remote sensing of environment (2008) 112, 3916

advertisement
1
2
3
4
5
6
7
8
9
10
11
12
13
ASSESSMENT OF RADIOMETRIC CORRECTION TECHNIQUES IN ANALYZING
VEGETATION VARIABILITY AND CHANGE USING TIME SERIES OF LANDSAT
IMAGES
1
14
gradual changes in vegetation cover via remote sensing data. Various sources of noise affect the
15
information received by satellites, making it difficult to differentiate the surface signal from noise
16
and complicates attempts to obtain homogeneous time series. We compare different procedures
17
developed to create homogeneous time series of Landsat images, including sensor calibration,
18
atmospheric and topographic correction, and radiometric normalization. Two seasonal time series of
19
Landsat images were created for the middle Ebro Valley (NE Spain) covering the period 1984–
20
2007. Different processing steps were tested and the best option selected according to quantitative
21
statistics obtained from invariant areas, simultaneous medium-resolution images, and field
22
measurements. The optimum procedure includes cross-calibration between Landsat sensors,
23
atmospheric correction using complex radiative transfer models, a non-lambertian topographic
24
correction, and a relative radiometric normalization using an automatic procedure. Finally, three
25
case studies are presented to illustrate the role of the different radiometric correction procedures
26
when analyzing and explaining gradual and abrupt temporal changes in vegetation cover, as well as
27
temporal variability. We have shown that to analyse different vegetation processes with Landsat
28
data, it is necessary to accurately ensure the homogeneity of the multitemporal datasets by means of
29
complex radiometric correction procedures. Failure to follow such a procedure may mean that the
30
analyzed processes are non-recognizable and that the obtained results are invalid.
31
Key-words. Landsat time series, TM-ETM+ cross-calibration, atmospheric correction, relative
32
normalization, vegetation change, Ebro Valley, Spain
Sergio M. Vicente-Serrano1*, Fernando Pérez-Cabello2 and Teodoro Lasanta1
Instituto Pirenaico de Ecología, CSIC (Spanish Research Council), Campus de Aula Dei, P.O. Box
202, Zaragoza 50080, Spain
2
Departamento de Geografía. Universidad de Zaragoza. C/ Pedro Cerbuna 12. 50009. Zaragoza.
Spain.
* svicen@ipe.csic.es
Abstract. The homogeneity of time series of satellite images is crucial when studying abrupt or
1
33
34
35
1. Introduction
36
surface characteristics over the past four decades (Cohen and Goward, 2004). Landsat images have
37
been widely used for land cover mapping and the creation of vegetation inventories at different
38
spatial scales (e.g., Bossard et al., 2000). Moreover, the systematic archiving of Landsat data makes
39
this information highly valuable for retrospective analyses of land surface characteristics. Together
40
with other types of satellite images such as NOAA-AVHRR (National Oceanic and Atmospheric
41
Administration-Advanced Very High Resolution Radiometer) and MODIS (Moderate Resolution
42
Imaging Spectroradiometer), multitemporal Landsat data have been widely used in recent decades
43
to identify changes in land cover (e.g., Lenney et al., 1996). The spatial and spectral resolution of
44
Landsat images makes these data highly suitable in analyzing both abrupt and gradual changes in
45
vegetation cover and monitoring environmental processes such as degradation and desertification
46
(Almeida-Filho and Shimabukuro, 2002), deforestation (Cohen et al., 2002; Huang et al., 2007),
47
habitat fragmentation (Millington et al., 2003), forest succession (Song and Woodcock, 2003; Song
48
et al., 2007), overgrazing (Jano et al., 1998; Pickup and Chewing, 1994), rangeland monitoring
49
(Hostert et al., 2003), and vegetation recovery after natural disturbance such as volcanism
50
(Lawrence and Ripple, 1999) and forest fires (Viedma et al., 1997; Lozano et al., 2007).
51
Nevertheless, there exist constraints in using Landsat data for multitemporal studies because of
52
problems in obtaining homogeneous time series, not affected by non-surface noise, and in which the
53
images are comparable between different dates since the data only report on surface conditions.
54
Multitemporal satellite-image datasets are affected by different sources of noise related to the
55
stability of sensors, changes in satellite responsivity, changes in illumination, atmospheric effects,
56
etc. These problems are not unique to Landsat images, but they are more difficult to overcome in
57
Landsat images compared with images compiled by other satellites because the low temporal
58
resolution of Landsat images makes it impossible to apply simple homogenization procedures such
The Landsat program for Earth observation has provided invaluable information on the Earth’s
2
59
as composite creation (Holben, 1986) and temporal filtering (Van Dijk et al., 1987), which
60
markedly reduce the non-surface noise.
61
Notable efforts have been made to reduce non-surface noise in Landsat images, including attempts
62
to calibrate the sensor to correct lifetime radiometric trends (Teillet et al., 2004; de Vries et al.,
63
2007), cross-calibrate the images obtained from different sensors (Teillet et al., 2006; Röder et al.,
64
2005), minimize atmospheric noise (Chavez, 1989; Ouaidrari and Vermote, 1999; Liang et al.,
65
2001), and reduce the influence of topography (Civco, 1989; Gu and Gillespie, 1998; Pons and
66
Solé, 1994); however, other studies have shown that the application of accurate sensor calibrations
67
and complex atmospheric corrections does not guarantee the multitemporal homogeneity of Landsat
68
datasets (Schroeder et al., 2006) because complete atmospheric properties are difficult to quantify
69
and simplifications are commonly assumed. This problem has led to the development of relative
70
radiometric normalization techniques based on adjustments to the radiometric properties of an
71
image time series to match that of a single reference image. In recent years, efforts have been made
72
to develop methods to select invariant pixels for a reliable application of relative normalization
73
techniques (Du et al., 2002; Chen et al., 2005; Canty et al., 2004).
74
Protocols have been proposed in processing multitemporal Landsat datasets (e.g., Hall et al., 1991;
75
Hill and Strum, 1991; Han et al., 2007), comprising the following steps: i) geometric correction, ii)
76
calibration of the satellite signal to obtain Top of the Atmosphere Radiance, iii) atmospheric
77
correction to estimate surface reflectance, iv) topographic correction, and v) relative radiometric
78
normalization between images obtained on different dates. Radiometric processing is recommended
79
to be done prior to geometric processing since this resampling step generally smoothness the data
80
set. Nevertheless, Landsat users in Europe commonly do not have access to data that have not
81
already been geometrically corrected (the Level 1 System Corrected -1G- from the European Space
82
Agency (ESA) is considered the standard product for most users since previous levels require
83
extensive processing). When these protocols are applied, several decisions must be made due to the
84
different procedures available at each step.
3
85
The high degree of interest in using multitemporal Landsat datasets is in contrast with the small
86
number of studies that have tested different procedures of radiometric correction with the aim of
87
obtaining temporally homogeneous images. The majority of studies have tested the influence of
88
atmospheric correction (Song and Woodcock, 2003; Norjamäki and Tokola, 2007) or relative
89
normalization (Yuan and Elvidge, 1996; Tokola et al., 1998; Heo and Fitzhugh, 2000; Olthof et al.,
90
2005) on the temporal stability of time series of Landsat images. In contrast, few studies have tested
91
the influence of each of the above steps in obtaining temporally stable time series of images and
92
few have assessed the relative importance of using simple or complex techniques at each step.
93
Existing studies have shown that relative normalization is the most critical step in obtaining
94
temporal stability in a series of images (Schroeder et al., 2006; Janzen et al., 2006), although the
95
calibration procedure also has a noticeable effect on the final results (Paolini et al., 2006) and
96
atmospheric correction is important in obtaining accurately estimate surface reflectance's and
97
accurate magnitudes of vegetation indices over time (Song and Woodcock, 2003).
98
Few studies have analyzed the role of complete radiometric correction protocols in processing
99
multitemporal Landsat data when a number of different vegetation processes are of interest.
100
Nevertheless, this is a crucial topic when analysing vegetation multitemporal dynamic using
101
Landsat data. Thus different results have been found for land classification (Song et al., 2001;
102
Paolini et al., 2006; Norjamäki and Tokola, 2007) and forest succession (Song and Woodcock,
103
2003; Schroeder et al., 2006) as a function of the radiometric correction applied.
104
The present study involves a complete evaluation of various radiometric correction processes
105
required to obtain accurate time series of Landsat imagery, including calibration, atmospheric
106
correction, topographic correction, and relative normalization. The objective is to identify the
107
influence of different processing steps on multitemporal spectral reflectance trajectories developed
108
with Landsat data and also to detect their performance to analyse different vegetation processes.
109
Three case studies related to different processes of vegetation change and variability are provided to
110
illustrate how different results can be obtained as a function of the employed radiometric correction
4
111
protocol and the ecological process of interest: land cover change, forest regeneration after fire and
112
ecosystem response to climate variability. The case studies were carried out in a complex ecological
113
region in which several natural and human-induced processes of changes in vegetation cover have
114
been recorded in recent decades.
115
116
117
118
2. Study area
The study area is the central Ebro Valley, Spain (Landsat Path 199 Row 31), located in the
119
northernmost semiarid region in Europe. Figure 1 shows the location of the study area, including a
120
detailed land cover map. Surrounded by mountain chains, the Ebro Valley has a Mediterranean
121
climate with continental characteristics, with marked spatial and seasonal variations in precipitation;
122
the dry season occurs during the summer months. The elevation ranges from less than 300 m above
123
sea level in the middle of the Ebro Depression to more than 2000 m a.s.l. in the Prepyrenees (North)
124
and 2000 m a.s.l. in the highest peaks of the Iberian Range (South). In the central part of the valley,
125
mean annual precipitation is 326 mm, with marked seasonality. The study area shows a strongly
126
negative water balance (precipitation minus potential evapotranspiration), greater than 900 mm. The
127
dominant land covers include steppes and herbaceous cultivation in dry farming areas. Coniferous
128
forests (mainly Pinus halepensis) cover the few slopes of the region. Vegetation distribution is
129
strongly controlled by aridity (Vicente-Serrano et al., 2006), and drought conditions have a marked
130
influence on vegetation cover and activity (Vicente-Serrano, 2007). The lithology of the area is
131
characterised by limestones and gypsums (Peña et al., 2002) that contribute to its aridity, since the
132
soils are unable to retain the water as a consequence of the high hydraulic conductivity.
133
134
135
136
137
138
3. Methodology
3.1. Satellite imagery database
To determine the patterns of vegetation variability and change, it is important to take into account
139
seasonal variations in the distributions of different vegetation types. In the middle Ebro Valley,
140
herbaceous species, shrubs, and forests show contrasting seasonal variations in vegetation activity
5
141
(Vicente-Serrano et al., 2006). The peak activity for herbaceous species and shrubs occurs in
142
spring; for forests in summer. These seasonal fluctuations make it difficult to capture the full range
143
of vegetation processes with just a single season of imagery.
144
Abrupt changes in vegetation activity commonly occur as a consequence of climate seasonality;
145
consequently, it is important that the capture dates of images are similar in different years. It is not
146
possible to combine images taken in different months, as this would have a strong influence on the
147
temporal homogeneity of the dataset.
148
We reviewed all of the available Landsat-Thematic Mapper (TM) and -Enhanced Thematic Mapper
149
(ETM+) images in the archives of the European Space Agency (ESA). Since most studies based on
150
Landsat data consider ETM+ and TM radiometry to be comparable, We used TM data from 1984
151
and, then, switched to the ETM+ data when it became available in 1999 due to the better calibration
152
of this sensor (Teillet et al., 2001); we subsequently switched back to using TM date due to the
153
ETM+ SLC failure in 2003. Frequent cloud cover in spring means that only the month of March has
154
a sufficient number of images to enable an analysis of vegetation activity. Months in summer pose
155
fewer problems in terms of obtaining reliable clouds-free images; therefore, we selected the month
156
of August to create a second time series. A total of 28 Landsat-TM and -ETM+ images taken
157
between 1984 and 2007 were acquired from ESA, 16 corresponding to the summer season and 12 to
158
spring. Table 1 lists the dates and types (TM or ETM+) of the selected images.
159
160
3.2. Geometric correction
161
The images were orthorectified using control points and a 1-m digital elevation model obtained
162
from stereo pairs of aerial photographs, and resampled to 30 m to match the TM and ETM+
163
resolutions. The image taken on 8 August 2000 with good visibility and being free of clouds, was
164
registered using orthorectified digital aerial photographs as a reference. The rest of the images were
165
co-registered to this image using control points. The images were orthorectified following the
166
method of Palà and Pons (1995), which includes elevation data in performing a polynomial
6
167
geometric correction. The X and Y root mean square error was less than 15 m (0.5 pixels) for all
168
images, guaranteeing a precise geometric match among them. After geometric correction, cloud-
169
covered and cloud shadows were manually digitized and eliminated.
