Response of sub-daily L-band backscatter to internal and surface canopy water dynamics
Vermunt, P.C.; Khabbazan, S.; Steele-Dunne, S.C.; Judge, Jasmeet; Monsivais-Huertero, Alejandro; Guerriero, Leila ; Liu, Pang Wei
DOI
10.1109/TGRS.2020.3035881
Publication date 2020
Document Version
Accepted author manuscript Published in
IEEE Transactions on Geoscience and Remote Sensing
Citation (APA)
Vermunt, P. C., Khabbazan, S., Steele-Dunne, S. C., Judge, J., Monsivais-Huertero, A., Guerriero, L., & Liu, P. W. (2020). Response of sub-daily L-band backscatter to internal and surface canopy water dynamics. IEEE Transactions on Geoscience and Remote Sensing, 1-16. https://doi.org/10.1109/TGRS.2020.3035881 Important note
To cite this publication, please use the final published version (if applicable). Please check the document version above.
Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons. Takedown policy
Please contact us and provide details if you believe this document breaches copyrights. We will remove access to the work immediately and investigate your claim.
This work is downloaded from Delft University of Technology.
Response of sub-daily L-band backscatter to
internal and surface canopy water dynamics
Paul C. Vermunt, Saeed Khabbazan, Susan C. Steele-Dunne, Member, IEEE,
Jasmeet Judge, Senior Member, IEEE, Alejandro Monsivais-Huertero, Senior Member, IEEE,
Leila Guerriero, Member, IEEE, and Pang-Wei Liu, Member, IEEE
Abstract—The latest developments in radar mission concepts
1
suggest that sub-daily synthetic aperture radar will become
2
available in the next decades. The goal of this study was to
3
demonstrate the potential value of sub-daily spaceborne radar
4
for monitoring vegetation water dynamics, which is essential
5
to understand the role of vegetation in the climate system. In
6
particular, we aimed to quantify fluctuations of internal and
7
surface canopy water and understand their effect on sub-daily
8
patterns of L-band backscatter. An intensive field campaign was
9
conducted in north-central Florida, USA, in 2018. A
truck-10
mounted polarimetric L-band scatterometer was used to scan a
11
sweet corn field multiple times per day, from sowing to harvest.
12
Surface canopy water (dew, interception), soil moisture, and plant
13
and soil hydraulics were monitored every 15 minutes. In addition,
14
regular destructive sampling was conducted to measure seasonal
15
and diurnal variations of internal vegetation water content. The
16
results showed that backscatter was sensitive to both transient
17
rainfall interception events, and slower daily cycles of internal
18
canopy water and dew. On late season days without rainfall,
19
maximum diurnal backscatter variations of >2 dB due to internal
20
and surface canopy water were observed in all polarizations.
21
These results demonstrate a potentially valuable application for
22
the next generation of spaceborne radar missions.
23
Index Terms—Sub-daily radar, vegetation, water content,
in-24
terception, dew, corn, L-band, ground-based, scatterometer, sap
25
flow, diurnal, backscatter
26
I. INTRODUCTION
27
G
LOBAL, daily to sub-daily monitoring of vegetation28
water dynamics is essential to address fundamental
29
questions surrounding the role of vegetation in the climate
30
system, and to provide information for a range of
appli-31
cations from agriculture and water management to weather
32
prediction [1]. Vegetation temporally stores water inside its
33
tissue and on its surface, and this water is transferred back
34
to the climate system through transpiration and evaporation.
35
Global evapotranspiration (ET) amounts from reanalysis data,
36
land surface model and diagnostic products disagree by up
37
P.C. Vermunt, S. Khabbazan and S.C.Steele-Dunne are with the Faculty of Civil Engineering and Geosciences, Delft University of Technology, 2628CN Delft, The Netherlands (e-mail: p.c.vermunt@tudelft.nl).
J.Judge is with the Department of Agricultural and Biological Engineering, University of Florida, Gainesville, FL 32611 USA.
A.Monsivais-Huertero is with the Escuela Superior de Ingenieria Mecanica y Electrica Unidad Ticoman, Instituto Politecnico Nacional, Mexico City 07340, Mexico.
L.Guerriero is with the Department of Civil Engineering and Computer Science, Tor Vergata University, 00133 Rome, Italy.
P.-W. Liu is with the Hydrological Science Laboratory in NASA’s Goddard Space Flight Center, Greenbelt, MD 20771 USA, and Science Systems and Applications, Inc., Lanham, MD 20706, USA.
to 50% [2], [3], and trends are uncertain [4]. Uncertainty 38
in ET partitioning is even more severe than uncertainty in 39
estimating ET itself [5] [6]. Lack of understanding of rainfall 40
interception by vegetation and its loss through evaporation is a 41
key limitation of current methods to estimate and partition ET, 42
and is essential for land surface modeling and understanding 43
the role of vegetation in land-atmosphere interactions [7], [8]. 44
Robust modeling of interception is hindered by holes in our 45
basic process understanding [6], and a lack of information 46
about surface canopy water (SCW), i.e. water storage on 47
vegetation surfaces as a result of dew formation or intercepted 48
precipitation [9]. Furthermore, there is a fundamental need for 49
leaf wetness monitoring to understand how projected changes 50
in climate will influence the timing, frequency, duration and 51
intensity of leaf wetting events and their effects on plant 52
function in terms of water relations, gas exchange, energy 53
balance, pathogens and pests, and reproduction [10]. 54
A new perspective on ET and leaf wetness may be pro- 55
vided by sub-daily synthetic aperture radar (SAR), which 56
gives direct insight into the mass balance of the vegetation. 57
Active microwave remote sensing has been found to be 58
sensitive to canopy water storage, depending on frequency, 59
polarization, incidence angle, and vegetation cover [11]–[17]. 60
This sensitivity has mainly been studied to account for the 61
confounding effect of vegetation on soil moisture retrieval e.g. 62
[18]–[20]. However, radar is also a valuable tool for vegetation 63
monitoring, and well-suited to many applications including 64
forest biomass and height [21], change detection [22], ecology 65
and plant physiology [23], and agricultural crop classification 66
and monitoring [24]. The launch of ESA’s Sentinel-1 mission 67
[25] in 2014 has accelerated the development of new applica- 68
tions for SAR in vegetation monitoring. By providing freely- 69
available data at an unprecedented high temporal resolution, 70
Sentinel-1 has stimulated rapid development of products for 71
monitoring natural and agricultural landscapes. However, the 72
repeat time of 6-12 days still limits the current state of the art 73
to products related to biomass, leaf area index (LAI), pheno- 74
logical stage, change (e.g. harvest, tillage) and anomaly de- 75
tection [26]–[31]. The Radarsat Constellation Mission (RCM) 76
was launched in 2019, and consists of three identical SAR 77
satellites. The resulting denser temporal sampling of RCM 78
(4-day revisit time) compared to Radarsat-2 (12-day revisit 79
time), is also expected to improve the potential of real-time 80
agricultural monitoring with the Radarsat series [32]. 81
The motivation for the current study is that the latest 82
daily SAR data will become available in the next decades.
