Malgré la précision des nouvelles mesures de SSA de la neige, ce dernier paramètre n’a pas été validé pour la modélisation du TR MOP du couvert nival. Suite au développement des instruments IRIS et SWIRcam, un deuxième objectif secondaire de la thèse est de valider et calibrer la SSA de la neige comme paramètre de taille de grains de neige pour la modélisation du signal MOP du cou- vert nival. Des études récentes (Langlois et al., 2010a; Brucker et al., 2011) ont démontré qu’il est possible de minimiser l’erreur de la modélisation du signal MOP provenant de la paramétrisation de la taille des grains de neige en optimisant un facteur multiplicatif à la SSA. Ce travail s’inscrit donc dans le cadre de cette thèse et a été publié dans la revue IEEE Transaction on Geoscience and Remote Sensing(Montpetit et al., 2013) et a aussi fait l’objet d’une présentation à la conférence Eastern Snow Conference 2012à Frost Valley (NY), États-Unis. Ce travail a permis de démontrer qu’il est possible de réduire l’erreur de la modélisation de la TB à 11.6 K en appliquant un facteur multiplicatif de 1.3 sur le paramètre de taille de grains de neige du modèle Microwave Emission Model of Layered Snowpacks(MEMLS) issu de la SSA mesurée à l’aide des instruments IRIS et SWIRcam. L’avantage de ce résultat comparativement aux études antérieures est que les mesures de la SSA avec l’IRIS et la SWIRcam sont plus précis et reproductibles et donc les résultats des simulations avec MEMLS sont eux aussi plus reproductibles et précis. Ces résultats sont présentés dans la section III.A de l’article Montpetit et al. (2013) présentée en section 5.0.2 de cette thèse. L’acquisition, le traitement, l’analyse de données, la production des résultats ainsi que la rédaction et la révision de l’article (Montpetit et al., 2013) ont été réalisé dans le cadre de cette thèse avec l’aide des co-auteurs de l’article.
4 Résultats : Modélisation de l’émission MOP du sol gelé
Une grande incertitude de la modélisation de la TBdu couvert nival est la contribution provenant du sol. Très peu d’études récentes se sont concentrées sur la modélisation du signal MOP du sol gelé dans les régions arctiques. Il a été démontré par Roy et al. (2013) que le sol peut avoir un impact sur le signal MOP même pour des fréquences plus élevées (jusqu’à 37 GHz) pour des couverts de neige moins dense et épais. Wegmüller et Mätzler (1999) ont démontré qu’il est possible de modéliser l’émissivité du sol à ces fréquences mais Roy et al. (2013) ont démontré que ce modèle nécessitait des modifications afin que la modélisation du TR MOP du manteau neigeux soit optimale. Le travail présenté dans l’article Montpetit et al. (2013) de la section 5.0.2 démontre aussi que le
modèle MEMLS. Toutefois, ce travail était incomplet dû au manque de données géophysiques et radiométriques précises du sol. Dans l’optique d’améliorer la modélisation du TR MOP du couvert nival, l’amélioration de la modélisation de l’émissivité MOP du sol gelé s’inscrit donc dans le cadre de cette thèse. Les données utilisées pour cette section du travail ont été acquises par l’équipe neige du CARTEL et des collaborateurs dans la région de la Baie James et Umiujaq (Québec), Canada, durant les hivers 2012-2013 et 2013-2014. Les mesures de la campagne Baie James 2012-2013 ont fait l’objet d’une présentation par affiche à la conférence Eastern Snow Conference 2013 à Huntsville (Ontario), Canada. Les résultats de l’analyse de la modélisation de l’émissivité du signal MOP du sol gelé sont aussi présentés dans un article présentement en révision dans le journal Geoscience and Remote Sensing Letters (Montpetit et al., En Revision). Ce travail démontre qu’il est possible calibrer le modèle de Wegmüller et Mätzler (1999) afin de minimiser l’erreur entre la TB mesurée et simulée à 4.65 K pour des fréquences de 10.7, 19 et 37 GHz en H-Pol et V-Pol. L’acquisition, le traitement et l’analyse des mesures ainsi que la production des résultats, la rédaction et la révision (en cours) de l’article ont été fait dans le cadre de cette thèse avec l’aide des co-auteurs de l’article.
