Visual Analytics of Multi-sensor Weather Information

Georeferenciation of Doppler Weather Radar and Weather Stations

Aitor Moreno, Andoni Gald

os Andoni Mujika and

Alvaro Segura

Vicomtech-IK4, Paseo Mikeletegui 57, San Sebasti

an, Spain

Keywords:

Weather Stations, Geovisual Analytics, Doppler Weather Radar, Scalar Fields.

Abstract:

This work presents a geovisual tool which integrates and georeferences data coming from some of the weather

instruments installed in the Basque Country: a Doppler weather radar and the weather station network com-

posed of around 100 multi-sensors stations (temperature, precipitation, wind...). The visualization of the raw

data coming from the weather radar is based on the generation of a set of 3D textured concentric cones (one

per elevation scan). The resulting 3D model is then integrated in the 3D digital terrain of the Basque Country.

For the weather stations, we have provided a Kriging based interpolation method to produce textures from the

scalar data measured at the weather stations. These textures are then mapped in the same 3D digital terrain

as before. The integrated visualization of the weather information enhances the understanding of the data. To

illustrate the proposed methods a use case is provided: matching the precipitation measured at ground level

with the radar scans.

1 INTRODUCTION

The combination of Geographic Information Systems

and Computer Graphics is the main characteristic of

the Geovisual Analytics (Andrienko et al., 2010). The

geovisual tools have a considerable potential in en-

vironmental monitoring and in the decision making

process (Tomaszewski et al., 2007). For years, the

utilization of weather 2D maps has helped to present

the data collected in the weather networks to the users

with a great success (Kraak and Ormeling, 2002).

The presentation of animated sequences introduces

the time to show the historical or predicted evolution

of a measured variable.

This paper presents the work carried out in the 3D

geovisualization of the weather multi-sensor network

installed in the Basque Country (Spain) by the Basque

Meteorology and Climatology Agency.

The presented geovisual 3D and interactive tool

helps to centralize and to enable the analysis of

large amounts of data collected from weather sensors,

spread around the territory. These sensors are aimed

to monitor the temperature, pressure, humidity, wind,

solar radiation, etc. There are other type of weather

devices, like a Doppler weather radar, a wind proﬁler

and several oceanic probes.

The ﬁrst section of this work introduces the

Basque Country weather network, including the

Doppler weather radar. The second section of this pa-

per presents the 3D visualization of raw data coming

from the Doppler weather radar. The nature of the

radar scans is volumetric, so we provide mechanisms

to produce a 3D model with the scanned 3D informa-

tion. These 3D models are integrated with the 3D dig-

ital terrain, constructed from the available geographic

information in the area.

The third section presents the tools to create inter-

polated images given a timestamp and a target instru-

ment (temperature, precipitation, etc.). The produced

images are overlaid over the same 3D reconstructed

digital model of the Basque Country territory.

The forth and ﬁfth sections present how the geovi-

sual interactive applications can help to handle all the

available information from the multi-sensor weather

network (radar or weather stations) and to combine

them to produce more helpful meteorological prod-

ucts.

Finally, the paper ends with the conclusions and

the future work.

2 BASQUE COUNTRY WEATHER

NETWORK DESCRIPTION

The Autonomous Community of the Basque Country

is a territory in northern Spain bordering with south-

329

Moreno A., Galdós A., Mujika A. and Segura Á..

Visual Analytics of Multi-sensor Weather Information - Georeferenciation of Doppler Weather Radar and Weather Stations.

DOI: 10.5220/0004677603290336

In Proceedings of the 5th International Conference on Information Visualization Theory and Applications (IVAPP-2014), pages 329-336

ISBN: 978-989-758-005-5

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

ern France. It spans an area of 7234 km

and is

crossed by a few mountain ranges. Established 1990,

the Basque Meteorology Agency, Euskalmet, has de-

ployed a large network of weather stations, includ-

ing a long-range radar, and provides past, present and

forecast meteorological information.

