Information about http://mpa.itc.it/papers/neteler_gisideas2004_in_press.pdf
MODIS TIME SERIES REMOTE SENSING FOR …
Tags: data density, epidemiological studies, geographical information system, geographical information systems, land surface temperature, land surface temperatures, logical applications, markus neteler, moderate resolution imaging, moderate resolution imaging spectroradiometer, modis data, mpba, povo trento, quality maps, quality pix, resolution imaging spectroradiometer, surface temperature maps, tick borne diseases, trento italy, vegetation indices,Pages: 7
Language: english
Created: Sun Aug 20 13:16:18 2006
Display cached document
Page 1
Page 2
Page 3
Page 4
Page 5
Page 6
Page 7
MODIS TIME SERIES REMOTE SENSING FOR
EPIDEMIOLOGICAL MODELLING
Markus Neteler
ITC-irst, SSI/MPBA
Via Sommarive, 13
38050 Povo (Trento), Italy
Email: neteler at itc dot it
ABSTRACT
This paper reports on the processing of time series of MODIS NDVI/EVI and LST satel-
lite data in a Geographical Information System (GIS). The required data preparations for
the integration of MODIS data in GIS is described with focus on the reprojection from
MODIS/Sinusoidal projection to national coordinate systems. To remove low quality pix-
els, the MODIS quality maps are utilized. We explain subsequent filtering of Land Surface
Temperature maps with an outlier detector to eliminate originally undetected cloud pixels.
Further analysis of time series is briefly discussed as well as related landscape epidemio-
logical applications in the field of tick-borne diseases.
1 INTRODUCTION
In epidemiological modelling, survey data are usually collected at sampling sites and then
regionalised in Geographical Information Systems (GIS). To enhance the spatial data density,
continuous field data such as land surface temperatures (LST), snow coverage, and vegetation
indices are commonly derived from satellite data. The launches of the new satellite systems
Terra (December 1999) and Aqua (May 2002) significantly improve the situation of data avail-
ability for scientific purposes and predictive epidemiological studies. The Moderate Resolution
Imaging Spectroradiometer (MODIS) is a key instrument on both Terra and Aqua satellites. As
they each deliver twice a day global coverages at 250m (Red, NIR), 500m (MIR) and 1000m
resolution (TIR), they are most interesting to support epidemiological studies. Usually one
week after acquisition the data sets are available to the public.
The orbit of Terra around the Earth is scheduled to pass from north to south across the
equator in the morning, while Aqua passes with reverted direction from south to north over the
equator in the afternoon. Terra, crossing the equator at about 10:30 a.m. local solar time, is in
a sun-synchronous orbit with a delay of 30 minutes with respect to LANDSAT-7. The further
orbital parameters are equal to those of LANDSAT-7.
MODIS is a whisk-broom sensor with 36 channels ranging from visible to thermal-
infrared (GSFC/NASA, 2003). Data are delivered at 250m (2 channels), 500m (5 channels) and
1000m resolution (29 channels). MODIS can be considered as a much enhanced successor of
the AVHRR instrument onboard the NOAA series of satellites. MODIS improves upon the per-
formance of AVHRR by providing both higher spatial resolution and greater spectral resolution,
and has been used as a major source for the design for AVHRR's planned operational successor
(VIIRS, http://www.ipo.noaa.gov/Technology/viirs_summary.html), cur-
rently anticipated for launch late in 2006. This paper focuses on two of the numerous MODIS
data products, the Land Surface Temperature (LST) and Vegetation Index 16-day composites
(NDVI and EVI).
International Symposium on Geoinformatics for Spatial Infrastructure Development in Earth and Allied Sciences 2004
2 EPIDEMIOLOGICAL RISK MAPS AND STUDY REGION
The methods described in this paper have been implemented to extend an epidemiological
study about the exposure risk to Lyme disease transmitted by ticks in the Autonomous Province
of Trento, Italian Alps. The predictions are carried out through the analysis of the distribution
of Ixodes ricinus (L.) nymphs infected with Borrelia burgdorferi s.l. with a model based on
tree-based classifiers. This model is supported by a Geographical Information System (Merler
et al., 1996; Rizzoli et al., 2002; Furlanello et al., 2003). In this project data on I. ricinus (L.)
density, assessed by dragging the vegetation in 438 sites during 1996 were cross-correlated with
the digital cartography of a GIS.
The area includes the Autonomous Provinces of Bolzano and Trento and the Province of
Belluno (Italy), a region of approximately 18,000 km2 area size. The complex terrain has an
elevation range from slightly above sea level to 3800 m (Bolzano: elev. range 3700m, mean
elev. 1800m, mean slope 26 ; Trentino: elev. range 3700m, mean elev. 1400m, mean slope
26 ; Belluno: elev. range 3200m, mean elev. 1500m, mean slope 27 ).