84
CapellaSpace [33] and Iceye [34] are currently populating
85
constellations of X-band (∼10 GHz) SAR systems in Low
86
Earth Orbit (LEO), capable of delivering hourly data with
87
resolutions of 10 m or less (e.g. [35]). An alternative is to
88
place a SAR instrument in a geostationary orbit. A mission of
89
particular interest is HydroTerra, one of the candidates bidding
90
to become the European Space Agency’s 10th Pathfinder
91
mission. HydroTerra, is a C-band (∼5 GHz) geostationary
92
(GEO) SAR mission which will deliver data at various spatial
93
and temporal resolutions to meet the science needs of users
94
[36]. L-band (∼1 GHz) geostationary SAR missions are also
95
under development [37]. With Medium Earth Orbit (MEO)
96
SAR, at altitudes between those of LEO and GEO, trade-offs
97
in system and orbit parameters allow a range of possibilities
98
in terms of spatial and temporal resolution [38]. The goal of
99
this study is to demonstrate the potential value of sub-daily
100
SAR to monitor internal and surface canopy water, and exploit
101
this emerging technology as a tool to address fundamental
102
challenges in our understanding of the role of vegetation in
103
the climate system.
104
Several studies have demonstrated that spaceborne radar is
105
capable of detecting plant water variations during the day.
106
Konings et al. [39], Paget et al. [40], and van Emmerik et
107
al. [41] used aggregated data from the non-sun-synchronous
108
RapidScat scatterometer (2014-2016) to demonstrate that the
109
diurnal cycles in Ku-band radar backscatter were discernible
110
over vegetated areas. Konings et al. [39] and Emmerik et
111
al. [41] related variations in the daily cycle of Ku-band
112
backscatter to vegetation water stress in the humid tropical
113
forests of Central Africa and the Amazon respectively. Other
114
studies used aggregated data from scatterometers in
sun-115
synchronous orbits to study diurnal variations in canopy water
116
[42]–[45]. Frolking et al. [42] found Ku-band backscatter over
117
the Amazonia to be up to 1.0 dB lower at 6:00 compared
118
to 18:00. This difference decreased when a major regional
119
drought continued. Schroeder et al. [43] and Friesen et al.
120
[44] observed diurnal differences in C-band backscatter in
121
the United States (∼10:00 and 22:30) and on a global scale
122
(∼9:30 and 21:30), respectively. In a study on North-American
123
grasslands, Steele-Dunne et al. [45] found mean differences
124
between evening (21:30) and morning (9:30) observations of
125
C-band backscatter between -0.6 and 0.6 dB, depending on
126
region and season. The limitation of radar data from current
127
satellites is that they are in sun-synchronous orbits, and data
128
only available at fixed times (e.g. ASCAT at 10 am/10 pm).
129
These times may not coincide with the ideal times to observe
130
plant water variations, do not provide enough detail to capture
131
and study daily cycles, and their data generally needs to
132
be aggregated to draw meaningful conclusions. The future
133
availability of spaceborne sub-daily SAR data therefore offers
134
a unique possibility to study vegetation water dynamics at an
135
unprecedented temporal resolution.