170
171
3.3. Calibration
172
A precise calibration is required to convert DN (Digital Numbers) to satellite radiances in W m –2 sr–
173
1
174
constant over time. A large variation, on the order of 20%, has been observed between the
175
prelaunch gain coefficients and postlaunch cross-calibration (Chander et al., 2004; Chander et al.,
176
2007). Moreover, the different spectral responses of TM and ETM+ sensors introduces
177
comparability problems between Landsat 5 and Landsat 7 images (Teillet et al., 2001).
178
For Landsat 5 imagery, the European Spatial Agency (ESA) has used constant calibration dynamic
179
ranges since 1984 (http://earth.esa.int/pub/ESA_DOC/landsat_FAQ/#_Toc69122047), embedded
180
within the product format. Chander et al. (2004) and Teillet et al. (2004) demonstrated an
181
exponential decay of the solar reflective bands of Landsat 5-TM since 1984, with some differences
182
between bands. We corrected the calibration coefficients embedded in the ESA products with
183
reference to the time after launch of Landsat 5 in 1984, applying the equations formulated by Teillet
184
et al. (2004). This procedure is recommended by the ESA in recalibrating Landsat products
185
(Saunier and Rodríguez, 2006). The coefficients of calibration for Landsat 7-ETM+ were obtained
186
according to upper and lower at-satellite radiance indicated in the Landsat-7 Science Data Users
187
Handbook (http://ltpwww.gsfc.nasa.gov/IAS/handbook/handbook_toc.html) using the values
188
corresponding to the date (before or after July 1, 2000) and the type of gain (high or low, as
189
embedded within the product format).
190
Although the Landsat 5-TM and Landsat 7-ETM+ bands are commonly considered to be
191
comparable, previous studies have demonstrated important differences between the two that may
192
affect comparability, suggesting the need to apply cross-calibration procedures to both sensors
m–1. This is highly problematic for Landsat 5 imagery because calibration coefficients are not
7
193
because for the majority of bands (2, 3, 4, and 7) the TM sensor underestimates the radiance values
194
regarding ETM+ (Teillet et al., 2001; Vogelmann et al., 2001).
195
Vogelmann et al. (2001) developed a simple procedure to cross-calibrate the Landsat-TM and
196
ETM+ images, applicable to Level 1G formats. These authors proposed empirically derived slope
197
and intercept values to convert Landsat 7 ETM+ DNs to Landsat 5 TM DNs based on two
198
simultaneous images taken on June 2, 1999. We used these slope and intercept values in the present
199
study to adjust the Landsat 7-ETM+ DNs to Landsat 5-TM DNs.
200
Satellite radiance was obtained for Landsat 5-TM quantized calibrated pixel values in DNs, Landsat
201
7-ETM+ quantized calibrated pixel values in DNs, and the cross-calibrated Landsat 7 ETM+
202
quantized calibrated pixel values in DNs to Landsat 5-TM quantized calibrated pixel values in DNs
203
according to:
204
Lsat   Grescale  DN  Brescale (1)
205
where Lsat is the satellite radiance in W m–2 sr–1 μm–1 for band . Grescale and Brescale are the
206
calculated band-specific rescaling factors. Satellite radiances were converted to Top Of the
207
Atmosphere (TOA) reflectances according to
208
TOA 
209
where is the TOA reflectance for band , d is the earth–sun distance in astronomical units,
210
ESUN is the mean solar exoatmospheric irradiance for band , and s is the solar zenith angle in
211
degrees. ESUN values were obtained from Chander and Markham (2003) for TM images and from
212
the
213
(http://ltpwww.gsfc.nasa.gov/IAS/handbook/handbook_toc.html).
  Lsat   d 2 (2)
ESUN cos  s
Landsat-7
Science
Data
User
Handbook
for
ETM+
images
214
215
3.4. Atmospheric correction
216
Although TOA reflectance values are widely used in inventory and ecosystem studies, they do not
217
take into account signal attenuation by the atmosphere, which strongly affects the
8
218
intercomparability of satellite images taken on different dates. Upward and downward irradiance is
219
modified by two atmospheric processes: absorption by gases and scattering by aerosols and water
220
molecules (Vermote et al., 1997a; Lu et al., 2002). There are noticeable daily variations in
221
atmospheric gas concentrations and aerosol volumes; these must be taken into account in
222
multitemporal studies. Moreover, the effect of atmospheric processes is spectral-dependent (Teillet
223
and Holben, 1993; Cachorro et al., 2000), affecting the magnitude of band ratio operations such as
224
vegetation indices (Myneni and Asrar, 1994).
225
Different models have been developed to minimize the noise introduced by atmospheric processes
226
on the signal received by the satellite, ranging from simple methods based on information contained
227
in the image [e.g., Dark Object Subtraction (DOS)-based methods (Chavez, 1996)] to complex
228
radiative transference models such as SMAC (Rahman and Dedieu, 1994), 6S (Vermote et al.,
229
1997a), MODTRAN (Berk et al., 1999), and ATCOR (Richter, 1996) that simulate the
230
atmosphere/light interactions between the sun surface and surface-sensor trajectories.
231
In this paper we tested two methods developed to minimize the atmospheric effects on Landsat time
232
series: i) one based on a modification of the DOS method that includes some atmospheric
233
information (Dark Target Approach -DTA-), and ii) one based on a complex radiative transfer code
234
that includes some atmospheric information commonly available in global climate databases and
235
some parameters obtained from the images themselves.
236
237
3.4.1. DTA method
238
Dark Target Approach (DTA)-based methods assume that some areas of the images have near-zero
239
reflectance. Although there are various DTA-based methods (Chavez, 1996), Song et al. (2001)
240
reported better results when including in the method atmospheric transmittance and Rayleigh
241
scattering. We therefore followed this approach in the present study. Surface reflectance (  ) can be
242
obtained as follows (Song et al., 2001):
243
 
e
 0 / uv
 ( Lsat   Lhaze )
(3)
( ESUN cos  s e  / u  Edown )
0
0
9
244
where Lhaze is the radiance (W m–2 sr–1 m–1) in areas with zero or near-zero reflectance, 0 is the
245
optical depth of the atmosphere, u0 is the cosine of the solar zenith angle, uv is the cosine of the
246
zenithal view angle, and Edown is the diffuse irradiance at the surface due to the scattered solar flux
247
in the atmosphere (W m–2 m–1). The most critical step is related to the calculation of Lhaze, for
248
which there are a number of alternatives (Chavez, 1989, 1996). In this study, we calculated Lhaze
249
according to the method of Song et al. (2001) and Schroeder et al. (2006):
250
Lhaze  Ldark  0.01 ESUN cos  s e  0 / u0  Edown e  0 / uv / 
251
where Ldark is the lowest radiance in the image whose value is taken from at least 1000 pixels
252
(Teillet and Fedosejevs, 1995). The estimation of 0 is complex as it requires in situ measurements,
253
which are unavailable for the dates of the selected images. We therefore used constant values of 0
254
for the entire set of images, based on USA standard values (Dozier, 1989) that are in agreement
255
with the values reported by Yang and Vidal (1990) for NE Spain. 0 values can be consulted in Pons
256
and Solé-Sugrañes (1994). Edown for a Rayleigh atmosphere was estimated as zero aerosol optical
257
depth at 550 nm using the 6S radiative transfer code, following Schroeder et al. (2006).


(4)
258
259
3.4.2. Radiative transfer model
260
Among the available codes, we used the Second Simulation of the Satellite Signal in the Solar
261
Spectrum (6S) code (Vermote et al., 1997a) to invert the TOA radiance and obtain surface
262
reflectance values from TM and ETM+ data assuming lambertian targets. 6S code takes into
263
account gaseous transmission and Rayleigh and aerosol transmission and intrinsic reflectance.
264
Although 6S enables the inclusion of detailed atmospheric conditions in several layers, as obtained
265
from radiosonde data, such data are commonly unavailable. 6S code also enables the user to restrict
266
the number of inputs and to assume certain constants. In this study, we used ozone concentrations
267
and column water vapor as gaseous inputs for 6S calculations because they are the most critical
268
parameters for atmospheric correction (Arino et al., 1997). Daily ozone concentrations (in atm cm–
269
2
) can be obtained from Total Ozone Mapping Spectrometer (TOMS) data, at a resolution of 1.25 
10
270
1.00,
from
the
NASA
GSFC
Data
Active
Archive
Center
271
(http://toms.gsfc.nasa.gov/ozone/ozone_v8.html). For the period for which ozone data were
272
unavailable (between 1994 and 1996), the data were calculated following Van Heuklon (1979) as a
273
function of the day of year:
274
O3  235  sin 2 (1.25  lat )  [150  40  sin( 0.9856  (d  30))  20  sin( 3  (long  20))]
275
where lat is the latitude in degrees, long is the longitude in degrees, and d is the day of year.
276
The column water vapor (in g cm–2) is obtained from NCEP reanalysis daily data at a resolution of
277
2.5  2.5 (http://dss.ucar.edu/datasets/ds090.0/). Elevation was also included by means of the
278
DEM described in Section 3.2.
279
The most critical parameter for atmospheric correction is the Aerosol Optical Thickness (AOT).
280
Although it is ideal is to obtain simultaneous measurements corresponding to the overpass of the
281
satellite, in the study area there are no systematic measurements of this parameter, this being the
282
case in most regions of the world. To solve this problem, various methods have been developed to
283
obtain reliable estimations of AOT (Liang et al., 1997; Wen et al., 1999). Most such methods are
284
based on the relatively strong aerosol radiance effect in low-surface-reflectance areas compared
285
with bright areas, as it is then possible to apply an inversion procedure in assessing AOT. In this
286
study, we chose to use the Dense Dark Vegetation (DDV) method (Liang et al., 1997). This
287
approach is based on the empirical relationship observed between the visible and infrared (IR)
288
bands in areas with dark and dense vegetation cover. The method is based on the weak influence of
289
aerosols on the mid-IR signal, as most aerosol particles sizes are smaller than the wavelength in this
290
part of the spectrum. Kaufman et al. (1996) proposed the following relationship for dark vegetation
291
canopies:
292
1  0.25 7
3  0.50  7
293
where 1 ,  3 , and  7 are the surface reflectances for bands 1, 3, and 7 of Landsat-TM images,
294
respectively.
(5)
(6)
11
295
We assumed that each Landsat scene has uniform AOT, and, following Song et al. (2001), dense
296
and dark vegetation areas were identified over the whole image ( 1  0.25 and NDVI (Normalized
297
Difference Vegetation Index) > 0.1). Average TOA reflectance values were obtained from bands 1,
298
3, and 7 in these regions, and band 7 was used to predict surface reflectance values for bands 1 and
299
3 according to the empirical relationship stated above. We iteratively ran the 6S code with the
300
ozone and water vapor values corresponding to each image and considering a continental aerosol
301
model. Following Song et al. (2001), AOT for 550 nm was defined in a range from 0.01 to 2.0 with
302
a step size of 0.01 for each iteration. AOT was set at 550 nm when modeled reflectance in bands 1
303
and 3 matched the reflectance obtained via the empirical relationship indicated in the equation.
304
Values of AOT for images with different dates ranged from 0.048 to 0.96, with generally higher
305
values for the August time series. AOT estimations, together with ozone and water vapor values,
306
were included as input for the 6S code to atmospherically correct the 28 selected images.
307
308
3.5. Topographic correction
309
Another source of artificial noise is the modification of illumination conditions by topography.
310
Although the effect of topography on illumination conditions is complex, even including reflections
311
from adjacent slopes (Proy et al., 1989), it is usually simplified for the purpose of analysis: shaded
312
areas show less than expected reflectance, whereas sunny areas show the opposite pattern.
313
Various methods are available to topographically correct satellite imagery (Civco, 1989; Pons and
314
Solé, 1994; Gu and Gillespie, 1998). In this paper, we tested two of the most commonly used
315
methods: the first assumes a lambertian behavior of the surface and the second considers non-
316
lambertian effects.
317
The illumination conditions can be modeled following the cosine law of spherical geometry (Civco,
318
1989):
319
  cos  S cos  n  sin  S sin  n cos(n  S )
(7)
12
320
where  is the cosine angle between the solar incident angle and the local surface normal,  S is the
321
solar zenith angle,  n is the zenith angle of the normal to the surface, n is the topographic aspect
322
angle, and  S is the solar azimuth angle.