The physical data collected by the sensors in the

network are stored, managed and retrieved to provide

meteorological products. The weather forecast to the

public is one of the most common ways to use the

meteorological products (in TV or in a web page).

The acquired information from the network is also

used to help in the decision making process in two

different ways. Firstly, the information is used to an-

alyze high risk potential hazard zones, especially, to

improve the ﬂood awareness in the territory, so pre-

ventive actions can be performed. Secondly, the his-

torical analysis of the available data is used to study

the evolution of the climate. The analysis of past

singular events is also important to extract important

knowledge and conclusions from the data and to help

to improve the numerical weather models.

2.1 Weather Radar

The Basque Weather service operates a dual Doppler

Weather Radar, located on top of Mount Kapildui,

1000 m. high and 100 km. away from the coast.

It is a Meteor 1500C model from Selex-Gematronik.

Among other variables, the radar computes the reﬂec-

tivity (dBZ), radial velocity (V) and spectral width

(W) ﬁelds every 10 minutes through two volumetric

and two elevation scans.

2.2 Multi-sensor Weather Stations

A weather monitoring network is composed of a set

of weather stations. Each one can be considered a

multi-sensor device, which provides periodical mea-

surements to the control center.

In the Basque Country, this weather station net-

work is composed of around one hundred network sta-

tions, each one with a variety of measurement instru-

ments. Some of them have thematic instruments, nor-

mally associated to the stations near the sea (currents,

salinity...) and the top of the main mountains. As for

the time resolution of the installed instruments, the

automated station delivers a new data package (which

includes all the measurements of all the installed in-

struments in the station) to the control center every 10

minutes.

The historical data from the Basque Country net-

work can be consulted and retrieved in the Euskalmet

web site

3 WEATHER RADAR DATA

Radar scans are typically represented as 2D images

in the form of either PPI (plan position indicator) or

CAPPI (constant altitude PPI) products.

In this work we don’t use these products. Since we

are aiming to visualize the complete radar volumes in

3D (not just individual slices from it), we use the raw

output from the weather radar, containing the whole

volume scan.

The process to construct a 3D model from the raw

data of the radar is addressed in the following subsec-

tions. Firstly, the raw output is processed and a set

of gray-scale images is created. From the metadata of

the volume scan, a 3D geometry is generated, com-

posed of a set of concentric cones. Then, the gray-

scale images are colored using a transfer function and

they are mapped as textured in the 3D cones, creating

the ﬁnal 3D model for the given volume scan. The

3D model is integrated with the digital terrain of the

Basque Country, resulting a 3D visual Geographic In-

formation System (Peuquet and Marble, 1990).

3.1 Volumetric Data Processing

The volumetric datasets acquired by the radar are

composed of 14 scans at increasing elevations (from

−1

◦

to 35

◦

). Such scanning process results in a

discrete sampling of the sky volume in which each

sample has elevation, azimuth and range coordinates.

Thus, samples can be addressed by spherical coordi-

nates.

The volumetric information is converted into a set

of gray-scale images, one per elevation. The scanned

values (dBZ, V or W) provided in the raw binary ﬁle

are not the actual values and some transformations

have to be applied to the raw values to get the actual

values. Then, the range of interest (window level) in

the values is mapped to the byte range (128 values) to

create the gray-scale image pixels.

The Figure 1 shows a composition of the 6 lowest

levels of a given volume scan. The left side of each

one is the closer side to the center of the volume (the

radar location) and all the pixels in the ﬁrst column

correspond to the same physical location due to the

nature of the polar coordinates.

The produced individual images require addi-

tional metadata to decode the information (encoded

Euskalmet web site (Basque Meteorology and Climatol-

ogy Agency): http://www.euskalmet.euskadi.net/

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

330

Figure 1: A composition of 6 gray-scale images corre-

sponding to the lowest elevation levels of a given radar scan.

Figure 2: A 3D model of the scanned volume is created as a

set of concentric cones. The radius and height of each cone

is calculated from the metadata of the radar scan.

variable, window level used in the byte range accom-

modation, numerical transformations, etc.) which can

be encoded in the ﬁlenames of the images. Any addi-

tional information could be still loaded from the orig-

inal raw data ﬁles.

