UNIVERSITY OF TARTU FACULTY OF SCIENCE AND TECHNOLOGY Institute of Physics

Tanel Voormansik QUALITY CONTROL OF ESTONIAN WEATHER RADAR NETWORK Master’s thesis (30 ECTS credits)

Supervisors:

Ph.D. Piia Post Associate Professor Institute of Physics University of Tartu

Ph.D. Dmitri Moisseev Radar Laboratory Director Dep. of Physics University of Helsinki

Tartu 2014

Table of Contents 1 Introduction .......................................................................................................................................... 3 2 Theoretical background........................................................................................................................ 4 2.1 Radar location effects.................................................................................................................... 4 2.1.1 RLAN effects ........................................................................................................................... 6 2.1.2 Wind turbine effects............................................................................................................... 7 2.2 Radar hardware and software – dual polarization technique....................................................... 8 2.3 Atmospheric conditions .............................................................................................................. 10 3 Current status of the Estonian weather radar network ..................................................................... 13 3.1 Location of the Harku weather radar .......................................................................................... 14 3.2 Location of the Sürgavere weather radar ................................................................................... 15 3.3 Measurement program ............................................................................................................... 16 3.4 Radar data availability ................................................................................................................. 19 3.5 Data quality – settings of the radar measurement ..................................................................... 20 4 Comparison to the neighbouring countries ....................................................................................... 24 5 Analysis of data quality problems based on radar reflectivity products long term accumulation .... 28 5.1 Methodology ............................................................................................................................... 28 5.1.1 Data used .............................................................................................................................. 28 5.1.2 Py-ART software for accumulation product generation ...................................................... 29 5.2 HydroClass based filtering of data............................................................................................... 30 6 Results and conclusions...................................................................................................................... 32 6.1 Warm conditions – August 2013 ................................................................................................. 32 6.1.1 Harku radar - unfiltered........................................................................................................ 32 6.1.2 Harku radar – filtered ........................................................................................................... 35 6.1.3 Sürgavere radar - unfiltered ................................................................................................. 36 6.1.4 Sürgavere radar - filtered ..................................................................................................... 38 6.1.5 Comparison with rain gauge measurements ....................................................................... 39 6.2 Cold contitions – December 2012 ............................................................................................... 42 6.2.1 Harku radar - unfiltered........................................................................................................ 42 6.2.2 Harku radar – filtered ........................................................................................................... 44 6.2.3 Sürgavere radar – unfiltered ................................................................................................ 45 6.2.4 Sürgavere radar - filtered ..................................................................................................... 46 6.3 Conclusions .................................................................................................................................. 46 7 Summary............................................................................................................................................. 48 Eesti ilmaradarivõrgu kvaliteedikontroll ............................................................................................... 51 Acknowledgements ............................................................................................................................... 54 References ............................................................................................................................................. 55

2

1 Introduction Weather radars are invaluable tools in todays weather services as they provide information about precipitation related weather phenomena with outstanding temporal and spatial resolution. With the introduction of dual polarization Doppler weather radars the reflecting particles can be described in even more detail. The radar data are used to serve society in a wide range of applications from aviation weather service to electricity providers. Quality assessment of the radar data is essential firstly in meteorological services in short term forecasting and nowcasting. Numerical weather prediction model outputs could also benefit from quality controlled data. The resulting accurate short term forecasting could be advantageous for a multitude of other fields: hydrology, transportation, insurance, agriculture and many more. Radar data quality issues depend mostly on radar technique, local conditions and climate. There have been many studies that have listed the most important problems in more detail and analyzed the possible solutions (Saltikoff et al, 2010a; Szturc et al, 2012). While some of the issues have effective solutions, there are still problems remaining. And although solutions have been proposed to some of the quality issues, they can not be applied to all weather radars the same. Even if some radars are technically identical (same signal processor, antenna, receiver etc.), they need to be examined individually, because the differences in local conditions and climate can affect the quality substantially. So it needs to be analyzed in each country separately, radar by radar. Even though Estonia has been covered by dual polarization Doppler weather radar measurements for almost five years, the quality control of the country’s radar network has not been carried out before. The aim of this thesis was to describe and analyze the dual polarization Doppler weather radar network in Estonia from the data quality point of view. This included the comparison of the network and radar scan strategies to the neighbouring countries of Estonia. Long term radar data accumulation needed to be carried out in different climatic conditions on both radars individually to specify the quality issues. The Python ARM Radar Toolkit (PyART), an open source module for working with radar data needed to be installed and modified for accumulation product generation and filtering of the data within the framework of this study.

3

2 Theoretical background The quality of a weather radar network depends on a number of factors and to evaluate the network as a whole, the details of the radars in the network need to be first examined individually and compared with each other. These details include the placement of the radar in relation to the surrounding area and in relation to the other radars in the network, hardware and software of the radar and how it has been set up. As the main purpose of the meteorological radar is to give information about hydrometeors in the atmosphere, one of the quality measures is the cleanness of the data or the amount of false echoes in the data. And as no two radars are placed identically, the same set up solution does not suit for the other radars in the network. Different climatic conditions on different locations also require individual approach. So there are some main topics and known problems that need to be looked into more thoroughly in order to evaluate the quality of the radar network and thus the data coming from the radars. Location, hardware and software of the radar and atmospheric conditions are the main contributors to the data quality and will be covered in the following paragraphs.