323
Illumination conditions corresponding to the date and timing of satellite overpass were determined
324
for each pixel of the Landsat images using the DEM described in Section 3.2 and by applying the
325
formulation stated above. The DEM was used at a spatial resolution of 15 m to guarantee the
326
reliability of the derived DTMs (Digital Terrain Models) in the topographic correction procedure.
327
To calculate the spatially distributed values of  n and n , we employed the Geographic Information
328
System (GIS) MiraMon.
329
Based on a lambertian assumption, the reflectance of a horizontal surface  h, can be determined by
330
Eq. 8.
331
 h,   
332
Lambertian models usually overestimate radiance for the high incidence angles of slopes facing
333
away from the sun (Vincini and Frazzi, 2003). Simple models have been proposed to take into
334
account the non-lambertian properties of the surface. These models are based on Minnaert’s theory,
335
in which a constant K, derived empirically for each band, enables the characterization of the non-
336
isotropic conditions of the surface. Minnaert’s original proposal has been modified in a number of
337
studies (Colby, 1991; Teillet et al., 1982; Vinzini and Frazzi, 2003); among these methods, the C-
338
correction (Teillet et al., 1982) has shown superior performance because it successfully retains the
339
spectral characteristics of each band and significantly reduces reflectance variability for
340
homogeneous vegetation classes (Riaño et al., 2003). Following C-correction, the reflectance
341
corresponding to a horizontal surface can be obtained as follows:
342
 h,    
 cos  S 

  
 cos  S  c
   c
(8)
 (9)


13
343
where c is obtained empirically from the entire image following c  bk / mk , where    bk  mk  .
344
This procedure removes the common dependence of   to  more efficiently than other methods
345
(Vinzini and Frazzi, 2003).
346
347
3.6. Relative radiometric normalization
348
The commonly assumed simplifications employed in atmospheric and topographic corrections mean
349
that they usually fail to completely remove non-surface noise. To obtain improved temporal
350
homogeneity of satellite imagery, it is common to apply relative normalization between images
351
(e.g., Yuan and Elvidge, 1996; Coppin and Bauer, 1994; Tokola et al., 1998). Relative radiometric
352
normalization is sometimes used as the sole correction procedure (Caselles and López-García,
353
1989), and is usually preferred to atmospheric correction for the purpose of change detection
354
(Olsson, 1995; Janzen et al., 2006). Thus, relative normalization can be applied directly to DNs,
355
radiance, TOA reflectance, or surface reflectance values.
356
Relative normalization is generally based on a linear comparison of image statistics for images
357
obtained on different dates. One of the images, commonly the most recent or least affected by
358
atmospheric effects, is considered as the reference to which the rest of the images are adjusted.
359
Among the methods proposed for relative normalization, linear regression is the most commonly
360
used and widely recommended (Yuan and Elvidge, 1996). The validity of the assumption of a linear
361
relationship between the reference image and the image to be normalized has been confirmed in
362
several studies (Schott et al., 1988; Caselles and López, 1989; Hall et al., 1991; Heo and FitzHugh,
363
2000). This assumption greatly simplifies the normalization process, based on matching the
364
reflectance values of the image to be normalized to the reference image as follows:
365
 reference,  a  b normalise, (10)
366
where a and b are the linear regression parameters.
367
The most critical decision to be made is the sampling of targets/pixels in estimating the regression
368
parameters, as it is necessary to identify constant reflectors between dates and to assume that
14
369
reflectance differences in these targets are due to non-surface noise. Several procedures can be
370
followed in selecting the constant targets, of which the most widely used is based on a visual
371
selection of non-variant targets [Pseudo-Invariant Features (PIFs)] (Schott et al., 1988). These areas
372
should contain minimum amounts of vegetation, be located in relatively flat areas to minimize the
373
effects of illumination differences, and cover a wide range of reflectance values (from dark to bright
374
areas) (Eckhardt et al., 1990). Targets such as sand, asphalt, and water are commonly selected as
375
PIFs (Caselles and García, 1989; Coppin and Bauer, 1994). In this paper, we followed a PIFs
376
normalization (Schott et al., 1988) based on the selection of 280 invariant pixels to be used for the
377
calculation of a and b in each TM and ETM+ band and image. PIFs were identified visually in
378
urban areas and areas of asphalt, water, dark vegetation and sand, for which an invariant reflectance
379
can be assumed over the study period. The last images for the series of March (13/03/2007) and
380
August (01/08/2006) were considered as references, and the rest were matched to these images.
381
We also used a simple radiometric correction method termed Temporally Invariant Cluster (TIC)
382
(Chen et al., 2005), based on the visual identification of centers of high frequency in density
383
scatterplots. The method assumes that these centers correspond to areas in which there are no
384
changes between the reference image and the image to be normalized (Chen et al., 2005). This
385
method is less time-consuming than PIFs because although identification of the TIC centers is not
386
automatic, it is faster than the identification of PIFs. Using this procedure, the values of only two
387
points are needed to obtain the regression line and the a and b coefficients that intersect the centers
388
of high density.
389
Finally, we used an automatic procedure for relative normalization. Automatic techniques have the
390
advantage of being less time-consuming than other techniques and maintaining criteria for target
391
selection. Pseudo-Invariant Feature Regression (PIFR) (Du et al., 2001, 2002) is an automatic
392
method that has provided excellent results in the relative normalization of Landsat images (Janzen
393
et al., 2006; Olthof et al., 2005; Paolini et al., 2006). The PIFR method applies Principal
394
Component Analysis (PCA) between each pair of images to obtain the PIFs. The first component
15
395
represents a least-square regression between overlap areas and the second residual variation. It is
396
recommended to remove outliers prior to model calculation, as they may affect model performance
397
(Heo and FitzHugh, 2000). Although the problem of outliers is minimized in our dataset (as cloud
398
cover was removed previously), the PIFR method removes outliers iteratively on the second
399
principal component to discard changed pixels. We used an iterative process in which the
400
standardized residuals obtained in the first run where used to remove those pixels that exceeded a
401
certain threshold. An initial threshold of +/–1.28 (10% probability according to the normal
402
distribution) was set to remove outliers after the first run. Thresholds corresponding to +/–1%
403
probability were added in subsequent runs to remove pixels to be excluded from the model. The
404
percentage of variance explained by the model was assessed for each run and a minimum of 95%
405
was chosen in selecting the model and the a and b parameters.
406
407
3.7. Validation
408
3.7.1. Calibration and relative normalization
409
To validate the calibration and relative normalization procedures, we identified Pseudo-Invariant
410
(PI) validation pixels for which it is assumed that reflectivity did not vary over time. The
411
identification of these pixels was independent regarding the PIFs used for relative normalization.
412
These pixels were selected from topographically corrected reflectance values and prior relative
413
radiometric normalization using a combined automatic and manual technique. First, the Coefficients
414
of Variation (CV) in each of the six Landsat reflective bands was calculated for each pixel in the
415
time series for March and August. Average CV values were obtained in the different bands of the
416
two series, with the aim of obtaining a unique image that represents the average variability of each
417
pixel independently of the spectral band and season. We considered as PI validation pixels those
418
with average CV values below 0.05. We sampled 300 such pixels, including those in urban areas,
419
dark coniferous forests, water bodies, areas of human infrastructure, and areas of bright sand. None
420
of the selected pixels was subjected to modifications in land cover over the period of interest;
16
421
reflectance values can be considered to have been stable over time. Temporal differences in
422
reflectance values may be attributed to non-surface noise related to factors such as calibration and
423
atmospheric influence.
424
The performance of TM and ETM+ calibration and the cross-calibration between ETM+ and TM
425
images was assessed by comparing the TOA reflectance obtained for the PI validation pixels. The
426
Mean Absolute Error (MAE) was calculated for each spectral band, considering as a reference the
427
last images in the time series of August (08/01/2006) and March (03/13/2007). PI validation pixels
428
were also used to assess the relative radiometric normalization methods.
429
430
3.7.2. Atmospheric correction
431
Atmospheric correction methods described in section 3.4 were applied to geometrically corrected
432
and calibrated/cross-calibrated images. Atmospheric correction applied to these images was
433
evaluated following two different procedures. The first approach involved a comparison of
434
atmospherically corrected surface reflectance values with near-simultaneous MODIS reflectances.
435
Although MODIS includes spectral bands with a number of differences compared with Landsat
436
bands (in terms of bandpasses and spectral response), they are generally considered to be equivalent
437
(Masek et al., 2006). Atmospheric correction of MODIS images can be performed with a greater
438
degree of robustness than Landsat images due to the improved onboard capabilities of MODIS
439
(Vermote et al., 1997b, 2002). MODIS surface reflectances were obtained from the MOD09GAV5
440
daily reflectance dataset (http://lpdaac.usgs.gov/modis/mod09ghkv4.asp) for the same region as that
441
covered by the Landsat dataset. The Landsat–MODIS comparison period is 2001–2007, for which 5
442
MODIS images were near nadir. Landsat TOA and surface reflectance values obtained from DTA
443
and 6S were aggregated to 500 m spatial resolution to match the MODIS reflectance dataset.
444
Atmospheric correction was also evaluated by comparing Landsat surface reflectance with field
445
surface reflectances obtained simultaneous with the overpass of Landsat 5. For this purpose, we
446
selected a homogeneous area characterized by xeric vegetation and very low degrees of herbaceous
17
447
and shrub cover. The study area (0°40'W, 41°27'N) covers 28.2 ha and is a plane platformal area
448
unaffected by topographic influence on illumination conditions. On 08/01/2006 and 03/13/2007 we
449
used an Ocean Optics USB2000 field spectroradiometer to obtain a random sample of field
450
reflectance measures (~200) simultaneous with the overpass of the Landsat-5 satellite (+/– 0.5 h).
451
This spectroradiometer makes 10 IFOV observations over the 400–900 nm bandwidth with a 0.3
452
nm sampling interval. A spectralon reference panel was used to obtain reflectance values. To ensure
453
a high signal-to-noise ratio, integration time was adjusted according to illumination conditions, and
454
each measure was calculated as the mean of 20 individual spectra. The spectral curves were
455
integrated to simulate the TM bands, taking into account the relative spectral response of each TM
456
band (Teillet et al., 2001) and compared with the TOA reflectance and DTA and 6S surface
457
reflectances obtained from Landsat imagery. We applied a one-way analysis of variance to
458
determine if DTA and 6S surface reflectance are significantly different from the field surface
459
reflectances. Tamhane’s pairwise comparisons test was used to produce the multiple comparisons
460
where the variances are unequal. The significance threshold was set at 0.05.
461
462
3.7.3. Topographic correction
463
Topographic corrections were assessed via temporal comparisons of the reflectance trajectories
464
before and after topographic corrections in an homogeneous Pinus halepensis forest located on
465
slopes of 20 in the central Ebro Valley (semi-arid conditions, average annual precipitation = 320
466
mm, 41°43'N,0°28'W). The forest has the same density in south-facing and north-facing slopes. Ten
467
pixels were selected in a south-facing slope of the forest, besides additional ten pixels in a north-
468
slope facing to assess the effects of topographic correction procedures on the derived trajectories.
469
470
471
472
473
474
475
4. Results
4.1. Sensor calibration and cross-calibration
Figure 2 shows the average TOA reflectance values for each image and band corresponding to the
time series of August for the PI validation pixels. Figure 2 also compares original TOA reflectances
18
476
obtained from ETM+ images with TOA reflectances obtained from ETM+/TM cross-calibration,
477
employing the equations of Vogelmann et al. (2001). The temporal trends reveal that comparisons
478
of TOA reflectance values between TM and ETM+ are highly problematic without prior cross-
479
calibration. In general, ETM+ images provide higher TOA reflectance values than TM-TOA
480
reflectance. Differences are greater for IR bands than visible bands, especially bands 4 and 5
481
(15.3% and 27.2% higher for average ETM+ than TM TOA reflectance in the August time series,
482
respectively). A similar behavior has been found for the March time series (results not shown).
483
The application of cross-calibration to Landsat 7-ETM+ images improves the temporal stability of
484
the series and comparisons between TOA reflectances. The procedure reduces TOA reflectance
485
values for most of the bands in ETM+ images, showing a better match with TM-TOA reflectances.