4.0.1 Article : In-situ passive microwave parameterization of sub-arctic frozen organic soils Plusieurs applications MOP telles que l’estimation de l’émissivité de surface ou l’analyse du cou- vert nival, nécessitent l’estimation de l’émission provenant du sol. De grandes incertitudes per- sistent dans la modélisation de l’émissivité du sol lorsqu’il est gelé. Dans cet article, une méthode empirique pour estimer l’émission de sols organiques de l’Arctique canadien à 10.7, 19 et 37 GHz est présentée. Cette méthode a été calibrée et validée à l’aide de mesures radiométriques in-situ à des angles d’incidences de 0o à 60o de sols du nord-est de l’Arctique canadien. Les valeurs de permittivités effectives retrouvées donnent une RMSE entre les TB mesurées et simulées de 4.8 K considérant toutes les fréquences. Cette méthode présente un très grand potentiel dans l’amé- lioration de l’estimation de la contribution du sol gelé à la TB du couvert nival. L’originalité de ce travail provient du fait que très peu d’études ont acquis le type de mesures radiométriques (TB multi-angles et multifréquences) pour évaluer l’émission de sols organiques en milieu Arctique. Les tableaux 4.1 et 4.2 montre les résultats principaux de ce travail.
Tableau 4.1 – Permittivité effective (ǫ′
ef f), rugosité de surface effective(σef f) ainsi que le paramètre de dépolarisation (β) retrouvés pour des sols organiques de l’Arctique canadien. — Effective permittivity (ǫef f), effective soil surface roughness (σef f) and depolarisation coefficient
(β) retrieved for organic soils of the canadian Arctic. f (GHz) ǫ′ ef f σef f β 10.7 3.20 0.19 1.08 19 3.45 0.72 37 4.53 0.42
Tableau 4.2 – Biais et RMSE entre lesTB mesurées et simulées pour six canaux pour des sites de sols organiques de l’Arctique canadien. — Bias and RMSE between modeled and measuredTB
for six channels for organic soil sites of the canadian Arctic.
Canal Biais (K) RMSE (K)
11V -1.3 1.6 11H -0.5 4.0 19V -2.2 4.4 19H -1.5 3.9 37V 0.6 5.8 37H 2.2 6.0 Tous -0.5 4.8 34
In-situ passive microwave parameterization of
1sub-arctic frozen organic soils
2Benoit Montpetit
∗, Alain Royer
∗, Alexandre Roy
∗and Alexandre Langlois
∗ ∗Centre
3
d’Application et de Recherche en T´el´ed´etection, Universit´e de Sherbrooke, Sherbrooke,
4
Qu´ebec, Canada
5
Abstract
6
Many applications such as passive microwave emissivity estimates or snow cover analysis need to take into
7
account the soil contribution to the overall signal emission. Soil emission modeling presents large uncertainties
8
when the soil is frozen. In this paper, an empirical retrieval method is presented, specifically for rough frozen soil
9
permittivity estimates at 10.7, 19 and 37 GHz. The method was tested and validated using in-situ passive microwave
10
measurements at incidence angles from 0 to 60o
of sub-arctic frozen organic soils in Northeastern Canada. The
11
retrieved permittivity values give an overall RMSE between the measured and simulated brightness temperatures
12
of 4.57 K for all frequencies combined. The method shows great potential to improve the estimation of the frozen
13
soil contribution to the measured passive microwave brightness temperature.
14
Index Terms
15
Passive Microwave, frozen soil, rough soil, dielectric soil properties, soil reflectivity modeling.