The density of available meteorological stations is highly varying and often concentrates
in the valleys. To improve the risk mapping at high resolution, other sources of climatic and
vegetational data are required. This demand can be fulfilled by using remote sensing data.
The integration of satellite data into epidemiological research enhances the spatio-temporal
resolution of climatological data in particular in mountainous regions where climatic stations
and ground surveys are unavailable or sparse (Hess et al., 2002).
3 MODIS DATA PREPROCESSING FOR GIS USAGE
MODIS data sets are delivered as "Base Level Swath Data" as well as "gridded data".
Grid data are projected in either Integerised Sinusoidal (ISIN, Level V003) or Sinusoidal (SIN,
Level V004) projections. Both data types require further preprocessing before they can be used
in a GIS. Especially ISIN is usually unsupported in most of "off the shelf" and free GIS and
image processing software. The "MODIS Reprojection Tool" (MRT, U.S. Geological Survey
2004) can be used to reproject both ISIN and SIN to a more common projection (e.g. UTM)
or to national grid systems such as Gauss-Boaga (Italy). Furthermore MRT allows for geo-
graphical subsetting. It writes the output to standard data formats such as Geo TIFF, and is
executable on various operating systems. In April 2004 the V003 product were removed from
the USGS archives as all existing MODIS data were reprocessed to V004. The V004 data qual-
ity is significantly higher, especially due to changes in the processing of inland water pixels
(Wan, 2003).
Further pre-processing steps after the reprojection comprise the pixel-wise application of
the quality map provided along each data product. This is followed by an outlier detection for
certain MODIS products to minimise the presence of low quality pixels not properly indicated
in the quality maps. The method is explained in greater detail in the following part.
3.1 Preprocessing of Land Surface Temperatures (LST) data
MODIS Land Surface Temperature and Emissivity (LST/E) products are mapping land
surface temperatures and emissivity values. The underlying algorithms use other MODIS
data as input, including geolocation, radiance, cloud masking, atmospheric temperature, wa-
ter vapour, snow, and land cover. Temperatures are extracted in Kelvin; accuracy of 1 Kelvin is
yielded for materials with known emissivities (Wan, 1999).
MODIS time series remote sensing for epidemiological modelling
After reprojection, we applied pixel-wise the related quality maps to the reprojected LST
maps. This step is required due to limitations in the official cloud detection algorithm as used
to create the land surface temperature quality maps. In particular, cloud detection during the
night pass is error-prone at cloud corners. Also other quality variations occur due to aerosol and
thin cloud presence which are not always indicated in the quality map. To overcome this prob-
lem, a simple outlier detection for the minimum temperatures was integrated into the proposed
procedure:
lower_boundary = 1st_Quartile - 1.5 (3rd_Quartile - 1st_Quartile) (1)
Here we considered only LST maps with a sufficient number of pixels. To avoid that a few
deviating pixels influence the overall outlier statistics, we were choosing as minimum the avail-
ability of at least 25 % valid LST pixels in a map. In case of less usable pixels the entire map
was rejected and not considered for the outlier detection. Quartiles and lower boundaries were
calculated monthly-wise for all accepted LST maps. The LST maps were filtered then on a
monthly base with the mean of the monthly lower boundary threshold values.
Due to the temperature differences in the day and night passes of MODIS the thresholds
were considered separately. The proposed outlier filter aims at removal of all pixels which con-
tain cloud top surface temperatures. Especially in the night overpasses these cloud contaminated
pixels sometimes remain undetected by the NASA quality algorithm.
3.2 Preprocessing of Vegetation Indices (NDVI/EVI)
The two MODIS vegetation indices, the classical Normalised Difference Vegetation Index
(NDVI) and a newly developed Enhanced Vegetation Index (EVI) are spectral measures of the
amount of vegetation present on the ground. The MODIS-NDVI describes the relative "green-
ness" of the Earth's vegetation on a scale of minus one (-1) to plus one (+1) and is intended to
continue the time series more of than 20 years of NOAA/AVHRR-derived NDVI. The magni-
tude of NDVI is related to the level of photosynthetic activity in the observed vegetation. The
EVI is MODIS-specific and offers improved sensitivity in high biomass regions while being
less sensitive to atmospheric aerosol scattering (especially smoke from burning vegetation). It
also minimises the influence of background interference caused by bare soil reflecting off the
ground (Huete et al., 1999).