486
Table 2 shows the MAE obtained from time series of each band. The error is calculated for each
487
image relative to the reference image in each dataset (01/08/2006 for August and 13/03/2007 for
488
March). Original ETM+ calibrated bands show higher MAE values than cross-calibrated bands,
489
both in visible and IR bands. These results support the validity of employing cross-calibrated ETM+
490
TOA reflectances in the following radiometric correction steps.
491
492
493
494
4.2. Atmospheric correction
Figure 3 shows the average values of TOA reflectances and DTA and 6S surface reflectances for
495
the PI validation pixels for the time series of August. DTA and 6S atmospheric-correction methods
496
yield different results in terms of the magnitude of reflectance values and temporal stability. DTA
497
shows clearly higher reflectance values than TOA reflectances and 6S surface reflectances. In
498
contrast, 6S shows lower values than TOA reflectances for visible bands but higher values for IR
499
bands. The behavior is similar in the March time series (not shown).
500
TOA reflectances are commonly higher than surface reflectances in visible bands because of
501
aerosol scattering and atmospheric Rayleigh scattering. Therefore, for visible bands it is expected
502
that atmospheric correction would reduce TOA reflectance values and that the reduction would be
503
spectrally dependant, being more pronounced in the blue band. The 6S method leads to reduced
19
504
reflectance values relative to TOA reflectances for visible bands (1–3), with the reduction being
505
greatest for band 1, in which aerosol scattering is more pronounced. In contrast, the DTA method
506
always overcorrects TOA reflectances for the visible bands.
507
The behaviors of the near- and middle-IR bands are expected to be different from those of the
508
visible bands because they are unaffected by aerosol scattering and because atmosphere molecular
509
absorption plays a major role. Given that absorption by gasses reduces the radiance received by the
510
sensor, it is expected that atmospheric correction leads to increased surface reflectance values for
511
near- and middle-IR bands relative to TOA reflectances. Both 6S and DTA surface reflectances
512
show higher values than TOA reflectances; however, DTA overcorrects for the influence of water
513
vapor absorption in band 4, as the influence of gas absorption in band 4 is similar to that in bands 5
514
and 7 (Arino et al., 1997).
515
Figure 5 shows, as an example, the relationship between TOA reflectances and DTA and 6S surface
516
reflectances derived for bands 1, 3, 4, and 7 in the image of 07/26/2001, versus the corresponding
517
MODIS surface reflectances for the same day in the region shown in Figure 4. Band 1 shows large
518
differences between TOA reflectances and MODIS surface reflectances. The DTA method
519
overcorrects surface reflectances, mainly in highly reflective areas, and 6S also shows higher values
520
than MODIS reflectances, although there is greater agreement relative to estimates obtained using
521
TOA and DTA. For bands 3 and 4, TOA reflectances underestimate the true values and DTA
522
clearly overestimates the surface reflectances relative to estimates obtained using MODIS. In
523
contrast, 6S and MODIS show strong agreement for most of the bands. For band 7, the difference
524
between TOA, 6S, and DTA surface reflectances and MODIS reflectances is lower than that
525
observed for the other bands; this is explained by the lower atmospheric influence in the mid-IR
526
region. The pattern apparent in the image of 26/07/2001 is also observed in other images taken
527
between 2001 and 2007, showing a high agreement between 6S and MODIS surface reflectances.
528
Although MODIS and Landsat images are corrected with the same 6S model, the used AOT
20
529
estimations are independent, showing high performance of the model to derive physically robust
530
surface reflectances under different input data.
531
Table 3 shows the average MAE obtained when comparing the MODIS surface reflectance and the
532
corresponding TOA reflectance and surface reflectances obtained from DTA and 6S in the five
533
analyzed images. 6S surface reflectances show a better match to MODIS reflectances than TOA
534
reflectances and DTA surface reflectances, especially for the visible and near-IR bands, in which
535
atmospheric aerosol and Rayleigh scattering are relatively pronounced. In contrast, for mid-IR
536
bands there is little difference in the MAE values obtained using the DTA and 6S methods; this
537
finding is attributed to the minor role of aerosols over these spectral regions.
538
Figure 5 shows a box plot of reflectance values collected in the field on 01/08/2006 and 13/03/2007
539
at the Mediana site (see Section 3.7.2). Also shown are simultaneous Landsat-TM TOA reflectances
540
and DTA and 6S surface reflectances collected over the same area. Field surface reflectance shows
541
greater spatial variability than that collected by satellite imagery. This phenomenon is commonly
542
observed in multi-scalar studies that employ remote sensing data, as finer spatial scales result in
543
greater spatial variability in reflectance values (Foody, 1991; Foody and Curran, 1994).
544
Independently of the range of reflectance values, the average 6S surface reflectances show a better
545
match with the field reflectances than DTA surface reflectances. The one-way analysis of variance
546
and the multiple comparisons between means do not statically show differences between the field
547
surface reflectances and the 6S surface reflectances in any of the four bands analysed for the two
548
day of measurements. On the contrary, the same analysis indicates statistically significant
549
differences between the DTA surface reflectances and the field surface reflectances for the different
550
bands. DTA, as previously shown with MODIS data, tends to overestimate the surface reflectance
551
values. The observed behavior is similar in the March and August images, and is the same as that
552
observed using MODIS data; no seasonal effects are inferred in the atmospheric correction process.
21
553
These results indicate that the 6S method is superior to the DTA method in terms of providing
554
physically coherent and robust surface reflectance values; accordingly, the 6S method is used for
555
subsequent radiometric corrections of the time series.
556
557
558
559
4.3. Topographic correction
Figure 6 shows the temporal evolution of March and August average reflectance values for bands 3
560
and 4 obtained in the Pinus halepensis forest indicated in section 3.7.3. The figure compares south-
561
facing and north-facing slopes, showing the time series of 6S surface reflectances and the surface
562
reflectance corresponding to a horizontal surface, as obtained via the Lambertian and C-correction
563
methods. In the series for March, non-topographically corrected reflectances show marked
564
differences between south-facing and north-facing slopes, mainly for band 4. Lambertian and non-
565
lambertian topographic correction improves adjustment of reflectances between south-facing and
566
north-facing slopes. The behavior observed after topographic correction provides a better match
567
with observed data, as in early spring the vegetation activity of these forests is very low, and no
568
differences should be observed in the reflectance values of north-facing and south-facing slopes.
569
Regarding the type of topographic correction, in March, a month in which the sun elevation angle is
570
low, band 3 shows similar results for the lambertian and non-lambertian models, whereas band 4
571
shows contrasting results. C-correction shows a good adjustment between north-facing and south-
572
facing slope reflectances, indicating the greater capacity of the model in correcting relief
573
perturbations under low sun-elevation angles.
574
The series for August shows higher reflectance values for south-facing than north-facing slopes.
575
These differences are less pronounced than those observed for March because of the higher sun-
576
elevation angle. Topographic correction leads to a reduction in the differences between north-facing
577
and south-facing slopes in band 3, with only minor differences observed between the lambertian
578
and non-lambertian models. For band 4, both topographic correction methods provide higher
579
reflectance values for north-facing than south-facing slopes. This pattern is more consistent with
580
real vegetation behavior because Pinus halepensis forests in the central Ebro Valley suffer water
22
581
stress during the summer months, and a north-facing exposure favors vegetation activity due to
582
lower evapotranspiration rates.
583
It is not only absolute reflectance values that are affected by topography: temporal variability is also
584
affected by topographic correction. This is clearly shown for the March time series, in which the
585
temporal variability of band 4 reflectance changes noticeably as a function of topographic
586
correction. Topographic effects are therefore spectrally dependant, and the temporal variability
587
observed in the series is strongly affected by topography during the season with a low sun-elevation
588
angle, as observed in March.
589
590
591
592
4.4. Relative radiometric normalization
Table 4 shows the Average MAE for each band relative the reference image in the dataset
593
(08/01/2006 for August and 03/13/2007 for March) in the time series for 6S, PIFs, PIFR, and TIC
594
reflectances for the PI validation pixels. For the August series, the PIFs method improves the
595
temporal homogeneity of the series for visible bands (1–3) relative to non-normalized reflectances,
596
but for near- and mid-IR bands (4–7) it yields similar errors to those in non-normalized images.
597
PIFR and TIC yield improved temporal homogeneity of the reflectance series, mainly for IR bands.
598
There are few differences in the MAE values between the two methods, although the PIFR method
599
generally yields lower MAE values.
600
For March, the differences between the non-normalized and normalized images are less pronounced
601
than those for August. PIFs yields inferior results compared with non-normalized images for bands
602
2–7. Although the PIFR method does not provide the best temporal homogeneity in certain bands, it
603
provides the best accuracy in terms of the average MAE errors between bands (0.0095 for non-
604
normalized images, 0.0083 for TIC, and 0.0075 for the PIFR method).
605
606
607
4.5. Case studies
23
608
Here, we provide a number of examples that illustrate the role of radiometric correction procedures
609
applied to time series of reflectance-based indices in evaluating land cover changes, forest
610
regeneration after forest fires, and ecosystem response to climate variability.
611
612
613
4.5.1. Case study 1: Detection of forest fire and subsequent recovery
614
The Zuera hills of the central Ebro Valley are residual Tertiary platforms with soils characterized by
615
a strong presence of limestone and gypsum and semiarid conditions (average annual precipitation =
616
475 mm). The area is covered entirely by Pinus halepensis forest that experiences occasional forest
617
fires. The most important historical fire burnt a total of 3093 ha in June of 1995. Following the fire,
618
no reforestation activities were undertaken, and the area was left to regenerate naturally. After 12
619
years of regeneration, a dense vegetation cover has formed. Most of the burnt area now supports a
620
high density of Pinus halepensis trees (235000 trees/ha upon some north-facing slopes with deep
621
soils), with ages between 10 and 12 years, heights between 1 and 2.5 m, and basal diameters
622
between 2 and 7 cm. The few areas in which Pinus halepensis has not regenerated are covered by
623
xeric shrubs.
624
The Normalized Burn Ratio (NBR) index (Miller and Yool, 2002) was calculated for each pixel and
625
image of the August series from TOA reflectances, DTA and 6S surface reflectances, and relative
626
radiometric normalized reflectances following the PIFs, PIFR, and TIC methods:
627
NBR 
628
where 4 and 7 are the reflectance values for bands 4 and 7.
629
This index is particularly effective in determining those areas affected by fire and burn severity
630
(Cocke et al., 2005; Wimberly and Reilly, 2007), and it has been used here to compare the different
631
radiometric correction procedures in the case that the occurrence of forest fire must be detected.
632
To analyze vegetation regeneration after forest fire, we used NDVI (Rousse et al., 1974), which is
633
widely used for this purpose (e.g., Díaz-Delgado et al., 2003; Viedma et al., 1997):
 4   7 (10)
4  7
24
 4  3
 4  3
634
NDVI 
635
where 4 and 3 are the reflectance values for bands 4 and 3.
636
Figure 7 shows the temporal evolution of average and standard deviation August NBR and NDVI
637
values obtained from TOA reflectances, DTA and 6S surface reflectances and PIFs, PIFR, and TIC
638
normalized reflectances for the burned areas in 1995. The fire event is well identified in 1995 by
639
NBR, regardless of the radiometric correction procedure. The six time series show a sharp decrease
640
in NBR values in 1995 and a progressive recovery to pre-fire values between 1997 and 2006. There
641
are no noticeable differences between time series as a function of the radiometric correction
642
procedure, and the magnitudes of NBR values are similar in all cases if directly calculated from
643
TOA reflectances or atmospherically corrected and normalized surface reflectances. Nevertheless,
644
noticeable differences have been found in the standard deviation values, because PIFR and PIFs
645
methods reduce the variability found of the NBR index of each image.
646
Time series of NDVI show important differences depending on the applied radiometric correction.
647
For 1995, all of the methods show a decrease in NDVI values after the fire. The behaviors of the
648
series show a marked change following the fire. The Pinus halepensis forests of the central Ebro
649
Valley show a highly stable interannual leaf vegetation activity, given the high resistance of these
650
trees to drought (Baquedano and Castillo, 2007). The occurrence of summertime hydric stress
651
caused by the calcareous and gypsum substrate and high evapotranspiration rates (Vicente-Serrano
652
et al., 2007) means that no vegetation activity is recorded in the underbrush species; only the Pinus
653
halepensis trees, which have the deepest root systems, maintain vegetation activity in summer. In
654
the period before the fire event, TOA, 6S, DTA, and PIFs reflectance-based NDVI values show
655
large variability, whereas the NDVI time series calculated from PIFR and TIC normalized
656
reflectances show a high degree of stability, more in accordance with reality. After the fire, the
657
NDVI values obtained from TOA reflectances are similar to those before the fire, indicating the
658
immediate (and therefore unrealistic) recovery of vegetation after the fire. NDVI values obtained
659
from 6S surface reflectances show a similar pattern. In contrast, NDVI values obtained after relative
25
660
normalization show a gradual increase following the fire; this pattern is observed for all three of the
661
applied methods (PIFs, TIC, and PIFR). Given that only Pinus halepensis trees are active in
662
summer and given the gradual recovery of tree density observed in the field, the pattern obtained
663
from PIFs, PIFR, and TIC reflectance is strongly consistent with the observed evolution of the
664
forest. In agreement with NBR, PIFR and PIFs methods also reduce the variability of the NDVI
665
values regarding 6S and TIC methods.
