CHARACTERIZING LANDSCAPE PROPERTIES IN

SOUTHEASTERN SPAIN BY COMBINING CLASSIFICATION AND

MULTI-DATE SPECTRAL MIXTURE ANALYSIS

Achim Röder 1 , Stella Hubert, Marion Stellmes

University of Trier, FB VI Geography/Geosciences,
Remote Sensing Department
Campus II
D-54286 Trier
Germany
+49.651.2014606
roeder@uni-trier.de
Peter Scarth
Remote Sensing Centre, Queensland Department of Environment and
Resource Management
Brisbane
Australia
Susana Bautista, Jordi Cortina
Universidad de Alicante, Faculdad de Ecologia
Alicante
Spain

Abstract
Mapping land use and land cover is one of the most frequently addressed topics
in remote sensing, owing to the importance of such information for a wide range
of applications that range from local land management to global climate models.
Local to regional scale data, such as those acquired by the Landsat sensors,
provide long-term monitoring capacities along with a scale that is consistent
with sub-national land management. These data allow mapping land use and
landing cover for a sequence of dates and comparing results both in terms of
overall rates as well as in a spatially explicit manner. Yet, it is now accepted that
for many environmental applications, the modification of ecosystems – even if in
categorical terms they remain stable – may prove equally important than
transformation from one land use type to another. Consequently, adapted data
processing and mapping strategies are need to consider both types of change.
Also, “translating” the derived information is essential to comply with
assessment frameworks such as “ecosystem services and functions”, which
may eventually form the basis for assessing trade-offs associated with land-use
decisions. In this study, we report first steps towards this goal and present a
combined approach for enhanced land use and land cover mapping based on a

1
Corresponding author

1
rigorous radiometric data processing scheme It incorporates a hybrid
unsupervised classification approach using available reference data for
histogram based class assignment, and a multi-date spectral unmixing
approach to further differentiate natural and semi-natural vegetation classes.
Results are discussed against the background of major socio-economic trends
and physical processes that are typical for many countries of the European
Mediterranean.

Introduction
Assessing the state of ecosystems is one of today’s foremost issues; yet,
depending on the individual perspective this can be highly subjective. The
Millenium Ecosystem Assessment promoted the concept of Ecosystem
Services and Functions (Millenium Ecosystem Assessment 2005), which groups
these into major groups of provisioning, regulating, preserving and cultural
services. This may serve as a conceptual framework to assess the state of
ecosystems with respect to the individual functions and services, while rating
can then be performed according to the respective individual management or
development goals. Many pressures are being exerted on ecosystem services
and functions, not the least arising from often competing demands by different
groups of stakeholders with individual priorities such as conservation, economic
development, food appropriation etc. In this context, land use and its change
have been identified as major driving forces which hence require adequate
attention (DeFries et al. 2004).
In the recent past, there have been manifold efforts to provide consistent land
use / land cover maps that cover national, continental or even global scales
(Loveland et al. 2000), and to combine such data with climatological data or
information on human settlement (Ellis and Ramankutty 2008). Such data sets
are essential to provide input to global models and support continental to global
policy initiatives. Yet, they are often limited to one particular date or period, or
are only available for relatively short time periods (e.g. MODIS land cover
product). In particular, they are often not suitable to match spatial scales
relevant for land managers, or at which individual land use decisions are being
taken and implemented. In this line, the importance of local scale studies to
understand human-environment systems has been reiterated. In addition to the
traditional concept of assessing land use and land cover change by diachronic
analyses, there is growing evidence that modifications of land cover, although
within a particular land cover class, might be of equal importance (Lambin and
Geist 2006). For instance, increasing or decreasing vegetation cover within one
broad vegetation class might have strong implications for fire risk, soil erosion
or carbon sequestration, among others.
Therefore, three major prerequisites emerge for land use / land cover change
analyses in a wider ecosystem perspective: a) they should ideally cover both,
transformation and modification processes; b) they should be linked to driving
forces to contribute to understanding of coupled human-environment systems;
and c) they should provide results in a way they can be linked to ESF to make
them useful in assessing tradeoffs between certain land use decisions.

2
In this study, we focus on the methodologies that might be used to address
mapping land use and land cover transformations and modifications in as much
as possible an operational way.

