Symposium on the hydrometeorological usage of data from commercial microwave link networks

Jun 25, 2019, 8:48 AM → Jun 28, 2019, 2:00 PM Europe/Berlin

large seminar room (KIT Campus Alpin)

Group picture on Day1

 

Background
The opportunistic usage of attenuation data from commercial microwave link (CML) networks for hydrometeorological applications has seen increasing adaption over the last years. This additional rainfall information can be a valuable complement to existing rain gauge and radar observation or it can provide significantly improved spatial coverage of rainfall observations in data scarce regions.

Objectives of the symposium
    1.    Present and discuss the newest research results
    2.    Foster exchange between the involved research groups
    3.    Initiate discussion with stakeholder, e.g. hydrological services and cellular operators
    4.    Identify pathways towards operational usage

Program overview

On Day1 the research groups involved in hydrometeorological usage of data from CML networks will give an overview of the current state of the art. In addition stakeholders will be able to present their view on the usage of CML data and on tackling the hydrometeorological challenge of rainfall observation in general. Day1 will present the unique opportunity for researchers and stakeholder to discuss requirements and pathways for future hydrometeorological CML applications.

On Day2 and Day3 researchers will present and discuss specific topics of the hydrometeorological usage of data from CML networks, e.g. processing methods, spatial reconstruction, data fusion, data formats, benchmark data sets, standard performance metrics, naming conventions, etc.. Discussion rounds will also focus on potential future collaborations and joint proposals.

Day4 will be used to conclude the tasks from Day2 and Day3 and to wrap-up.

 

Privacy_Policy_HyMet_CML_symposium.pdf

Tuesday, June 25

8:48 AM Arrival with local bus line number 2

1 Opening

Speaker: Prof. Harald Kunstmann

2 Keynote 1: How can urban water management benefit from high coverage by cellular networks?

Speaker: Dr Vojtěch Bareš (Czech Technical University in Prague)

3 Keynote 2: GPCCs contributions and recommendations towards an integrated global high-quality and accessible precipitation monitoring

Speaker: Dr Andreas Becker (DWD)

10:30 AM Group picture outside

10:35 AM Coffee

Overview presentations: HyMet CML research overview of the different countries and institutes 1

12+3 minutes

4 Country-wide rainfall estimation with CML data in Germany: The past, the present and the future

Speaker: Dr Christian Chwala (KIT (IMK-IFU) / Uni Augsburg)

5 Wet Antenna Attenuation Estimation by Monitoring the Reflection Coefficient

Estimating near ground precipitation from permanent recordings of received signal fading in commercial microwave links (CMLs) is a promising and emerging technique in weather observation. It is particularly valuable for regions with sparse or even nonexistent weather radar and rain gauge coverage to support hydrological modeling and to enhance regional severe weather warning.

Wetness on the antenna cover due to rainfall or dew leads to signal attenuation that occurs in addition to the actual rain induced path attenuation of interest. Lack of knowledge of the wet antenna attenuation causes overestimation of the link attenuation derived rain rate. Hence, the wet antenna effect is a significant source of error in the processing of link attenuation data into rain rate. Identifying and estimating it improves the accuracy of CML based precipitation mapping.
We developed an approach to estimate the wet antenna attenuation by monitoring the connection-side reflection coefficient of the antenna. Using this information leads to a feasible procedure for correcting the wet antenna effect in path attenuation time series and consequently improving the quality of the rain rate data. To demonstrate the practicability of this approach we present wet antenna detection and correction based on data from a new atmospheric microwave transmission experiment where both the link's transmit-receive signal level (TRSL) and the reflection coefficient of customary CML reflector antennas are simultaneously measured.

Speaker: Mr Clemens Moroder (Technical University of Munich)

6 Inherent limitations in 2-D rain retrieval from CMLs

One of the main attractions in in using CMLs for hydrology is in their potential for creating 2-D near-ground rainfall maps. Since 2008, various methods have been presented for 2-D rain retrieval from CMLs and their operation has been demonstrated in different regions around the world. In this talk I will present results of our theoretical study under the IMAP project of the achievable spatial resolution that can be achieved by CML-based rainfall mapping, as well as its accuracy. In particular, we have studied the potential accuracy improvement that can be achieved when using a CML as a line rather than a point for rainfall mapping based on spatial interpolation.

Speaker: Prof. Hagit Messer-Yaron

7 Other-than-rainfall monitoring of moisture, fog and dew

Air-moisture maps were generated based on commercial microwave links data over N. Israel and compared with ECMWF ERA-Interim maps. Skill scores were calculated utilizing 39 weather stations. High correlations range, 0.75–0.9, were found between the new links’ humidity fields and the stations. The numbers of best correlated links are found to be higher compared to ECMWF, and in the measure of standard deviation the links performed better. However, for the mean humidity, the ECMWF is doing better. The links’ humidity thus provides a more accurate picture of the observed moisture as compared to current weather prediction product. Other-than-rainfall monitoring of fog and dew will be reviewed

Speaker: Prof. Pinhas Alpert (Department of Geophysics, Tel Aviv University, Israel)

11:55 AM Lunch

Overview presentations: HyMet CML research overview of the different countries and institutes 2

12+3 minutes

8 Commercial microwave link rainfall estimation for the Netherlands and upscaling to developing regions

Commercial microwave links (CMLs) from cellular communication networks have been used for rainfall monitoring in the Netherlands since 2003. We provide an overview of our work in the Netherlands with an emphasis on country-wide application using CML data from T-Mobile NL from networks with different sampling strategies. The open-source R package RAINLINK is described, explaining the processing steps of the rainfall retrieval and mapping algorithm. Recent as well as planned updates are discussed. Both CML-based and satellite rainfall maps are compared with gauge-adjusted radar data. Finally, upscaling of CML rainfall monitoring to developing regions is presented.

