved LST estimates. However, clouds obscure the radiation emanating from the surface, so infrared data records can only provide a clear-sky view of the surface. To have estimates for “all-weather” conditions, LST_cci is also providing a 25-year data record derived from microwave observations, which can see through most clouds, therefore providing a view of the surface independent of cloud coverage.

The microwave LST is derived from the Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave/Imagers (SSM/I) and the Special Sensor Microwave Imager Sounder (SSMIS) instruments. These sensors are operational weather satellites not originally designed with the instrumental and orbital stability constrains required to study decadal climate trends. To mitigate these limitations for climate studies, different aspects of data selection and algorithm development aspects have been considered, including:

1) Brightness temperatures. The Fundamental Climate Data Record (FCDR) of Microwave Imager Radiances from the Satellite Application Facility on Climate Monitoring (CM SAF) is selected to provide the observations. It corrects calibration anomalies and several known issues with the instruments to ensure the long-term stability of the FCDR, providing an inter-calibrated and homogenized data record of brightness temperatures.

2) Retrieval algorithm. To reduce the dependence on ancillary data, the processing is based around a neural network algorithm having only as inputs the observations at 18, 22, 36, and 85 GHz, together with pre-calculated emissivity values to account for the large variations in microwave emissivity. While most of the surface temperature information comes from the 18 and 36 GHz window channels, the 22 GHz, being close to a water vapor line, and the 85 GHz, more impacted by the atmosphere, allow to take into account atmospheric variations without the need to rely on related ancillary data.

3) Time correction. The orbital drift of the DMSP platforms results in variations of the overpass local time during the lifetime of the instrument. To facilitate climate studies, a temperature adjustment is available to have all LST estimates corrected to a fixed local time (6AM/PM).

The current processing is based around the inversion of the observations from SSM/I onboard the DMSP F13 satellite, and SSMIS onboard F17, covering respectively the periods 1996-2008 and 2009-2020. Twice per day LST estimates are available, coinciding with the descending (~6AM) and ascending (~6PM) overpasses of the instrument. The estimated uncertainty suggests precisions ranging between 2.0 and 3.5 K depending on the observing conditions. Comparing with the infrared retrievals, the uncertainties are larger, reflecting the stronger dependence of the microwave radiances in a more varying surface emissivity. Adding the time-correction offset permits to have the previous LST estimates adjusted to represent the LST at a fixed local time of 6 AM/PM. This correction improves the time stability of the data record, although adds a new component of uncertainty (~1.0-2.0 K) related to the errors of the time correction.

Current validation efforts are targeting comparisons with point measurements and larger scale estimates from atmospheric reanalysis. Comparing with in situ LST from a selection of ground stations, Root Means Square Differences take values between 2.0 and 5.0 K, but these figures need to be approached with caution due to the challenge of comparing the station point measurement with the ~25 km footprint of the microwave LST estimates. To analyze the stability of the long time series, the LST has been globally compared with skin temperature from ERA5. The first analyses suggests that the SSM/I and SMIS time-corrected LST data records are reasonably stable with time when considered as two independent time series, and that the discontinuities in the time-corrected data record associated with the change of instrument are greatly reduced when the data record is considered as a full time series.

The SSM/I and SSMIS LST data record is publicly available and can be downloaded from the ESA CCI Open Data Portal. Daily, monthly, and yearly LST estimates are provided, separated by descending and ascending orbits. The LST values are offered as a L3C product where the original swath L2 estimates have been remapped to a regular grid of 0.25 degrees to facilitate the use of the data record for global applications, but the original L2 values can also be provided if requested. The LST adjustment is included in the L3C files, and can be added by the user to have the time series at any location providing an LST estimate at a fixed 6AM/PM local time.

Climate data records (CDRs) from Earth observation (EO) are generated using data from multiple satellite instruments. Long-term stability in the CDR is essential for detecting climate signals; for products that are dependent on a pre-processing cloud detection step, continuity in cloud mask performance between different sensors (potentially with different channels or characteristics) is essential. This is of particular relevance in the retrieval of surface temperature: for example, a change in the cloud hit rate over time could lead to apparent warming (increased hit rate) or cooling (decreased hit rate) in either a local or global temperature trend that is an artefact of the cloud detection performance over time, rather than a real climate signal.

The Land Surface Temperature (LST) Climate Change Initiative (CCI) single-sensor CDR is comprised of data from the Along-Track Scanning Radiometer instruments (ATSR-2, AATSR), the Moderate Resolution Imaging Spectroradiometer (MODIS Terra) and Sea and Land Surface Temperature Radiometer (SLSTR-A) instruments. We use a match-up database (MD) of satellite and ceilometer observations over 25 years (1996-2020) to assess cloud masking stability for three cloud detection algorithms: 1) A Bayesian cloud detection scheme developed by the University of Reading and 2) A naïve probabilistic cloud detection algorithm developed by the University of Leicester and 3) the operational cloud mask for each sensor.

