GNSS-Scatterometry Challenges in Lake Ice Remote Sensing

Delay-Doppler Maps (DDMs) have already shown promising potentials in the sea ice remote sensing, including ice recognition and sea ice thickness estimation. However, this kind of products is yet to be tested for lake ice remote sensing, which looks more challenging than that of sea ice, especially for mid-latitude ones.

Generally speaking, two main approaches have been established so far for GNSS-R remote sensing: altimetric and scatterometric approaches. Although these two approaches are somehow connected to each other, the methodologies can be way apart. Accordingly, ice remote sensing using GNSS-R can be, and has been, conducted in both ways.

Among many scatterometric-based research items for ice remote sensing, Qingyun Yan has successfully carried out and published several papers on sea ice remote sensing by analyzing DDMs obtained from TechDemoSat-1 (TDS-1) satellite over northern regions. He has also introduced the feasibility of sea ice thickness measurement by extracting brightness temperature values from DDMs reflectivity observations. However, no GNSS scatterometry experiments have been yet done over lake ice to examine the GNSS-R capability of lake ice remote sensing by means of DDMs.

As part of an experiment that I’m currently involved in regarding the DDMs potential of lake ice remote sensing, I started exploring challenges that one may face in lake ice remote sensing using scatterometric data. Lake ice covers, especially mid-latitude ones, as discussed in some previous posts of this blog (e.g., this one), are, in general, highly influenced by mid-winter warmings and melt/refreeze events, which significantly change the reflectivity properties of the lake ice reflective layer(s). Therefore, I came up with two main challenges as follows.

Firstly, in GNSS scatterometry, it’s usually assumed that we’re dealing with a single scattering regime, i.e., only one surface contributes to the reflection. This assumption in mid-latitude frozen lakes is improper because, as mentioned above as well as in my master’s thesis, mid-winter warming events leads to the appearance of some wet layers inside and over the ice column causing interferences in reflections. In other words, depending on the water content of each contributing layer, which controls the dielectric permittivity of overlaying layers, these layers may cause a multi-reflection regime, which differs the scenario from what has been defined in the GNSS-R tutorial.

secondly, melt/refreeze events during the freezing season change the ice-water interface roughness significantly, which may change the amount of the reflected power received by the satellites. In the literature, some theoretical models, such as Kirchhoff Approximation and Integral Equation Method, are applied to simplify assumptions regarding model the roughness of reflective surfaces. However, ice-water interfaces at frozen lakes are less likely to be adopted with those models; therefore, non-specular reflections are very likely to happen from lake ice towards GNSS-R receivers.

Furthermore, topography and EIRP effects have been discussed here an here, respectively; however, we’re actively exploring for other potential sources of inconsistencies. Feel free to reach us out and contribute if you have ideas on that.

Importance of Effective Isotropic Radiative Power (EIRP) for GNSS-R Lake Ice Remote Sensing

GNSS-R applications for lake ice remote sensing is still under study, and there are many vague points in this subject, especially for those like me who are “encore vert” in this topic. Among various corrections and calibrations must be made on GNSS-R data, EIRP looks decisive; however, a genius solution has been recently proposed for that: dynamic calibration.

Postdoctoral researcher with the University of Michigan Tianlin Wang, who is known with his insightful research items on microwave instrumentation of GNSS-R systems, has recently published a paper expanding the concept and application of the dynamic calibration of EIRP for GNSS-R remote sensing. This research item, which has been conducted in co-authorship with a number of GNSS-R giants, such as Dr. Chris Ruf and Dr. Scott Gleason, is in the following of a conference paper that had been presented at IGARSS 2019. In this research, authors have discussed the necessity of various calibrations should be applied to normalized bistatic radar cross section (NBRCS) measurements as it is a key to obtain physical insights into the mechanism of GNSS-R scattering from reflective surfaces. Among those calibrations, the one applied to EIRP has been discussed as one of the most challenging as it has to overcome multiple major challenges including variations in the transmit power, inaccuracies in antenna gain measurements, and the flex transmit power. Being mentioned as the most important motivation of this research, the latter challenge, i.e., flex transmit power, will cause changes in CYGNSS level-1 NBRCS values, which may limit the mission’s potential by flagging out up to 37% of observations. In this dynamic calibration method, direct GPS signals are used to estimate the EIRP in the direction of specular reflection point, and by implication, calibrate the NRBCS.

It is my understanding that this correction is crucial specially for CYGNSS as it is classified among conventional GNSS-R approaches (vs. interferometric GNSS-R), in which the narrower bandwidth of the public codes (for more information on iGNSS-R and cGNSS-R, and their application in ice altimetry, see this paper). Moreover, the calibration quality of NBRCS could be extremely decisive as the roughness of the reflective surface increases. Roughness can be really challenging in lake ice GNSS-R remote sensing as I have mentioned here and here. Moreover, under this LinkedIn post, which was about the same topic, two GNSS-R researchers (Ben and Lucy) discussed various sources of uncertainty in CYGNSS remote sensing specifically for lake ice studies. Tianlin’s paper seems to open a new way to improve CYGNSS ability in lake ice remote sensing.