Speaker: Dr Aart Overeem (Royal Netherlands Meteorological Institute / Wageningen University & Research)

9 Rainfall detection by wireless microwave links in Italy: the story so far

This contribution overviews the state-of-the-art of the research dedicated to rainfall monitoring through wireless microwave links in Italy. Early studies on the application of tomography to retrieve 2D rainfall fields from a mesh of terrestrial microwave links date back to the nineties. A pioneering experimental work was carried out in 2001 in the framework of the MANTISSA project. More recently, the pervasive growth of cellular mobile networks as well as the usage of Ku- and Ka-band telecommunication satellites have boosted this research area. Currently, several projects are active in Italy: some of them exploit signals transmitted across terrestrial networks (RAINBO and MOPRAM), whereas some others make use of satellite links (NEFOCAST and Living LAB). In particular, data collected from Commercial Microwave Links (CML) are key assets of RainBO and MOPRAM. The former is aimed at developing and improving methodologies and tools to mitigate the impact of extreme rainfall events over the Emilia-Romagna region (Bologna and Parma provinces mainly), by implementing, among others, a new monitoring infrastructure based on CMLs. The MOPRAM project (MOnitoring of PRecipitation through A network of Microwave radio links) aims at developing an innovative and efficient tomographic technique to track the spatial variability of rainfall fields for hydrological applications. The 2D rainfall maps retrieved from CML data are used to feed hydrological models and estimate the flow of river basins in two pilot areas in Lombardy.

Speaker: Roberto Nebuloni (Consiglio Nazionale delle Ricerche (CNR))

10 Opportunistic Use of Data from Satellite Communication Networks for Rainfall Estimation

There are more than 300,000 satellite ground terminals across Europe and 2 million worldwide which are being used for providing broadband internet services. To maintain a certain level of quality of service, gateway stations (of satellite operators) continuously monitor links between satellite and ground terminals (enabled by the bidirectional link). Carrier-to-Noise ratio (C/N) parameter denotes quality of the received signal and the link. The C/N parameter is highly dependent on the link condition and is mainly affected by the rain attenuation at the operational Ka/band frequencies. By using appropriate signal processing and machine learning techniques, we are able to estimate the rainfall from C/N measurements to a high accuracy.
In this presentation, we will briefly overview our relevant research activities, which were conducted in collaboration with University of Luxembourg. We will present operational set-up used for rainfall estimation. It includes 35 satellite ground terminals in southwest of Germany (Eifel region) which receive services from ASTRA 2F satellite located at the orbital position of 28:2 E. The links from satellite to ground terminals operate at K-Band (19:70-20:20 GHz). C/N data from these terminals are collected every 5 minutes and stored in a database. We will introduce a tool developed for data collection, processing and visualization.

Speaker: Dr Ahmad Gharanjik (Databourg Systems SARL-S)

11 3-D Rain Rate Estimation from Integrated Measurements of Commercial Wireless and Satellite Links: an overview

Abstract

In the last decade, various algorithms have been developed to provide accurate rainfall maps from measurements of rain-induced attenuation on commercial wireless links (CWLs).
These solutions are able to give precise results but they also require dense terrestrial microwave networks, which have non negligible installation and operating costs.
A cheaper alternative for rainfall estimation is represented by broadcast satellites links (BSLs).
However, estimation of the rain-induced attenuation on satellite links requires complex signal processing techniques, due to the physical structure of these links. Furthermore, to the best of authors' knowledge, data provided by satellite links only cannot be used to properly estimate a precise rainfall map of an area.

To overcome these problems, we studied a mixed approach based on data given by both CML and BSL, able to provide a three-dimensional (3-D) rain rate map of a monitored area.
The proposed joint approach gives us remarkable appealing advantages:
1. Efficiency: both CWLs and BSLs exploit already existing wireless infrastructure, at no extra costs for the required equipments, the installation and operating conditions.
2. Coverage: CWL coverage can be improved by including satellite terminals already installed at domestic premises for TV reception. Additionally, more satellite devices can be purposely installed in areas not adequately covered by terrestrial microwave links, where the deployment of conventional methods of observation, as rain gauges and weather radars, is impractical.
3. Diversity: measuring the signal levels coming from different links, terrestrial and satellite, provides a diversity gain which is the key to improve the accuracy and reliability of the overall joint system.
4. Accuracy: the numerical results obtained by simulations corroborate the effectiveness of the proposed mixed 3-D strategy and quantify the improvements over the conventional systems based on CWLs only.
\end{itemize}

Our contribution lies in the description of the data processing schemes developed to retrieve these maps.
In particular, the first part consist in pointing out the problems connected to the satellites links and a signal processing approach able to provide precise data of estimated rainfall.
Then, a summary of the algorithm will given and preliminary results obtained by the proposed approach will be presented, in order to show the effectiveness of our approach.

