Radio spectrum monitoring and enforcement actions are key to deliver radio protection and ensure that the different radio services operate in compliance with the in-force radio regulations and to secure an efficient and effective spectrum use, free from Radio Frequency Interference (RFI) . Notwithstanding these essential actions, RFI is an inevitable consequence of the operation of radio systems, and it is a growing threat that has affected numerous Earth Observation (EO) satellite missions in recent years, especially passive microwave sensors . These interferences vary widely in power level, radiation pattern, polarization, bandwidth, time duration, duty cycle, etc., and as a consequence there is no single algorithm that can optimally detect all levels and types of radiometric contamination .

The European Space Agency (ESA) and other space agencies rely on microwave radiometers for a wide variety of Earth Observation (EO) missions, operating in frequency bands from 1 to 700 GHz. RFI can negatively impact these radiometers in three different ways: High power RFIs can potentially damage the instrument, but have low occurrence probability; medium power RFIs cause data loss and gaps; and low power RFIs, mostly undetected, can distort scientific data, potentially compromising geophysical retrievals and weather forecasts.

It is thus of utmost importance to detect such RFI occurrences, not only to address their impact on the users' side, but also to support respective radio protection actions. However, instances of interference can vary widely not just in magnitude as mentioned above, but also in whether it is possible or appropriate to assign responsibility. Albeit the role of the International Telecommunications Union (ITU) in facilitating international coordination of radio protection actions, national administrations, by means of their National Regulatory Authorities (NRA), are responsible for overseeing the spectrum use and enforce the law in the territory under their jurisdiction. The bands typically used in EO missions fall into EESS (Earth Exploration-Satellite Service), including passive allocations with either exclusive access or shared access with active services. Service disruptions should always be reported to NRA who are responsible for conducting the required investigations and actions on the field to secure integrity of the spectrum so it can be effectively and efficiently used. In the case of exclusive access bands, clear violations, from the law enforcement standpoint, are easier to act upon. In case of shared access bands, the regulatory framework is still to be enforced. If the regulatory framework is enforced and service disruptions still occur, those findings are relevant in closing the loop for future enhanced spectrum policy, planning and assignment conditions. Comprehensive details about radio protection policies and practices can be found in . There are numerous instances of the process to locate and shut down illegal transmissions ongoing, but new RFI sources appear constantly and the procedure to detect, attribute, and shut down illegal operators takes time.

In addition to clear violations of ITU regulations, many instances of apparent RFI come from legal operations in nearby bands or simply other users in a non-exclusive band. Furthermore, the spectral response of a radiometer can vary significantly from its specified bandwidth, and in this case apparent violations may be caused by radiometer sensitivity outside of the exclusive-use band due to being manufactured outside specified tolerances. Many space agencies now publicly provide measured spectral response functions (SRFs) for MW radiometers and it is a recognized best practice , but it is rare for heritage sensors and indeed most of the instruments studied here do not have measured SRFs publicly available. In other words, the cause of apparent RFI may be due to instrument design, illegal operators, or be a simple consequence of sharing a crowded part of the spectrum. No matter their source, satellite operators and downstream users need to address the inevitable presence of unwanted emissions.

Regardless of the source of RFI, and although the RF environment is constantly changing, it remains critically important for the planning of future EO missions to know the current status of contamination at bands of interest. For example, the L-band has been widely studied , and this has led to tangible changes in mission design and requirements for subsequent L-band missions. Nevertheless, the ESA RFI Monitoring & Information Tool (ERMIT) showed that the decreasing trend in worldwide contamination suddenly changed in 2024 when the Russo-Ukrainian war and the Myanmar civil war led to massive contamination in all observations over Europe and most of the South and South East Asia . C-band contamination was first discovered in AMSR-E measurements . The presence of RFI in this band led to inclusion of an extra C-band channel on the follow-on AMSR2 mission for the purpose of RFI mitigation . RFI surveys conducted with different methodologies have only been conducted up to and including the 23.8 GHz band , showing contamination in the 18.7 GHz band but not at higher frequencies. reports RFI up to the 36.5 GHz band, albeit not widespread. Despite the best efforts of space agencies to predict the potential contamination issues of future EO missions, the RFI environment could change rapidly as services require more and more bandwidth, technology becomes more accessible and thus widespread, active services are aiming at higher frequencies, and the spectrum becomes more crowded with different users. Spectrum use, in the past, present and future, by Earth observation missions has been reviewed in .