Test area and data sets

A test site in South-Eastern Spain was selected for this task, which corresponds
to one full Landsat frame (path 199/row 33) and comprises parts of the
provinces of Valencia, Alicante, Albacete, Cuenca and Murcia. Altogether, it is
representative of a wide range of historic and recent change processes typical
for many countries of the European Mediterranean in general, as well as Spain
in particular. Evidently, accession to the European Union in 1986 has been a
primary driver of economic developments ever since.
In general, landscape change in the region is characterized by both,
intensification and extensification processes occurring often in close proximity
(Hobbs et al. 1995). On the one hand, there is a strong tendency towards
urbanization, aggravated by increasing touristic uses. Both are concentrated
along the coastal zones and cause a strongly increasing consumption of
freshwater (and land). At the same time, an intensive agricultural use is noted in
suitable locations, which adds to the problem of water consumption. On the
other hand, in less favorable areas land abandonment occurs, often
accompanied by ageing rural societies. This process in turn is often followed by
shrub encroachment associated with high risk of wildfires. The latter are
particularly evident in the northern part of the test area where, depending on fire
frequencies, different states of stability, disturbance and recovery may be
observed, ranging from low, primarily herbaceous cover, to dense mattorrals
and forests. These are mainly composed of pine tree species, which are also a
source of timber used commercially. This is contrasted by Stipa steppes
occurring further south, where significantly lower precipitation prevents
development of dense vegetation cover.
A major factor responsible for this land cover pattern is the elevation, ranging
from sea level to ~1300m a.s.l. in higher mountain ranges and around 750m
a.s.l. towards the Meseta higher plains further west. Precipitation shows a
gradient that is mostly oriented in North-South direction with amounts ranging
from more than 500 mm in Ayora region to around 300 mm around Alicante. In
the plain areas, mostly fertile, alluvial soils support agricultural use, while
mountainous regions are commonly characterized by shallow soils forming on
calcareous rocks of the Betic Cordillera.
Individual treatise of transformation and modification processes poses specific
requirements to the data used. Most commonly, land use classification
approaches rely on multi-temporal data for specific periods, since these allow
incorporating seasonality information to discriminate agricultural uses or
different vegetation types. On the other hand, assessing modifications requires
either hypertemporal data or, if undertaken at local scale, a time series of
images acquired at comparable phenological dates. For this reason, we have
set up a data base consisting of 31 Landsat-TM and -ETM+ images and that
covers the time span from 1984 to 2007. Wherever possible, one image per
year was acquired for the period of maximum photosynthetic activity (preferably

3
in a time window between April and June), and complementary images from
other seasons where added for different dates, in particular for 1986 to
characterize the state prior to accession to the EU, and for 2007 representing
the current situation. In addition, MODIS single day reflectance and SRTM
digital elevation data were used.

Data pre-processing
Geometric processing
For data acquired from Eurimage ™ we adopted a geometric correction
approach based on cross-correlation windows between an uncorrected Landsat
image and an illuminated digital elevation model (Hill and Mehl 2003). More
than 2000 ground control points were retrieved this way and enabled a highly
accurate correction of a master image, including the elevation model to account
for non-systematic distortions. Subsequently, the same algorithm was employed
to find ground control points for the remaining images. Data provided by USGS
were downloaded as level 1G orthorectified products. Thorough checks showed
these data to be highly consistent with the data rectified as described above,
such that they could be incorporated without further processing.

Radiometric processing
Radiometric processing is essential if data from different dates are to be jointly
evaluated and quantitatively analyzed. We have therefore employed a full
radiative transfer model approach based on a modified 5S code (Tanré 1990),
which was extended to account for topography-induced illumination effects.
Using a digital elevation model, the sun-surface-sensor constellation at the time
of image acquisition is characterized and shaded elements as well as visible
sky fraction are calculated. Then, the information is incorporated into the
radiative transfer model; direct irradiation is corrected using a cosine correction,
while diffuse irradiation terms are separated into their isotropic and a anisotropic
component using the anisotropy index by Hay and McKay (1985). The latter is
corrected using a cosine correction, while the first is corrected making use of
the visible sky fraction to weight the amount of incoming diffuse irradiance (Hill
and Mehl 2003, Röder et al. 2008). Sensor calibration constants were based on
in-flight calibration data in the case of Landsat-7 ETM+, or on vicarious
calibration data for Landsat-5 (Helder et al. 2008).
The transfer model is parameterized using Angstrom coefficients to determine
aerosol optical depth of the atmosphere, and by characterizing gaseous
transmission based on the Modtran code (Berk et al. 1999) as a function of
water vapor concentrations. In the absence of specific measurements, an
approach utilizing MODIS single day reflectance data (MOD09GA) was
developed. This draws from the documented high quality of the radiometric
correction chain implemented for MODIS (Liang et al. 2002). It involves an
adjustment of MODIS bands to match TM/ETM+ bandpaths, and application of
a Gaussian smoothing filter and subsequent resampling of the Landsat data, to
match the low pass effect and lower geometric resolution of the MODIS data.
Assuming that after this procedure both data sets are fully congruent,
signatures for areas with different brightness levels were extracted. Using a