Acknowledgement

This work is supported by Fondo per le Agevolazioni alla Ricerca and Fondo Aree Sottoutilizzate (FAR-FAS) 2014 of the Tuscany Region, Italy, under agreement No. 4421.02102014.072000064 SVI.I.C.T.PRECIP. (Sviluppo di piattaforma tecnologica integrata per il controllo e la trasmissione informatica di dati sui campi precipitativi in tempo reale).

Fruitful discussions with prof. Hagit Messer of the University of Tel Aviv are greatly acknowledged.

Speaker: Fabio Saggese (University of Pisa)

Short presentations: HyMet CML research overview of the different countries and institutes

12 Combining microwave links and meteosat second generation data for rainfall studies in Kenya

Speaker: Kumah Kwabena Kingsley (University of Twente / ITC )

13 Rain rate retrieval test from millimeter wave link measurement in Beijing

Speaker: Congzheng Han

14 Improve rainfall measurement using CML in Melbourne

Speaker: Jayaram Pudashine (Monash University / Wageningen University & Research)

15 First results of CML rainfall estimation in Palestine (presented by C. Chwala)

Speakers: Christian Chwala (KIT (IMK-IFU) / Uni Augsburg), Dr Sameer Shadeed (An-Najah National University )

16 CML rainfall estimation in Burkina Faso (presented by C. Chwala)

Speakers: Christian Chwala (KIT (IMK-IFU) / Uni Augsburg), Prof. François Zougmoré (University of Ouagadougou - Laboratoire de Matériaux et Environnement LA.M.E)

Posters with coffee and cake: HyMet CML research overview of the different countries and institutes

17 Applying CML rainfall estimation at regional and urban scale in Nigeria, Sri Lanka and Bangladesh

We present the results of three pilot studies for high-resolution rainfall monitoring using CML networks over Nigeria, Sri Lanka and Bangladesh in collaboration with local gsm operators. We focus in particular on several densely populated urban areas with high link network density. We also discuss some of the challenges and opportunities regarding continental-scale rainfall monitoring using microwave links from cellular communication networks. We show comparisons with satellite products and rain gauges.

Speaker: Thomas van Leth (Wageningen University & research)

18 Opportunistic use of commercial microwave links and meteosat second generation data for rainfall studies in Kenya

Rainfall and its spatiotemporal variation are essential for many operational and research applications in Kenya. However, the existing meteorological infrastructure in Kenya is poorly suited for adequate rainfall monitoring. Here, we present the commercial microwave link (CML) data from Kenya and highlight the opportunity that lies in the combination of CMLs and MSG data for rainfall monitoring in Kenya.

Speaker: Kumah Kwabena Kingsley (University of Twente / ITC )

19 Rain Rate Retrieval Test from Millimeter-wave Measurement in Beijing

Accurate rainfall monitoring is important, and Rain Rate Retrieval Test from 25 GHz, 28 GHz, and 38 GHz Millimeter-wave Link Measurement in Beijing Congzheng Han, Yongheng Bi, Shu Duan, and Gaopeng Lupreviously it has been shown that microwave backhaul link between adjacent base station towers can be used for rainfall estimation. However, with deployment of 5G using millimeter technology, there is opportunity for dense rainfall monitoring network using both backhaul and data transmission links. This paper presents our millimeter-wave measurement results during rainy days in Beijing, China. Our measurement design is an exemplary Line-of-sight (LOS) transmission link. We show that it is possible to retrieve path averaged rain rate from the measurement and to use both millimeter-wave transmission link and backhaul link to assist rainfall monitoring. Compared to the local rain measurement from rain gauge and disdrometer, we show that the correlation value of millimeter-wave link retrieved average rain rate varies between 0.6 and 0.9 for different rainfall events. There is room for improvement, and we find that if we can monitor the received signal for a sufficient long period of time, we can quantify the bias due to fading and work out a better estimate of the rain retrieval model's reference level.

Speaker: Dr Congzheng Han (Institute of Atmospheric Physics, Chinese Academy of Sciences)

20 Rainfall retrieval using commercial microwave links in Melbourne, Australia

In this study, we present the first evaluation study of rainfall retrieval using commercial microwave link data for the oceanic temperate climate using one year of received signal level data collected from 115 microwave links (recorded at 15-minutes interval with minimum, maximum and average sampling strategy). The open-source package RAINLINK was applied to retrieve the rainfall intensities confirming that the package can be applied for the average logged data as well. Rainfall intensities for all links are validated with the path-average rainfall intensities obtained from the gauge-adjusted radar product and automatic rain gauges. Comparison of our results showed that average sampling outperforms Min/Max sampling regarding both the occurrences and quantity at a 15-minutes sampling rate for the Melbourne climate.

Speaker: Jayaram Pudashine (Monash University / Wageningen University & Research)

21 The challenges along the CML data processing chain and how we tackle them

Speaker: Christian Chwala (KIT (IMK-IFU) / Uni Augsburg)

Posters with coffee and cake: Specific HyMet CML research topics

22 Analysis of RSSI of signals sampled from an operational E-Band wireless communication network

Received Signal Strength Indicator (RSSI) refers to a measurement of the power of a received radio signal, and it is usually expressed in decibels relative to a milliwatt (dBm) from zero to -120dBm and the closer it is to zero, the stronger the signal is. This work represents an extensive analysis of RSSI of signals sampled from an operational E-Band wireless communication network. The data was gathered and analyzed during the past year (Feb. 2018-Feb. 2019). The network consists of 34 hops, covering the center of the city of Rehovot, a city located in the center of Israel. The analysis covers both dry periods which reveal periodicity phenomenon of the RSSI values and wet periods which thanks to the high sampling time resolution (30 seconds) rain cell movements can be seen in a street level scale.