2.1 Radar location effects One of the most influential aspects is the location of the radar. The decision where to locate the radar needs very thorough investigation as the operational radar stations are stationary and moving them to a new location could be very costly and time consuming. But there is no such thing as ideal position for a radar, it is about finding the optimal location that would fulfill the needs of the users of the radar data. When we think of weather radar, we usually think that the radar shows us information about meteorological targets, but actually we might see echoes caused by other objects in the way of the electromagnetic wave as well. This phenomena is often called clear-air return that incorporates all the non-meteorological sources of radar echo. While it is often possible to learn about the weather from these non-meteorological targets (Rinehart, 1997), we would generally like to see as little echoes in the radar display in clear weather situations as possible. How much of these we see in each radar depends largely on the location - the surrounding buildings, landscape (mountains, sea, etc) can all contribute to this. Quite often the radar echoes are strongly contaminated by signals other than precipitation. Unwanted signals in a radar are generally described as noise and clutter. The definiton of clutter depends on the function of the radar. In weather radar, clutter includes 4

mainly returns from ground, sea, buildings, birds and insects (Radartutorial, 2014a). False echoes caused by non-moving objects can be easily mitigated by the use of Doppler filter in Doppler weather radars – objects that have zero or near zero velocity can be easily detected and removed from the data (Keränen et al, 2010). There are still several other echo sources at both land and at sea that remain in radar data after suppression of static ground echo. One of these can be related to surface clutter – ground and sea returns can have radial velocity large enough to not be filtered out in simple Doppler filtering. Wind can cause movement in the trees so that they do not appear stationary and echoes from sea waves can also have radial velocity (Radartutorial, 2014a). In addition to waves, there is another possible source of echoes from sea – ships are also efficient reflectors of electromagnetic radiation and they appear as sharp spots in weather radar images. Ships interfere even when they lie a couple of degrees off the radar beam and this makes them insensitive to vertical gradient based recognition – they could be more easily confused for local showers or hail (Peura, 2002). Fig 1 and Fig 2 illustrate the current ship lines in the Baltic Sea and this hints that ship reflections might be a serious problem especially in atmospheric conditions when the radar beam is being bent more towards the Earth than in normal conditions and probably more in Harku radar, as it is located very close to the sea.

Fig 1. RORO (Roll-on/Roll-off ships) and ferry map in the Baltic Sea (http://www.baltictransportmaps.com/rofemap.html#?z=1&x=0&y=0)

5

Fig 2. Container ship map of Baltic Sea (http://www.baltictransportmaps.com/contmap.html#?z=1&x=0&y=0)

2.1.1 RLAN effects Another problem that is becoming more and more important these days and what is rather new even in the short history of radar meteorology is related to the Radio Local Area Network (RLAN) interference with weather radars. This affects mostly C-band radars that are operating in the same frequency region with some wireless internet devices. The C-band weather radars can operate in the frequency range of 5250–5725 MHz while most of the Cband radars belong to the 5600–5650 MHz range (Leck, 2009), which is also in the range of some RLAN devices – in World Radio-communication Conference (WRC-2003), the decision was made to allow wireless acess systems (WAS) use the band 5470-5725 MHz (ITU-R 229, 2003). The RLAN bands consist of 10 channels, each with 18 MHz bandwidths. The weather radar, with its typical 0.5 or 1 MHz bandwidth for a 2 µs or 1 µs pulse, respectively, will see these RLAN signals as additive white noise and not at a particular Doppler frequency which would be the case if the RLAN bandwidths were smaller than about 1 MHz (Joe et al, 2005). Joe et al (2005) tested the effect of RLAN signals to operational C-band weather radar from 0.5 and 10 km ranges and the measurements of a radar were affected on both distances. On the shorter distance, the effect was stronger – the signal from one RLAN device was seen in 30° sector and not only in the main lobe but in the side lobes of the radar. So the RLAN signal was picked up even when the antenna was not directly pointed at the RLAN source.

6

But this interference should theoretically not be a problem, as radar and WAS technologies are expected to co-exist in the same environment by using a frequency abandonment protocol of the RLANs. The RLANs are required to implement a Dynamic Frequency Selection (DFS) system in which radio-frequencies are monitored and the RLAN selects frequencies that are not used. The DFS has two main modes – before transmitting on a given channel, the RLAN device is on Channel Availability Check (CAC) for 1 minute and is on receive only mode. If no radar is detected, it can start using the channel. The other DFS mode is called In-Service Monitoring (ISM) – while using the channel, the RLAN should still constantly monitor this channel for radar signals (Tristant, 2009). And if a signal is then detected, the RLAN device must vacate the channel for 30-minute period and before re-using the channel, it must continuously monitor the channel for a 10-minute period (Joe et al, 2005).

