Enhancing SMAP Radiometer Performance with Deep Learning

Motivation

NASA’s SMAP mission has produced a decade of global soil moisture observations from an L-band radiometer. Two effects limit the quality of that record: radio frequency interference from ground-based transmitters in and near the protected band, and slow drifts in radiometer calibration. Both are difficult to handle with fixed, threshold-based algorithms, because interference appears in many forms and calibration references are limited.

Approach

This project applies deep learning to both problems, working directly with SMAP mission data:

• High-resolution RFI detection — convolutional architectures that flag interference at finer time-frequency resolution than conventional detectors, recovering observations that would otherwise be discarded.
• Learning-based calibration — calibration models that use reduced reference information together with two-dimensional spectral features to track instrument behavior over time.
• Validation against mission products — assessing the effect of improved RFI handling and calibration on downstream soil moisture retrievals.

Selected outcomes

• Radio Frequency Interference Detection for SMAP Radiometer Using Convolutional Neural Networks, IEEE JSTARS, 2022.
• Microwave Radiometer Calibration Using Deep Learning With Reduced Reference Information and 2-D Spectral Features, IEEE JSTARS, 2024.
• High-Resolution Radio Frequency Interference Detection in Microwave Radiometry Using Deep Learning, IGARSS, 2023.