666
667
668
669
670
4.5.2. Case study 2: Identification of areas of dry land converted to irrigated land
671
drylands to highly modernized irrigated lands, intensively cultivated in the summer months for
672
corn, rice, alfalfa, and various vegetables. Figure 8 shows the average August NDVI values for a
673
sample of 58 pixels transformed in 1993 from dry herbaceous cultivation (winter cereals: wheat and
674
barley) to irrigated cultivation in the Monegros II irrigated areas. The series were obtained from
675
TOA reflectances, DTA and 6S surface reflectances, and relative normalized reflectances using the
676
PIFR, PIFs, and TIC methods.
677
For all radiometric correction procedures, the dry land converted to irrigated land in 1993 is readily
678
identified in the time series. The post-conversion trends of the various time series are similar for the
679
different methods, although the NDVI series derived from TOA reflectances does not record the
680
commonly observed interannual variability in vegetation activity within irrigated lands that arises
681
from the change in cultivation type. Greater differences are observed between the series in the
682
period prior to conversion. Cereals in drylands are harvested in June, meaning that in August the
683
soil is devoid of vegetation cover, and no interannual variability is expected in NDVI values.
684
Therefore, the variability in NDVI values obtained from TOA, DTA and 6S surface reflectances
685
prior to irrigation are interpreted to represent non-surface noise. The differences in the standard
686
deviation values are less important among the different methods than those observed in the previous
687
case study, but the values are in general lower for the PIFR and PIFs procedures.
During the 1990s, hundreds of hectares of land in the central Ebro Valley were transformed from
26
688
689
690
691
692
4.5.3. Case study 3: Response of steppes and areas of dryland cultivation to variability in
precipitation
Precipitation plays an important role in generating interannual variability in herbaceous cover in
693
steppes and areas of wheat and barley cultivation within the Ebro Valley, and as such has been
694
widely analyzed in field studies (Austin et al., 1998) and studies based on NOAA-AVHRR images
695
(Vicente-Serrano, 2007). The maximum vegetation activity of steppe regions is recorded in March–
696
April, being mainly dependent on soil moisture, which is highly variable and determined by the
697
precipitation accumulated over winter (Austin et al., 1999). A strong correlation was expected
698
between reflectance-based vegetation and/or moisture indices and wintertime precipitation. We
699
calculated a vegetation water index [Normalized Difference Infrared Index (NDII)] (Hunt and
700
Rock, 1989), which is a normalized ratio, using Landsat bands 4 and 5:
701
NDII 
702
where 4 and 5 are the reflectance values for bands 4 and 5, respectively. Because band 5 is located
703
in a strong water-absorption region, reflectance is inverse to leaf water content: low index values
704
indicate low water content. NDII was calculated from TOA reflectances, DTA and 6S surface
705
reflectance, and PIFs, PIFR, and TIC normalized surface reflectances for the steppes and dry
706
herbaceous lands of the centre of the Ebro valley.
707
Figure 9 shows the temporal evolution of the average and standard deviation NDII values for the
708
March time series and the November–February precipitation recorded at the observatory in
709
Zaragoza, which is representative of that for the entire region. The choice of radiometric correction
710
not only affects the magnitude of NDII values, but also the degree of variability in the series. NDII
711
values obtained from TOA reflectances and DTA and 6S surface reflectance show similar
712
behaviors. Although the maximum and minimum NDII values correspond to those years in which
713
maximum and minimum wintertime precipitation was recorded, respectively; the degree of
714
agreement between NDII and wintertime precipitation is relatively low for NDII values obtained
715
from relative normalized reflectances. Pearson coefficient of correlation values between NDII and
 4  5
 4  5
(11)
27
716
wintertime precipitation are R = 0.70, R = 0.71, and R = 0.73 (p < 0.01) for NDII values derived
717
from TOA and 6S and DTA surface reflectances, respectively.
718
The NDII values obtained from relative normalized surface reflectances show noticeable
719
differences between the PIFs, PIFR, and TIC methods. The PIFs approach shows the highest
720
temporal variability, with high NDII values recorded for years without excessive humidity, such as
721
1990. Although the correlation between NDII and wintertime precipitation is stronger than that
722
using TOA and 6S surface reflectance values (R = 0.74, p < 0.01), it is weaker than that obtained
723
using the PIFR and TIC methods (R = 0.77 and R = 0.81, p < 0.01, respectively). The NDII values
724
obtained from PIFR and TIC reflectance show good agreement with the wintertime precipitation
725
series. It is also observed in the time series of standard deviations that the PIFR method reduces the
726
deviation around the means in comparison with other methods, reducing the residual scatter in this
727
very spatially homogeneous region.
728
729
730
731
5. Discussion and conclusions
In this paper, we assessed the performances of different radiometric-correction procedures in
732
obtaining physically reliable and homogeneous time series of Landsat images. We tested the
733
different steps involved in the process, including the calibration of sensors, atmospheric and
734
topographic corrections, and relative radiometric normalization. We demonstrated that each step is
735
critical and necessary if the physical robustness and homogeneity of the time series are to be
736
maintained.
737
For sensor calibration, we documented the difficulties that exist in directly comparing the
738
reflectance values obtained from TM and ETM+ sensors. In general, higher reflectance values are
739
obtained from ETM+ than TM reflectances for the majority of bands, in accordance with previous
740
studies (Teillet et al., 2006; Vogelmann et al., 2001). Although most studies based on Landsat data
741
consider ETM+ and TM reflectances to be comparable, with the outcome that cross-calibration
742
procedures are rarely applied, we demonstrated that a change in satellite causes a noticeable
743
instability in the time series. Moreover, important magnitude differences exist as a function of
28
744
Landsat band, which may have noticeable effects when band-based operations are applied, such as
745
the calculation of vegetation indices. These results indicate the need of TM/ETM+ cross-calibration
746
to improve the temporal homogeneity of Landsat series. Despite the limitations of Landsat data
747
provided by ESA in 1G formats, the application of cross-calibration coefficients developed by
748
Vogelmann et al. (2001) shows great potential in reducing the temporal inhomogeneities related to
749
differences in sensor response, although not in all bands (e.g., bands 1 and 7).
750
We found that the temporal homogeneity of the series is unaffected by the method employed for
751
atmospheric correction. This may suggest that the atmospheric correction step is not critical in the
752
case that the objective is to create multitemporal Landsat series. Other studies have reported similar
753
results; for example, Schroeder et al. (2006) found no differences in the temporal stability of
754
Landsat series obtained after applying DTA and 6S atmospheric correction techniques. The authors
755
reported the same error for the two procedures (RMSE = 0.024) and no difference in TOA
756
reflectance (RMSE = 0.023). Moreover, we have shown that the application of atmospheric
757
correction only fails to produce homogeneous multitemporal series of Landsat surface reflectances.
758
This has been observed for both simple atmospheric correction methods such as DTA and complex
759
models such as 6S. Based on these results, we can confirm that the temporal stability of a series
760
must be a secondary criterion when applying atmospheric correction: the physical robustness of
761
reflectance values must be the prime factor. The subsequent application of relative normalization
762
techniques is required to solve the problem of temporal homogeneity.
763
We have shown that the DTA method overcorrects reflectance values for all bands, regardless of the
764
atmospheric conditions and season. This overcorrection has a noticeable impact on the calculation
765
of vegetation indexes such as NDVI, resulting in a general underestimation and leading to
766
subsequent problems in extracting biophysical properties and undertaking multitemporal studies of
767
vegetation cover (Song and Woodcock, 2003). Therefore, the use of TOA reflectances and DTA
768
surface reflectances is inappropriate for multitemporal analyses for which an accurate estimation of
769
vegetation parameters or robust vegetation indices are required.
29
770
Improved physical performance was obtained in using the 6S model, which reduces the magnitude
771
of reflectance values for visible bands (1–3) and increases the magnitude in near- and mid-IR bands
772
(4–7), as is physically expected due to the contrasting influences on the signal of aerosols and the
773
molecular atmosphere (Arino et al., 1997). Here, we used indirect AOT estimates from the image
774
itself. This is the most critical parameter used in the model (Ouaidrari and Vermote, 1999), as some
775
authors have reported contrasting surface reflectance values depending on whether direct AOT
776
measures or indirect estimations were used (Song et al., 2001; Schroeder et al., 2006). Nevertheless,
777
other studies have reported similar 6S surface reflectances regardless of whether AOT was obtained
778
indirectly from the DDV method or via direct measurements (Song and Woodcock, 2003). Masek et
779
al. (2006) compared DDV-AOT estimates with AERONET observations at 18 sites in North
780
America, showing a reasonable agreement between image-based AOT estimates and observations
781
(R = 0.81). In our study, a comparison of quasi-simultaneous MODIS surface reflectances and two
782
samples of surface reflectances obtained in the field also demonstrated the strong performance of
783
the 6S model in estimating surface reflectance values. Given that MODIS images are
784
atmospherically corrected using AOT estimates, which are obtained with a high degree of accuracy
785
(Chu et al., 2002), our results also indicate the excellent accuracy of the AOT estimates obtained for
786
the Ebro Valley using Landsat data.
787
Topographic normalization is seldom taken into account in multitemporal processing protocols
788
applied to Landsat data. We found noticeable differences in the temporal evolution of surface
789
reflectance values between north-facing and south-facing slopes upon a relatively flat, directly
790
related to illumination effects. Our results suggest that the spatial comparability of images is not
791
assured without the application of topographic corrections, which ensure that the evolution of
792
surface reflectance values between north-facing and south-facing slopes are comparable. In
793
considering topographic correction methods, there are few differences between the lambertian and
794
non-lambertian procedures. Nevertheless, the non-lambertian C-correction provides a better match
795
between the temporal evolution of surface reflectance values for north-facing and south-facing
30
796
slopes. Given the readily satisfied requirements for topographic correction (requiring only a digital
797
elevation model at the same spatial resolution of the Landsat images), it is highly recommended that
798
this step is included in multitemporal processing protocols, especially for areas with highly complex
799
topography in which the sun-surface-sensor geometry is also complex.
800
We have shown that relative radiometric normalization is essential to ensure the homogeneity of
801
multitemporal Landsat datasets. Previous studies have compared the temporal stability of
802
multitemporal Landsat datasets before and after relative radiometric normalization. In most of these
803
studies, normalization was applied without prior atmospheric correction (e.g., Olsson, 1993; Yuan
804
and Elvidge, 1996; Olthof et al., 2005), indicating the irrelevance of this step in ensuring the
805
temporal homogeneity of the series. Recent studies have also compared the homogeneity of
806
multitemporal atmospherically corrected Landsat datasets before and after relative radiometric
807
normalization (Janzen et al., 2006; Schroeder et al., 2006), confirming the need to apply this step
808
because atmospheric correction alone is unable to remove the non-surface noise in the series. The
809
results of this paper are in accordance with the findings of these studies; nevertheless, it is also
810
obtained evidence of the need for the prior cross-calibration and atmospheric correction of images.
811
In contrast, accurate biophysical parameters cannot be obtained from inversion models (González-
812
Sanpedro et al., 2008), and it is not possible to calculate the true magnitude of band-based indices
813
such as NDVI (Myneni and Asrar, 1994).
814
Among the three methods tested in applying relative radiometric corrections, the most manual—that
815
based on the visual identification of PIFs—provided the worst results. The temporal stability of the
816
series was improved when using the PIFR and TIC methods, which yielded similar results, although
817
the low time and cost requirements of PIFR and its automatic nature and automatic operation make
818
it preferable over TIC. Moreover, the PIFR method reduced the deviation scatter around the mean
819
values in more depth than the other methods for the different vegetation and moisture indices
820
indicating a higher success in the radiometric correction given the spatial homogeneity of the case
821
study analysed.
