TL;DR
Scientists have created the first global, long-term (since 1992) satellite radar dataset, called CScat, by merging data from multiple C-band and Ku-band sensors. This new dataset fills a critical gap, enabling robust analysis of land surface changes over decades, which was previously impossible due to data limitations.
What Problem Does This Paper Address?
Satellite radar backscatter is a powerful tool for studying land surface moisture, vegetation, and roughness, offering all-weather capabilities. However, a major limitation for long-term trend analysis has been the lack of a single, continuous global radar dataset spanning more than two decades. Existing datasets are often limited in time or frequency band, making it difficult to study decadal changes in Earth's land surface. This paper addresses this by creating the CScat dataset, which provides a consistent, high-resolution C-band radar backscatter record from 1992 to the present.
Methods
The core of this work involves merging data from three key satellite scatterometers: ERS (1992-2001, C-band), QSCAT (1999-2009, Ku-band), and ASCAT (2007-present, C-band). The primary challenge was bridging the six-year gap between ERS and ASCAT. To achieve this, the researchers developed a novel two-step approach:
- Signal Rescaling: A new method was devised to unify the backscatter signals from different sensors, especially crucial for short overlapping periods. This method is designed to be robust against trends and sudden changes in the radar signal, unlike previous CDF matching or linear regression techniques which struggled with limited overlap or abrupt shifts.
- Monthly Signal Difference Correction: After rescaling, monthly differences between the C-band and scaled Ku-band signals were modeled and corrected. This was done using decision tree regression, with climatic variables like precipitation, skin temperature, and snow depth as predictors. This approach accounts for the different sensitivities of C-band and Ku-band signals to various environmental factors.
Datasets
The primary datasets used are:
- ERS (1992-2001): European Remote Sensing satellite, C-band (5.3 GHz).
- QSCAT (1999-2009): Quick Scatterometer, Ku-band (13.4 GHz).
- ASCAT (2007-present): Advanced Scatterometer, C-band (5.255 GHz).
- ERA5-Land: Reanalysis data for monthly precipitation, skin temperature, and snow depth, used as predictors for signal difference correction.
- ESA ERS-2 data set (2001-2011): Used for independent validation of the merged CScat data.
The CScat dataset itself is a newly introduced benchmark, providing a continuous, global, high-resolution (~8.9 km) C-band radar backscatter record.
Key Insights and Ideas
- Bridging the Gap is Crucial: The six-year gap between C-band ERS and ASCAT data was a significant barrier to long-term analysis. Filling this gap with a modeled Ku-band equivalent is key to creating a truly long-term dataset.
- New Rescaling Method is Robust: The developed signal rescaling method is a significant advancement, proving effective even with short sensor overlap periods, which is a common challenge in satellite data fusion.
- Climatic Variables Explain Signal Differences: The study demonstrates that precipitation, skin temperature, and snow depth are significant drivers of the differences observed between C-band and Ku-band radar signals, allowing for their correction.
- C-band vs. Ku-band Sensitivity: The research highlights how the different wavelengths of C-band and Ku-band radar lead to varying sensitivities to atmospheric conditions (like rain) and surface features (like vegetation canopy and snow), which is fundamental to understanding the signal differences.
Key Results
- The CScat dataset provides a continuous global C-band radar backscatter record from 1992 to the present at ~8.9 km resolution.
- The data merging process achieved high accuracy, with regional Pearson r values between the C-band and corrected Ku-band signals exceeding 0.93 and RMSE values below 0.16 in overlapping years.
- Pixel-level validation showed median r values of at least 0.80 after correction, with significant improvements observed in northern high latitudes.
- Independent validation against the ESA ERS-2 dataset showed consistent monthly dynamics, with Pearson r values ranging from 0.79 to 0.98 across regions.
- The study identified three main types of pixels based on which climatic variable best predicted signal differences: precipitation (31%), skin temperature (64%), and snow depth (5%).
Limitations and Future Work
- Climatic Data Accuracy: The accuracy of the merged signal depends on the accuracy of the ERA5-Land climatic data used for modeling signal differences. Future work could explore other climate datasets like MERRA-2 or CHIRPS, which may offer better performance in specific regions.
- High-Latitude Snow Depth: Snow depth is a critical predictor in high-latitude regions like northern Alaska, but its accurate mapping is challenging.
- Tropical Rainfall Estimation: Similarly, rainfall estimation in tropical regions with sparse climate stations is difficult.
- Other Predictors: The authors suggest exploring other remote sensing-based layers, such as NDVI (Normalized Difference Vegetation Index) and aerosol products, to potentially improve signal difference modeling.
- Sensor Fusion Expansion: Including data from other sensors like Oceansat-2 Scatterometer (OSCAT) could further extend the overlapping periods and improve the merging method.
- Advanced Machine Learning: Exploring more advanced machine learning techniques for modeling signal differences is a promising avenue.
- Combined Predictors: Using all three climatic variables (precipitation, skin temperature, and snow depth) together in a single regression model could enhance accuracy.
- Higher Resolution: Generating a new version of the dataset using only QSCAT and ASCAT at their native higher resolution (4.45 km) is planned.
- Signal Separation: Future work aims to separate the merged radar signal into soil moisture and vegetation components using radiative transfer models, enabling direct relation to soil or vegetation dynamics.
Summary updated