These vegetation indices can be integrated into epidemiological models to reflect vegeta-
tion dynamics. The 16-day composite product minimises cloud cover problems by substituting
a cloud covered pixel with a later uncontaminated pixel within a 16-days period. These com-
posites reflect the current vegetation status at a sufficient temporal resolution. In order to use
these MODIS map products in epidemiological applications, the beforehand reprojected NDVI/
EVI maps have to be filtered using the related quality bit pattern maps. These quality maps
contain information about pixel quality, aerosol contents, cloud, water or snow presence etc. As
they are encoded in 16 bit, a bit pattern module was developed for GRASS GIS to perform this
operation (r.bitpattern).
4 MODIS TIME SERIES ANALYSIS
The study has been carried out with GRASS GIS software (Neteler and Mitasova, 2004).
The recent implementation of general time series processing (r.series) for GRASS raster
maps supports univariate statistics for a series of MODIS scenes. By selecting various time
ranges and operators, various indicators can be calculated.
International Symposium on Geoinformatics for Spatial Infrastructure Development in Earth and Allied Sciences 2004
Table 1: Number of observations for calculation of monthly mean temperatures (2001)
from Cavedine meteorological station and related MODIS/Terra data.
Data Month of year 2001 Act./Pot.
source 1 2 3 4 5 6 7 8 9 10 11 12 Observ.
Meteo 721 649 722 697 721 697 721 721 697 720 697 721 8484/8760
MODIS 10 32 15 29 23 8 29 35 24 35 32 20 292/730
4.1 Validation and applications of daily LST data
To validate the usability of MODIS/Terra data in epidemiological studies as an enhance-
ment of data availability, the monthly mean temperatures of selected meteorological stations
and the related MODIS data at the same coordinates have been investigated. Fig. 1 shows
monthly mean temperatures of Cavedine meteorological station (Trentino, Italy) compared to
mean LST results from MODIS/Terra. Despite missing pixels due to cloud and high aerosol
presence the curves are matching surprisingly well except for January (1), February (2) and
July (6). Table 1 reports on the number of observations used for the calculation of the monthly
mean temperatures. Only in case of months with nearly continuous cloud cover the mean tem-
peratures deviate significantly. Fig. 2 shows a boxplot which confirms that the data distribution
of both data types is similar. A two-sample Kolmogorov-Smirnov test on both distributions
results in D=0.1667 and p-value=0.9985, confirming that the distributions significantly match.
It is important to note that LST temperatures are not identical to air temperatures as measured
by meteorological stations. However, efforts have been undertaken to derive air temperatures
Monthly mean temperatures 2001 at Cavedine, Trentino, Italy
30
Meteo station (black)/ MODIS LST (red) monthly means
25
Meteo station data
20
MODIS/Terra data
15
10
5
0
-5
2 4 6 8 10 12
Months in year 2001
Figure 1: Comparison of monthly mean temperatures (2001) from Cavedine meteorolog-
ical station (hourly data) and related MODIS/Terra V003 data (at max. two
values per day). The standard deviation is indicated (bar for station, dot for
MODIS).
MODIS time series remote sensing for epidemiological modelling
Meteostation 2001, MODIS/Terra 2001,
Lago di Cavedine Lago di Cavedine
20
20
Monthly mean temperatures
Monthly mean temperatures
15
15
10
10
5
5
0
-5
0
Figure 2: Boxplot of monthly mean temperatures (2001) from Cavedine meteorological
station (hourly data) and related MODIS/Terra data (max. two values per day).
from surface temperatures (e.g., Goetz et al. 2000).
An application related to the distribution of tick-bourne diseases is the calculation of "au-
tumnal cooling". This index is calculated by linear regression to describe the autumnal temper-
ature decline from August to October (northern hemisphere). Sites of tick-bourne encephalitis
(TBE) appear to be characterised by a high rate of autumnal cooling, relative to the annual max-
imum of the monthly mean LST level in midsummer (Randolph et al., 2000). Using the raster
map time series calculator of GRASS (r.series) this index can be easily derived for all years
of available MODIS data.
4.2 Applications of NDVI/EVI 16-day composites
The temporal dynamics of vegetation are an important indicator of vegetation type and,
consequently, moisture conditions on the ground. This is an important predictor of suitable
habitat for I. ricinus (Randolph, 2001). Two applications can be identified: the direct integra-
tion of NDVI/EVI values into presence/absence models for infectious diseases. Then it is also
promising to calculate temporal NDVI/EVI differences in order to determine the spatial pat-
terns of spring duration etc. These patterns may have influence on both the life-cycle of (small)
mammals as hosts and ticks as vectors.