16
I. INTRODUCTION
17
F
OR many years, scientists have studied bare soil reflectivity modeling ( [1]; [2]; [3]; [4]; [5]), partly18
motivated by the ability to retrieve soil moisture from satellite data; such as the Soil Moisture and
19
Ocean Salinity (SMOS) mission [6]. The future Soil Moisture Active and Passive (SMAP) mission [7]
20
will also study soil moisture from satellite-based data, but will also study the soil state (frozen/thawed) in
21
northern regions [8]. The majority of these studies mainly focus on L-band rather than higher frequencies
22
( [9]; [10]; [5]; [11]; [12]), given the ability of such wavelengths to penetrate vegetation and snow. For
23
cryosphere studies frequencies up to 37 GHz are commonly used [13]. Some studies have shown that, even
24
at these high frequencies, the soil contribution to the emitted signal at the surface of the snowpack has to
25
be estimated ( [14]; [15]). A major issue with the estimation of the soil contribution to the emitted signal
26
at the surface is the estimation of frozen soil permittivity where the real part drastically changes when the
27
soil freezes and the very low imaginary part controls the penetration depth [16]. Different models exist
28
but still need soil in-situ measurements (soil bulk density, volumetric moisture, temperature, etc.) that are
29
complex to extract in remote arctic soils.
30
In this paper, we present a simple method to retrieve passive microwave soil properties from bright-
31
ness temperature measurements using the semi-empirical model developed by Wegm¨uller & M¨atzler [3]
32
(hereafter referred to as WM99) and validate these properties using an independent dataset taken at the
33
calibration site and another site in northern Qu´ebec (see Section II). We first describe the study sites and
34
the geophysical and radiometric measurements. Then, the models and the optimization method will be
35
detailed. Finally, the optimization and validation results will be presented and discussed.
36
II. STUDY SITES AND MEASUREMENTS
37
The soil and radiometric data for this study were first collected in the James Bay area, Qu´ebec (53o26’N,
38
76o46’W, 186 m a.s.l.) during three intensive measurement periods (IMP) in January (8th to 12th, IMP1),
39
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. XX, NO. X, JULY 2014 2 0 10 20 30 40 50 60 235 240 245 250 255 260 θ (o) Tb (K) 10.7V Meas. 10.7V Sim. 10.7H Meas. 10.7H Sim. 19V Meas. 19V Sim. 19H Meas. 19H Sim. 37V Meas. 37V Sim. 37H Meas. 37H Sim.
Fig. 1. Typical angular T b measurements (points) acquired at the BJCal site and the resulting simulations (lines) using the WM99 model with the optimized parameters of Table I at 10.7 (black circles and lines), 19 (light gray squares and lines) and 37 (dark gray triangles and lines) GHz at V-Pol (filled shapes and full lines) and H-Pol (empty shapes and dotted lines). Tsoilef f= −5 ± 2.5oC.
February (12th to 17th, IMP2) and March (19thto 23rd, IMP3) of 2013. More data was also acquired near
40
Umiujaq, Qu´ebec (56o33’N, 76o30’W, 74 m a.s.l.) during one intensive period in January (21st to 28th of 41
2014). The bare soil measurements were done in clearings with minimal influence from the environment
42
(topography, vegetation, etc.) to the measured microwave brightness temperature (T b). A total of 8 sites
43
were selected in this study where the soil composition mainly consisted of organic matter.
44
The soil T b measurements were acquired using surface-based radiometers on a mobile sled at 10.7,
45
19 and 37 GHz in both horizontal (H-Pol) and vertical (V-Pol) polarizations. Calibrations were done on
46
a regular basis throughout the winter season using a warm and cold target as described by Asmus and
47
Grant [17]. The downwelling T bskywere estimated with an atmospheric absorption microwave model [18]
48
implemented in the Helsinki University of Technology (HUT, [19]) model, using the 29 atmospheric layers
49
above surface of the North American Regional Reanalysis (NARR) data. The measured Tb at frequency
50
f and polarization p can then be described by:
51 T bsoil(f, p) = (1 − Γf,p)T ef f soil + Γf,pT b ef f sky(f, p) (1)
whereΓf,p is the rough soil reflectivity at polarization p and T ef f
soil is the effective soil physical temperature. 52
Temperature profiles were taken using a Traceable 2000 digital temperature probe with an accuracy of
53
0.1 oC for depths of 0 to 10 cm with an interval of 2 cm. Since the soil is considered a homogeneous
54
medium for this study, the effective soil temperature was estimated to be the averaged temperature over
55
the first 5 cm.