The objective of this study is to survey RFI contamination in several key passive microwave bands from 6 up to 200 GHz using the same methodology, independent on the sensor characteristics (conical scan, across-track scan, …), in all bands commonly used by microwave radiometers using software developed under an ESA contract: the Earth Observation RFI Scanner (EORFIScan). This system is designed to be placed between the data generated from EO missions and the end users, such as the Numerical Weather Prediction (NWP) centres, with the goal of detecting and filtering RFI in order to avoid the assimilation of RFI-contaminated data. These new flags are appended to the original EO products. For this analysis, a comprehensive library of RFI detection algorithms is used, supported by external information and RFI databases. As the system analyses the incoming EO products, the results are stored in internal RFI databases that are used to improve future detection of RFI signals in the products. These databases show a picture of the RFI environment in each frequency band analysed. L-band is not analysed since EORFIScan is an upgrade of the Ground Detection RFI System (GRDS), which already surveyed this frequency band , and frequency bands beyond 200 GHz the number of sensors working at these bands is reduced and RFIs are less likely to happen.

It is worth emphasizing that proper validation of RFI detection methods is difficult because no source of truth is available. Traditional metrics like a false alarm ratio are then hard to define, in contrast to a parameter like precipitation for which rain gauges offer a local measure of truth. In this study, RFI detection is assessed using simulations from an NWP model, following the idea that NWP model simulations provide information on the expected natural variability, which can be used as reference for RFI detection. This allows some validation of the methods, both instantaneously and in aggregate, but does not lead to specific conclusions on aspects like detection limits or false alarms. Furthermore, even an excellent RFI detection method will miss contamination that is within instrument or geophysical noise levels. This type of low-level RFI would be effectively undetectable on an instantaneous basis. Therefore, the lack of detection does not preclude the existence of RFI in a given band. This is especially true at higher frequencies where natural variability is larger due to more significant atmospheric signals.

This paper is structured as follows. Section  describes the system architecture, the RFI detection techniques utilized, and the threshold computation procedure, Sect.  shows the RFI maps of the passive microwave bands from 6 up to 200 GHz, which are discussed in Sect. . Section  describes the results validation, and Sect.  draws some conclusions.

This section presents a brief overview of the system architecture, describes which instruments were analysed during which time period, describes the RFI detection techniques applied in the analysis, and explains the procedure to compute the thresholds of each detection technique.

In order to evaluate the status of the RFI environment, RFI probability maps are used. Neither AMSR2, AMSU-A or MWHS-2 data is gridded, so EASE2 grid with 25 km spacing has been used for the database. Observations are assigned to the closest grid point. Then, the probability of RFI is calculated at each grid point as PRFI=D/O, where D is the number of RFI detections, and O is the total number of observations.

The maps are computed for each month along the temporal scope of each sensor analysis using both ascending and descending data, and both H and V polarization (if available). In the following, we present maps of mean RFI probabilities generated from these monthly maps, with the aim to highlight areas that are consistently flagged as RFI-affected. While these average maps may not capture more transient events that are only present during certain times, they provide a representative overview of the main affected areas. Average maps are a good trade-off between showing RFI contamination and mitigating false alarms.

Furthermore, after discarding measurements potentially affected by sun glint, observations over 350 K were stored along with their latitude, longitude and date per each frequency band. Then, they were grouped into clusters using k-means method.

Figures  to show mean RFI probability maps for selected channels from the four instruments considered, covering the frequency range 6.9 to 183 GHz. Maps from AMSR2 have been prioritized over AMSU-A since two polarizations are available, and consequently the resulting maps are less noisy. Furthermore, the higher the frequency, the noisier the maps, since the spatial resolution is higher, the atmospheric effects on the TB become more intense, and the atmospheric contributions are less smooth than the land or sea dynamics particularly because cloud and precipitation sensitivity generally increases with frequency.