5 CONCLUSIONS AND FUTURE RESEARCH
The usability of MODIS LST and NDVI/EVI data for epidemiological studies appears to
be promising. More that 3 years of data are available now, extending the 20 years of time series
of AVHRR/NOAA. All available MODIS/Terra data sets have been processed for the study area.
To enhance the temporal resolution of LST maps, the integration of the MODIS/Aqua data is
ongoing which then provides four maps per day from May 2002 onwards (in case of cloudfree
International Symposium on Geoinformatics for Spatial Infrastructure Development in Earth and Allied Sciences 2004
conditions). The study on TBE presence and its relationship to "autumnal cooling" is currently
ongoing for the Trentino (Northern Italy) study area.
Further work is ongoing to relate the density of rodents as tick hosts to the temporal
variations of the EVI over the year.
ACKNOWLEDGEMENTS
M. Neteler was supported by the FUR-PAT Project WebFAQ. The meteorological data
have been provided from the Autonomous Province of Trento. Dr. A. Rizzoli/CEA is acknowl-
edged for her support with ticks related data of the ECODIS project.
REFERENCES
Furlanello, C., Neteler, M., Merler, S., Menegon, S., Fontanari, S., Donini, A., Rizzoli, A., and
Chemini, C., 2003. GIS and the Random Forest Predictor: Integration in R for Tick-borne
Disease Risk Assessment. In Hornik, K. and Leisch, F., editors, Proceedings of the 3rd
International Workshop on Distributed Statistical Computing (DSC2003, Vienna, Austria,
March 20-22).
Goetz, S., Prince, S., and Small, J., 2000. - In: S.I. Hay and S.E. Randolph and D.J. Rogers,
2000. Remote sensing and Geographical Information Systems in epidemiology, chapter
Advances in satellite remote sensing of environmental variables for epidemiological ap-
plications, pages 289307. Academic Press.
GSFC/NASA, 2003. MODIS technical specifications. Electronic document:
http://modis.gsfc.nasa.gov/about/specs.html.
Hess, G., Randolph, S., Arneberg, P., Chemini, C., Furlanello, C., Harwood, J., Roberts, M.,
and Swinton, J., 2002. - In: P.J. Hudson and A. Rizzoli and B.T. Grenfell and H.
Heesterbeek and A. P. Dobson, 2002. The Ecology of Wildlife Diseases, chapter Spatial
aspects of disease dynamics, pages 102118. Oxford University Press.
Huete, A., Justice, C., and Leeuwen, W., 1999. MODIS vegetation index (MOD 13).
Algorithm theoretical basis document ATBD13. Electronic document:
http://modis.gsfc.nasa.gov/data/atbd/atbd_mod13.pdf.
Merler, S., Chemini, C. F. C., and Nicolini, G., 1996. Classification tree methods for analysis
of mesoscale distribution of Ixodex ricinus (Acari: Ixodidae) in Trentino, Italian Alps.
Journal of Medical Entomology, 33(6):888893.
Neteler, M. and Mitasova, H., 2004. Open Source GIS: A GRASS GIS Approach. Number 773
in SECS. Kluwer Academic Publishers, Boston, 2nd edition.
Randolph, S., 2001. The shifting landscape of tick-borne zoonoses: tick-borne encephalitis
and lyme borreliosis in europe. Phil. Trans. Royal Soc. London B, 356:10451056.
Randolph, S., Green, R., Peacey, M., and Rogers, D., 2000. Seasonal synchrony: the key to
tick-borne encephalitis foci identified by satellite data. Parasitology, 121:1523.
MODIS time series remote sensing for epidemiological modelling
Rizzoli, A., Merler, S., Furlanello, C., and Genchi, C., 2002. Geographical information system
and bootstrap aggregation (bagging) of tree-based classifiers for lyme disease risk
assessment in Trentino, Italian Alps. Journal of Medical Entomology, 39(3):485492.
U.S. Geological Survey, 2004. MODIS Reprojection Tool V3.1. Software:
http://edc.usgs.gov/programs/sddm/modisdist/index.shtml.
Wan, Z., 1999. MODIS Land-Surface Temperature. Algorithm theoretical basis document
(LST ATBD): LST calculations. Electronic document:
http://modis.gsfc.nasa.gov/data/atbd/atbd_mod11.pdf.
Wan, Z., 2003. MODIS Land-Surface Temperature products users' guide. Electronic
document:
http://www.icess.ucsb.edu/modis/LstUsrGuide/usrguide.html.
International Symposium on Geoinformatics for Spatial Infrastructure Development in Earth and Allied Sciences 2004