Signal and image processing for air pollution retrieval using remote

sensing technique over usm campus, malaysia

H. S. Lim, M. Z. MatJafri K. Abdullah,N. Mohd. Saleh and J. M. Rajab

School of Physics, Universiti Sains Malaysia, 11800 Penang, Malaysia

E-mail: hslim@usm.my, mjafri@usm.my, khirudd@usm.my, nasirun@usm.my and
jasim_rijab@yahoo.com
Tel: +604-6533888, Fax: +604-6579150

1
 ABSTRACT

In this study, we applied a newly developed algorithm to digital camera imageries to

determine the concentrations of particulate matter of size less than 10 micron (PM10) from
remotely sensed data. A conventional digital camera Kodak DC 290 was used as a sensor to
capture digital images of our reference targets. The objective of this study is to evaluate the
performance of a digital camera for providing useful remotely sensed data for air pollution
studies. Digital camera imageries were captured during a period of three days in 2006 in
the Universiti Sains Malaysia (USM) campus. The techniques were used to analyze multi-
temporal images. The digital imageries used in this study were captured under different
solar radiation, solar angle and weather conditions. The digital camera imageries were
separated into three bands assigned as red, green and blue bands for multispectral
algorithm analysis. The digital numbers of the three bands were converted into irradiance
and then reflectance. The relationship between the reflectance and the corresponding air
quality data was determined using regression analysis. An algorithm was developed based
on the aerosol properties to correlate the atmospheric reflectance and PM10. The in situ
PM10 data were collected simultaneously with the acquisition satellite image using a
DustTrak meter. A high correlation coefficient (R) was obtained in this study between the
measured and predicted PM10 values. The surface irradiance measurements were collected
using a spectroradiometer. Multiple regression algorithm was employed using the digital
imagery data for retrieval of the air quality measurements. This study indicated that the
digital camera can be a very good tool for air quality study.

Index Terms—PM10, Digital camera, DustTrak meter

1. INTRODUCTION
Presently, most data from satellite or airborne camera are in image form, thus most research
studies involve image processing analysis. Signal processing can contribute significantly in
extracting information from the remotely sensed waveforms or time series data. In remote sensing,
data captured by satellite or the airborne sensors have to be processed before they can be
transformed into a useable from. Software programmes are used to process the data in a computer.
Since data mainly contain images, they have to be analysed by using image processing techniques.
This paper tries to explain the potential and useful of digital image analysis based on image
processing technique for environmental pollution studies. Air pollution causes a number of health
problems and it has been linked to illnesses and deaths from heart or lung diseases. Some studies
showed that satellite data could be useful for revealing climatic and environmental implications of
global air pollution [1]. Air pollution in Asian cities has grown with the progressing
industrialization and urbanization.
Remote sensing has been widely used for environmental applications such as for air quality
and water quality studies. A considerable number of remote sensing systems have been placed into
orbit that have the capability to measure concentrations of many important air pollutants,
including ozone, NO2, CO, and SO 2 {e.g., TOMS, Global Ozone Monitoring Experiment [GOME],
Improved Limb Atmospheric Spectrometer [ILAS], and SBUV}. The atmosphere particles
associated with these events can significantly reduce visibility and can have detrimental health
effects [2]. Several studies have shown empirical relationships between satellite data and air
pollution [3, 4]. In fact, air quality can be measured using a ground instrument such as an air
sampler. But these instruments are quite expensive and a limited number of the air pollutant
stations can be set up in each area. So, they cannot provide a good spatial distribution of the air
pollutant over a city.
The objective of this study is to investigate the use of digital camera data to detect the air
pollutant using remote sensing and digital image processing technique. A normalization technique
applied in multi temporal data analysis. The PM10 data were collected using a handheld DustTrak
Aerosol Monitor 8520 simultaneously with the acquisition of the digital images. In-situ
measurements of corresponding air pollution parameters collected at the air pollution station in
University Sains Malaysia, Penang, were needed for algorithm calibration.