GNSS Reflectometry Using Smartphones; A Potential for Lake Ice Thickness Monitoring

SNR-based GNSS-Reflectometry has shown an enormous potential for ground-based altimetry purposes as the frequency of SNR oscillations is directly connected to antenna height from the reflective surface. This technique, which is usually called in the literature as GNSS Interferometric Reflectometry, has been already tested for retrieving snow depth, monitoring water level changes, and measuring lake ice thickness. The considerable advantage of this technique can be listed as the low costs of equipment since a simple standard GNSS receiver would be enough for running a GNSS-IR experiment.

Recently, Cemali Altuntaş, a researcher from Yildiz Technical University, Turkey, has evaluated the potential of android smartphones for the GNSS-IR experiment and antenna height measurement. This well-structured research, which is going to be published in a couple of weeks by Digital Sensor Processing, Elsevier, analyzed the SNR-based retrieved heights obtained from a single-frequency GNSS receiver embbeded in a Xiaomi Mi 8 Lite smartphone, compared with those recorded by a Trimble NetR9 geodetic receiver, and then validated with in-situ measurements. Results show a stunning performance by the smartphone unit compared to the geodetic receiver in terms of height residuals.

As I am not an electrical specialist, I have no clue on why a low-cost GNSS receiver designed for daily routine jobs shows better results in comparison with a standard geodetic receiver, which has been always used for highly professional geodetic observations. Although the positioning precision of Xiaomi Mi 8 Lite has been tested and shown to get close and be comparable to geodetic receivers, but in terms of SNR-based reflectometry, I would be grateful if a specialist could make it a bit clear whether it is related to the noise amount, antenna gain, or any other possible reasons.

Considering the advantageous potentials of smartphone GNSS receivers for reflectometry purposes, one may expand the mobile-based GNSS-IR to a wider range of remote sensing applications, especially those requireing massive data, which can be collected by non-specialists individuals. As an example, GNSS-IR has shown a promising potential for the freshwater ice thickness measurement using a single geodetic receiver. In a study, we placed a GNSS antenna on the ice surface allowing it to receive reflected GNSS signals from the ice-water interface to measure lake ice thickness. Lake ice is known as a very common place for many people who live in cold regions, such as Canada, to spend hours over for, for instance, ice fishing. Therefore, smartphone-based GNSS-IR may offer a new opportunity for individuals allowing them to just leave their smartphones in a short distance away from themselves on the ice surface during their on-ice activities in order to collect reflected GNSS signals from the lake ice. This contribution may be aggregated by a central server to provide a wide coverage of lake ice thickness data with a demonstrated accuracy in our paper.

Can Small Changes in Water Surface Topography be Detected by GNSS Reflectometry?

Precipitation detection over ocean using various radar remote sensing techniques have been fully understood and tested for several decades. But in terms of GNSS Reflectometry, researchers had been skeptical about its potential as a rain detector. Recently, a groundbreaking paper has been published by Dr. Milad Asgarimehr and his colleagues addressing this question cropping up in another paper published three years ago.

Having done his Ph.D. with Technische Universität Berlin, Berlin, Germany, Milad is currently doing his postdoctoral research on improving GNSS-R observations with AI at the German Research Centre for Geosciences (GFZ). Although the GNSS-R capability of rain detection had been doubted at the beginning, Milad’s recent study suggested that precipitation affects the power of reflected GNSS signals, which can be critical in GNSS-R experiments as we prefer to seek for unaffected data.

Regardless of microwave remote sensing, the effect of rain on ocean waves and currents has been simulated in laboratory experiments. As an example, this study designed a circulating wind-wave tank to simulate rain drops effects on ocean surface waves showing that ring waves caused by rain drops can amplify the surface roughness at centimetre scales although it can attenuate gravity waves over ocean surfaces. However, these laboratory studies could not fully explain the mechanism of rain effects on ocean surfaces in real environment.

In this recent study, Milad tested the rain effect by analyzing polarimetric GNSS observation at a GFZ coastal GNSS-R station at Onsala Space Observatory in Sweden equipped with two side-looking antennas at two different polarizations (RHCP and LHCP). Among multiple results obtained from this experiment, including sea surface salinity changes due to rain events, the key one was discerning rain drops in the power of RHCP and LHCP reflections over a calm sea. Analyzing the I/Q components of the reflected signals, authors has shown that the received signal power at a low-wind-speed condition (<5 m/s) is reduced due to the diffuse scattering caused by rain-drop-made ring-waves over sea surface.

