The Detection and Classification of Radio Frequency Interference in the Soil Moisture Active Passive Dataset

Loading...
Thumbnail Image

Date

2023-05

Journal Title

Journal ISSN

Volume Title

Publisher

The Ohio State University

Research Projects

Organizational Units

Journal Issue

Abstract

Soil moisture data is used in weather forecasting, drought detection, flood anticipation, and crop monitoring. NASA's Soil Moisture Active Passive (SMAP) mission orbits the Earth and operates a L-Band radiometer measuring the natural emission from the Earth to retrieve soil moisture. Despite the L-Band being a protected part of the spectrum to allow for passive observation of the Earth, radio frequency interferences (RFI) are observed in the SMAP measurements. This RFI is generated by illegal emissions within the protected band or by transmitters in adjacent frequency bands. Detecting RFI is critical, as it corrupts the radiometer measurements and can potentially bias the soil moisture retrievals, if the RFI is undetected. The detection is possible thanks to nine detection algorithms that are implemented in ground processing. Despite the overall good performance of the detection algorithms, some undetected RFI, also defined as residual RFI, are still noticeable in the SMAP measurements. This research is divided into two parts. The first part of the work focuses on classifying RFI sources using deep learning to provide a better understanding of the RFI environment. In this part of this study, the brightness temperatures for each SMAP radiometer measurement, or footprint, are treated as an image which will be the input to a deep learning neural network to classify the types of RFI into three groups: no RFI, wideband RFI, and narrowband RFI. The first results confirmed that using a neural network to classify different RFI types is possible and the SMAP footprints were successfully classified with and accuracy of 98%. The second part of the project investigates the implementation of image processing techniques to detect residual RFI by using mean filtering to detect and remove residual RFI in SMAP data. The filter was used on weekly max hold filtered brightness temperature maps, i.e., after RFI detection was performed and convolves over the global image, identifying regions with large spatial variations. Initial results for the mean filtering algorithm demonstrated the potential of this technique to detect spatial areas contaminated with residual RFI sources.

Description

Keywords

NASA SMAP mission, neural networks, image classification, RFI detection, RFI classification, mean filtering, image processing

Citation