31
822
In summary, for analyzing abrupt vegetation changes such as forest fire events or land cover
823
changes based on different band indices, it is unnecessary to apply complex radiometric correction
824
procedures to Landsat datasets: the use of simple TOA reflectances provides good results.
825
Nevertheless, when the spectral signal is not sufficiently strong to minimize noise from real
826
vegetation dynamics and change, it is necessary to undertake highly accurate radiometric
827
corrections. To analyze vegetation dynamics before and after events that involve change, such as
828
the rate of vegetation recovery after a fire or the role of climate variability on vegetation activity, it
829
is necessary to accurately ensure the homogeneity of multitemporal Landsat datasets via complete
830
radiometric correction procedures that include sensor calibration and cross-calibration, atmospheric
831
correction, topographic correction and relative radiometric normalization using objective statistical
832
techniques. Failure to follow such a procedure may mean that the analyzed processes are not
833
recognizable and that the obtained results are invalid.
834
835
836
837
Acknowledgements
838
Spanish Commission of Science and Technology and FEDER, and “Programa de grupos de
839
investigación consolidados” financed by the Aragón Government. Authors would like to thank to
840
the four anonymous reviewers for their helpful comments.
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
This work has been supported by the research project CGL2005-04508/BOS, financed by the
References
Almeida-Filho, R. and Shimabukuro, Y.E. (2002): Digital processing of a Landsat-TM time series
for mapping and monitoring degraded areas caused by independent gold miners, Roraima
State, Brazilian Amazon. Remote Sensing of Environment 79: 42-50.
Arino O., Vermote E.F. and Spaventa V., (1997): Operational Atmospheric Correction of Landsat
TM Imagery, Earth Observation Quarterly 56-57: 32-35.
Austin, R.B., Cantero-Martínez, C., Arrúe, J.L., Playán, E. and Cano-Marcellán, P., (1998): Yieldrainfall relationships in cereal cropping systems in the Ebro river valley of Spain. European
Journal of Agronomy, 8: 239-248.
32
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
Austin, R.B., Playán, E. and Gimeno, J., (1999): Water storage in soils during the fallow: prediction
of the effects of rainfall pattern and soil conditions in the Ebro valley of Spain. Agricultural
Water Management, 36: 213-231.
Baquedano, F.J. and Castillo, F.J. (2007): Drought tolerance in the Mediterranean species Quercus
coccifera, Quercus ilex, Pinus halepensis, and Juniperus phoenicea. Photosynthetica 45:
229-238.
Berk, A., Anderson, G.P., Bernstein, L.S., Acharya, P.K., Dothe, H., Matthew, M.W., AdlerGolden, S.M., Chetwynd, J.H., Richtsmeier, S.C., Pukall, B., Allred, C.L., Jeong, L.S. and
Hoke, M.L., (1999): MODTRAN4 radiative transfer modeling for atmospheric correction.
Proceedings of SPIE Optical Spectroscopic Techniques and Instrumentation for
Atmospheric and Space Research III, Denver, Co, USA, 18 July 1999.
Bossard, M., Feranec, J., and Otahel, J. (2000): CORINE Land Cover. Copenhagen (EEA).
http://terrestrial.eionet.eu.int. Technical Guide — Addendum 2000. Technical report No 40.
Cachorro, V.E., Durán, P., Vergaz, R. and De Frutos, A.M., (2000): Estudio de la influencia de los
aerosoles sobre la reflectancia de los canales 1 y 2 del sensor AVHRR NOAA y el NDVI.
Revista de Teledetección 13: 1-13.
Canty, M.J., Nielsen, A.A., and Schmidt, M. (2004): Automatic radiometric normalization of
multitemporal satellite imagery. Remote Sensing of Environment 91: 441-451.
Caselles, V. and López-García, M.J., (1989): An alternative simple approach to estimate
atmospheric correction in multitemporal studies. International Journal of remote Sensing,
10: 1127-1134.
Chander, G. and Markham, B. (2003). Revised Landsat-5 TM radiometric calibration procedures
and postcalibration dynamic ranges. IEEE Transactions on Geoscience and Remote Sensing,
41, 2674-2677.
Chander, G., Helder, D.L., Markham, B.L., Dewald, J.D., Kaita, E., Thome, K.J., Micijevic, E. and
Ruggles, T.A. (2004). Landsat-5 TM reflective-band absolute radiometric calibration. IEEE
Transactions on Geoscience and Remote Sensing, 42, 2746-2760.
Chander, G., Markham, B.L. and Barsi, J.A., (2007). Revised Landsat 5 Thematic mapper
Radiometric calibration. IEEE Transactions on Geoscience and Remote Sensing, 4, 490-494.
Chavez, P.S., (1989): Radiometric calibration of landsat Thematic Mapper multispectral images.
Photogrammetric Engineering and Remote Sensing, 55: 1285-1294.
Chavez, P.S., (1996): Image-based atmospheric corrections-revisited
Photogrammetric Engineering and Remote Sensing, 62: 1025-1036.
and
improved.
Chen, X., Vierling, L. and Deering, D., (2005): A simple and effective radiometric correction
method to improve landscape change detection across sensors and across time. Remote
Sensing of Environment 98: 63-79.
Chu, D.A., Kaufman, Y.J., Ichoku, C., Remer, .A., Tanré D. and Holben, B.N., (2002): Validation
of MODIS aerosol optical depth retrieval overland. Geophysical Research Letters, 29, DOI:
10.1029/2001GL013205.
33
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
Civco, D.L., (1989): Topographic normalization of Landsat Thematic Mapper digital imagery.
Photogrammetric Engineering & Remote Sensing 55: 1303-1309.
Cocke, A.E., Fulé, P.Z. and Crouse, J.E., (2005): Comparison of burn severity assessments using
Differenced Normalized Burn Ratio and ground data. Journal of Wildland Fire 14: 189-198.
Cohen, W.B., Spies, T.A., Alig, R.J., Oetter, D.R., Maiersperger, T.K. and Fiorella, M., (2002):
Characterizing 23 years (1972-95) od stand replacement disturbance in Western Oregon
forests with Landsat imagery. Ecosystems, 5: 122-137.
Cohen, W., and Goward, S. (2004): Landsat's role in ecological applications of remote sensing.
BioScience, 54: 535−545.
Colby, J.D., (1991): Topographic normalization in rugged terrain. Photogrammetric Engineering
and Remote Sensing, 62: 151-161.
Coppin, P.R. and Bauer, M.E., (1994): Processing of multitemporal Landsat TM imagery to
optimize extraction of forest cover change features. IEEE Transactions on Geoscience and
remote Sensing, 32: 918-927.
Díaz-Delgado, R., Lloret, F. and Pons, X., (2003): Influence of fire severity on plant regeneration
by means of remote sensing imagery. International Journal of Remote Sensing 24: 17511763.
Dozier, J., (1989): Spectral signature of alpine snow cover from the Landsat Thematic Mapper,
remote Sensing of Environment, 28: 9-22.
Du, Y., Cihlar, J., Beaubien, J. and Latifovic, R., (2001): Radiometric normalization, compositing,
and quality control for satellite high resolution image mosaics over large areas. IEEE
Transactions on Geoscience and Remote Sensing, 39: 623-634.
Du, Y., Teillet, P. and Cihlar, J., (2002) : Radiometric normalization of multitemporal highresolution satellite images with quality control and land cover change detection. Remote
Sensing of Environment, 82: 123-134.
Eckhardt, D.W., Verdin, J.P. and Lyford, G.R., (1990): Automated update of an irrigated land GIS
using SPOT HRV imagery. Photogrammetric Engineering and Remote Sensing, 56: 15151522.
Foody, G. M., (1991):, Soil moisture content ground data for remote sensing investigations of
agricultural regions. International Journal of Remote Sensing, 12: 1461–1469.
Foody, G. M., and Curran, J. P., (1994): Scale and environmental remote sensing. In Environmental
Remote Sensing from Regional to Global Scales, edited by G. M. Foody and P. J. Curran
(New York: Wiley), pp. 223–232.
González-Sanpedro, M.C., Le Toan T., Moreno, J., Kergoat, L. and Rubio, E. (2008): Seasonal
variations of leaf area index of agricultural fields retrieved from Landsat data. Remote
Sensing of Environment, doi:10.1016/j.rse.2007.06.018
34
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
Gu, D., and Gillespie, A. (1998): Topographic normalization of Landsat TM images of forest based
on subpixel Sun-canopy-sensor geometry. Remote Sensing of Environment 64: 166-175.
Hall, F.G., Strebel, D.E., Nickeson, J.E. and Goetz, S.J. (1991): Radiometric rectification: toward a
common radiometric response among multidate, multisensor images. Remote Sensing of
Environment 35: 11-27.
Han, T., Wulder, M.A., White, J.C., Coops, N.C., Alvarez, M.F. and Butson, C., (2007): An
efficient protocol to process landsat images for change detection with tasselled cap
transformation. IEEE Geoscience and Remote Sensing Letters, 4: 147-151.
Heo, J. and FitzHugh, T.W. (2000). A standardized radiometric normalization method for change
detection using remotely sensed imagery. Photogrammetric Engineering and Remote
Sensing, 66, 173-181.
Hill, J. and Sturm, B., (1991): Radiometric correction of multitemporal Thematic mapper data for
use in agricultural land-cover classification and vegetation monitoring. International
Journal of Remote Sensing, 12: 1471-1491.
Holben, B., (1986): Characteristics of maximum value composite images from temporal AVHRR
data. International Journal of Remote Sensing. 6: 1271-1328.
Hostert, P., Röder, A. and Hill, J., (2003): Coupling spectral unmixing and trend analysis for
monitoring of long-term vegetation dynamics in Mediterranean rangelands. Remote Sensing
of Environment, 87: 183-197.
Huang, C., Kim, S., Altstatt, A., Townshend, J.R.G., Davis, P., Song, K., Tucker, C.J., Rodas, O.,
Yanosky, A., Clay, R. and Musinsky J., (2007): Rapid loss of Paraguay’s Atlantic forest and
the status of protected areas – A Landsat assessment. Remote Sensing of Environment, 106:
460-466.
Hunt, E. R. and Rock, B. N. (1989): Detection of changes in leaf water content using near and
middle-infrared reflectances. Remote Sensing of Environment 30: 43-54.
Jano, A.P., Jefferies, R.L. and Rockwell, R.F., (1998): The detection of vegetational change by
multitemporal analysis of Landsat data: the effects of goose foraging. Journal of Ecology,
86: 93-99.
Janzen, D.T., Fredeen, A.L. and Wheate, R.D. (2006). Radiometric correction techniques and
accuracy assessment for landsat TM data in remote forested regions. Canadian Journal of
Remote Sensing, 32, 330-340.
Kaufman, Y.J., Tanré, D., Remer, L.A., Vermote, E.F., Chu, A. and Holben, B.N., (1996):
Operational remote sensing of tropospheric aerosol over the land from EOS-MODIS.
Journal of Geophysical Research, 102: 17051-17067.
Kogan, F. N., (1998): Global drought watch from space. Bulletin of the American Meteorological
Society 78, 621–636.
Lawrence, R.L. and Ripple, W.J., (1999): Calculating change curves for multitemporal satellite
imagery: Mount St. Helens 1980-1995. Remote Sensing of Environment, 67: 309-319.
35
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
Lenney, M.P., Woodcock, C.E., Collins, J.B. and Hamdi, H. (1996). The status of agricultural lands
in Egypt: the use of multitemporal NDVI features derived from Landsat TM. Remote
Sensing of Environment, 56, 8-20.
Liang, S., Fallah-Adl, H., Kalluri, S., JáJá, J., Kaufman, Y. and Townshend, R.G., (1997): An
operational atmospheric correction algorithm for Landsat Thematic Mapper imagery over
land. Journal of Geophysical Research, 102: 17173-17186.
Liang, S., Fang, H. and Chen, M. (2001): Atmospheric correction of Landsat ETM+ land surface
imagery-Part I: Methods. IEEE Transactions on Geoscience and Remote Sensing 39: 24902498
Lozano, F.J., Suárez-Seoane, S. and de Luis, E. (2007): Assessment of several spectral indices
derived from multi-temporal Landsat data for fire occurrence probability modelling. Remote
Sensing of Environment 107: 533-544
Lu, D., Mausel, P., Brondizio, E. and Moran, E., (2002): Assessment of atmospheric correction
methods for Landsat TM data applicable to Amazon basin LBA research. International
Journal of Remote Sensing,23: 2651-2671.