136
Ground-based radar experiments on sub-daily variations in
137
backscatter have a longer heritage, and have shown that radar
138
backscatter is sensitive to vegetation water changes during the
139
day. Using a truck-mounted dual-pol radar spectrometer
scan-140
ning densely vegetated sorghum fields, Ulaby and Batlivala
141
[46] combined 13 data acquisitions within ten days, where 142
each acquisition was conducted at a different time of day. 143
Aggregating these data, they found clear diurnal variations 144
which they attributed to vegetation. Brisco et al. [16] used a 145
truck-mounted Ku-, C-, and L-band quad-pol scatterometer to 146
study sub-daily fluctuations in backscatter measurements of a 147
wheat canopy. They measured three full days, spread over two 148
years. The results showed that, for C- and L-band, the diurnal 149
backscatter variation correlated well with measured vegetation 150
water content in the vegetative stage of the crop, while the 151
correlation with soil moisture was higher when the plants were 152
senescing. In addition, they showed that bulk vegetation water 153
content (including surface and internal water content) and 154
HH-polarized C-band backscatter with 20◦ incidence angle in 155
the vegetative stage of wheat following a diurnal cycle with 156
maxima just after sunrise and minima between solar noon and 157
sundown. Using results from the same experiment, Gillespie 158
et al. [17] evaluated the effect of dew. The presence of dew 159
was assessed by visual inspection, and backscatter patterns 160
between two nights with and without dew were compared. 161
They concluded that dew has an effect on C-band signals in 162
particular, and that dew and internal canopy water effects can 163
be differentiated according to timing and strength of response. 164
Other ground-based experiments focused on trees [47], [48]. 165
Bouten et al. [47] measured the vertical attenuation profile 166
of a Douglas fir stand before and after rain events. The X- 167
band microwave generator and receiver were mounted on two 168
towers, 12.5 m apart, and the beam of the transmitter was 169
pointed in the direction of the receiver. They found a clear 170
increase in attenuation after canopy wetting. Moreover, they 171
estimated canopy surface water storage from precipitation and 172
throughfall measurements, and found high correlations with 173
increments of the vertically integrated attenuation profiles. 174
De Jong et al. [48] analyzed the relation between vertically 175
polarized X-band backscatter and rainfall interception for a 176
single ash tree. Backscatter observations during 14 rain storms 177
were averaged, and compared to dry situations. The results 178
showed a logarithmic increase of backscatter with cumulative 179
precipitation, supported by physical model simulations. In 180
addition, several studies have observed a diurnal cycle in trunk 181
dielectric constant, which has been related to tree water status 182
and sap flow [49]–[54]. 183
These ground-based experiments successfully demonstrated 184
that radar backscatter is sensitive to variations in total vegeta- 185
tion water content. However, the limited datasets leave many 186
open questions in terms of the sensitivity of radar backscatter 187
to surface versus internal water content, the influence of 188
phenological stage and providing a quantitative link to water 189
transport processes in the vegetation. 190
The aim of this study was to quantify fluctuations of internal 191
and surface canopy water and their effect on sub-daily patterns 192
of L-band backscatter with a view to demonstrating the 193
potential value of sub-daily spaceborne radar backscatter for 194
monitoring vegetation water dynamics. An intensive field cam- 195
paign was conducted over an entire growing season of corn, 196
combining temporally dense radar backscatter observations 197
with continuous observations of leaf surface wetness, surface 198
variables, and frequent destructive vegetation sampling. These
200
data were analyzed during the early-, mid- and late-season
201
to study how backscatter in each polarization is affected by
202
variations of internal and surface water content, and how this
203
sensitivity varies as the crop develops. Data from the fully
204
grown canopy were also combined to obtain an average daily
205
cycle, and insight into the influence of surface canopy water
206
on the amplitude and timing of the daily cycle of backscatter.
207
II. MATERIALS ANDMETHODS
208
A. Study site
209
The experiments were conducted in the spring of 2018 at the
210
Plant Science Research and Education Unit (PSREU) of the
211
University of Florida and the Institute of Food and Agricultural
212
Sciences (UF—IFAS) at Citra, Florida (29.410N, 82.179W).
213
Sweet corn (Zea mays L. var. rugosa) was sown in rows on
214
April 13 and harvested on June 18. The average plant density
215
was 7.9 plants m-2. The soil at the field site consists of >90%
216
sand [55], [56], which allows for high infiltration rates. Early
217
in the growing season, the corn field was irrigated several times
218
with a center-pivot irrigation system. Irrigation was applied in
219
the evening to minimize evaporative losses.
220
B. Radar backscatter
221
Observations of radar backscatter (σ0) were made with the
222
University of Florida L-band Automated Radar System
(UF-223
LARS). A full description of the UF-LARS can be found in
224
[57]. The UF-LARS operates at a center frequency of 1.25
225
GHz, and is designed to collect four polarization combinations
226
(HH, VV, HV and VH) simultaneously. The system was
227
mounted on a Genie manlift, and scanned the corn field with
228
an antenna height of 14 m, and an incidence angle of 40◦.
229
The backscatter coefficients are computed using the Single
230
Target Calibration Technique (STCT) [58], [59]. To suppress
231
fading, 27 independent samples are averaged [60]; nine
sam-232
ples were taken at 30 MHz increments from 1130-1370 MHz,
233
for each of the three azimuthal scans at -9◦, 0◦, and +9◦.
234
The total error was estimated by [60] to be 1.71 dB, and
235
includes a systematic error of 1.49 dB and a random error
236
of 0.85 dB (fading). The systematic error was estimated, via
237
error propagation, by combining the measurement errors of
238
calibration target geometry (1.4 dB), ranges between antenna
239
to terrain (0.17 dB) and calibration target (0.35 dB), and
240
incidence angle (0.32 dB) [57].
241
Ground range and azimuth resolutions were calculated
242
based on the 3 dB beamwidth of 14.7◦ in the E-plane and
243
19.7◦ in the H-plane [57], and are provided in Table I. The
244
resulting single-scan footprints in HH, VV, and cross-pol were
245
40.0, 39.7, and 29.1 m2, respectively. Sampling areas (section
246
II-D) and in-situ sensors (section II-C) were located outside
247
the arc swept by the radar, to avoid disturbing the scene.