The 6.4–7.7 GHz band is not exclusively allocated for passive applications and therefore passive sensors cannot claim protection. This regulatory framework can be linked to the significant contamination documented previously . It is separated into the 6.9 and 7.3 GHz sub-bands. At the 6.9 GHz sub-band contamination is widespread in the US, and India. Some contamination is scattered in South America (with a denser detection area in Brazil), in Africa (with a denser detection area in South Africa), Europe (note the dense detection area over Ukraine and the detection from possible oil rigs in the North Sea), and the sea contamination in the Yellow Sea in front of China probably due to marine traffic. Point sources of 6.9 GHz contamination over sea had previously not been identified in the literature. In the 7.3 GHz sub-band contamination is present in all populated continents. Both North and South America show a few scattered detections, with strong detections in Central America. Europe also shows a few RFI sources, mostly in the North Sea due to oil rigs, and a large contaminated area over Ukraine. The Middle East is heavily contaminated in Turkey and Syria, with some contamination seen in Saudi Arabia. Indonesia and Thailand show strong signatures across their entire areas due to a telecommunication service using this band. Some isolated RFI sources can be seen in the rest of Oceania. Also note the reflection of geostationary satellite signals over the sea near Japan, in the Mediterranean Sea and the Atlantic Ocean near Europe, and in the Pacific Ocean between Hawaii and the US.

Only a part of the 10.7 GHz band is exclusive for passive services (10.68 to 10.70 GHz), whereas AMSR2 has a nominal bandwidth from 10.60 to 10.70 GHz. RFI contamination is mostly present in Europe due to the reflection of geostationary TV broadcast satellites over the sea . The reflections are seen as stripes following the descending orbit direction. The same signatures can be seen as descending stripes in the Bay of Bengal near the coast, which were not observed in previous analyses. Some strong signatures can be seen in the UK, and some contamination is visible along the Nile River, in Nigeria, on the east coast of China, Mexico, Middle East, Central and Western Europe, Australia and widespread in Japan and Turkey.

The 18.7 GHz frequency band is not exclusive for Earth Observation, but despite this the RFI contamination maps are relatively clean. RFI contamination is found on the US coast, also including Hawaii. No other regions seem to be contaminated. However, the analysis focused on 2022; since then, strong contamination has been observed by the recent Compact Ocean Wind Vector Receiver onboard the International Space Station (COWVR-ISS) instrument . It is true that COWVR-ISS has a larger bandwidth than AMSR2, and the detected RFIs could be out of the AMSR2 bandwidth. This fact plus a growing presence of contaminated areas makes an RFI detection and mitigation system increasingly attractive at this frequency band.

One area where we can zoom in is near the Norwegian coast in the North Sea and Arctic Ocean, where several spots of contamination at different bands can be observed. They closely match the location of the 2022 oil rig maps observed in Fig.  and it seems reasonable to view this as causal. These signatures are only observed by AMSR2. They are also visible at 6.9, 7.3 and 36.5 GHz, but not at 10.7 GHz, and 89 GHz (see Sect. ). A possible explanation could be that the RFI Index technique is causing false positives due to the coefficients associated with certain bands. Bands where RFI is erroneously detected (tentatively 18.7 GHz) are explained by the coefficients in a and b corresponding to the Tb of the channels where the RFI is present being negative, contributing positively to the RFI Index (see Eq. ). On the contrary, in bands that do not show this “cross-contamination” effect, the coefficients in a and b corresponding to the Tb of the channels where the RFI is present is positive, contributing negatively to the RFI Index. Figure  shows that the upper RFI detected area at 7.3 GHz and the lower RFI detected area at 6.9 GHz seem to be “cross-contaminating” the 18.7 and 36.5 GHz maps, since patterns have similar shapes. It could be the opposite way, but it is less likely that an oil rig is transmitting at 36.5 GHz, since commercial radio links are common at 6.9 and 7.3 GHz bands.