This new MD provides a unique opportunity to evaluate cloud mask performance over time for different locations across the globe with differing biomes and atmospheric regimes. Ceilometer data from Alaska, Ny Alesund and the Southern Great Plains spans all sensors in the CDR, with further data from the Tropical West Pacific, Greenland, UK, Puerto Rico and France covering two or more sensors. We examine the stability in cloud masking performance over time, focusing on the step changes between sensors and day/night differences, assessing the suitability of these cloud detection approaches for the construction of CDRs. Analysis of the impact on LST of any changes in cloud detection artefacts will also be examined.

Surface reflectance is one of the key products used in developing several land Essential Climate Variables, such as Vegetation Indices, Albedo and LAI/FPAR, it is therefore seminal to detecting trends in the biosphere and land surface and has been classed by NOAA as a “Fundamental Climate Data Record (FDCR) for Land”. Building a long-term surface reflectance data record of climate quality requires combining data from different instruments, sensors and satellites, accounting for different spatial resolutions and spectral characteristics, assuring consistent calibration, and correcting for atmospheric and directional effects.

In this work, we use robust reflectance data records and inter-comparison methods that we have developed over the past several years (consisting of atmospheric correction, directional effect correction and spectral normalization) to establish and verify the inter-consistency of the reflectance products from the AVHRR sensors on-board NOAA 7, 9, 11, 14, 16, 17,18, 19 and METOPB, the MODIS sensors on-board Aqua and Terra, the VIIRS sensors on-board Suomi-NPP and JPSS1, the TM/ETM+ on Landsat 5/7, the OLI on Landsat 8 and the MSI on Sentinels 2A,B. The resulting dataset being the 35+ years surface reflectance product and vegetation indices from 1981 to present of increasing spatial, spectral and accuracy performance, which provides a solid basis for deriving a number of Terrestrial Essential Climate Variables.

Satellite SAR data archives are vital assets for environmental and geohazard investigations, enabling multi-decadal observation of surface processes and dynamics. In this work, we exploit ERS-1/2, ENVISAT and Sentinel-1 long time data series and hosted on-demand processing tools based on advanced multi-temporal InSAR workflows that are made available in ESA’s Geohazards Thematic Exploitation Platform (GEP) to showcase the value of these archives for ground deformation monitoring. Data series with some tens and up to more than three hundreds of input scenes are processed using cloud resources to retrieve coherent target datasets. Each of them is provided with associated information on long term deformation velocities and full deformation histories capable to capture temporal trends and their evolution over the last three decades, in response to changed environmental and climate conditions, and/or urban growth and boosted need for land and water resources. Case studies in southern Italy (e.g. [1]) and central Mexico (e.g. [2]) are used to provide a representation of different geological and anthropogenic hazards, including land subsidence due to groundwater resources exploitation and aquifer depletion in expanding cities and metropolises, hydrocarbon extraction in rural and sub-urban regions, slow moving landslides and erosion landforms, volcano lava flow cooling and compaction, as well as settlement of recently built coastal protection infrastructure. The integration with other geospatial datasets – either from in situ measurements or national repositories (e.g. landslide inventories, urban geostatistical layers) or other Earth Observation-based products (e.g. Copernicus Land Monitoring Service–Coastal Zones and land cover datasets) – allows for an effective valorization of the information extracted from SAR deformation time series, to achieve a correct interpretation of (geo-)hazard processes and a value-added understanding of their impacts on local environment, population and infrastructure. The availability of SAR big data stacks ensuring continuity of observation across long periods and cloud based processing tools proves essential to retrieve robust and reliable estimates and information on long-term deformation patterns and trends, which often cannot be depicted via short-term assessments based only on a few months or 1-2 years of observations. Thus, long-term preservation of SAR data archives from both heritage and current emissions is crucial to guarantee their availability in the future.

[1] Cigna F., Tapete D. 2021. Sentinel-1 Big Data Processing with P-SBAS InSAR in the Geohazards Exploitation Platform: An Experiment on Coastal Land Subsidence and Landslides in Italy. Remote Sensing, 13 (5), 885. https://doi.org/10.3390/rs13050885

[2] Cigna F., Tapete D. 2021. Satellite InSAR survey of structurally-controlled land subsidence due to groundwater exploitation in the Aguascalientes Valley, Mexico. Remote Sensing of Environment, 254, 1-23, https://doi.org/10.1016/j.rse.2020.112254