3.2 3D Representation and

Visualization

Previous works have dealt with the visualization of

the volume scans acquired from a weather radar. We

can see an extensive classiﬁcation and review of the

main techniques and methods in (Ernvik, 2002).

The geometry of the radar scans can be seen as a

set of conical sweeps at different elevations. The work

of (Sundaram et al., 2008) et al. adds a rectiﬁcation

step to convert the conical grid into a rectilinear grid.

This approach suits better for the traditional methods

of volume visualization: indirect (isosurfaces, march-

ing cubes...) or direct (volume rendering).

As the rectiﬁcation process is very time consum-

ing, our approach do not create such rectilinear grid.

Figure 3: Color mapping of the ﬁrst level of a scanned vol-

ume by the weather radar. The images are given in polar co-

ordinates, with the weather radar located in the upper side

of the images.

Instead, a set of geometrical and concentric cones (see

Figure 2), centered in the radar location is created

(Peng and Lingda, 2007). The height and radius of

each of the cones are determined by the elevation an-

gle of such scan.

Each of the images is then used as texture of the

corresponding cone. As the images are gray-scale,

a coloring function (transfer function) is used. The

Figure 3 shows how a colored RGBA texture (on

the right) is created by applying a transfer function

(shown in the middle) to the gray-scale image (on the

left) (Ginn, 1999). The black area in the ﬁnal colored

texture corresponds to the alpha channel, allowing a

proper 3D visualization of the scene when all the tex-

tures of the set of concentric cones are rendered to-

gether.

3.3 Ground Clutter Removal

Given the topography of the Basque Country, the

lower scans are affected by the surrounding moun-

tains and other topographical elements, adding almost

constant noise to the data, which should be ignored.

This constant noise is known as ground clutter.

In Mount Kapildui clean scans can be obtained at

elevations greater than 1

◦

. The elevations between

−1.0

◦

and 0.5

◦

contain noticeable ground clutter, but

they can not be discarded since they provide valuable

information.

As the ground clutter is in theory constant in time,

its effects in the lowest scans can be reduced by sub-

tracting a ﬁxed mask to the retrieved data.

However, given the variability of radar echoes

caused by topography, a single scan of a clear sky is

not enough to create a reliable clutter ﬁlter. In order

to avoid this problem, the ﬁnal clutter mask was cre-

ated as the average image of 6 radar scans taken at

different times with a clear sky.

The resulting mask effectively removes the

ground clutter from the volume scans, but it may not

ﬁlter correctly all the ground echoes due to the inher-

VisualAnalyticsofMulti-sensorWeatherInformation-GeoreferenciationofDopplerWeatherRadarandWeatherStations

331

ent variability of the radar echoes in the topography.

3.4 GEOREFERENCED 3D

VISUALIZATION

A fully interactive viewer requires the presentation of

a virtual scene to the users, which is composed of two

main elements: i) a model coming from the weather

radar data and ii) the terrain where the weather radar

is located. The most common techniques to integrate

terrain and radar data involve the overlapping of 2D

images: i) the terrain image or map, where colors rep-

resent the height and ii) the image of the weather radar

(Toussaint et al., 2000). Usually, the radar images are

typically represented as 2D images in the form of ei-

ther PPI (plan position indicator) or CAPPI (constant

altitude PPI) (James et al., 2000).

In our work, we aimed to a 3D visualization of the

volume scans and the terrain where the data has been

acquired. As presented before in the subsection 3.2,

the visualization of the weather radar data is achieved

by creating a 3D model composed of the textured 3D

cones. The resulting 3D model correspond to the 3D

visualization of the given volume radar data and it can

be used in interactive 3D applications.

For the 3D digital terrain, we have constructed a

3D model of the whole Basque Country. It was cre-

ated as a combination of highly detailed digital eleva-

tion data and a set of properly adjusted high resolution

orthophotographs (Jenson and Dominque, 1988), pro-

vided by the Basque Government. This large amount

of data has been prepared and transformed into a

PagedLOD model, ready to be loaded and rendered

by the OpenSceneGraph graphics library at interac-

tive frame rates.