248
For most of the season, 32 averaged σ0 observations were
249
obtained per day. For the last eight days of the season, the
250
number of acquisitions was reduced to 16 to avoid radio
251
frequency interference with other microwave sensors in the
252
field. HV and VH polarizations were averaged, and further
253
shown as average cross-pol. All in-situ sensors and vegetation
254
sampling areas were located outside the footprint of the 255
instrument. 256
TABLE I
GROUND RANGE AND AZIMUTH RESOLUTIONS
range resolution (m) azimuth resolution (m) HH-pol 8.5 4.7
VV-pol 6.2 6.4 cross-pol 6.2 4.7
C. Hydrometeorology 257
Meteorological data were obtained from the Florida 258
Automated Weather Network (FAWN)1. The 18-meter tall 259
FAWN weather station was located <600 m east from the 260
experimental site. Observations of rainfall, air temperature on 261
2 m height, solar radiation, relative humidity and wind speed 262
are available every 15 minutes. Reference evapotranspiration 263
(ETo) was calculated from these data using an hourly version 264
of the Penman-Monteith approach [61]. 265 266
Sap flow is the flux of water through the plant, as water 267
extracted by the roots is transported to the leaves to replenish 268
water lost through transpiration. In large trees, the time lag 269
between transpiration and sap flow measured at the base of 270
a stem can be on the order of several hours, while the time 271
lag between transpiration and sap flow at the crown is much 272
smaller [62]–[64]. For corn, we observed that the time lag 273
between calculated reference evapotranspiration and sap flow 274
on the stem was on the order of minutes. Therefore sap 275
flow is a useful indicator of the timing and strength of the 276
daily transpiration cycle which drives internal canopy water 277
dynamics during the day. 278
Sap flow rates of four representative plants were measured 279
with SGEX-19 Dynagage sap flow sensors (Dynamax Inc., 280
Houston, TX, USA), which were installed close to other 281
sensors, just outside the radar footprint. The measurements are 282
based on the stem heat balance method [65]–[67]. Part of the 283
lower stem tissue was continuously heated by external heater 284
strips, and heat convection carried by the sap was measured. 285
The sensors were enclosed by insulating and water-resisting 286
materials, based on the method described in [68]. The built- 287
in sap flow calculator of the Dynagage Flow32-1K system 288
was used to estimate the sap flow rate [g h−1] for each plant 289
every 15 minutes. Sap flow [mm 15min−1] was calculated by 290
averaging over the four sensors, converting the average weight 291
of water to volume using the density of liquid water at 25◦C 292
(0.997 g cm−3), and multiplying the results for an ‘average’ 293
plant with the plant density. Because installation of a SGEX- 294
19 sensor requires a stem diameter of at least 15 mm, the 295
sensors were first installed on May 18. 296
Data gaps, e.g. due to battery failure or poor contact, 297
were filled using a linear relationship between sap flow and 298
the transpiration component of FAO crop evapotranspiration 299
(KcbET o) [69]. Here, ET o is the reference
evapotranspi-300
ration, i.e. the evapotranspiration from a hypothetical,
well-301
watered grass reference surface, calculated using
meteorolog-302
ical data from FAWN and the FAO Penman-Monteith method.
303
Kcb is the basal crop coefficient for transpiration of a sweet
304
corn canopy at potential rate. The multiplication with Kcb
305
converts hypothetical evapotranspiration of a grass surface into
306
transpiration of a sweet corn canopy, assuming no limitation
307
of water [69]. Linear regression between sap flow (F) and
308
KcbET o was described by
309
F = 0.7222 × KcbET o − 0.001 (1)
with R2 = 0.871, based on n = 2389 observations.
310
Leaf wetness due to dew and interception was monitored
311
using three PYTHOS31 dielectric leaf wetness sensors. These
312
sensors are designed to approximate the thermodynamic
prop-313
erties of leaves, and output a voltage signal proportional to the
314
dielectric of a 1 cm zone above the upper side of the sensor,
315
which is proportional to the amount of water on the sensor
316
[70]. This mV output is then converted automatically to a scale
317
called ’counts’, to ensure that sensor outputs are universal
318
regardless of the excitation voltage of the used data logger. For
319
the EM50 data logger used here, counts = voltage(mV)/0.733
320
[70]. The sensors were attached to a wooden pole in between
321
two rows in the early season. They were reattached to the corn
322
plants once the stems were strong enough. The sensors were
323
installed at different heights to capture the vertical distribution
324
of water droplets in the canopy. The empirical model of Cobos
325
et al. [71] was used to estimate the mass of water (Mw)
326
deposited on the sensor surface [g m−2]:
327
Mw= 1.54 × exp(5.8 × 10−3× counts) (2)
Estimation of the mass of water on the canopy, Surface Canopy
328
Water (SCW), was performed in two steps. First, regular
329
measurements of leaf height, length and width were conducted,
330
and leaf areas were estimated by assuming that corn leaves
331
have an elliptical shape with the assumption that corn leaves
332
approximate the shape of an ellipse:
333
Aleaf = π × l ×
w
4 (3)
where Aleaf is the leaf area [m−2], l is leaf length [m] and
334
w is leaf width [m]. Second, it was assumed that the wetness
335
of a leaf at any height could be approximated as that of the
336 nearest sensor. 337 SCW = ρplant× n X i=1 Aleaf i× Mwi (4)
where SCW is surface canopy water per square meter of
338
ground [kg m−2], ρplant is the average number of plants per
339
m−2, Mwi is the water mass on the sensor closest to leaf i
340
[kg m−2], and n is the number of leaves per plant.
341
Root zone soil moisture was measured with ten Decagon
342
EC-5 sensors, which were installed in two pits at five different
343
depths: 5, 10, 20, 40 and 80 cm. The pits were located 40
344
meters apart, but centered between the same two rows.