56
Among the 8 sites analyzed, a James Bay site, measured on February 13th 2013, was considered for
57
model calibration purposes (hereafter referred to as the BJcal site) since it was the only site where a wide
58
range of incidence angle (0o to 60o) was measured with the surface-based radiometer. The site consisted
59
of a bare soil area where the snow was removed (20 m long by 5 m wide) to eliminate any possible
60
contribution coming from the snow walls around the bare soil surface. Measurements over the 20 m long
61
transects were taken in order to establish the variability of the soil T b measurements. The averaged soil
62
temperature over the first 5 cm was -5 ± 2.5 oC. The same site was revisited during the winter IMPs and
other sites were measured during the 2013 winter campaign for validation purposes (hereafter referred to
64
as BJval sites). The BJval sites were also clearings and soil temperatures varied from -13oC to -5oC. Three
65
other validation sites were measured during the 2014 winter campaign in Umiujaq (hereafter referred to
66
as UMIval sites). The UMIval sites were clearings and soil temperatures varied from -17o C to -10oC.
67
III. METHOD
68
The WM99 model describes the rough soil reflectivity at a frequency f and polarization p (Γf,p) by
69
its smooth Fresnel reflectivity in H-Pol (ΓF resnel
f,p ) which depends on the incidence angle (θ) and the
70
real part of the soil permittivity (ǫ′), weighted by an attenuation factor that depends on the standard
71
deviation in height of the surface (soil roughness, σ), the measured wavenumber (k) and a polarization
72
ratio dependency factor. As shown in [14], a modification to this model was applied for this study using
73
a β factor to take into account the frequency dependency of the polarization reflectivity ratio. The WM99
74
model for incidence angle lower than 60o is therefore described by:
75 Γf,H(θ, ǫ′f, σ) = Γ F resnel f,H (θ, ǫ′f) exp(−(kσ) √ −0.1 cos θ) (2) Γf,V(θ, ǫ′f, σ) = Γf,H(θ, ǫ′f, σ) cos βfθ (3)
The first part of the optimization process consisted in obtaining the soil permittivity (ǫ′
f) for each 76
frequency and surface roughness ( σ) that minimized the error between simulated and measured T b in
77
H-pol using Eq. 2 for all the incidence angles between 0o and 60o. The optimization algorithm used is
78
the least squared error minimization algorithm of Levenberg-Marquardt ( [20], hereafter referred to as
79
LM63). Afterwards, using the soil properties of the first step, the βf factor was determined with Eq. 3
80
using the LM63 algorithm to minimize the error between the V-pol T b measurements and the simulations.
81
Once the optimized parameters were obtained, these retrieved parameters were inserted in the WM99 and
82
the soil T b were modeled and compared to the measured T b of the BJval sites. Finally, these parameters
83
were validated against T b measurements acquired at another northern sites (UMIval sites).
84
IV. RESULTS
85
Table I shows the optimization results using the LM63 algorithm at H-Pol and V-Pol. The final RMSE
86
obtained for each optimization is given. Since the soil roughness is considered as a geophysical parameter
87
specific to the soil, it can be considered frequency independent ( [3]; [21]) and a different soil permittivity
88
per frequency was obtained with the LM63 optimization on H-Pol measurements. It should be noted that
89
since no soil surface roughness measurements were acquired, the surface roughness value (σ) obtained in
90
Table I is an effective parameter retrieved through model inversion and should not be considered as a true
91
geophysical parameter. An optimization per frequency was done for the βf factor in V-Pol to determine
92
if the βf factor was frequency dependent. Given the different values of βf obtained, it can be concluded
93
that the βf factor is frequency dependent as mentioned by [14].
94
TABLE I
TABLE1: FROZEN SOIL PERMITTIVITY AND ROUGHNESS(EQ. 2)AND THEβfFACTOR(EQ. 3)RETRIEVED THROUGH THELM63 OPTIMIZATION PROCESS USING THEH-POLT bANDV-POLT bRESPECTIVELY,AT10.7, 19AND37 GHZ IN THEWM99. THE
RESULTINGRMSEOF THE OPTIMIZATION PROCESS IS ALSO GIVEN.