2. STUDY AREA
The selected air quality station was located in Universiti Sains Malaysia (USM) campus at the
longitude of 1000 17.864’ and latitude of 50 21.528’ (Figure 1). The site consists mainly of undulating
land and has many assets that make it an ideal of University campus. USM is situated in the
northeast district of Penang Island (Figure 1).

Figure 1: Study area and Air Quality Station.

 3. METHODOLOGY
The digital images were captured using a digital camera model DC Kodak 290 (Figure 2) during
there consecutive days from 5/1/2006 – 7/1/2006 started from 8.00 am. to 7.00 p.m, hourly. The air
pollution station selected in this study was based on the several reasons, 1). it was in the USM
campus, so it was easy to capture the digital images hourly. 2). a construction site was located near
to the station, so we expected that the PM10 levels would be high when the work was in progress.
3). the USM bus stop was near to the station, so we expected that the CO levels would be high
during the busy times in the morning and evening and 4). there is a building near the station, so we
can stick colour papers on the wall as reference surfaces. The air quality station belongs to ASMA
is shown in Figure 1. The digital images were captured hourly because the air qualities data were
also collected hourly.
We stuck four colour papers on the wall of a building (red, green, blue and black papers).
The colour papers were used as reference targets. Later we selected the black colour paper as our
reference target because it produced the highest accuracy based on the values of the correlation
coefficient (R) and root-mean-square (RMS). The camera axis was at 90 0 with the plane of the
colour paper. We captured the digital images of the reference targets by using the digital camera at
near and far distances from the targets. The far distance between the building and the camera was
100 meter. The air pollution station was located between the reference target and the digital
camera. The spectral reflectance values of the four colour papers were measured using a handheld
spectroradiometer.
The digital imageries were separated into three visible wavelength bands namely red, green
and blue bands. The techniques used in this study were (i) semi empirical, (ii) far-near and (iii)
normalization flat-filed techniques. Over a simple black target, the observed atmospheric
reflectance, is the sum of reflectance of aerosols and Rayleigh contributions (Equation 1). This
simplification, however, is not valid at short wavelengths (less than 0.45 pm) or large sun and view
zenith angles [5]. In this study, a simpler form than equation (1) was used in this study (Equation
2). This equation was also used by other researchers [6].
 Figure 2: The skylight parameter model [8].

The atmospheric reflectance was obtained by subtracting the reference reflectances from
the total reflectance observed by the digital camera. Then the retrieved atmospheric reflectance
values were used for regression analysis to determine the relationship between atmospheric
reflectance and PM10 concentration.

4. ALGORITHM MODEL
The atmospheric reflectance due to molecule, R r, is given as equation (3) [7]

 r Pr ()
Rr  (3)
4 s  v
 Rs – TRr = Ratm (1)
Rs – Rr = Ratm (2)
where:
Rs = reflectance recorded by satellite sensor
Rr = reflectance from surface references
Ratm = reflectance from atmospheric components (aerosols and molecules)
T = transmittance

where
τr = Rayleigh optical thickness
Pr(  ) = Rayleigh scattering phase function
µv = Cosine of viewing angle
µs = Cosine of solar zenith angle

We assume that the atmospheric reflectance due to particle, R a, is also linear with the τ a [9,
10]. The linear relationship was found between both aerosol and molecule scattering [7].
 a Pa ()
Ra  (4)
4 s  v
where
Pa(  ) = Aerosol scattering phase function
τa = Aerosol optical thickness

Atmospheric reflectance is the sum of the particle reflectance and molecule reflectance, R atm [11].
Ratm=Ra+Rr (5)
where
Ratm=atmospheric reflectance
Ra=particle reflectance
Rr=molecule reflectance

 P ()  r Pr () 
R atm   a a  
 4 s  v 4 s  v 
1
R atm   a Pa ()   r Pr () (6)
4 s  v
The optical depth is given as in equation (7) [12]. From the equation, we rewrite the optical depth
for particle and molecule as equation (8)
  s (7)
where
τ = optical depth
σ = absorption
s = finite path
   a  r

 r   r r s (8a)

 p  p ps (8b)

Equations (8) are substituted into equation (6). The result was extended to a three bands algorithm
as equation (9)
Form the equation; we found that PM10 was linearly related to the reflectance for band 1

and band 2. This algorithm was generated based on the linear relationship between τ and

reflectance. Other researcher also found that the PM10 was linearly related to the τ and the

correlation coefficient for linear was better that exponential in their study (overall) [13]. This

means that reflectance was linear with the PM10. In order to simplify the data processing, the air

quality concentration was used in our analysis instead of using density, ρ, values.