Site-345
specific calibration was performed yielding a linear regression
346
R2 of 0.993. Soil moisture was similar in both pits. Hence, 347
the presented results are the averages over the two pits. 348
Linear interpolation between the measurements was applied 349
to visualize root zone soil moisture. 350
Soil water potential was monitored using two T4e pressure 351
transducer tensiometers [72]. These were installed 40 meters 352
apart, close to the soil moisture pits. The centers of the ceramic 353
cups were located at a depth of 20 cm. The presented results 354
are the averaged signals of both tensiometers. 355
D. Vegetation sampling and monitoring 356
Vegetation water content (VWC) and dry biomass (md) were 357
measured using destructive sampling. Four sampling areas 358
were established outside the arc swept by the radar, and outside 359
the in-situ sensor locations, at the beginning of the season. 360
Each sampling time, two representative samples were taken 361
from each of these four sampling areas. Any surface water 362
present on the plant tissue was removed with paper towel 363
before weighing. From the eight samples, the stems, leaves, 364
tillers, tassels, and ears were separated, weighed, and oven- 365
dried on 60◦C for 4-5 days in early season to 7 days in late 366
season. The dry samples were weighed again, and VWC [kg 367
m-2] was derived from equation 5:
368
V W C = (mf− md)ρplant (5)
where mf is the average fresh weight or fresh biomass of 369
the eight samples [kg], md is the average dry weight or dry 370
biomass of the eight samples [kg], and ρplantis the number of 371
plants per square meter of ground. Gravimetric water content, 372
Mg, is the mass of water per unit mass of fresh biomass 373
(equation 6): 374
Mg=
mf–md
mf
(6) Equations 5 and 6 were applied for each of the plant con- 375
stituents (i.e. leaves, stems, tillers, ears) separately. 376
Sampling was conducted before sunrise (6am) to minimize 377
the effect of transpiration on the measurements that represent 378
seasonal variability of VWC and Mg. These predawn measure- 379
ments were scheduled three times per week. On one of these 380
three days, one extra sampling was performed during the day 381
in order to capture diurnal variations. This second sampling 382
was at 6pm, which would be the time of the corresponding 383
evening pass for a sun-synchronous satellite such as SMAP 384
[73]. 385
Plant growth stages were visually identified three times per 386
week, using the Biologische Bundesanstalt, Bundessortenamt 387
and Chemical industry(BBCH) scale for corn [74]. The cut 388
samples were used to measure plant heights. Leaf area index 389
(LAI) was calculated by multiplying the averaged, estimated 390
leaf areas by plant density. 391
III. RESULTS 392
A. Hydrometeorology 393
Fig. 1 shows the hydrometeorological and soil moisture 394
conditions during the growing season. The first three weeks 395
Fig. 1. Time series of (a) rainfall, irrigation and reference evapotranspiration (ETo), (b) presence of surface canopy water resulted from dew, irrigation or rain, (c) volumetric root zone soil moisture content, (d) surface soil moisture content only, (e) soil water potential at 20 cm depth.
evapotranspiration (ETo) and an absence of precipitation (Fig.
397
1a). On several days, midnight irrigation was applied to control
398
soil moisture content (Fig 1a-d), leading to a soil water
399
potential which is favorable for root water extraction as soon as
400
the roots reach deep enough (Fig 1e). These conditions allowed
401
for high rates of transpiration. Water films on leaf surfaces
402
were detected every morning (Fig 1b) as a result of dew
403
formation, interception of sprinkler irrigation or a combination
404
of both, and disappeared at around 10:00 every morning.
405
The mid-season weather conditions featured frequent,
tropi-406
cal rainfall and thunderstorms (Fig. 1a). This resulted in water
407
droplets on the canopy for long periods during the day (Fig.
408
1b), and several sharp increases in root zone soil moisture
409
content (Fig. 1c). Limited rain between May 22 and May 27
410
led to a temporary reduction in root zone soil moisture content
411
and potential (Fig. 1d-e).
412
A dry period with high temperatures and solar radiation,
413
started on June 1. This produced high evaporative demand
414
(ETo in Fig. 1a), which resulted in a rapid decrease of soil
415
moisture in the root zone. Despite the limited root zone
416
soil moisture, leaf surfaces were wet every morning, mainly
417
because of dew formation (Fig. 1b). A substantial rain event 418
on June 10 ended the dry period. 419
B. Vegetation development and water content 420
The sweet corn crop development is illustrated in Fig. 2. 421
Corresponding explanations of the BBCH phenology codes 422
can be seen in Table II. Fig. 2(a) shows the plant height 423
and dry biomass accumulation [kg m-2] of the total plant and
424
individual plant constituents during the life cycle, based on 425
destructive vegetation sampling data. Fig. 2 (b) and (c) show 426
how the water content of the plant and its constituents vary 427
during the growing cycle. Fig. 2(b) shows the mass of water 428
stored in [kg m-2], a measure commonly used in microwave
429
remote sensing. Fig. 2(c) shows this water storage in terms 430
of gravimetric moisture content, which is the mass of water 431
per total mass of the plant. This is more closely related to the 432
relative water content used by plant physiologists. 433
First plant emergence was observed on April 19, six days 434
after sowing. Although leaf and stem dry biomass (md) 435
d 2 2 g
Fig. 2. Seasonal patterns of (a) dry biomass and maximum canopy height, (b) predawn vegetation water content and leaf area index, and (c) predawn gravimetric water content, including the contributions of dominant plant constituents to total. Important phenological stages are represented by BBCH codes, which are explained in Table II.
height (Fig. 2a), stems held substantially more water (Fig.
2b-437
c). At the end of the vegetative stage, 65% of all VWC was
438
stored in the stems.
439
In the reproductive stage, ear formation coincided with
440
VWC decreases in all other constituents, especially in the
441
stems. From May 30 to June 6, water storage in the stems
442
decreased by almost 30%: -0.8 kg m-2, as the ears formed and
443
seperated from the main stem (Fig. 2b). Leaf senescence of the
444
lowest leaves occurred from June 2 onward. The reproductive
445
stage largely coincided with the dry period shown in Fig. 1.
446
The corn was harvested five days after the last sampling.