Frequency Soil Soil RMSE RMSE
(GHz) Permittivity Rougness T b(H-Pol) βf T b(V-Pol)
(cm) (K) (K)
10.7 3.20 1.08 0.93
19 3.45 0.19 1.23 0.72 0.75
37 4.53 0.42 0.43
Figure 1 presents the typical angular T b measurements acquired at the James Bay site (BJcal). The
95
results of the modeling using the WM99 model and the parameters of Table I are also given in Figure
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. XX, NO. X, JULY 2014 4 220 225 230 235 240 245 250 255 260 265 220 225 230 235 240 245 250 255 260 265 Tb Measured (K) Tb Simulated (K) 10.7 GHz V−Pol 10.7 GHz H−Pol 19 GHz V−Pol 19 GHz H−Pol 37 GHz V−Pol 37 GHz H−Pol RMSE(10.7V)= 1.58 K RMSE(10.7H)= 4.00 K RMSE(19V)= 4.42 K RMSE(19H)= 3.92 K RMSE(37V)= 5.78 K RMSE(37H)= 5.95 K RMSE(All)= 4.65 K
Fig. 2. Validation results between modeled and measured T b for a frozen bare soil using the WM99 model and the parameters of Table I at 10.7, 19 and 37 GHz at both H-Pol and V-Pol for the James Bay sites (BJcal and BJval) (black) and the Umiujaq sites (UMIval) (gray). The RMSE of the different channels are given. The solid black line corresponds to the 1:1 line.
1. Figure 1 shows the decrease in T b with increasing incidence angle. This means that the reflectivity
97
increases with incidence angle (Eq. 1). This result is similar to what was previously obtained by [3]. It
98
also shows that the decrease is stronger in H-Pol compared to V-Pol. This can be explained by a stronger
99
increase of the surface reflectivity with incidence angle in H-Pol compared to V-Pol.
100
Figure 2 shows the results of comparison between the simulations using the WM99 model optimized
101
with the parameters of Table I against the measured T b of the James Bay sites (BJcal and BJval sites)
102
acquired throughout the 2013 winter, and the Umiujaq sites (UMIval). The measured and simulated T b
103
are, in general, in agreement with a RMSE ranging from 1.6 to 6.0 K (overall RMSE = 4.65 K). There is
104
some scatter especially at 19 and 37 GHz (overall R2
=0.64) which could be explained by the roughness
105
variability (see discussion below) and the uncertainty in the downwelling T b where the surrounding
106
environment can be slightly more important than the estimated T bsky.
107
The RMSE obtained at James Bay (4.58 K) is similar to the one of the Umiujaq sites (4.49 K) which
108
suggests that the optimization process could be applied to other sites characterized by similar northern
109
organic frozen soil composition. Table II summarizes the errors obtained between the simulations and the
110
measurements for both the James Bay and Umiujaq campaigns and two campaigns combined. The mean
111
bias appears relatively small at -0.45 K. The overall RMSE is comparable to the measured T b variability
112
over a 20 m transect which was of 5.18 K, 4.41 K and 6.79 K at 37, 19 and 10.7 GHz respectively. This
113
variability can be explained by the surface roughness variability. As a matter of fact, over this transect,
114
keeping the same permittivity as obtained above (Table I) and fitting an optimized roughness value for
115
each site, over a range of 0 to 2 cm, reduces the RMSE from 6.07 K to 0.37 K. The range of surface
116
roughness obtained in this optimization process was 0.01 ≤ σ ≤ 2 cm, 0.36 ≤ σ ≤ 2 cm and 0.01 ≤ σ ≤
117
0.23 cm at 10.7, 19 and 37 GHz respectively.
118
It is difficult to compare the real part of the permittivity values (ǫ′
f) obtained in this study to previous 119
studies because of all the factors and conditions (soil types, soil ice fraction, soil temperatures, frequencies,
120
etc.) that influence these values. This information is not always available. As reference, from 1 to 100 GHz
121
and at a temperature of 0 oC, ǫ′
f of pure ice is constant at ≈ 3.2 (with a weak temperature dependence),
122
where ǫ′
f of free water decreases following a power function from 90 to 6 [16] (according to [22], we
123
get ǫ′
f of pure water at 0 oC of 85.9, 38.3, 20.5 and 10.2 at 1.4, 11, 19 and 37 GHz respectively).