The 23.8 GHz frequency band is exclusive for passive services, but the deployment of communication services for 5G services in the nearby band 24.25–27.5 GHz might cause RFI contamination in this band. This band was identified for use by 5G in 2019 . Analysis of the AMSR2 measurements at 23.8 GHz highlighted instances in which the measured brightness temperature was between 350 and 400 K which indicates presence of RFI. Some of those observations are suspected to be caused by sun glint. These measurements were scattered along a constant latitude that changed from month to month, and matched the regions prone to sun glint shown in and Sect. 3 of whose latitude also change along the year. Some of these observations even showed abnormally high values compared to the surrounding area in the higher frequency bands. These observations were discarded from the analysis. The rest of those observations over 350 K were grouped in clusters and are listed in Table . The largest cluster of RFI signatures is located in Yuzhong County, Lanzhou, Gansu, China. These observations are, to our knowledge, the first confirmed detection of RFI in the 23.8 GHz band. The COWVR-ISS also reported RFI in this frequency band , but since the instrument's bandwidth is 75 MHz larger than the allocated band for passive observations, it is not possible to ensured if it is an in-band or out-of-band emission. Adjacent channels include mobile and fixed communications, and amateur communications. All RFI observations have been located in a World map in Fig. . These detections are not observed in the RFI Probability maps since the number of contaminated observations is proportionally low to the total number of observations collocated to a cell. Furthermore, the smaller antenna swath makes adjacent cells not to contain any contaminated observations, and therefore they do not show the typical “bump” shape. As consequence of both, the effect of these observations in the maps can not be distinguished from the detection noise.

The 31.3–31.8 GHz is exclusive for passive services in region 2, but no clear signs of RFI contamination can be seen except than false positives in already previously mentioned areas such as sea-ice missed classifications and coastal up-welling areas. The band from 33.5–34.5 GHz is shared but results do not suggest man-made RFI contamination. RFI has been found in this frequency band with COWVR , but the larger bandwidth of COWVR (1975 MHz) also covers part of the adjacent bands. Therefore, these reported sources cannot be confirmed as in-band.

The 36–37 GHz band is shared with other services. RFIs have been reported in this frequency band in November 2018 caused by a radar facility in Roi Namur in the Kwajalein atoll, Marshall Islands. The RFI signal only affected the gain of the Sentinel-3 instrument, and the radiometric data quality was not affected . No clear RFI signatures were detected in the considered dataset. AMSR2 data showed 1 single observations in 2022 over 350 K which could not be linked to sun glint effects. The only remaining observation is located at 37.41° N, 37.61° E in Araban, Gaziantep province, Turkey. This RFI observation has been located in a World map in Fig. .

At higher frequencies, there were no clear detections of RFI signals in the considered dataset, regardless of whether the bands are exclusive or shared. The signatures shown in the maps seem to respond to geophysical signatures rather than RFI contamination due to their large extension, location, and variability throughout the year. Most bands show false detections at the sea-ice borders as a result of missed classifications in both hemispheres, even after using a sea ice concentration algorithm that performs well in the ice marginal zone . Threshold could have been adjusted to reduce the false positives in the ice edges, but it would had come at the expense of misdetections in this area. Although not many RFI are expected here, some ships cross this area and their presence should not be neglected. False detections can also be seen at the borders of the Greenland ice sheet, or in the Himalayan mountains. Many coastlines, such as the Mauritanian coast, the Angola-Namibia coast, the Chilean-Peruvian coast, the north-west coast of Australia, the Yemen-Oman coast, the Somalia coast, the Baltic Sea coastline, the Yellow Sea coastline, and the California coast show a high RFI probability in many frequency bands (such as 18.7, 23.8, 36.5, or 89 GHz) and instruments. shows correspondence between these areas and the so-called upwelling regions of the ocean, of which the four main ones are the Chilean-Peruvian coast, the California coast, the Angola-Namibia coast, and the Mauritanian coast. The water anomaly observed by this effect in the Canary and California currents can reach 4 K after long time averaging and could explain the detection in these areas . For analogous reasons to the sea ice edge false alarms, softening the thresholds might result in misdetections in marien regions such as the Yellow sea of the seas around Japan. Other false positives are observed in South America and Africa can be seen in all frequency bands such as the Lakes Titicaca and Poopó. The RFI probability changes over the year with a stable decreasing trend until mid-year, when it starts to increase again. This could be caused by a reduction on the water level during summer until the rainy period in winter.