Reviewing this paper, I came up with this question if we could study ice bottom roughness using GNSS-R tools as either a ground-based or a satellite payload sensor. As discussed in a previous post, mid-winter warm temperature may cause changes in ice roughness, and this variation can be significant enough to be discerned by reflected GNSS signals. In addition, the water underneath the ice is so calm that satisfies the weak diffuse scattering regime condition, similar to the low-wind-speed condition Milad suggested in his paper. To see if GNSS-R has the potential to be used as a roughness indicator, the same steps can be taken as Milad has; however, a very big challenge is ahead of us in lake ice remote sensing: multi-layer scattering! In most GNSS-R studies, the reflection is assumed to be from a single reflective layer, e.g, sea water surface, but in lake ice, as shown in my master’s thesis, there are usually multiple reflective layers unless we experience consecutive days with a cold temperature so we can neglect undesired L-band reflections from layers rather than the ice-water interface.

To be specific, our Haliburton experience, which has been explained in Chapter 4 of my master’s thesis, clearly reflects the temperature effect in the power retrieval process (see this post). But aside from the temperature effect, Milad’s paper shows that GNSS-R is able to detect topography changes in small amplitudes. In other words, if the temperature is cold enough to avoid wet layers, ice bottom roughness may change the power of reflected signals as GNSS-R has well shown the potential to discern the rain-caused roughness in the order of ~5 to 50 mm. However, the other big challenge we may face is ancillary data about real ice bottom roughness, which seemingly necessitates another lengthy fieldwork. Dr. Grant Gunn has recently conducted a research on freshwater ice roughness and capacity (will be bulished by Jun 2021) summerizing multiple methods of ice roughness retrieval and explaining in-situ measurements they have done for this purpose in the Straits of Mackinac region of the Great Lakes between Michigan’s upper and lower peninsulas.

CYGNSS and Spatial Interpolation; a Review

“Why talk to others when you can talk to yourself?” This is how Clara Chew captioned her fantastic YouTube video in LinkedIn to start suggesting a spatial interpolation method for CYGNSS data.

First of all, I should say what a great idea form one of my most inspiring and favorite researchers. Establishing a YouTube channel to talk to ourselves about GNSS reflectometry is what we’d need, and to be honest, that’s the main encouragement for myself to kick off this website. Although I am currently the only writer here, but this platform is going to become an online magazine for GNSS-R researchers to freely read and write others’ opinions. It’s yet to be widely advertised, but the core idea is the same: let’s just talk to ourselves : )

About the video, let’s start from the end; this method is not going to be used for every application. “If you are trying to do some sort of analysis where you need exact knowledge of reflectivity at a certain point,” Clara clarifies “you might not want to use this method”. It sounds very true for my current research focusing on Qinghai Lake, Tibet Plateau, where I restricted Fresnel Zones to only the central region of the lake to make sure that I am not receiving reflections from the surrounding lands or even the shore lines. So, no interpolation is required at any level. But aside from the Qinghai’s research, it’s been a few months that I’ve been thinking about expanding my CYGNSS explorations to a wider region, say central desert of Iran, and meanwhile, one of my greatest concerns has been how to do data interpolation over a large area. Clara has also mentioned it somewhere in her video that “trust me, it won’t be too frustrating”, but it is for me :)) because I haven’t started yet.

But when I’m saying my concern is on how to do it, I wouldn’t say that choosing the interpolation method is not a big concern, as it really is, but it is not the biggest issue for me, because I had practiced a method years ago. The story took place in 2013-2015, when I was working on geodynamics of the Earth, and my specific research at that time was calculating strain tensors. I found out that routine methods for interpolating crustal movement values, e.g. IDW and Spline, are way off the road. I spent couple of months on that, and realized that “deterministic interpolation methods” do not perfectly work for those problems which require “geo-statistical interpolation methods”, such as Kriging. That was the point.

In short, despite deterministic methods which only consider distance as the effective parameter in weighting procedure, geo-statistical interpolation methods also take the correlation into account as a weighting function. This is exactly what Clara suggested: “correlation coefficients“. In Kriging, for example, a covariance matrix is created among node points based on statistical constraints, and then, that matrix plays a key role in establishing the weighting function. The same idea can be seen in another powerful interpolation method called Least-Square Collocation, which is widely used in geoid problems and potential field anomaly interpolations.

I employed Ordinary and Universal Kriging interpolation techniques for estimating strain matrix components, and improved the accuracy up to 70%. As Clara has remained the discussion open at the end of her video, I would invite her to a collaboration in order to explore the ability of geo-statistical methods of interpolation in CYGNSS cases.