TABLE II
MEAN BIASES ANDRMSEBETWEEN SIMULATED AND MEASUREDT bAT10.7, 19AND37 GHZ AT BOTHH-POL ANDV-POL FOR THE JAMESBAY ANDUMIUJAQ MEASUREMENT CAMPAIGNS. THE OVERALL MEAN BIAS ANDRMSEARE ALSO GIVEN.
Channel James Bay Umiujaq All
Bias (K) RMSE (K) Bias (K) RMSE (K) Bias (K) RMSE (K) 10.7V -1.36 3.15 -0.73 3.40 -1.26 1.58 10.7H -0.10 3.77 -2.83 5.03 -0.53 4.00 19V -2.19 4.31 -2.24 4.97 -2.20 4.42 19H -1.34 3.57 -2.50 5.44 -1.52 3.92 37V 1.36 6.01 -3.26 4.38 0.63 5.78 37H 2.97 6.27 -2.02 3.83 2.18 5.95 10.7 -0.73 3.47 -1.78 4.29 -0.89 3.62 19 -1.76 3.95 -2.37 5.21 -1.86 4.18 37 2.17 6.14 -2.64 4.12 1.41 5.87 All -0.11 4.67 -2.26 4.56 -0.45 4.65
Rautiainen et al. [23] showed that at L-band the soil permittivity varied from 3.3 to 3.8 in frozen soils
125
having temperature ranging between -6 oC to -3.7 oC but measured in-situ permittivity values of 6.3 to
126
7.1 for a frozen wetland site in Northern Finland [24]. Schwank et al. [25] showed that for L-band, the
127
frozen soil permittivity varies from 3.5 to 4.5, while Hallikainen et al. [26] showed that it varies from 5
128
up to 8 in the 10 to 18 GHz frequency range. Puilliainen et al. [27] used a fixed value of 6 to simulate
129
the frozen soil T b at 19 and 37 GHz. Finally, Mironov et al. [28] developed a temperature dependent
130
permittivity model and showed that the permittivity could vary from 3 to 4.5 in the 10 to 16 GHz range
131
for a frozen soil at -25oC. Here, we obtained values of 3.20 at 10.7 GHz, 3.45 at 19 GHz and 4.53 at 37
132
GHz for soils between -13oC to -5oC (the first 5 cm). This is in agreement with these different studies
133
where for lower frequencies, we have a lower soil permittivity. The retrieved values of this study might
134
have been different if another rough bare soil reflectivity model was used (e.g. [2], [21]) but such models
135
had too many parameters to inverse in the optimization process to be an efficient method of retrieving
136
effective soil parameters.
137
As mentioned above, one of the main difficulties in the retrieval process is the combined effect of
138
permittivity and the surface roughness. Figure 3 shows the sensitivity of the mean measured-simulated
139
T b (V & H) RMSE on the soil permittivity and the surface roughness. The simulations conditions are
140
those of the BJcal site.
141 ε’ (10.7 GHz) 5 4.5 4 3.5 3 2.5 2 ε’ (19 GHz) 5 4.5 4 3.5 3 2.5 2 2 4 6 8 10 12 14 16 18 20 ε’ (37 GHz) σ (cm) 0 0.25 0.5 0.75 1 5 4.5 4 3.5 3 2.5 2 RMSE (K)
Fig. 3. Sensitivity test of the T b RMSE (in K, color scale) on the soil permittivity (ǫ′) and the surface roughness (σ) on the BJcal site measurements (Figure 1), and for 3 frequencies.
Figure 3 shows that the sensitivity to the surface roughness is weaker than for the permittivity. Note
142
that the pattern for 1 ≤ σ ≤ 2 cm (not shown) can be extrapolated from the one around σ=1 cm, and
143
that for σ=0 cm, a discontinuity results from the abrupt change from the Fresnel reflectivity to the rough
144
surface reflectivity (Eq. 2). For σ < 1 cm, it appears that it exist an ensemble of (ǫ′, σ) couples that 145
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. XX, NO. X, JULY 2014 6
provides an ensemble of solutions with low RMSE (dark blue in Fig. 3). However, when combining the
146
three frequencies showing different behaviour for the lowest RMSE values, the range where the minimum
147
RMSE combining all three frequencies occurs for a limited range of σ, 0.1 < σ < 0.23 cm. The optimized