NWP models compare millions of observed radiances with model-simulated radiances every day as part of the data assimilation procedure. For microwave radiances at ECMWF, this process employs the RTTOV-SCATT radiative transfer model , simulating radiative effects from the surface, atmospheric gases, clouds, and precipitation – a so-called all-sky simulation. The observed radiance (O) is compared to the model's short-range forecast simulation of the observation, known as the model background (B), interpolated in time and space to the observation location and with the same viewing geometry. This is the first step of the assimilation of the observations, which will subsequently lead to updated model fields. In addition to offering information to update the model state, analysis of departures between observations and model (O-B) can be instructive for monitoring the quality of observations themselves. It is worth noting that in this study the full spatial resolution of radiances are analyzed (i.e. with no averaging applied), in contrast to typical pre-processing of observations before assimilation at ECMWF that includes averaging (superobbing) for most microwave data.

Simulations of the NWP model background do not include artificial sources. The background thus represents a realization of a pristine RF environment, one with natural sources of emission only. Analysis of departures can therefore aid in the identification and quantification of RFI sources, as the natural range of O-B variability is known very well in most cases. For example, over the ocean at low frequencies that are primarily sensitive to SST, O-B variations are typically within about 1 K. This provides a tight constraint on expected behavior, allowing abnormal radiative signals to stand out as likely contamination. Instantaneous departures can show these anomalous scenes, but the real power of NWP-based monitoring is in the aggregate, with signals able to emerge even in noisy sampling over many days and weeks. Crucially, RFI should manifest as a positive O-B signal only, whereas other mismatches between the model and observations should be distributed between positive and negative departures. For example, a misplaced cloud in the model background will show up at a cloud-sensitive frequency band as a positive blob next to a negative one, whereas a region of RFI should be overwhelmingly positive. Even if RFI is coincident with cloud and precipitation, the large amount of sampling afforded by NWP-based analysis means that these signals should still be clear in time-averaged maps and histograms.

In this validation analysis, only results from AMSR2 low frequency channels are presented, covering March and October of 2022. This is to focus on frequency bands in which significant contamination was found. Fuller analysis can be found in the technical report from , and comprehensive analysis of AMSR2 performance in the ECMWF system can be found in .

This paper described an assessment of key passive microwave bands exploited for Earth Observation from the 6.4–7.3 GHz band to 174–191 GHz. First the EORFIScan software has been described, which applies multiple RFI detection techniques to EO brightness temperature products. This part includes a description of the RFI detection techniques and the methodology to compute the thresholds. The software has then been used to analyze a year of AMSR2, AMSU-A, MWHS-2, and AMR-C data in order to cover all main EO passive microwave bands beyond the L-band. The methodology has been assessed for AMSR2 and AMSU-A via comparisons with NWP model equivalents, with results presented for the 6–19 GHz bands on AMSR2, where the large majority of corroborated RFI was found. RFI probability maps for 2022 have been computed as the number of detections over the number of observations accounting for all polarizations and orbital passes.

The results have shown heaviest contamination in the 6.4–7.3 GHz band, more moderate contamination at 10.7 and 18.7 GHz bands. The identified areas of systematic contamination have been assessed using all-sky NWP departures, and in most cases are found to correlate strongly with positive departures as would be expected for skilful flagging of RFI. Several new sources are observed in these bands with respect to the literature and corroborated by the NWP-based analysis. The emergence of 6.9 GHz contamination over sea is particularly troubling, as this is a key band for all-weather SST and sea-ice measurements , and will be flown on the upcoming CIMR mission from ESA as well as on the successor instrument to AMSR2, AMSR3. First RFI signatures have been found at 23.8 and 36.5 GHz in AMSR2.

Consideration of RFI detection systems in these bands is strongly advised as part of future mission planning, as the contamination observed already has an impact on users and may be expected to worsen in the upcoming years. At bands from 50 GHz and up, no significant contamination has been detected in the dataset under consideration. However, the constantly evolving threat of RFI in EO bands would benefit greatly from monitoring in real time, especially for bands of consequence for NWP and other operational applications. It is clearly suboptimal to discover new sources of contamination a year or two after those radiances have been ingested into a variety of operational models and retrievals, and thus real-time monitoring would provide significant benefit to a wide community of users.

It is reasonable to expect that high-frequency technology will become more accessible, low-frequency bands will become more crowded, and new services will require greater bandwidths in the near future. As a result, higher frequency bands may start showing contamination and RFI at lower frequency bands could become more widespread than it already is, which will threaten their use for Earth observation. Given that new missions take years from design until the end of their operational lifetime, new EO missions should include RFI detection and mitigation systems to be prepared for the expected worsening of the RFI environment.