4
minimization algorithm, the radiative transfer model was used to seek the
Angstrom coefficients and water vapour concentrations that allowed to best
match corrected Landsat reflectances with MODIS reference reflectances. The
parameter combination found this way was then utilized to correct the Landsat
image in its original resolution and including the information on topography-
related illumination effects. In this way, a corrected Landsat master image was
prepared and used to subsequently correct the remaining Landsat scenes using
pseudo-invariant features and the same matching approach. The resulting time
series was found to be quantitatively consistent, with limitations mainly resulting
from extreme topographic effects with images acquired at low sun elevation
angles, i.e. early or late in the year.

Image classification
Histogram-based classification of non-natural areas
Mapping transformation of land use or land cover classes was based on two
multi-temporal data sets, covering the time frame of 1986 (23/06/1986 and
26/08/1986) and 2005-2007 (26/5/2005, 03/07/2007, 04/08/2007). The
classification strategy was based on the CORINE land cover data (Coordinated
Information on the European Environment) available from the European
Environment Agency (http://www.eea.europa.eu/data-and-maps/data). It is
available for 1990, 2000 and an update for 2006 is underway. The data sets are
based on photo-interpretation of aerial photographs, Landsat-7 ETM+ and
SPOT 4/IRS 6, all with a minimum mapping unit of 25 hectares. These data sets
were aggregated to the 8 target classes “Urban, Rainfed Agriculture, Irrigated
Agriculture, Permanent Crops, Bare Ground, Herbaceous, Transitional
Woodland, Woodland”.
Our approach builds on the assumption that, despite reported deficiencies of
the CORINE data base, broad land use / land cover categories are adequately
represented. For both multi-temporal stacks, principal components were
calculated and the first 9 and 12 bands were retained. To provide additional
discriminative power, NDVI was calculated for every image and added as an
additional band to the respective layer stack. The respective data sets were
used as input for an unsupervised classification using an ISODATA clustering
algorithm, which yielded 100 spectral classes each. Then, histograms were
calculated for each of these classes, representing the frequencies of thematic
classes of the CORINE reference, using the CORINE 1990 and 2000 layers for
both dates. For each spectral class, the thematic class was assigned if it
contributed more than 60% to the overall pixels. In case of more ambiguous
frequency distributions, classes were manually checked and assigned using a
combination of the majority approach described before and contextual
information. As classification of urban areas poses specific requirements, it was
decided to include the respective CORINE areas in the classified maps.
While no validation of results was possible for the early date, the recent
classification was validated using a stratified random sampling approach. 1011
points were generated in relation to the respective classes (excluding the Urban
class), exported to Google Earth ™ and labelled with the apparent class.
Depending on acquisition conditions, image quality etc., some classes are

5
difficult to detect visually. Therefore, validation was carried out at three levels; at
the first, all classes were used; at the second, agricultural classes were merged
while natural/semi-natural classes remained unchanged; at the third, only
agriculture, woody and non-woody cover was evaluated. Considering
uncertainties in mapping validation data, seasonal difference between
acquisition dates etc., results were considered satisfactory with kappa values of
0.65, 0.75 and 0.9, respectively.