The terrain data and the radar scans are correctly

georeferenced, since the radar data includes the corre-

sponding UTM coordinates and therefore, a seamless

visualization of the 3D terrain model and the radar in-

formation at the same time is achieved without major

inconveniences. The union of radar and topographic

data clearly highlights the presence of ground clutter

around the highest mountain ranges (see Figure 4).

The Figure 5 shows some visualization examples

of volume scans with different variables. Each mea-

sured variable has its own transfer function, follow-

ing the standardized coloring functions used by the

commercial products (like Rainbow 5, from Gema-

tronik). The left column shows the reﬂectivity (dBZ)

of two different timestamps and viewpoints using the

trasfer function shown in Figure 3. The right column

shows the velocity (V) in the same timestamps and

viewpoints. They use a custom transfer function (a

gradient from blue to red).

Figure 4: Unﬁltered Kapildui radar volumetric information

visualization using a reﬂectivity color map. In the left im-

age, the rain areas can be seen in blue as well as ground

clutter. In the right image, a close up of the ground clutter

is shown, matching the mountain causing it, which proves

that the georeferenciation of the volume radar and the 3D

terrain is accurate.

3.5 Animation support

The interactive visualization of a single radar volume

merged with the 3D geographic model enhances the

understanding of the radar data. The users can nav-

igate through the 3D world and visually inspect the

volume and its relationship with the terrain (moun-

tains, valleys). With the visualization of a sequence

of consecutive radar volumes, the users’ knowledge is

increased dramatically, since the temporal axis gives

additional visual information. Hidden information in

the full set of sequential 2D slides, which compose the

radar scans, emerges when several data are visualized

in an animated way. Some of most appealing retrieved

information refers to the evolution of the rain clouds,

the visual inspection of the trajectories of the storms

and the effect of the mountains in the evolution of the

rain clouds.

The animation support requires to have a quite

large amount of consecutive radar data, which will be

loaded in runtime. As the data amount could be ran-

domly huge, it is not feasible to precalculate all the 3D

model of the radar scans. Therefore, a fast on-demand

construction of the 3D models is required.

4 GEOVISUALIZATION OF

SCALAR FIELDS

This section introduces the visualization techniques

used to display the information acquired from the in-

struments installed on the weather station network.

Since the values of temperature and precipita-

tions are known only in the points where the weather

stations are, an interpolation method has to be set

in order to color the whole map. Kriging methods

(Cressie, 1993) and its variations are widely used

(Mair and Fares, 2011) (Hartkamp et al., 1999) to

interpolate the measurements from the weather net-

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

332

Figure 5: Weather radar 3D visualization over the Basque Country 3D digital terrain. Reﬂectivity (dBZ) in the left column

and Velocity (V) in the right column.

work. These methods obtain the value in a point com-

bining the values of the neighbor known values and

the distances to those points.

Let (x

, x

, ..., x

) ⊂ X ⊂ R

be the points where

the temperature is measured. In the probabilistic

model used by ordinary kriging, the temperature in

a speciﬁc point, t(x), is considered the realization of

a random variable T (x) and two degrees of station-

arity are assumed. This implies that the mean of all

random variables T (x) is the same and that the corre-

lation between two random variables depends only in

the distance between the points and not in their posi-

tions.

E{T (x)} = m ∀x ∈ X

γ(T (x

), T (x

)) = γ(h) ∀x

, x

∈ X

Where x

= x

+ h and γ(T (x

), T (x

)) =

VAR(T (x

) − T (x

)) is the variogram function.

In ordinary kriging, the random variable T (x) is

estimated by

T (x). It is the linear combination of the

random variables referred to the known points and the

weights w

(x) are obtained from the stationarity as-

sumptions.