1
Ratm   a  a sPa ()   r  r sPr ()
4 s  v
s
R atm ( 2 )   a ( 2 ) PPa (,  2 )   r ( 2 )GPr (,  2 )
4 s  v
P  a 0 R atm (1 )  a1 R atm ( 2 ) (9)
where
P = Particle concentration (PM10)
G = Molecule concentration
Ratmi= Atmospheric reflectance, i = 1, 2 and 3 are the band number
aj= algorithm coefficients, j = 0, 1, 2, … are then empirically determined.
 5. DATA ANALYSIS AND RESULTS
Digital processing involves many possible procedures in which the digital numbers which represent
the image are manipulated. Many computer processes involve resampling which must keep in mind
the basic resolution of the data. The PCI Geomatica version 9.1.8 digital image processing software
was used in the image-processing analysis. The digital images were separated into three bands (red,
green and blue). The average DN for each digital image captured at near and far distances from the
reference targets were extracted. All the DN values were converted into irradiance using equations
10, 11 and 12.The DNs were converted into irradiance using the digital camera coefficients
calibrated for each band (Equation 10, 11 and 12). And then the irradiances were converted to
reflectance using the equation 12 for each band. The solar angles and Earth-Sun distance were
calculated corresponding to the time the digital image acquisition.
The calibrated digital camera coefficients are
y1 = 0.0005x1 + 0.0738 (10)
y2 = 0.0007x2 + 0.0517 (11)
y3 = 0.0006x3 + 0.0497 (12)
where
y1 = irradiance for red band (Wm-2 nm-1)
y2 = irradiance for green band (Wm-2 nm-1)
y3 = irradiance for blue band (Wm-2 nm-1)
x1 = digital number for red band
x2 = digital number for green band
x3 = digital number for blue band
A handheld spectroradiometer was used to measure the sun radiation at the ground surface. The reflectance
values was calculate using equation (13) below.
y( )
R 
E s ( )
(13)

where
y (λ) = radiance of each visible bands (Wm-2 nm-1)
Es (λ) = sun radiation at the ground surface using a hand held spectroradiometer
(Wm-2 nm-1)
After that, the reflectance recorded by the digital camera was subtracted by the reflectance
of the known surface and we obtained the reflectance caused by the atmospheric components. The
atmospheric reflectances were obtained using equation 2 and later for used algorithm regression.
 The relationship between the reflectance and the corresponding air quality data for the
pollutant was carried out using regression analysis. The proposed model produced high correlation
coefficient, R2 and low root mean square error, RMS for PM10 (Figure 2).
The proposed model produced the correlation coefficient of R=0.8125 and RMS value of
6.32561 µg/m3 using black colour paper as reference target.

115

R=0.8125 110
RMS=6.32561 µg/m3
105

Estimated PM10 (g/m3)

100

95

90

85

80

75
70 80 90 100 110 120

Measured PM10 (g/m3)

Figure 3: Relationship between measured and estimated PM10 (g/m3)

6. CONCLUSION
Preliminary, the analysis indicated that digital camera imagery could provide useful
remotely sensed data for air pollutant measurements using a multiband algorithm. In this
paper, the computation for PM10 requires the surface radiations values. This new finding
will be further investigated and evaluated with the aim to improve the accuracy. We found
that the black colour paper was the best reference surface used in this study because it
produced the highest accuracy compared to the results using green, blue and red colour
papers. This proposed technique gives an alternative way to provide low cost digital
imageries for environmental air pollution application. Further study must be carried out to
increase the digital camera capacity to estimates air quality.

7. ACKNOWLEDGEMENTS
This project was carried out using the USM short term grants and Science Fund (04-01-05-SF0035).
We would like to thank the technical staff who participated in this project. Thanks are extended to
USM for support and encouragement.
in this project. Thanks are extended to USM for support and encouragement.