Speaker: Ms Mor Hadar (Tel-Aviv University)

23 Classifying fog depth using data from commercial microwave links

Developing tools for reliable spatial mapping of fog with limited effort and low implementation costs is desirable. Commercial microwave links (CMLs) that form the infrastructure for data transmission between cellular base stations have been proven to be most useful for weather monitors including fog and in particular, rainfall sensing . Previous work had demonstrated the ability to generate 2-D fog maps on a national scale from existing commercial microwave network actual measurements based on measurements of the signals level in a given network of CMLs complemented with records from humidity gauges deployed in the region to improve the resulting product. Our paper focuses on fog level classifications into thick, medium or light fog.
Previous works used several fog events for performance evaluation. Here, those events and several more are used for verification of the classification method and to demonstrate the operation of the proposed algorithm for generating a reliable fog levels mapping.
The classification performance was verified against visibility stations spread across the country. We found our proposed algorithm to have outstanding fit to meteorological observations.

Speaker: Mr David Tsidkiahu (Student )

24 Estimating raindrop size distributions using microwave link measurements

We present a novel method of using two or three collocated microwave link instruments to estimate the three parameters of a gamma raindrop size distribution (DSD) model. This allows path-average DSD measurements over a path length of several kilometers as opposed to the point measurements of conventional disdrometers. Our model is validated in a round-trip manner using simulated DSD fields as well as a five laser disdrometers installed along a path. Different potential link combinations of frequency and polarization are investigated. We also present preliminary results from the application of this method to an experimental setup of collocated microwave links measuring at 26 GHz and 38 GHz along a 2.2 km path. Simulations show that a DSD retrieval on the basis of microwave links can be highly accurate. We found that a triple link retrieval provides no added benefit over a dual link retrieval. In practice, the accuracy and success of the retrieval is highly dependent on the stability of the base power level.

Speaker: Thomas van Leth (Wageningen University & research)

25 Path Averaged Rainfall as a Single Point: Unfulfilled Potential or a Good Enough Convention?

Most studies represent the rainfall measured by a CML as a single Virtual Rain Gauge (VRG) in the center of the CML path. We compare the 2-D rain retrieval performances of IDW-based spatial interpolation methods, where CMLs are represented either by one or multiple VRGs. A synthetic rain field was produced sampled by a synthetic CML network. If the size of a rain-cell is sufficiently larger than the average length of the CMLs, representing a CML by more than a single VRG, negligibly improves performances. However, if the dimensions are of the order of the length of the CMLs, using several VRG with the iterative algorithm, utilizing neighboring samples for path distribution assessment, improves performances.

Speaker: Adam Eshel (Tel Aviv University)

26 Performance evaluation of rain level estimation from CMLs

The Cramér–Rao bound (CRB), a common measure of performance estimation, which expresses
a lower bound on the variance of unbiased estimators of an unknown parameter,
represents the case of estimating “no rain” by R ̂=0, i.e. one step estimation.
Commonly, rain estimation is performed in two steps:

1. Detection, model selection
2. Estimation of selected model parameters

We propose a performance evaluation measure for the two steps estimation case.
This performance evaluation measure takes into account estimation errors of the true model parameters, miss detection errors caused when choosing the wrong model and thus not estimating the true parameters, and false alarm errors caused when choosing the wrong model and thus estimating wrong parameters.

Speaker: Taeer Weiss

27 Urban Humidity Field from Commercial Microwave Links

The humidity in the atmosphere has a crucial role in a wide range of atmospheric processes determined by the water vapor concentration in the air. The accuracy of weather forecasts, especially the prediction of rain, is largely determined by the humidity field measured at low atmospheric levels, where most of the atmospheric water sinks and sources. At this level, the absolute humidity variation can be large due to the land covers’ variability. One of the main manmade land covers which has a great impact on the humidity field is the city. Large urban areas are developing and covering more land, causing a significant change in the humidity field above the surface. The total effect of the city is noticeable at a few meters height above the surface, where the city’s water sinks and sources are summed up at the "urban canopy" level. Measuring the general effect of the city requires a wide deployment of instruments at the canopy level, a requirement that is not satisfied by the currently available tools for measuring humidity. A new method for measuring the humidity based on the cellular network is addressing exactly this issue of measuring the humidity at the city canopy level. This method is based on the fact that water vapor in the air attenuates the signal between two antennas (link). A significant attenuation occurs around the resonance line for water vapor of 22.23 GHz. This value is close to the frequency of many links deployed by the cellular companies which are located at ~30m above the ground, a good height for measuring the city canopy. For comparison, the weather stations are located according to WMO rules normally at 2m above the ground. The humidity field around Tel Aviv was retrieved from the cellular links, interpolated using IDW interpolation and analyzed. The calculations were performed for different atmospheric conditions. The results show a well-noticed impact of the city on the humidity field over the Tel-Aviv metropolitan region. Most of the time, the absolute humidity was found to be higher around the city as compared to the city center. The results are compared to weather stations and ERA-Interim, data sources that have a low spatial resolution and/or rely on instruments that are not located at the canopy level. In summary, the new method for measuring the humidity based on the cellular network can provide a better description of the humidity field at the city canopy level and a good assessment of the urban effects on the environment and on rain in particular.