2.1.2 Wind turbine effects The effect of wind farms has also increased on weather radar measurements in the recent years. Wind turbines built nowadays are large structures, which often reach more than 150 m above the ground and they are built both on- and offshore. Although radars and wind turbines have coexisted already for many decades it is only in the last years that the interference problem has received considerable attention. It is because in recent years the number of wind turbines has increased, as have their sizes and at the same time radars have become more sensitive (Norin and Haase, 2012). There are three more common effects caused by wind turbines – clutter contamination (continuous false echoes), beam blockage and erroneous Doppler measurements. Operating wind turbines generate both static (clutter with zero or near-zero radial velocity) and dynamic clutter (clutter with larger radial velocities). Static velocities can be easily filtered by using Doppler filtering, while dynamic clutter can be harder to detect as clutter. Because radars are sensitive to wind turbines, the location of the radar in relation to wind farms is an important issue. WMO (World Meteorology Organisation) has proposed a guidance statement on weather radar and wind turbine siting, which is presented in Table 1. From the table it is clear that the location of the radar in relation to wind turbines is of great importance in order to get high quality data without wind turbine interference. Dynamic clutter is more difficult to remove, as it can have radial velocities similar to precipitation echoes and can therefore be mistaken for precipitation. What makes the removal of wind turbine clutter even harder is the fact that echoes from wind turbines are highly variable in

7

time since the amplitude of the scattered signal depends very much on the angle of the wind turbine and its blades (Norin and Haase, 2012).

Table 1. WMO guidance statement on weather radar and wind turbine siting (WMO, 2010). Range

0-5 km

Potential Impact

Guideline

The wind turbine may completely or

Definite Impact Zone: Wind

partially block the radar and can

turbines should not be installed in

result in significant loss of data that

this zone.

cannot be recovered.

5-20 km

Multiple reflection and multi-path

Moderate Impact Zone: Terrain

scattering can create false echoes

effects will be a factor. Analysis and

and multiple elevations. Doppler

consultation is recommended. Re-

velocity measuremens may be

orientation or re-siting of individual

compromised by rotating blades.

turbines may reduce or mitigate the impact.

20-45 km

Generally visible on the lowest

Low Impact Zone: Notification is

elevation scan; ground-like echoes

recommended.

will be observed in reflectivity; Doppler velocities may be compromised by rotating blades. Generally not observed in the data

> 45 km

Intermittent Impact Zone:

but can be visible due to propagation Notification is recommended conditions.

2.2 Radar hardware and software – dual polarization technique Radar hardware and software can greatly contribute to the quality of the measurements and also to the possibilities of detecting and mitigating the errors. Up until the recent years most operational weather radars were single polarization Doppler weather radars, which were able to detect rainfall intensity based on the energy of the backscattered horizontally polarized electromagnetic wave from the precipitating particles and also the radial speed and direction of the precipitating particles based on the Doppler effect. But with the arrival of the dual polarization technique to the weather radars the available information expanded significantly. As the name suggests, Dual polarization Doppler weather radars can transmit and receive both horizontally and vertically polarized EM waves. Most of the dual polarization 8

operational radars work in the HH and VV configuration – in HH the radar transmits horizontally polarized wave and also detects horizontally polarized component of the backscattered signal and in VV it transmits vertically polarized wave and receives vertically polarized component of the signal (Radartutorial, 2014b). By comparing these reflected power returns in multiple ways (ratios, correlations, etc.), we are able to obtain information about the size, shape and density of cloud and precipitation particles (CIMMS, 2014a). Some of the fundamental variables measured by polarimetric radars (in addition to conventional Doppler radar horizontal plane reflectivity (Z) and radial velocity (V)) are Differential Reflectivity (ZDR), Correlation Coefficient (RhoHV) and Specific Differential Phase (PhiDP). The differential reflectivity is a ratio of the reflected horizontal and vertical power returns. It is a very good indicator of the shape of the reflecting particle. So it is also a good estimate of average drop size. Differential reflectivity is defined as ( ) where

is the returned power in horizontal polarization and

(1) is the returned power in

vertical polarization (NOAA, 2014a). Correlation coefficient is a correlation between the reflected horizontal and vertical power returns (values between 0 and 1). It is a good indicator of regions where there is a mixture of precipitation types (snow and rain). From quality point of view it is a good estimator of whether the reflecting objects are of meteorological or non-meteorological origin, because hydrometeors have RhoHV generally over 0.8. Non-meteorological objects (birds, insects, etc.) on the other hand have low RhoHV values, because their shapes are complex and highly variable so horizontal and vertical pulses will behave very differently with these objects (NOAA, 2014b). Specific differential phase is a range derivative of the differential phase (PhiDP) and measures the phase change per kilometer between the horizontally polarized and vertically polarized electromagnetic waves. That means it is immune to attenuation and hail (hailstones are usually spherical). KDP is calculated as ( ) (

where

and

( )

(2)

)

refer to measurements range 1 and range 2 from the radar, where