T (x) =

∑

i=1

(x) ∗ T (x

)

In our case, universal kriging (Huijbregts and

Matheron, 1971), also known as regression kriging

(Goovaerts, 2000), has been implemented. In this

method, the variable t(x) is divided into a determin-

istic component and the residual, that is treated as a

random variable.

t(x) = m(x) + r(x)

Then, the deterministic component is approxi-

mated by a plane using least squares method and the

residual is computed with ordinary kriging. For this

latter, since the variogram function needed for the

kriging method cannot be computed, a spherical vari-

ogram model (Cressie, 1993) is used.

Once prediction for the values at random loca-

tions are obtained, the interpolated numerical values

have to be visualized somehow. A transfer color func-

tion is used to predicted values to a speciﬁc color. In

this way, anyone can take visual indications about the

warmer zones or the areas where the precipitation has

been higher. The 2D colored images break the link

with the actual terrain. One way to solve this issue

is to generate a 3D scene. A reconstructed 3D ter-

rain, textured with the colored image, georeferences

the weather data and the terrain where they have been

measured (see Figure 6). As the texture is overlaid

over the terrain, the resolution of such textures is im-

portant. A balance between the computational effort

(the interpolation method has be called for each pixel

in the image) and the quality of visualization output

has to be found.

The Basque Country can be embedded in a 150

km. × 150 km. square. Provided a 1024 × 1024

texture, we have approximately a 150 m. per pixel

resolution, which is enough to get high quality vi-

sual representation of the scalar ﬁeld. In a commodity

computer, the computational time is below 2 seconds.

The interpolation techniques also predict values

outside the Basque Country territory, as the square

texture covers parts of the neighboring provinces. For

those outer regions, the number of near weather sta-

VisualAnalyticsofMulti-sensorWeatherInformation-GeoreferenciationofDopplerWeatherRadarandWeatherStations

333

Figure 6: Temperature (

◦

C) rendered in an overlay tex-

ture and ﬁtted to the Basque Country administrative bound-

aries. In the bottom ﬁgure, the pins show the location of the

weather stations and a 2D graph is attached to the selected

station showing the evolution of the temperature within the

selected day.

tions is very limited and therefore, the interpolated

values are less accurate. In fact, in the corners of the

texture, the method can be perceived as an extrapola-

tion method. To limit the impact of such undesired be-

havior, an alpha mask with the Basque Country shape

is used to limit the texture to the correct boundaries.

This technique also has an impact in the performance,

since the pixels out of the mask are not calculated us-

ing the interpolation method.

The upper image in the Figure 6 shows an scalar

ﬁeld (temperature) mapped in a 3D model of the

Basque Country. The texture resolution is 1024 ×

1024 and an alpha mask has been added with the ad-

ministrative boundaries. The bottom image shows the

location of the weather stations as 3D pins. Although

the network is composed of almost 100 stations, not

all of them have the same instruments. And addi-

tionally, at a given time, some stations can be out-

of-service or their data could have been discarded due

to the quality control over the data and the electronics

in the weather station. In the shown case, the number

of available stations for the temperature instrument is

68 at that precise instant.

5 GEOTOOL USER INTERFACE

This section introduces the visual geotool imple-

mented to seamlessly visualize the weather radar vol-

ume scans and the weather station network.

Figure 7: The user interface of the geotool. In this case, a

weather station is selected and the evolution of the temper-

ature during the selected day. Also, the weather radar data

is displayed in the 3D environment.

The amount of available data is stored in a Post-

greSQL DataBase, queried from a QT application.

The temporal nature of the data requires to arrange a

scroll panel where the available timestamps are shown

(see bottom part of the Figure 7).

There are two main timelines: the weather radar

and the weather station network. The timeline for the

weather radar conﬁgures which one of the available

variables is shown, i.e, reﬂectivity (dBZ), velocity (V)

or spectral width (W). In a similar way, the timeline

for the weather stations conﬁgures which instrument

is shown in the overlaid texture over the terrain.

Some zooming functionality is added to the time-

lines to allow the user to inspect and navigate the

database. The available data in the database spans

for 7 consecutive days in our tests (around 1000 sam-

ples) but it could be extended to the whole exist-

ing database in the Euskalmet meteorological agency

(several years for the weather station network).