Multi-date spectral mixture analysis

Given the relevance of vegetation communities in the context of fire risk and
post-fire dynamics, or nutrient flows in sparsely vegetated Stipa steppes, an
important goal was to further differentiate semi-natural and natural areas. We
have adopted a multi-date spectral mixture analysis approach (Kuemmerle et al.
2006) based on a simple representation of the spectral feature space and its
temporal dimension. Non-natural areas where masked out, and from the
remaining areas every 20th pixel was retained. Then, a composite image was
calculated, incorporating the selected pixels from all images of the respective
target period to represent the full spectral dimensionality. Following an
orthogonal subspace projection to reduce the number of bands to one less than
the number of end-members, a simplex-based method was applied to identify
the extremes of the feature space. The method assumes that the volume of a
simplex constructed within a data cloud is maximized if the purest endmembers
present in an image are found. The respective pixels are found by constructing
the volume based on an initial set of randomly selected endmembers, and then
replacing each endmember with each other pixel (Winter 1999). This way, a
simple three endmember set was set up, including spectra for
photosynthetically active green vegetation, a background spectrum, and a zero
reflectance endmember to account for shading effects. The model was applied
to all individual data sets, and a subsequent shade-normalization was carried
out.
To validate the results, aerial photos were employed to map woody cover and
then results were integrated to match the 30 m resolution of the Landsat data.
Validation was carried out by comparing unmixing results of the late summer
image on a pixel-per-pixel basis, for a 3x3 window and for the best fitting pixel
within this window. This way, potential inaccuracies in different pre-processing
steps and inconsistencies between satellite imagery and aerial photos could be
accounted for and an overall estimation of plausibility of the unmixing model
could be attained. With scatter plots close to the 1:1 line and r2 of .87, .85 and
.96, respectively, results were considered satisfactory. The model was then
applied to all images, and a rule-base was set up to assign each pixel to one of
the following classes based on seasonal differences: bare ground, herbaceous
vegetation, woody cover, dense woody cover. Where seasonal patterns
indicated agricultural use, a re-assignment of the respective pixel to the
permanent crop class was carried out. Figures 1 and 2 show the resulting
maps.

6
Figure 1: Result of the hybrid classification approach for 1986

7
Figure 2: Result of the hybrid classification approach for 2007

8
Change analysis and interpretation
The two maps generated according to the methodology described before
allowed assessing major changes in land use and landing cover for the period
1986 to 2007. While a change matrix was employed to illustrate overall change
rates between classes (table 1).

Table 1. Matrix illustrating land use / land cover change in the test area between 1986
and 2007, expressed in percent of total area. Changes larger than 1% are highlighted
in red, diagonal elements represent unchanged areas
    2007 
    Urb.  Arab.  Irr.  Perm Bare  Herb.  Woody  Dense   Wat. 
.  W. 
Urb.  1.46  0.3  0  0.04  0.01  0.02  0.01  0.01  0 
Arab.  0.18  15.46  0.66  3.97  0.26  2.61  0.33  0.03  0.02 
Irr.  0.2  0.5  1.1  0.94  0.01  0.05  0.09  0.04  0 
Perm.  0.35  7.17  0.69  7.79  0.18  1.87  1.33  0.73  0.03 
Bare  0.15  2.19  0.12  1.6  0.27  6.17  1.2  0.27  0.04 
1986 
Herb.  0.13  3.35  0.19  3.23  0.08  9.05  6.44  0.77  0.02 
Woody  0.06  0.39  0.12  1.3  0.02  0.68  4.89  4.47  0.02 
Dense  0.01  0.03  0.06  0.44  0.0  0.07  0.62  3.27  0.02 
W. 
Wat.  0  0.01  0.01  0.02  0.01  0.01  0.02  0.02  0.16 

From table 1, two major trends can be discerned that are associated with
largely different change syndromes. The first is the regeneration of previously
burned areas during the observation period. As fires are frequent in the northern
region of the test area, the first date of image acquisition is likely to show
already different stages of post-fire succession. With the process of
regeneration advancing, these can be perceived as a “shift up” along a class
gradient “Bare” – “Herbaceous” – “Woody” – “Dense Woody”. Figure 3 shows
the corresponding subset from the classification images.