Masek, J.G., Vermote, E.F., Saleous, N.E., Wolfe, R., Hall, F.G., Huemmrich, K.F., Feng Gao,
Kutler, J. and Teng-Kui, L., (2006): A Landsat surface reflectance dataset of North America,
IEEE Geoscience and Remote Sensing Letters, 3: 68 – 72.
Miller, J.D. and Yool, S.R. (2002): Mapping forest post-fire canopy consumption in several
overstory types using multi-temporal Landsat TM and ETM data. Remote Sensing of
Environment 82: 481-496.
Millington, A.C., Velez-Liendo, X.M., Bradley, A.V., (2003): Scale dependence in multitemporal
mapping of forest fragmentation in Bolivia: implications for explaining temporal trends in
landscape ecology and applications to biodiversity conservation. ISPR Journal of
Photogrammetry & Remote Sensing 57: 289-299.
Myneni, R.B. and Asrar, G., (1994): Atmospheric effects and spectral vegetation indices. Remote
Sensing of Environment, 47: 390-402.
Norjamäki, I. and Tokola, T., (2007): Comparison of atmospheric correction methods in mapping
timber volume with multitemporal Landsat images in Kainuu, Finland. Photogrammetric
Engineering and Remote Sensing, 73: 155-163.
Olsson, H. (1993): Regression functions for multitemporal relative calibration of Thematic Mapper
data over boreal forest. Remote Sensing of Environment 46: 89-102.
Olsson, H., (1995): Reflectance calibration of Thematic Mapper data for forest change detection.
International Journal of Remote Sensing 16: 81-96.
Olthof, I., Pouliot, D., Fernandes, R. and Latifovic, R., (2005): Landsat-7 ETM+ radiometric
normalization comparison for northern mapping applications. Remote Sensing of
Environment, 95: 388-398.
Ouaidrari, H. and Vermote, E.F., (1999): Operational atmospheric correction of Landsat TM data.
Remote Sensing of Environment, 70: 4-15.
36
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
Palà, V. and Pons, X. (1995). Incorporation of relief in polynomial-based geometric corrections.
Photogrammetric Engineering and Remote Sensing, 61, 935-944.
Paolini, L., Grings, F., Sobrino, J.A., Jiménez Muñoz, J.C. and Karsebaum, H., (2006): radiometric
correction effects in landsat multi-date/multi-sensor change detection studies. International
Journal of Remote Sensing, 27: 685-704.
Peña, J.L., Pellicer, F., Julián, A., Chueca, J., Echeverría, M.T., Lozano, M.V., and Sánchez, M.
(2002): Mapa geomorfolo´gico de Aragón. Consejo de Protección de la Naturaleza de
Aragón, 54 pp + 3 maps
Pickup, G. and Chewings, V.H., (1994): A grazing gradient approach to land degradation
assessment in arid areas from remotely-sensed data. International Journal of Remote
Sensing, 15: 597-617.
Pons, X. and Solé-Sugrañes, L., (1994): A simple radiometric correction model to improve
automatic mapping of vegetation from multispectral satellite data. Remote Sensing of
Environment, 48: 191-204.
Proy, C., Tanre, D. and Deschamps, P.Y. (1989): Evaluation of topographic effects in remotely
sensed data. Remote Sensing of Environment 30: 21-32.
Rahman, H. and Dedieu, G. (1994). SMAC: a simplified method for the atmospheric correction of
satellite measurements in the solar spectrum. International Journal of Remote Sensing, 15,
123-143.
Riaño, D., Chuvieco, E., salas, J. and Aguado, I., (2003): Assessment of different topographic
corrections in landsat-TM data for mapping vegetation types. IEEE Transactions on
Geoscience and Remote Sensing, 41: 1056-1061.
Richter, R. (1996). A spatially adaptive fast atmospheric correction algorithm. International
Journal of Remote Sensing, 17, 1201-1214.
Röder, A., Kuemmerle, T. and Hill, J., (2005): Extension of retrospective datasets using multiple
sensors. An approach to radiometric intercalibration of Landsat TM and MSS data. Remote
Sensing of Environment. 95: 195-210.
Rousse, J.W., Haas, R.H., Schell, J.A., Deering, D.W. and Harlan, J.C. (1974): Monitoring the
vernal advancement of retrogradation of natural vegetation, NASA/GSFC, Type III, Final
Report, Greenbelt, MD, 371 pp
Saunier, S. and Rodríguez, Y., (2006): Landsat Product Radiometric calibration. Technical note.
ESA. Available in http://earth.esa.int/pub/ESA_DOC/GAEL-calibration-proceeding.pdf
Schott, J.R., Salvaggio, C., and Volchok, W.J. (1988). Radiometric scene normalization using
pseudoinvariant features. Remote Sensing of Environment, 26, 1-6.
Schroeder, T.A., Cohen, W.B., Song, C., Canty, M.J. and Yang, Z. (2006). Radiometric correction
of multi-temporal Landsat data for characterization of early succesional forest patterns in
western Oregon. Remote Sensing of Environment, 103, 16-26.
37
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
Song, C. and Woodcock, C.E., (2003): Monitoring forest succesion with multitemporal Landsat
images: factors of uncertainty. IEEE Transactions on Geoscience and Remote Sensing,41:
2557-2567.
Song, C., Woodcock, C.E., Seto, K.C., Lenney, M.P. and Macomber, S.A., (2001): Classification
and change detection using Landsat TM data: When and how to correct atmospheric
effects?. Remote Sensing of Environment, 75: 230-244.
Song, C., Schroeder, T.A. and Cohen, W.B., (2007): Predicting temperate conifer forest
successional stage distributions with multitemporal Landsat Thematic Mapper imagery.
Remote Sensing of Environment, 106: 228-237.
Teillet, P.M. and Holben, B.N., (1993): Toward operational radiometric calibration of NOAAAVHRR imagery in visible and infrared channels. Canadian Journal of Remote Sensing 20:
1-10.
Teillet, P.M., Guindon, B. and Goodeonugh, D.G., (1982): On the slope-aspect correction of
multispectral scanner data. Canadian Journal of Remote Sensing, 8: 84-106.
Teillet, P.M. and Fedosejevs, G., (1995): On the dark target approach to atmospheric correction of
remotely sensed data. Canadian Journal of Remote Sensing. 21: 374-387.
Teillet, P.M., Barker, J.L., Markham, B.L., Irish, R.R., Fedosejevs, G. and Storey, J.C., (2001):
Radiometric cross-calibration of the Landsat-7 ETM+ and Landsat-5 TM sensors based on
tandem data sets. Remote Sensing of Environment, 78: 39-54.
Teillet, P.M., Helder, D.L., Ruggles, T.A., Landry, R., Ahern, F.J.., Higgs, N.J., Barsi, J., Chander,
G., Markham, B.L., Barker, J.L., Thome, K.J., Schott, J.R. & Palluconi, F.D. (2004). A
definitive calibration record for the Landsat-5 thematic mapper anchored to the landsat-7
radiometric scale. Canadian Journal of Remote Sensing, 30, 631-643.
Teillet, P.M., Markham, B.L. and Irish, R.R., (2006): Landsat cross-calibration based on near
simultaneous imaging of common ground targets. Remote Sensing of Environment, 102:
264-270.
Tokola, T., Löfman, S. and Erkkilä, A., (1998): Relative calibration of multitemporal Landsat data
for forest cover change detection. Remote Sensing of Environment, 68: 1-11.
Van Dijk, A., Callis, S.L., Sakamoto, C.M. y Decker, W.L., (1987): Smoothing vegetation index
profiles: an alternative method for reducing radiometric disturbance in NOAA/AVHRR
data. Photogrammetric Engineering and Remote Sensing. 53: 1059-1067.
Van Heuklon, T.K., (1979): Estimating atmospheric ozone for solar radiation models, Solar Energy
22: 63-68.
Vermote, E.F., Tanré, D., Deuzé, J.L., Herman, M. and Morcrette, J.J., (1997a): Second simulation
of the satellite signal in the solar spectrum, 6S: an overview. IEEE Transactions on
Geoscience and Remote Sensing, 35: 675-686.
Vermote E.F., El Saleous N.Z., Justice C.O., Kaufman Y.J., Privette J., Remer L., Roger J.C. and
Tanré D., (1997b): Atmospheric correction of visible to middle infrared EOS-MODIS data
38
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
over land surface, background, operational algorithm and validation, Journal of Geophysical
Research, 102: 17131-17141.
Vermote E.F., El Saleous N. and Justice C. (2002): Atmospheric correction of the MODIS data in
the visible to middle infrared: First results, Remote Sensing of Environment 83: 97-111.
Vicente-Serrano, S.M., Pons, X. and Cuadrat, J.M., (2004), Using Landsat and NOAA satellites on
soil moisture mapping and comparison with meteorological data. Application in the central
Ebro river valley (NE of Spain). International Journal of Remote Sensing. 25: 4325-4350.
Vicente-Serrano, S.M., Cuadrat, J.M. and Romo, A. (2006): Aridity influence on vegetation
patterns in the middle Ebro valley (Spain): evaluation by means of AVHRR images and
climate interpolation techniques. Journal of Arid Environments, 66: 353-375.
Vicente-Serrano, S.M. (2007). Evaluating the Impact of Drought Using Remote Sensing in a
Mediterranean, Semi-arid Region. Natural Hazards, 40, 173-208.
Vicente-Serrano, S.M., Lanjeri, S. and López-Moreno, J.I. (2007): Comparison of different
procedures to map reference evapotranspiration using geographical information systems and
regression-based techniques. International Journal of Climatology, 27: 1103-118.
Viedma, O., Meliá, J., Segarra, D. and García-Haro, J. (1997): Modeling rates of ecosystem
recovery after fires by using landsat TM data. Remote Sensing of Environment 61: 383-398.
Vincini, M. and Frazzi, E., (2003): Multitemporal evaluation of topographic normalization methods
on deciduous forest TM data. IEEE Transactions on Geoscience and Remote Sensing, 41:
2586-2590.
Vogelmann, J.E., Helder, D., Morfitt, R., Choate, M.J., Merchant, J.W. and Bulley, H. (2001).
Effects of Landsat 5 Thematic Mapper and Landsat 7 Enhanced Thematic Mapper Plus
radiometric and geometric calibrations and corrections on landscape characterization.
Remote Sensing of Environment, 78, 55-70.
de Vries, C., Danaher, T., Denham, R., Scarth, P. and Phinn, S. (2007): An operational radiometric
calibration procedure for the Landsat sensors based on pseudo-invariant target sites. Remote
Sensing of Environment 107: 414-429.
Wen, G., Tsay, S., Cahalan, R.F. and Oreopoulos, L., (1999): Path radiance technique for retrieving
aerosol optical thickness over land. Journal of Geophysical Research, 104: 31321-31332.
Wimberly, M.C. and Reilly, M.J. (2007): Assessment of fire severity and species diversity in the
southern Appalachians using Landsat TM and ETM+ imagery. Remote Sensing of
Environment 108: 189-197.
Yang, C. and Vidal, A., (1990): Combination of digital elevation models with SPOT-1 HRV
multispectral imagery for reflectance factor mapping. Remote Sensing of Environment, 32:
35-45.
Yuan, D. and Elvidge, C.D., (1996): Comparison of relative radiometric normalization techniques.
ISPRS Journal of Photogrammetry & Remote Sensing, 51: 117-126.
39
1217
Table 1. Dates of the Landsat 5-TM and Landsat 7-ETM+ images used in this study
March
Date
11/03/1989
30/03/1990
06/03/1993
09/03/1994
28/03/1995
17/03/1997
20/03/1998
23/03/1999
17/03/2000
10/03/2003
07/03/2005
13/03/2007
August
Sensor
TM
TM
TM
TM
TM
TM
TM
TM
ETM+
ETM+
TM
TM
Date
20/08/1984
07/08/1985
13/08/1987
02/08/1989
24/08/1991
10/08/1992
29/08/1993
03/08/1995
24/08/1997
14/08/1999
08/08/2000
26/07/2001
30/08/2002
27/08/2004
14/08/2005
01/08/2006
Sensor
TM
TM
TM
TM
TM
TM
TM
TM
TM
TM
ETM+
ETM+
ETM+
TM
TM
TM
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
Table 2. Mean Average Error for each band relative to the reference image in the dataset
(08/01/2006 for August and 03/13/2007 for March)
August
March
Original Cross-calibrated Original Cross-calibrated
Band 1
Band 2
Band 3
Band 4
Band 5
Band 7
0.0142
0.0122
0.0112
0.0111
0.0169
0.0106
0.0132
0.0110
0.0090
0.0062
0.0088
0.0095
0.0115
0.0083
0.0106
0.0123
0.0146
0.0090
0.0125
0.0070
0.0087
0.0074
0.0073
0.0076
Table 3. Average Mean Absolute Error (MAE) between MODIS and Landsat reflectances. Average
MAE values were obtained for the following images: 07/26/2001, 08/30/2002, 03/10/2003,
03/07/2005, and 08/01/2006.