Speaker: Mr Yoav Rubin (Tel Aviv University)

Stakeholder presentations: Stakeholder views on HyMet challenges and CML networks

28 The mobile industry view on CML for rainfall observation: A GSMA perspective

The presentation will introduce the existing work and future plans of the GSMA in the field of mobile technology and mobile data applied to weather services in developing countries. The session will focus specifically on the GSMA’s approach, as a private sector stakeholder, to provide value for mobile operators via sustainable business models for CML data sharing and value addition. I will also present our view on the value CML data can provide to the users by improving key services such as weather forecasts, weather insurance and decision agriculture. I will use examples from our partnership with Wageningen University and mobile operators in Nigeria, Bangladesh and Sri Lanka to test and validate CML opportunistic weather monitoring.

Speaker: Mr Daniele Tricarico (GSMA)

29 An overview of the real-time radar-based precipitation analysis of Deutscher Wetterdienst and plans for including CML data

In this presentation, we will introduce the Deutscher Wetterdienst (DWD) radar network and real-time analysis within the hydrometeorological production system RADOLAN. The key feature is the adjustment of radar-derived precipitation estimates to gauge-based measurements with a temporal resolution of one hour on a 1 km2 raster covering the German territory. The RADOLAN system has been developed in close cooperation with the hydrological agencies of the German federal states and provides real-time high-resolution precipitation data since June 2005.

Within the framework of a project financed by the „Strategic Alliance of German Federal Agencies ‚Adaptation to Climate Change‘“, DWD reprocessed the radar-based precipitation estimates starting in 2001. This RADKLIM data is available from the Climate Data Center of DWD. Using the long-term climatological analysis RADKLIM as an example, we show some measurement-specific challenges in deriving radar-based precipitation products illustrating the potential benefit of using additional CML data in the process.

Within the current project ‚HoWa-innovativ‘, together with several partners, DWD will include CML-derived precipitation information into a demo version based on the operational RADOLAN suite with the goal to provide optimized real-time precipitation information. Together with the hydrological forecasting authorities of the federal state of Saxony this demo product will be used to evaluate the potential of CML data for application in flood forecasting in small catchments.

Speaker: Dr Tanja Winterrath (Deutscher Wetterdienst )

30 Rain monitoring with commercial microwave link networks in Sweden

Since 2015 SMHI, in collaboration with Ericsson and Tre, have been deriving the rain rate from microwave link data in Gothenburg in a semi-operational setup. We present how we measure rain with different networks of rain gauges, radar and microwave link in Sweden and discuss their strengths and weaknesses and how microwave links can help in gaining accurate rain rate measurements. Recently we also extended the microwave link trail to Stockholm and as a result we present verification results from both Gothenburg and Stockholm, where we not only look at the rain rate, but also how it effects hydrological modelling. We will also show the live demo that shows rain rate maps for Gothenburg and Stockholm in near-real time, where processing takes less than one minute.

Speaker: Remco van de Beek (SMHI, Sweden)

31 Automatic classification of fading events in large microwave networks

TBA

Speaker: Dr Jonas Hansryd (Ericsson AB )

4:50 PM Coffee

32 Podium discussion

6:09 PM Transfer with local bus line number 2 to conference dinner

7:00 PM Conference Dinner

Wednesday, June 26

Specific research topics: Long-term and/or large-scale analysis

33 Methodology improvement of microwave link derived rain rate in Sweden

Microwave link derived rain rates show good correlation with ground observations, but can often suffer from large biases when applying only a standard ITU-R equation. The ITU-R equation describes the relation between path attenuation and rain rate, but other effects like wetting of the antenna can occur. Since 2015 SMHI has a semi-operational processing method that tries to correct these effects. Recently a new version of the algorithm was developed that was based on microwave link data gathered between May and November 2015 in Gothenburg, Sweden. These data were used to derive a attenuation- rain rate equation by directly comparing these data with rain gauge data. The resulting equation was tested on an independent dataset with a similar climate, namely Stockholm, for the period of 28 Juli-31 October 2018. The results are very promising and will be presented.

Speaker: Remco van de Beek (SMHI, Sweden)

34 The challenge of processing one year of German-wide CML data

Since September 2017 we gather data from around 4000 CMLs distributed over all landscapes of Germany at a temporal resolution of one minute.
We present a fast, parallelized workflow with a processing scheme composed from methods from literature, that were adjusted to our data set. This includes the comparison of different methods to classify wet and dry periods and to compensate for wet antenna attenuation.
We derive rainfall estimates which correlate well with a radar based, gauge-adjusted rainfall data set
from the German Meteorological Service on an hourly, monthly and seasonal scale.