9
 Figure 3: Post-fire regeneration apparent from change detection analysis between
1986 (top) and 2007 (bottom) in the Northern Ayora area
A second major group of change is attributed to land use transformations
resulting in an overall increase in intensely used land by irrigated agriculture or
– most importantly - permanent crops. The latter class can be shown to
increase at the expense of arable land as well as “natural” classes such as
“Bare Ground”, “Herbaceous” and “Woody Dominated Cover”. The
transformation of natural/semi-natural areas to agricultural use as well as
tendencies towards intensification are a direct consequence of the notable
socio-economic development in Spain. During the period covered, these are
strongly determined by the accession to the European Union and its benefitting
from different European funding and subsidy schemes. One area where these
effects are apparent is the region around the city of Valencia, which is today
renowned for tree crops of orange, almond and olive trees. While this
development certainly improved the economic situation of many people, it is
evident that the associated increase in water uptake has to be acclaimed
critically, especially given current climate change scenarios for Spain where
water scarcity is considered a major risk. Figure 4 illustrates this process for the
abundant orange plantations of the Levante, close to the city of Valencia.

10
 Figure 4: Intensification of agricultural uses and urbanization between 1986 (left) and
2007 (right) close to the city of Valencia
While these enlarged subsets of both maps illustrate the processes subsumed
in the change matrix, change maps support a spatially explicit evaluation of
change trajectories. Figures 5 and 6 show these maps for non-natural and
natural areas, respectively.
Besides the examples shown above, another issue typical for many
Mediterranean regions is the process of land abandonment in marginal regions.
In the test area, this is coincident with a significant increase in population
figures along the coast, which result from internal migration from abandoned
villages, but is also connected to European migration patterns of people moving
to Mediterranean countries, often after reaching retirement age, and of course
to the touristic attractiveness of seaside resorts. The latter is difficult to assess
using satellite data alone, in particular given that the urban class was directly
incorporated from the respective CORINE data sets with their rather coarse
minimum mapping units. Yet, the effect of land abandonment can be attributed
to the transformation from “Arable Land” to the “Herbaceous”, and to a lesser
degree the “Woody” class, subsumed in the change map by the “arable to
natural” class. In the latter cases, the abandonment of regions that had been
used for marginal agricultural use has caused the encroachment of shrub
vegetation reflected in the shift to the indicated classes.

11
 Figure 5: Change from or towards non-natural classes between 1986 and 2007

Opposed to major changes between classes resulting from different factors, the
natural/semi-natural areas in the southern part of the test site remain largely
stable. This area is generally characterized by less precipitation and accordingly
sparser vegetation, and frequently occupies mountainous ridges unsuitable for
intensive use. Therefore, there is no economic appeal to convert these areas,
while vegetation cover is too sparse for wildfires to play a major role. Hence, the
only change visible is a shift up of one class classes for some areas (figure 6).
It is also interesting to note that areas classified as “Dense Woody Cover” are
hardly converted to other land uses, which is indicative of the high priority
attributed to maintaining and preserving such areas due to their high ecological
importance. Again owing to climatological reasons, these areas occur mainly in
the northern part of the test area and in higher altitudes.

12
 Figure 6: Change within natural/semi-natural areas between 1986 and 2007

Conclusions and further steps

The approach presented here is based on routinely available Landsat data to
address land use and land cover in a highly heterogeneous environment. Based
on a dedicated data pre-processing scheme we suggest the combination of a
classification approach with a multi-date spectral unmixing approach is capable
of operationally mapping land use and land cover information using available
reference information and image-based endmembers. We have analysed
results for both dates and the resulting change trajectories in relation to driving
factors responsible for landscape alterations, both physical and socio-economic.
Results emerging from the data analysis are congruent with processes such as
land abandonment and associated deterioration of marginal agricultural lands,
an increase allocation of land to production of export-oriented cash crops, and
typical succession patterns in areas prone to wildfires. On the other hand,
stability is encountered particularly where no economic interests spur land use
transformation, or where no fires occurred for longer periods.
Referring back to the overall research question, it is apparent that the
information products derived here are more indicative of transformation than
modification processes with their often transitional character. This issue may be
addressed using time series analysis of densely populated time series using
data from comparable phenological stages. The potential of this approach has
been demonstrated for a period from 1978 to 2001 for a part of the test area

13
(Röder et al. 2008). Further steps therefore include applying the approach to all
semi-natural and natural areas of the full test area, and incorporating derived
trend parameters into the overall analysis.
The combined analysis of land use / land cover modifications and
transformations will then be analysed in relation to available auxiliary
information (e.g. census data, water runoff information etc.), with the aim to
relate patterns to landscape change processes beyond the interpretative
manner presented here. This will also form the basis upon which to attempt the
“translation” of land use and land cover to ecosystem services and functions,
and their subsequent economic and non-economic valuation (Hein et al. 2006)
Such spatially explicit, quantitative assessment of ecosystem properties is
ultimately considered an excellent basis to develop land management or
restoration strategies that consider the trade-off between different ways of using
terrestrial resources.