TOA
BAND 1
BAND 2
BAND 3
BAND 4
BAND 5
BAND 7
DTA
0.06
0.03
0.04
0.08
0.10
0.09
6S
0.16
0.08
0.07
0.07
0.04
0.05
0.02
0.03
0.02
0.03
0.05
0.04
1237
1238
1239
40
1240
1241
1242
1243
1244
Table 4. Average MAE for each band relative the reference image in the dataset (08/01/2006 for
August and 03/13/2007 for March) in the time series for 6S, PIFs, PIFR, and TIC reflectances.
AUGUST
band 1
band 2
band 3
band 4
band 5
band 7
MARCH
band 1
band 2
band 3
band 4
band 5
band 7
6S
0.0080
0.0114
0.0113
0.0121
0.0123
0.0130
6S
0.0131
0.0076
0.0096
0.0097
0.0079
0.0088
PIFs
0.0045
0.0057
0.0063
0.0104
0.0110
0.0125
PIFs
0.0063
0.0115
0.0112
0.0104
0.0110
0.0100
TIC
0.0060
0.0076
0.0065
0.0071
0.0063
0.0042
TIC
0.0038
0.0076
0.0114
0.0081
0.0100
0.0086
PIFR
0.0048
0.0054
0.0069
0.0054
0.0050
0.0046
PIFR
0.0050
0.0080
0.0093
0.0042
0.0079
0.0107
41
1245
1246
1247
1248
1249
Figure 1. Location of path 199, row 31 and the spatial distribution of the main land-cover types
42
1250
1251
0.5
0.5
Band 1
Band 2
0.4
Reflectance
Reflectance
0.4
Original
Cross-calibrated
0.3
0.2
0.1
0.3
0.2
0.1
L5-TM
L7-ETM+
L5-TM
L5-TM
0.0
0.5
Band 3
Band 4
0.4
Reflectance
Reflectance
0.4
0.3
0.2
0.1
0.3
0.2
0.1
L5-TM
L7-ETM+
L5-TM
L5-TM
0.0
L7-ETM+
L5-TM
0.0
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
0.5
0.5
Band 7
Band 5
0.4
Reflectance
0.4
Reflectance
L5-TM
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
0.5
0.3
0.2
0.3
0.2
0.1
0.1
L5-TM
L7-ETM+
L5-TM
L5-TM
L7-ETM+
L5-TM
0.0
0.0
1252
1253
1254
1255
1256
L7-ETM+
0.0
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
Figure 2. Time series of August average TOA reflectance values for the PI validation pixels for
each band in the TM and ETM+ images and cross-calibrated ETM+ images.
43
0.5
0.5
Band 1
0.4
Reflectance
0.4
Reflectance
Band 2
TOA
6S
DTA
0.3
0.2
0.1
0.3
0.2
0.1
L5-TM
L7-ETM+
L5-TM
L5-TM
0.0
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
0.5
Band 3
Band 4
0.4
Reflectance
Reflectance
0.4
0.3
0.2
0.1
0.3
0.2
0.1
L5-TM
L7-ETM+
L5-TM
L5-TM
0.0
L7-ETM+
L5-TM
0.0
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
0.5
0.5
Band 7
Band 5
0.4
Reflectance
0.4
Reflectance
L5-TM
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
0.5
0.3
0.2
0.3
0.2
0.1
0.1
L5-TM
L7-ETM+
L5-TM
L5-TM
L7-ETM+
L5-TM
0.0
0.0
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
1257
1258
1259
1260
1261
L7-ETM+
0.0
84 85 87 89 91 92 93 95 97 99 00 01 02 04 05 06
19 19 19 19 19 19 19 19 19 19 20 20 20 20 20 20
Figure 3. Time series (August) of average TOA reflectances and 6S and DTA surface reflectances
for the PI validation pixels.
44
1262
1263
1264
1265
1266
1267
Figure 4. Relationship between TOA (left) and DTA (middle) and 6S (right) surface reflectances
from the 07/26/2001 Landsat-ETM+ image and corresponding MODIS-derived surface reflectances
45
01/082006
13/03/2007
0.5
0.5
0.3
0.2
0.1
0.1
TOA
DTA
TOA
DTA
0.1
TOA
DTA
6S
0.0
FIELD
TOA
DTA
FIELD
6S
0.4
Reflectance
0.4
0.3
0.2
DTA
6S
0.2
0.0
0.0
TOA
0.3
0.1
0.1
FIELD
6S
Band 4
0.4
0.2
DTA
Band 3
0.5
0.3
TOA
0.6
0.5
0.0
FIELD
0.2
0.5
0.1
0.0
1268
1269
1270
1271
1272
1273
0.3
0.1
Band 4
Reflectance
Reflectance
0.2
0.2
0.6
Band 3
0.4
0.3
0.3
6S
0.6
0.5
0.4
0.0
FIELD
6S
0.4
Reflectance
FIELD
0.5
0.1
0.0
0.0
0.5
Reflectance
Reflectance
Reflectance
0.2
Reflectance
0.4
0.4
Band 2
Band 1
Band 2
Band 1
0.3
0.6
0.6
FIELD
TOA
DTA
6S
FIELD
TOA
DTA
6S
Figure 5: Box plot of field surface reflectances measured at the Mediana site on 01/08/2006 and
13/03/2007, and TOA reflectances and DTA and 6S surface reflectance values. The median, first
and third quartiles are indicated in the shaded boxes, and the bars represent 10th and 90th centiles.
1274
46
March
0.04
A)
B)
C)
Band 3
0.03
0.02
0.01
0.18
0.00
Band 4
0.16
0.14
0.12
0.10
0.08
89 90 93 94 95 97 98 99 00 03 05 07
19 19 19 19 19 19 19 19 20 20 20 20
89 90 93 94 95 97 98 99 00 03 05 07
19 19 19 19 19 19 19 19 20 20 20 20
89 90 93 94 95 97 98 99 00 03 05 07
19 19 19 19 19 19 19 19 20 20 20 20
August
0.07
Band 3
0.06
A)
B)
0.05
C)
Northern slope
Southern slope
0.04
0.03
0.02
0.24
0.01
Band 4
0.22
0.20
0.18
0.16
0.14
19
8
19 4
8
19 5
8
19 7
8
19 9
9
19 1
9
19 2
9
19 3
9
19 5
9
19 7
9
20 9
0
20 0
0
20 1
0
20 2
0
20 4
05
19
8
19 4
8
19 5
8
19 7
8
19 9
9
19 1
9
19 2
9
19 3
9
19 5
9
19 7
9
20 9
0
20 0
0
20 1
0
20 2
0
20 4
05
1275
1276
1277
1278
1279
1280
1281
19
8
19 4
8
19 5
8
19 7
8
19 9
9
19 1
9
19 2
9
19 3
9
19 5
9
19 7
9
20 9
0
20 0
0
20 1
0
20 2
0
20 4
05
0.12
Figure 6. Band 3 and Band 4 6S surface reflectance (A), reflectance of a horizontal surface
(lambertian model) (B), and reflectance of a horizontal surface (C-correction model) (C) for a Pinus
halepensis forest within the semi-arid central Ebro Valley. Reflectance values represent the average
of 10 pixels in a south-facing slope and 10 pixels in a north-facing slope
47
1282
0.8
TOA
0.6
TOA
0.6
NDVI
NBR
0.4
0.2
0.4
0.0
0.2
-0.2
0.8
DTA
0.6
DTA
0.6
NDVI
NBR
0.4
0.2
0.4
0.0
0.2
-0.2
0.8
6S
0.6
NDVI
NBR
0.4
0.2
0.0
6S
0.6
0.4
-0.2
0.2
0.8
PIFs
0.6
PIFs
NDVI
NBR
0.4
0.2
0.0
0.6
0.4
-0.2
0.2
0.8
0.4
NDVI
NBR
0.6
0.2
0.0
0.4
PIFR
PIFR
-0.2
0.2
0.8
TIC
0.6
TIC
0.4
NDVI
NBR
0.6
0.2
0.6
0.4
0.0
-0.2
0.2
19
84 985 987 989 991 992 993 995 997 999 000 001 002 004 005 006
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
19
84 985 987 989 991 992 993 995 997 999 000 001 002 004 005 006
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
Standard deviation (NBR)
0.20
0.18
0.16
0.14
0.12
PIFR
TIC
PIFs
6S
0.10
0.08
0.06
19
84
19
85
19
87
19
89
19
91
19
92
19
93
19
95
19
97
19
99
20
00
20
01
20
02
20
04
20
05
20
06
Standard deviation (NDVI)
0.20
1283
1284
1285
1286
1287
1288
1289
0.18
0.16
0.14
0.12
0.10
0.08
0.06
19
84
19
85
19
87
19
89
19
91
19
92
19
93
19
95
19
97
19
99
20
00
20
01
20
02
20
04
20
05
20
06
Figure 7. Average and standard deviation August NBR and NDVI values obtained from TOA
reflectances and DTA, 6S, PIFs, PIFR, and TIC surface reflectances for the burned areas of Montes
de Zuera in 1995. Time series of standard deviation values in the 6S, PIFs, PIFR and TIC methods
are also shown.
48
1.0
0.8
0.8
0.6
0.6
NDVI
NDVI
1.0
0.4
0.2
0.2
TOA
PIFs
0.0
1.0
1.0
0.8
0.8
0.6
0.6
NDVI
NDVI
0.0
0.4
0.2
0.4
0.2
DTA
0.0
PIFR
0.0
1.0
1.0
0.8
0.8
0.6
0.6
NDVI
NDVI
0.4
0.4
0.2
0.4
0.2
6S
0.0
19
TIC
0.0
84 985 987 989 991 992 993 995 997 999 000 001 002 004 005 006
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
19
84 985 987 989 991 992 993 995 997 999 000 001 002 004 005 006
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
Standard deviation (NDVI)
0.30
0.25
0.20
0.15
PIFR
TIC
PIFs
6S
0.10
0.05
0.00
84
19
1290
1291
1292
1293
1294
1295
1296
1297
85
19
87
19
89
19
91
19
92
19
93
19
95
19
97
19
99
19
00
20
01
20
02
20
04
20
05
20
06
20
Figure 8. Average NDVI and standard deviation obtained from TOA reflectances and DTA, 6S,
PIFs, PIFR, and TIC surface reflectances for those parts of the Monegros II area transformed from
non-irrigated arable lands to irrigated lands. Time series of standard deviation values in the 6S,
PIFs, PIFR and TIC methods are also shown. OJO PIFs
49
1298
0
-0.1
-0.2
-200
-0.2
0.3
300
0.2
100
0.0
0
-0.1
-100
-0.2
0.2
-200
300
6S
0.1
200
100
0.0
0
-0.1
-100
-0.2
-200
19
90
19
93
19
94
19
95
19
97
19
98
19
99
20
00
20
03
20
05
20
07
-200
300
PIFR
NDII
0.1
-100
0.1
200
100
0.0
0
-0.1
-100
-0.2
-200
0.2
400
TIC
0.1
NDII
200
Precipitation
DTA
0.2
Precipitation
-100
100
0.0
Precipitation
0
200
300
Precipitation
0.0
300
PIFs
0.1
NDII
100
-0.1
NDII
200
Precipitation
NDII
0.1
NDII
0.2
300
TOA
Precipitation
0.2
200
0.0
100
0
-0.1
-100
-0.2
-200
19
90
19
93
19
94
19
95
19
97
19
98
19
99
20
00
20
03
20
05
20
07
Standard deviation (NDII)
0.20
0.15
PIFR
TIC
PIFs
6S
0.10
0.05
0.00
1299
1300
1301
1302
1303
1304
1305
1306
Figure 9. Average and standard deviation NDII obtained from TOA reflectances and DTA, 6S,
PIFs, PIFR, and TIC surface reflectances for steppes and non-irrigated arable lands of Monegros
(bold lines) and total wintertime precipitation (November–February) (thin lines). Time series of
standard deviation values in the 6S, PIFs, PIFR and TIC methods are also shown. OJO PIFs
50
Download