447
The results of the seven days of twice-daily destructive
448
vegetation sampling are shown in Fig. 3. The figure shows
449
the internal canopy water differences between 6:00 and 18:00
450
for the total plant and the most important constituents (by
451
biomass). The smallest ∆ Mg was observed on May 16, when
452
cloud and rain limited transpiration. Significant decreases in
453
internal water content were observed in the early season, as
454
a result of high atmospheric demand for evapotranspiration
455
(Fig. 1(a)), a shallow root zone, and a relatively dry upper
456
soil (Fig. 1(c)). These differences in Mg translate to small
457
∆VWC (Fig. 3b) due to the limited fresh biomass in the early
458
vegetative stages. In the reproductive stage, diurnal moisture
459
TABLE II
CROP DEVELOPMENT STAGES
BBCH Stage of development Dates 13 Leaf development – 3 leaves (V) Apr 25 21 Start of tiller formation (V) May 7 30 Start of stem elongation (V) May 18 51 Start of tassel emergence (V) May 23 63 Male: start pollen shedding. Jun 1
Female: stigmata tips visible (R)
71 Start of grain development: Jun 11 kernels at blister stage (R)
V=vegetative stage. R= reproductive stage.
g g 2
Fig. 3. Change (6:00 minus 18:00) in internal canopy water content of total plant and dominant constituents, represented as (a) gravimetric moisture loss and (b) equivalent weight of moisture loss [kg m-2].
losses in stems increased, while such losses decreased for ears. 460
At this stage, ears grow and store water, while the internal 461
water content of the stems starts to decrease. These sub-daily 462
variations were substantial compared to the seasonal predawn 463
Mgvariations (Fig. 2c). It should be noted that maximum sub- 464
daily moisture variations may be higher than the difference 465
between 6:00 and 18:00. 466
C. Backscatter 467
1) Seasonal variations in backscatter: Backscatter coef- 468
ficients (VV, HH, average cross-pol) are shown in Fig. 4. 469
Backscatter increased in all polarizations with the growth of 470
the crop. Co-polarized backscatter increased from <-14 dB 471
after planting to about -5 dB when plants reached half of 472
their total biomass, while cross-polarized backscatter increased 473
from <-32 dB to about -16 dB. 474
The influence of early season irrigation events (Fig. 1) is ap- 475
parent in all polarizations (Fig. 4). Sensitivity to wetting events 476
(irrigation and rainfall) decreased as the canopy grew and σ0
477
became increasingly sensitive to wet biomass (Appendix A). 478
The increasing trend in σ0due to vegetation growth tapers
479
off around May 20. These high values, 3-4 days prior to 480
plant VWC and LAI maxima can be explained by the heavy 481
rainstorms around May 20. Precipitation from these storms in- 482
creased both canopy surface wetness and soil moisture, which 483
produced high σ0 values in all polarizations. Nonetheless, 484
Fig. 4. Time series of observed L-band co- and cross-polarized backscatter.
decrease in σ0 from June 1 corresponds with the drop in soil
486
moisture (Fig. 1) and the sharp reduction in stem water content
487
(Fig. 2b). Backscatter increased again following small rain
488
events and the formation and separation of ears.
489
2) Early season: Two three-day periods in the early season
490
are highlighted in Fig. 5. Fig. 5 (a-c) shows a period
11-491
13 days after emergence when bare soil exposure was still
492
considerable and plant height was just 15-20 cm. Fig. 5 (a)
493
shows co- and cross-polarized backscatter. Fig. 5 (b) shows
494
raw data counts from two leaf wetness sensors positioned 7cm
495
above the ground, as well as the sap flow. Fig. 5 (c) shows
496
the precipitation at the nearby weather station and the soil
497
moisture observed at 5cm depth. The irrigation event on April
498
30 lead to an increase in soil moisture at 5cm depth, and a
499
sharp increase in σ0 of up to 5dB. Clear cyclic variations of
500
2-3 dB are observed in σ0, particularly in σ
V V. These cannot
501
be explained by the 5cm soil moisture, but seem to follow the
502
accumulation and dissipation of dew on the vegetation and
503
soil surface as indicated by leaf wetness sensor (LWS) data in
504
Fig. 5 (b). The LWS counts increase during the night as dew
505
accumulates on the sensor. The LWS counts decrease rapidly
506
after sunrise as the increase in solar radiation allows the dew
507
to evaporate. On each of the days shown in Fig. 5, and also
508
in Fig. 1 (b), the dew has generally dissipated by 10:00 am. It
509
is important to note that, in addition to forming on the leaves
510
and the LWS, dew also forms as a film of water on the soil
511
surface. It is clear from Fig. 5 (c) that it is insufficient to
512
infiltrate the soil and reach the sensor at 5cm. However,
L-513
band backscatter is dominated by surface scattering from the
514
soil at this stage ( [75]–[78], and Fig. 10), and the difference
515
between a wet (e.g. 0.3 cm3cm−3) and dry (e.g. 0.1 3cm−3)
516
soil can produce differences of up to 3.5 (HH) and 6 (VV)
517
dB [79]. We postulate, therefore, that the accumulation and
518
dissipation of this film of water on the soil surface is the most
519
important reason behind the cyclic σ0variations in Fig. 5 (a).
520
The effect of dew on the topsoil is also clear in VV where
521
the irrigation event on April 30 increases σV V from -18 dB to
522
-13 dB, a value at which it stays due to the presence of dew
523
until sunrise the following morning. Moreover, σV V ramps
524
up as dew accumulates in the early hours of May 2, before
525
decreasing again at sunrise.