The 3D virtual terrain is centered in the geotool.

The selected volume data coming from the weather

radar can be conﬁgured: global and per cone visi-

bility / transparency. The textures interpolated from

the weather station network have similar options: vis-

ibility and transparency. There is also a selector to

choose the instrument to visualize: temperature, pre-

cipitation, solar radiation, pressure or any other avail-

able instrument in the database.

An additional panel is added to show metadata.

For the weather radar, the metadata embedded in the

raw ﬁles is shown which included useful information

for the meteorologists. For the weather station net-

work, information about the selected instrument and

the selected station (if any) is shown. The user can in-

spect the installed instruments on a given station and

show in a 3D panel the evolution of a variable along

the selected day (see Figure 7).

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

334

Figure 8: Two examples of georeferenciation between the weather radar and the precipitation measured at the ground level.

In both cases, the location of the water in the atmosphere matches the pattern obtained by the Kriging interpolation.

As the nature of the data is inherently 4D, we have

provided animation support for some of the typical

VCR functionality on the data: Play and stop with

some speed up functionality, go to the next data or to

the previous data. The interactive visualization of the

3D environment is kept while the animation is run-

ning, which is useful to inspect freely the evolution of

the weather in the region.

6 GEOVISUAL ANALYSIS:

PRECIPITATION

Having multiple monitoring devices produces lot of

information for the meteorologists. Analyzing all the

data using long tables is not efﬁcient, or at least, it

is difﬁcult to globally understand the data. The uti-

lization of textured maps with the scalar ﬁelds helps

to visualize in a concise way all the existing data in a

given timestamp.

The weather radar monitors the atmosphere and

thus, it detects the amount and type of the meteors

(ultimately, the water in the atmosphere). This in-

formation is related to the measured amount of pre-

cipitation by the weather stations. The correlation of

such variables should show that fact. But it is difﬁ-

cult to analyze the raw data from the weather radar

and cross-reference the values with tabular data of the

precipitation measured by the weather stations.

The presented geotool can load a 3D representa-

tion of the weather radar and visualize the precipi-

tation measured at the ground level by the weather

station network as a texture on the 3D terrain. The

Figure 8 shows that the composition of both datasets

ﬁts perfectly. The shape produced by the precipita-

tion matches the shape of the meteors in the atmo-

sphere. Although the acquisition time is almost iden-

tical, there is a 10-20 minutes delay since the weather

data is measuring the meteor in the atmosphere and

the weather stations are measuring the water already

fallen to the ground.

The georeferenciation of both datasets can be seen

clearly in an animated loop. The volume captured by

the weather radar advances towards the East and we

can see the same pattern in the shape of the interpo-

lated texture from the stations.

7 CONCLUSIONS

Meteorologists receive tons of data coming from mul-

tiple sources. It is important to provide tools to help

them to analyze such amount of data. Visual Analyt-

ics has been proved to provide concise thematic visual

outputs from large datasets.

This work has provided a geovisual tool to help

in the integration and georeferenciation of the data

coming from the weather instruments installed in the

Basque Country: a weather radar and a weather sta-

tion network composed of around 100 multi-sensors

stations.

The generation of the 3D models from the raw

radar information provides a pseudo volumetric visu-

alization at interactive rates. Users can analyze visu-

VisualAnalyticsofMulti-sensorWeatherInformation-GeoreferenciationofDopplerWeatherRadarandWeatherStations

335

ally where the meteors are located in the atmosphere

and where it is expected to rain. Additionally, cross-

referencing the radar information with the precipita-

tion measured in the weather stations provide a unique

visualization and understanding of the process.

As future work, the massive introduction of de-

vices like the smart phones or tablets makes interest-

ing to port the presented geovisual tool to the Web

(Van Ho et al., 2012).

Additionally, the utilization of volume render-

ing techniques can provide further analysis methods

to the meteorologists, even in the Web environment

(Congote et al., 2011).