Temperature Effect on Reflected GNSS Signals from Mid‐Latitude Lake Ice; Review and Discussion

Our poster was presented on the first IAG’s conference-workshop on “Geodesy for Climate Research” which is currently (March 29-31, 2021) being held online. The idea of this poster, as can be understood from the title, was on examining the temperature effect on GNSS reflectometry over lake ice.

During the freezing season of 2019-2020, a GNSS-R experiment was conducted at a mid-latitude frozen lake called MacDonald Lake to examine the ability of GNSS interferometric reflectometry for lake ice remote sensing. The initial plan was analyzing GPS signals reflected from the lake to monitor ice formation and generate an ice thickness chart for those dates. However, due to natural features of temperate lake ice covers, such as warm temperatures and mid-winter melt/refreeze events, ice-thickness estimation using GNSS-R was not as straightforward as other routine GNSS altimetry applications conducted before over snow fields or sea surfaces. Meanwhile, the footprints of other lake ice features, such as slush and wet snow, were appeared on the recorded GNSS-R data.

Among those side-parameters, temperature was chosen for this poster. One may think that, by the word temperature effect, I am talking about the tropospheric effects on GNSS signals, such as delay and bending effects, which do matter and have been evaluated for near surface GNSS-R receivers, but I do not mean that. Except for the tropospheric effect, and also the electrical aspects of GNSS receivers, temperature does not directly impact reflected GNSS signals. The point is that temperature causes mid-winter melt/refreeze events in lake ice and, by implication, changes the roughness patterns of reflective layers. Therefore, roughness is the key driver affecting the reflected signals, which is itself affected by temperature variations.

To be specific, for this experiment, we used the SNR-analysis approach in which daily SNR files were extracted and their frequency were calculated to retrieve the antenna heights above reflective surfaces, which could be the ice-water interface, slush layers, or top of the wet snow pack. More details can be found in Chapters 2 and 4 of my master’s thesis. Meanwhile, the power spectrum related to those frequencies showed to be varying over the time, i.e., as shown in Fig 2 in the poster, the SNR amplitude values were not necessarily the same among different dates. Plotting this ups and downs against daily temperatures, as shown in Fig 3 of the poster, we see a harmony, which can be meaningful.

Around Dec 10-15, the freeze-up event happened and a large spike appeared; afterwards, the SNR power values dwindled to almost a third or less and that sharp spike never showed up again. This might be explained by saying when a thin layer of ice appears at the beginning of the freeze-up period, a calm ice/water surface appears whose roughness is much less than that of the lake water in the absence of the ice just before the freeze-up. So, it makes sense if we say the surface became very reflective and a huge reflection happened towards the GNSS antenna. This explanation has been also used in previous GNSS reflectometry studies, e.g. Qingyun Yan‘s research items on sea ice remote sensing through investigating DDMs. However, this reflectivity does not last for a long time as roughness conditions of the reflective layer(s) changes due to lake ice evolution, precipitation, wet layers appearance, and so on. Nevertheless, an agreement can be seen between temperature variations and SNR spectrum vales, so that when the temperature is close to or above zero, small jumps can be seen in SNR amplitudes. To better illustrate this co-variation, Fig 4 is plotted in the poster to illustrate SNR jumps against temperature.

As shown there, when temperature rises, red lines appear in “dAmplitude” values meaning that there are downward jumps in the SNR spectrum. It is not always true, but it mostly happens. Besides, a delay may be seen in this co-variability, which may be because of the time needed for melt/refreeze events to proceed. Meanwhile, one of our co-authors, Justin, who is currently doing his Ph.D. on lake ice remote sensing with the University of Waterloo, raised questions about my explanation and said “as shown in Fig 3, an increase in temperature mostly causes an increase in SNR amplitudes”. This statement retracts from the validity of the poster idea. “This is also agree with what we have in the physics of remote sensing”, Justin added “where wet layers appeared due to melt event leads to a higher dielectric constant and results in stronger returns”.

Justin’s criticism elicited my response by referring to bi-static radar properties where receivers and transmitters are not collocated; so an increase in dielectric constant does not necessarily bring about an increase in reflection strength towards the antenna. Instead, roughness has the main effect even if it happens at very small scales, as recently shown in Milad Asgarimehr‘s studies on the effect of small-scale roughness changes on GNSS signals reflected off oceans. However, my explanation may not look satisfying as those warm days are only some occasions; so, significant trends yet to be derived from this new idea with a short time series. It should be noted that Justin justly mentioned to the refreeze event as a concomitant of the melt event and its significant impact on the roughness, because I hadn’t included the refreeze event and only mentioned to the melt event in the first draft of the poster.

The bottom line is that climate change may show it’s effects on GNSS signals reflected off frozen lakes; however, further investigations are required for achieving to a firm conclusion. Longer time series may be required for future studies to extract more meaningful trends and correlations.