Speaker: Maximilian Graf (Karlsruhe Institute of Technology, IMK-IFU, Kreuzeckbahnstr. 19, 82467 Garmisch-Partenkirchen, Germany)

35 Rainfall estimation accuracy of a nation-wide instantaneously sampling commercial microwave link network: error-dependency on known characteristics

Commercial microwave links are installed and maintained for the purpose of telecommunication. Hydrometeors between transmitting and receiving antennas cause the microwave signal to be attenuated. From signal attenuation, the path-averaged rainfall intensity can be calculated. A 7-month dataset of instantaneously logged signal powers from almost 2000 unique links in The Netherlands is analyzed. Rainfall intensities are calculated with the RAINLINK-package with a novel preprocessing module, enabling the pack-age to be applied on instantaneously logged data from now on. Rainfall intensities per link are validated with the path-averaged rainfall intensities according to a gauge-adjusted radar product. Both the overall performance and the dependence of errors on link characteristics and measurement conditions are evaluated. The coefficient of variation decreases from 3.70 to 2.32 and the correlation increases from 0.30 to 0.63 from instantaneous to daily estimates of rainfall accumulations. The coefficient of variation is also smaller during heavy rainfall. Errors are largest for path lengths shorter than 2 km, for observations during the late night and early morning, and observations during colder months (when solid or melting precipitation could occur and dew is more likely to form on the antennas). Comparison of our results with those of earlier studies show that min/max sampling (widely employed in network management systems) outperforms instantaneous sampling regarding detection of both quantity and occurrence of rain at a 15 minute sampling rate in the Dutch climate.

Speaker: Dr Aart Overeem (Royal Netherlands Meteorological Institute / Wageningen University & Research)

10:10 AM Coffee

Specific research topics: CML time series processing

36 Comparison of the achievable accuracy of CML based rain-retrieval algorithms based on min/max measurements vs. instantaneous measurements

Many different algorithms and approaches for the retrieval of rain using CMLs measurements have been presented in the past. In general, these algorithms can be divided into two groups, based on the different type of the available measurements: A) algorithms that make use of the standard network management system 15-minute min/max measurements; B) algorithms that make use of instantaneous measurements (sampled at intervals of seconds to hours).

In this study we show that the maximum attenuation measurements hold more information regarding the rain than a set of instantaneous attenuation measurement sampled at same intervals. That is, a time series of maximum attenuation values per 15-minute intervals can potentially result in more accurate rain estimation than a time series of instantaneous attenuation measurements sampled at the same 15-minute intervals. We show this by presenting a mathematical proof for a general heuristic case, and then, by demonstrating our conjectures on real-world scenarios using actual CMLs.

Speaker: Dr Jonatan OSTROMETZKY (Columbia University in the City of New York)

37 Detecting rainfall events in CML attenuation time series using convolutional neural networks

CML data might exhibit high signal fluctuations, even during dry periods. At the same time the number of newly available CMLs is rising fast. It is necessary to develop a technique to recognise the pattern of rainfall in CML signal levels, that is (1) generalizing to previously unknown sensors, (2) stable in time and (3) showing better performance than established methods. We therefore introduce convolutional neural networks to the task of wet/dry classification and test them on data from a German-wide CML network.

Speaker: julius polz (KIT/IMK-IFU)

11:10 AM Coffee

Specific research topics: Error sources and corrections

38 Electromagnetic Phenomena behind the Wet Antenna Effect

This talk will give an introduction to fundamental antenna parameters and characteristics like gain, directivity, efficiency, and reflection coefficient. It will debate the impact of wetness on the antenna cover on these parameters and on the link attenuation as a whole.
The impact of water on electromagnetic wave propagation is explained based on the canonical example of a thin water layer at plane wave incidence. To show this impact for a more realistic example, numerical field simulation and measurement results from a directive horn antenna are discussed. The power dissipation properties of water, which are characterized by the loss tangent, are considered in particular.
That leads to the discussion of the constituent parts of the wet antenna effect due to wetness on the cover of highly directive reflector antennas, as they are used in commercial microwave links (CMLs).
From this a model is derived, to describe the wet antenna effect from a physical point of view.

Speakers: Clemens Moroder (Technical University of Munich), Dr Uwe Siart (Technical University of Munich)

39 Long-term rainfall-runoff experiment at a small urban catchment using dense CML network

This contribution presents the ability of individual CMLs to provide relevant QPEs for rainfall-runoff simulations, specifically, the influence of CML characteristics and position on the predicted runoff. The analysis is based on a 3-year-long experimental data set from a small (1.3 km2) urban catchment located in Prague, Czech Republic. QPEs from real world CMLs are used as inputs for an urban rainfall-runoff model. The prediction performance is assessed by comparing simulated and measured stormwater discharges. The results show that model performance is related mainly to the CML sensitivity to rainfall given by their frequency and path length and by the CML position. The runoff predictions are mostly highly overestimated. The largest volume errors are associated with the shortest CMLs. On the other hand, shortest CMLs, located with or close to catchment boundaries, reproduce especially during heavy rainfal, the discharge temporal dynamics (e.g. timing of peak flows) better than longer less biased CMLs.The feasibility of CML use for runoff predictions is largely compromised by bias in CML QPEs. Therefore our future research is focused on bias reduction based on independent hydrometeorlogical observations: specifically ) adjusting by existing precipitation ground observations, ii) discharge observations using backwards-hydrology approach.