ACKNOWLEDGEMENTS

This work has been funded by the Basque Govern-

ment’s ETORTEK Project research programs.

Authors thank Basque Country Government for

the Open Data initiative, which provided part of the

geographic information and the weather data used in

this work.

REFERENCES

Andrienko, G., Andrienko, N., Demsar, U., Dransch, D.,

Dykes, J., Fabrikant, S. I., Jern, M., Kraak, M.-J.,

Schumann, H., and Tominski, C. (2010). Space, time

and visual analytics. International Journal of Geo-

graphical Information Science, 24(10):1577–1600.

Congote, J., Segura, A., Kabongo, L., Moreno, A., Posada,

J., and Ruiz, O. E. (2011). Interactive visualization of

volumetric data with webgl in real-time. In 16th Inter-

national Conference on Web 3D Technology, Web3D

2011, pages 137–146.

Cressie, N. A. (1993). Statistics for Spatial Data. Wiley-

Interscience.

Ernvik, A. (2002). 3D Visualization of Weather Radar Data.

Technical Report 3252, Linkping University, Depart-

ment of Electrical Engineering.

Ginn, E. W. L. (1999). From PPI to Dual Doppler Images

- 40 Years of Radar Observations at the Hong Kong

Observatory. In Proceedings of the 32nd Session of

the ESCAP/WMO Typhoon Committee.

Hartkamp, A., de Beurs, K., Stein, A., and White, J. (1999).

Interpolation techniques for climate variables. In

CIMMYT, editor, NRG-GIS Series 99-01, chapter 26.

Huijbregts, C. and Matheron, G. (1971). Universal kriging.

In Proc. of International Symposium on Techniques

for Decision-Making in Mineral Industry, page 159

169.

James, C. N., Brodzik, S. R., Edmon, H., Houze, R. A.,

and Yuter, S. E. (2000). Radar data processing and

visualization over complex terrain. Wea. Forecasting,

15:327 – 338.

Jenson, S. K. and Dominque, J. O. (1988). Extracting topo-

graphic structure from digital elevation data foar geo-

graphic information system analysis. Photogrammet-

ric Engineering and Remote Sensing, 54(11).

Kraak, M.-J. and Ormeling, F. (2002). Cartography: Visu-

alization of Geospatial Data. Pearson Education.

Mair, A. and Fares, A. (2011). Comparison of Rainfall

Interpolation Methods in a Mountainous Region of a

Tropical Island. Journal of Hydrologic Engineering,

16(4):371+.

Peng, C. and Lingda, W. (2007). 3D representation of

radar coverage in complex environment. International

Journal of Computer Science and Network Security,

7(7):139 – 145.

Peuquet, D. J. and Marble, D. F. (1990). Introductory Read-

ings in Geographic Information Systems. Taylor and

Francis.

Sundaram, V., Ru, Y., Benes, B., Zhao, L., Song, C. X.,

Park, T., Bertoline, G. R., , and Huber, M. (2008). An

integrated system for near real-time 3D visualization

of NEXRAD Level II Data using TeraGrid. In Tera-

Grid 08 - The 3rd Annual TeraGrid Conference, Las

Vegas, NV., pages 1 – 8.

Tomaszewski, B. M., Robinson, A. C., Weaver, C., Stryker,

M., and Maceachren, A. M. (2007). Geovisual analyt-

ics and crisis management. In Proceedings of the 4th

International ISCRAM Conference, May 13-16, 2007,

pages 173–179.

Toussaint, M., Malkomes, M., Hagen, M., Hller, H., and

Meischner, P. (2000). A real time data visualization

and analysis environment, scientiﬁc data management

of large weather radar archives. Physics and Chem-

istry of the Earth, Part B: Hydrology, Oceans and

Atmosphere, 25(1012):1001 – 1003. First European

Conference on Radar Meteorology.

Van Ho, Q., Lundblad, P.,

Astr

om, T., and Jern, M. (2012).

A web-enabled visualization toolkit for geovisual an-

alytics. Information Visualization, 11(1):22–42.

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

336