8. REFERENCES

[1] Akimoto, H., 2003, Global Air Quality and Pollution”, Science, 302: 1716 – 1719.
[2] Science Applications International Corporation, (2002) An Examination of the Application
of Remote Sensing Data in Detecting and Characterizing Air Pollution Transport and Emissions,
Report, 21-30, [Online] available: http://www.epa.gov/ttn/chief/eiip/pm25inventory/remsens.pdf.
[3] Ung, A., Weber, C., Perron, G., Hirsch, J., Kleinpeter, J., Wald, L. and Ranchin, T., 2001.
Air Pollution Mapping Over A City – Virtual Stations And Morphological Indicators. Proceedings
of 10th International Symposium “Transport and Air Pollution” September 17 - 19, 2001 – Boulder,
Colorado USA, [Online] available: http://www-
cenerg.cma.fr/Public/themes_de_recherche/teledetection/title_tele_air/title_tele_air_pub/air_polluti
on_mappin4043.
[4] Weber, C., Hirsch, J., Perron, G., Kleinpeter, J., Ranchin, T., Ung, A. and Wald, L. 2001.
Urban Morphology, Remote Sensing and Pollutants Distribution: An Application To The City of
Strasbourg, France. International Union of Air Pollution Prevention and Environmental Protection
Associations (IUAPPA) Symposium and Korean Society for Atmospheric Environment (KOSAE)
Symposium, 12th World Clean Air & Environment Congress, Greening the New Millennium, 26 –
31 August 2001, Seoul, Korea. [Online] available: http://www-
cenerg.cma.fr/Public/themes_de_recherche/teledetection/title_tele_air/title_tele_air_pub/paper_urb
an_morpho.
[5] Vermote, E. and Roger, J. C., 1996, Radiative transfer modeling for calibration and
atmospheric correction, Advances in the Use of NOAA AVHRR Data for Land Applications, 49-72.
[6] Popp, C., Schläpfer, D., Bojinski, S., Schaepman, M. dan Itten, K.I. (2004) Evaluation of
Aerosol Mapping Methods using AVIRIS Imagery. R. Green (Editor), 13th Annual JPL Airborne
Earth Science Workshop. JPL Publications, March 2004, Pasadena, CA, 10.
[7] Liu, C. H., Chen, A. J. and Liu, G. R., 1996, An image-based retrieval algorithm of aerosol
characteristics and surface reflectance for satellite images, International Journal Of Remote
Sensing, 17 (17), 3477-3500.
[8] Asmala, A and Hashim, M., 1997, Determination of Haze from Satellite Remotely Sensed
Data: Some Preliminary Results, [Online] available:
http://www.gisdevelopment.net/aars/acrs/1997/ps3/ps3011.shtml.
[9] King, M. D., Kaufman, Y. J., Tanre, D. dan Nakajima, T., 1999, Remote sensing of
tropospheric aerosold form space: past, present and future, Bulletin of the American
Meteorological society, 2229-2259.
[10] Fukushima, H., Toratani, M., Yamamiya, S. and Mitomi, Y., 2000, Atmospheric correction
algorithm for ADEOS/OCTS acean color data: performance comparison based on ship and buoy
measurements. Adv. Space Res, Vol. 25, No. 5, 1015-1024.
[11] Vermote, E., Tanre, D., Deuze, J. L., Herman, M. and Morcrette, J. J., 1997, Second
Simulation of the satellite signal in the solar spectrum (6S), [Online] available:
http://www.geog.tamu.edu/klein/geog661/handouts/6s/6smanv2.0_P1.pdf.
[12] Camagni, P. and Sandroni, S., 1983, Optical Remote sensing of air pollution, Joint Research
Centre, Ispra, Italy, Elsevier Science Publishing Company Inc.
[13] Retalis, A., Sifakis, N., Grosso, N., Paronis, D. and Sarigiannis, D., 2003, Aerosol optical
thickness retrieval from AVHRR images over the Athens urban area, [Online] available:
http://sat2.space.noa.gr/rsensing/documents/IGARSS2003_AVHRR_Retalisetal_web.pdf.