Speaker: Dr Vojtěch Bareš (Czech Technical University in Prague)

40 Rainfall estimation from Cellular Networks data based on local climatology.

The current study proposes a new method to process Commercial Microwave Links (CML) data for rainfall estimation.
Since the concept was introduced in the mid-2000, several papers have discussed and illustrated how rain-induced fluctuations could be used to quantify rainfall.
Most previous work on the subject is based on applying a power-law relationship between the attenuation over the link and rain rate, with the coefficients (a,b) of the relationship being either provided by the ITU, or derived from drop size distribution information for the region of interest.
Our method is based on matching the Intensity-duration-frequency (IDF) curves derived from CMLs to IDFs derived from rain gauges, comparing statistical characteristics of rainfall.
The calibration of CMLs based on local climatology is conceptually appealing as it does not require direct fitting of estimated rainfall intensity to ground observations and line-integrated rainfall better fit IDF concept
It is therefore also suitable for regions with sparse rain gauge networks, regions where CML rainfall information has the greatest potential to predict rainfall

Speaker: Yury Lechinsky (KLL)

12:30 PM Lunch

Specific research topics: Spatial reconstruction and data fusion

41 Stochastic reconstruction of precipitation fields using commercial microwave link information

The difficulty when using path-averaged rain rates derived from commercial microwave links (CMLs) is, that they give non-linear constrains for the precipitation field. To address this challenge, we apply Random Mixing to stochastically simulate precipitation fields. We generate precipitation fields as linear combinations of unconditional spatial random fields, where the spatial dependence structure is described by copulas. The weights of the linear combination are optimized in such a way, that the observations and their spatial structure are reproduced. Random Mixing allows to simulate precipitation field ensembles of any size, where each ensemble member is in concordance with the underlying observations.

Speaker: Dr Barbara Haese (University of Augsburg, Institute of Geography, Augsburg, Germany )

42 Key parameters for the tomographic reconstruction of rainfall fields

In this contribution, we present the results of the simulations carried out during the testing phase
of the tomographic reconstruction algorithm, developed in the framework of the MOPRAM project. To this purpose, we exploited a large data set of radar rain maps (assumed as ground truth) and assumed a realistic distribution of microwave links over the area of interest. We considered both stratiform and convective precipitation events, and evaluated the quality of the reconstructed rain field as a function of the number of links and base functions. In order to feed the algorithm with realistic CML data, we simulated different sampling times and quantization levels of the “measured” Received Signal Strength Indicator (RSSI).

Speaker: Mr Angelo Sileo (Politecnico di Milano)

43 Finding an optimal grid resolution

Creating an interpolated gridded map based on local measurements is commonly done in a straightforward way using IDW or kriging. The chosen output resolution is usually the same as the resolution of a field to be compared with (i.e. interpolated rain gauges and radar) or it is defined by the required parameters of a model. But how to choose an optimal resolution based on the input data? We developed a general technique based on the locations of the measurd variable (microwave link or rain gauge locations) and the variograms of the variable being assessed (rain rate), where the user defines what quality level is required. The results allow estimation of the optimal resolution as well as the possibility of adding quality flags to each grid cell of the final gridded Product.

Speaker: Remco van de Beek (SMHI, Sweden)

2:50 PM Coffee

44 3-D Rain Rate Estimation from Integrated Measurements of Commercial Wireless and Satellite Links

Abstract

In the last decade, various algorithms have been developed to provide accurate rainfall maps from measurements of rain-induced attenuation on commercial wireless links (CWLs), such as [1].
These solutions provide precise results but they also require dense terrestrial microwave networks, which have non negligible installation and operating costs.
A cheaper alternative for rainfall estimation is represented by broadcast satellites links (BSLs).
However, to the best of authors' knowledge, these approaches are able to estimate the rainfall rate on a single point only [2]-[4].
To gain all the benefits provided by both the cited schemes, we propose an adaptation of the state-of-the-art algorithm in [1] which is able to integrate the data provided by both the CWLs and BSLs, yielding a three-dimensional (3-D) rain rate map of a monitored area.
Our contribution lies in the technical description of the data processing schemes developed to retrieve these maps, with particular interest in the 3-D system model.
In the following, we summary the machinery of our approach. Let us consider a monitored zone, where $N = N_w + N_s$ communication links are active, with $N_w$ and $N_s$ are the number of CWLs and BSLs, respectively. It is worth noting that rain is assumed only present below the $0^\circ$C isotherm height. We divide then the vertical dimension into $H$ fixed heights, from ground level up to the $0^\circ$C isotherm height. Hence, by implementing the algorithm in [1] for each height, we obtain a set of $H$ rain maps, one for each height value, yielding to a 3-D description of the phenomena.

The effectiveness of novel approach is assessed using a simulator able to set up a configuration of the coordinates for both CWL and BSL terminals, and an instance of a simulated rain in a randomly-generated position.
Preliminary results are given in Figures 1 and 2 (presented into the attached file) which illustrate different scenarios in the presence of the same rain conditions. In the scenario of Fig. 1, the estimation is performed using only 21 CWLs, while Fig. 2 presents the estimation results using 13 BSLs and 8 CWLs. Though the position of the rain column with diameter 3 km is correctly estimated in both cases, it can be nevertheless noted that the joint utilization of both CWL and BSL provides a more accurate estimation of the rain intensity.
To show a more general prove of effectiveness, the near-to-the-ground RMSE between actual and estimated rain rate is plotted for different scenarios. In Fig. 3, the RMSE is shown as a function of the number of wireless links, randomly chosen from the arrangement presented in Fig. 1. The curves are plotted for different numbers of satellite receivers, i.e., 0, 4, 8 and 16 randomly-positioned on the maps. It is apparent that even a few number of satellites can be of great help in reducing the RMSE.