526
Fig. 5 (d-f) shows the measurements of one week later 527
when the maximum plant height has increased to 37 cm (May 528
7) and 43 cm (May 9), LAI is around 0.57 and VWC is 529
increasing from 0.16 kg m-2(May 7) to 0.23 kg m-2(May 9). 530
While soil moisture values are comparable to those observed 531
the week before, the σ0 values in Fig. 5 (d) are around 4 532
dB higher than those in Fig. 5 (a) in all polarizations. From 533
Fig. 10, this can be attributed to the increase in vegetation 534
scattering in all polarizations, an increase in double-bounce in 535
VV, and an increase in vegetation ground scattering in HH and 536
cross-pol. In other words, the backscatter is increasing due to 537
plant growth , and microwave interactions with the vegetation 538
are becoming increasingly important. Sap flow values are 539
higher than in Fig. 5 (b) due to increase in plant area and 540
transpiration. The LWS have been repositioned at 10cm and 541
20cm to accommodate the growing plant, and a sensor was 542
added at 30cm on May 7. 543
Irrigation events on May 6 and 9 lead to sharp increases 544
in soil moisture and σ0. Increases of 8 dB (VV), 4 dB (HH) 545
and 5-6 dB (cross-pol) were observed in response to the event 546
on May 9. On the May 7 and 9, initial rapid increases in 547
LWS counts due to interception of irrigation were followed 548
by more gradual increases as dew accumulated during the 549
night. Steady dew accumulation is also observed during the 550
night of May 7-8. On all three days, the accumulated moisture 551
dissipated quickly after sunrise. The cyclic variations in σ0are
552
clearer than they were in Fig. 5, and their correspondence with 553
the LWS data is even more striking. The σV V is particularly 554
responsive to the presence of water on the soil and vegetation. 555
This may be due to the important role of double-bounce in 556
σV V at this time. 557
3) Mid-season: Fig. 6 shows two periods in the mid-season. 558
Note that average σ0increased significantly since early season,
559
as a result of plant growth. In the time period shown in 560
Fig. 6 (a-c), the corn had started to tassel, and leaves had 561
almost reached final sizes (Fig. 2). Fig. 10 shows that, for this 562
growth stage, σ0is dominated by vegetation scattering. There 563
are limited contributions from double-bounce in VV and HH, 564
and the vegetation-ground term in HH. This is consistent with 565
earlier research [75]–[78]. 566
Fig. 5. Early season patterns of co- and cross-polarization backscatter (upper row), raw data counts from the leaf wetness sensors and sap flow (middle row), and surface soil moisture and precipitation (lower row). The left figures show a 3-day period when plants reach 15-20 cm, while the right figures show a 3-day period when plants reach 43-65 cm. Note that the vertical axes of the left and right backscatter plots are different.
intercepted by the almost fully grown leaves and had a
568
negligible impact on soil moisture. This suggests that σ0
569
variations in Fig. 6a can be attributed to variations in SCW
570
and internal VWC (Fig. 6a-b). Cross-pol backscatter, which is
571
sensitive to leaf moisture content, increased rapidly in response
572
to interception in the evenings of May 24 and May 25. The
573
presence of dew in the early hours of May 23 and 25, and
574
interception on May 24 and 25 resulted in elevated values of
575
σcross. Rapid dissipation of dew in the early morning on May
576
23 and 25 produced a ∼2dB drop in σcross. The difference
577
in response of the three polarizations to SCW is particularly
578
noticeable during the interception and dew events early on
579
May 25, and could be explained by their relative sensitivities to
580
different canopy constituents. Note that estimated interception
581
sometimes exceeds measured rainfall. This could be due to
582
(1) the simplistic model used to convert sensor output to
583
full canopy interception (see section II-C), (2) spatial rainfall
584
variability (rainfall was collected 600 meters from studied
585
field), (3) accuracy of rainfall data (∼0.25 mm) or (4) spatial
586
heterogeneity of interception itself due to e.g. variations in
587
plant architecture.
588
Backscatter in all polarizations reflects variations in internal
589
water content. Recall from Fig. 3 that internal water losses
590
were high in this period (∼0.5 kg m-2) because of a relatively
591
high atmospheric water demand (Fig. 1). The rise in sap flow
592
and transpiration resulted in a decrease in σ0, and water uptake
593
in the evening resulted in an increase. 594
Fig. 6(d-f) illustrates the observations of one week later, 595
after a four-day period of heavy rainfall (Fig. 1). The last 596
rain event on May 30 was followed by two dry and hot days, 597
resulting in a decrease in surface soil moisture. The limited 598
variation in σV V and σHH in response to the sharp increases 599
in SCW and soil moisture suggests that co-pol backscatter 600
saturated at ∼-4 dB. Despite high rates of sap flow on May 601
31, diurnal cycles of σV V and σHH were not observed. This 602
saturation was probably caused by a combination of a wet field 603
(Fig. 1) and a peak in VWC (Fig. 2). May 31 was characterized 604
by high evapotranspiration rates, causing canopy surface water 605
to disappear, and soil moisture to decrease. Meanwhile, stem 606
water content started to drop significantly (Fig. 2). These 607
losses of water led to a decrease in co-pol backscatter, which 608
resulted in observed diurnal cycles of σ0again (Fig. 6d). Dips
609
in cross-polarized backscatter (May 31) and all polarizations 610
(June 1) coincide with the dissipation of dew and peaks of sap 611
flow. 612
4) Late season: Fig. 7 shows observations from two periods 613
in the late season during which the corn plants experienced 614
the lowest root zone water availability of the season (Fig. 1(c 615
and e)). Recall from Fig. 3 that diurnal water fluctuations in 616
Fig. 6. Mid-season patterns of co- and cross-polarization backscatter (upper row), surface canopy water and sap flow (middle row), and surface soil moisture and precipitation (lower row). The left column shows a 3-day period when plants reach 125-140 cm, while the right column shows a 3-day period when plants reach 180-189 cm. Note that the vertical axes of backscatter and surface canopy water are different from Fig. 5.
Nonetheless, the plants were able to recover from water losses
618
after solar noon; predawn Mg did not decrease with higher
619
rates than in the wet period before (Fig. 2). Simulations in Fig.
620
10 suggest that σcrossand σV V were dominated by vegetation
621
scattering, while σHH still had limited sensitivity to
ground-622
related terms.
623
The diurnal VWC cycles were discernible in σ0 in all
624
polarizations, particularly on the days without rainfall. On
625
June 5, there is a noteworthy decrease of almost 4dB in all
626
polarizations. This coincides with a significant loss of internal
627
water content (Fig. 3) due to transpiration (sap flow in Fig.