References
BERK, A., ANDERSON, G. P., ACHARYA, P. K., CHETWYND, J. H., BERNSTEIN,
L. S., SHETTLE, E. P., MATTHEW, M. W., and ADLER-GOLDEN, S. M.,
1999, MODTRAN4 user’s manual, Air Force Research Laboratory, Space
Vehicles Directorate, Air Force Material Command.
DEFRIES, R., FOLEY, J. A., and ASNER, G. P., 2004, Land-use choices: balancing
human needs and ecosystem function. Frontiers in Ecology and Environment, 2,
249-257.
ELLIS, E. C., and RAMANKUTTY, N., 2008, Putting people in the map:
anthropogenic biomes of the world. Frontiers in Ecology and the Environment,
6, 439-447.
HAY, J. E., and MCKAY, D. C., 1985, Estimating solar irradiance on inclined surfaces:
A review and assessment of methodologies. International Journal of Solar
Energy, 3, 203-240.
HEIN, L., VAN KOPPEN, K., DE GROOT, R. S., and VAN IERLAND, E. C., 2006,
Spatial scales, stakeholders and the valuation of ecosystem services. Ecological
Economics, 57, 209-228.
HELDER, D. L., MARKHAM, B., THOME, K. J., BRARSI, J. A., CHANDER, G., and
MALLA, R., 2008, Updated radiometric calibration for the Landsat 5 Thematic
Mapper reflective bands. IEEE Transactions on Geoscience and Remote
Sensing, 46, 3309-3325.
HILL, J., and MEHL, W., 2003, Geo- und radiometrische Aufbereitung multi- und
hyperspektraler Daten zur Erzeugung langjähriger kalibrierter Zeitreihen.
Photogrammetrie – Fernerkundung – Geoinformation, 1, 7-14. .
HOBBS, R. J., RICHARDSON, D. M., and DAVIS, G. W., 1995, Mediterranean-type
ecosystems: opportunities and costraints for studying the function of
biodiversity. In Mediterranean-type ecosystems: The function of biodiversity,
edited by G. W. Davis, and D. M. Richardson (Berlin, Heidelberg, New York:
Springer).
KUEMMERLE, T., RÖDER, A., and HILL, J., 2006, Separating grassland and shrub
vegetation by multidate pixel-adaptive spectral mixture analysis. International
Journal of Remote Sensing, 27, 3251-3271.

14
LAMBIN, E. F., and GEIST, H. J., 2006, Land Use and Land Cover Change. Local
Processes and Global Impacts (Berlin, Heidelberg, New York: Springer Verlag).
LIANG, S., FANG, H., CHEN, M., SHUEY, C., WALTHALL, C., DAUGHTRY, C.,
MORISETTE, J. T., SCHAAF, C., and STRAHLER, A. H., 2002, Validating
MODIS land surface reflectance and albedo products. Methods and preliminary
results. Remote Sensing of Environment, 83, 149-162.
LOVELAND, T. R., REED, B. C., BROWN, J. F., OHLEN, D. O., ZHU, Z., YANG,
L., and MERCHANT, J. W., 2000, Development of a global land cover
characteristics database and IGBP DISCover from 1 km AVHRR data.
International Journal of Remote Sensing, 21, 1303-1330.
MILLENIUM ECOSYSTEM ASSESSMENT, 2005, Ecosystems and human well-
being: synthesis (Washington: Island Press.).
RÖDER, A., DUGUY, B., ALLOZA, J. A., VALLEJO, R., and HILL, J., 2008, Using
long time series of Landsat data to monitor fire events and post-fire dynamics
and identify driving factors. Remote Sensing of Environment, 112 (1), 259-273.
TANRÉ, D., 1990, Description of a computer code to simulate the satellite signal in the
solar spectrum: 5S code. International Journal of Remote Sensing, 11, 659-668.
WINTER, M. E., 1999, N-FINDR: an algorithm for fast autonomous spectral end-
member determination in hyperspectral data: Imaging Spectrometry V, p. 266-
275.

15