Acknowledgement

This work is supported by Fondo per le Agevolazioni alla Ricerca and Fondo Aree Sottoutilizzate (FAR-FAS) 2014 of the Tuscany Region, Italy, under agreement No. 4421.02102014.072000064 SVI.I.C.T.PRECIP. (Sviluppo di piattaforma tecnologica integrata per il controllo e la trasmissione informatica di dati sui campi precipitativi in tempo reale).
Fruitful discussions with prof. Hagit Messer of the University of Tel Aviv are greatly acknowledged.

References

[1] O. Goldshtein, H. Messer, and A. Zinevich, "Rain rate estimation using
measurements from commercial telecommunications links," Signal
Processing, IEEE Transactions on, vol. 57, pp. 1616 -- 1625, 05 2009.
[2] F. Giannetti, R. Reggiannini, M. Moretti, E. Adirosi, L. Baldini, L. Facheris, A. Antonini, S. Melani, G. Bacci, A. Petrolino, and A. Vaccaro, "Real-time rain rate evaluation via satellite downlink signal attenuation measurement," Sensors, vol. 17, p. 1864, 08 2017.
[3] F. Giannetti, M. Moretti, R. Reggiannini, and A. Vaccaro, "The NEFOCAST system for detection and estimation of rainfall fields by the opportunistic use of broadcast satellite signals," Aerospace and Electronic Systems, IEEE Magazine, 2019.
[4] A. Gharanjik, K. V. Mishra, B. S. Mysore, and B. Ottersten, "Learning-based
rainfall estimation via communication satellite links," 06 2018, pp.

Speaker: Fabio Saggese (University of Pisa)

45 Signal Processing Approach to Spatial Reconstruction and Data Fusion

TBA

Speaker: Prof. Hagit Messer-Yaron

3:50 PM Coffee

Specific research topics: Diversity

46 Rainfall retrieval with new generation of cellular backhaul network operating at E-band

New generation of E-band commercial microwave links (CMLs) operating at frequency band 70 - 86 GHz is gradually completing cellular backhaul networks and especially in cities often replacing older devices. CML rainfall retrieval methods developed during last decade have been designed and tested for frequency bands 20-40 GHz, where path attenuation caused by raindrops is almost linearly related to rainfall intensity and does not strongly depend on drop size distribution. This contribution assesses potential of E-band CMLs for rainfall retrieval by combining theoretical calculations and observations retrieved from E-band CMLs operating within cellular backhaul of T-Mobile, CZ. The initial results show that i) number of E-band CMLs has increased dramatically during last three years, ii) CML rainfall retrieval at E-band is sensitive to drop size distribution and spatial variability of rainfall, and iii) E-band CMLs are about three times more sensitive to rainfall than older 20-40 GHz devices, which enables to reliably measure rainfalls even with short (sub-kilometre long) CMLs. Longer CMLs, on the other hand, experience outages during heavier rainfalls, nevertheless, they can measure reliably light rainfalls, which has not been mostly feasible with old generation of devices.

Speaker: Martin Fencl (Czech Technical University in Prague)

47 TBA

Speaker: Yoav Rubin

Thursday, June 27

Community topics: Standards

48 Naming conventions

49 Performance metrics

10:15 AM Coffee

Community topics: Data sets and tools

50 Open high-resolution signal strength data from a commercial microwave link network in Gothenburg, Sweden

High-resolution signal strength data from commercial microwave links is not readily available to the research community. This may hinder breakthroughs in processing or applications. Moreover, access restrictions limit the need of a common reference and benchmark when e.g. comparing methodologies. To encourage broader scientific study and enable community benchmarking, we will therefore openly release a microwave link data set from Gothenburg, Sweden. The data set covers the period June-August 2015 and constitutes received and transmitted signal strength time series at 10-sec. resolution from 728 sub-links (363 links) in the Hi3G network in and around Gothenburg, covering ca. 2855 km2. Along with this we will also release conventional rainfall data (e.g. from local 1-min. gauges) to enable comparison and algorithm development. The presentation will briefly present the data set and the plans for releasing it openly. Our hope is that this data can become a useful benchmark against which all current and future algorithms can be compared. With this we also encourage others to share similar data sets in different climate zones.

Speaker: Dr Remco van de Beek (SMHI)

51 pycomlink - a Python toolbox for CML data processing

52 R packages

11:35 AM Walk/hike for lunch and coffee with community discussion

Community topics: Future collaboration and joint proposals

53 Cost Action OPENSENSE: Opportunistic precipitation sensing network

As already known, our joint proposal (OPENSENSE) submitted to COST Action was denied at the beginning of June 2019. The community should discussed future steps about this proposal during the meeting, since the next call is open until September 5th 2019. The decission about resubmission should be made in GaPa.

4:20 PM Coffee

Community topics: Conclusions

Friday, June 28

54 Flexible time for concluding Day2 and Day3 tasks and for wrap-up