628
7 (b)). The minimal change in soil moisture at this time, and
629
the fact that the decrease is consistent across polarizations
630
suggests that this is a decrease in vegetation scattering due to
631
the observed drop in internal water content. From midnight
632
on June 9 to noon on June 10, soil moisture barely changes.
633
Backscatter on the other hand, especially VV and
cross-634
pol, increases with dew accumulation during the night and
635
decreases as dew dissipates and transpiration leads to internal
636
water content losses during the day. A similar response is
637
observed in the response to dew and transpiration in the early
638
hours of June 11. Again, the minimal variation in soil moisture
639
and the consistency across polarizations suggest that this is a
640
response to internal and surface canopy water dynamics rather
641
than sensitivity to soil moisture.
642
Precipitation events on June 4, 6 and 10 (Fig. 7 (c) and 643
(f)) resulted in spikes in interception (SCW in Fig. 7(b) and 644
(e)). The precipitation event of June 10 led to a substantial 645
and prolonged increase of soil moisture. The limited effect 646
this prolonged increase had on backscatter (particularly σV V 647
and σcross) confirms the strong reduction to soil moisture 648
sensitivity at this stage. Given the lack of sensitivity to surface 649
soil moisture in VV and cross-pol, it is likely that these 650
backscatter increases are primarily in response to interception 651
rather than moisture on the soil surface. 652
5) Mean daily cycles: Fig. 5-7 show that sub-daily varia- 653
tions in σ0 included rapid variations due to the interception 654
of intermittent precipitation events, and slower variations due 655
to dew formation and dissipation and internal water content 656
variations. To minimize the influence of random individual 657
precipitation events and gain some insight into the average 658
daily cycle, data were averaged over a 21-day period between 659
May 23 and the last day of the experiment, June 23. This 660
is the period in which σ0 did not increase anymore as a
661
result of crop growth (Fig. 4). Previous studies and model 662
simulations suggest that the effect of soil moisture on σ0 in
663
all polarizations is limited in this period (Fig. 10, [75], [76], 664
[78]). 665
Fig. 7. Late season patterns of co- and cross-polarization backscatter (upper row), surface canopy water and sap flow (middle row), and surface soil moisture and precipitation (lower row). Maximum canopy height was stabilized at 205 cm during this period. Note that the temporal density of backscatter is less than in previous figures.
flow, soil moisture and precipitation over this period. Clear
667
daily cycles can be observed in σ0, sap flow and SCW. Peaks
668
of VV and cross-pol coincide with the peak in SCW, and
669
the start of the sap flow/transpiration cycle. After sunrise,
670
the increase in net radiation drove transpiration and led to
671
dissipation of dew from the canopy. Backscatter dropped
672
on average with 0.7 dB (VV), 0.6 dB (HH), and 1.0 dB
673
(cross-pol) between sunrise and 15:00. After 15:00, there is
674
a downward trend in sap flow and an upward trend in σ0.
675
Most rainfall events occurred during daytime and they explain
676
the fluctuations in averaged SCW during the afternoon. The
677
peak values in rainfall at 09:45 and 12:15 were due to two
678
major convective rainfall events, each of which resulted in a
679
significant increase in average soil moisture but only a modest
680
and transient effect on average SCW and σ0.
681
To exclude the effect of rainfall completely, the four days
682
without any rainfall within this period were plotted separately
683
in Fig. 9. Note that there was a decreasing trend in σ0 during
684
this period due to the loss of internal water content of the
685
stems in this growth stage, and the limited root zone soil
686
moisture availability between June 5 and 9. Also note that
687
sap flow was high in this period, so high sub-daily variations
688
of internal water content are expected. Temporal patterns were
689
similar to those in Fig. 8, although the timing of the σ0minima
690
are slightly different. Cross-polarized backscatter changed
691
inflection again after the peak hours of evapotranspiration, 692
while VV-polarized backscatter changed inflection with the 693
start of dew formation. In both Fig. 8 and Fig. 9, nocturnal 694
increase is only observed in σV V and σcross. In the absence 695
of precipitation, the average diurnal difference in σ0 on these 696
four days was 2.4 dB (VV), 1.6 dB (HH) and 2.0 dB (cross- 697
pol). 698
IV. DISCUSSION 699
Consistent with previous studies (e.g. [80]), L-band sensi- 700
tivity to scattering from vegetation correlated with the buildup 701
of VWC during the season (Fig. 2). With low vegetation, early 702
season σ0patterns in all polarizations were consistent with soil
703
moisture responses to wetting events (irrigation, precipitation), 704
and even showed strong similarities with dew deposition on 705
the topsoil (Fig. 5). Similar wetting events, with similar soil 706
moisture responses, showed a much smaller effect on σ0 in
707
all polarizations in mid and late season (Fig. 4 and Fig. 7). 708
In mid and late season, and particularly beyond May 18, 709
differences between σV V and σHH were minimal. This can 710
only be explained by the predominance of volume scattering, 711
i.e. direct vegetation scattering, since stem attenuation and 712
scattering, as well as double bounce are polarization dependent 713
[81]. This predominance of vegetation scattering is confirmed 714
00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 6.1 5.8 5.5 5.2 4.9 0 [d B ] ( co -p ol ) (a) Backscatter HH VV 18.2 17.9 17.6 17.3 17.0 0 [d B ] ( cr os s-po l) Av cross-pol 0.00 0.05 0.10 0.15 0.20 0.25 SC W [k g m 2 ]
(b) SCW (left) and sap flow (right) SCW 0.00 0.02 0.04 0.06 0.08 0.10 Sa p fl ow [m m 15 m in 1 ] sap flow 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 0.1350 0.1425 0.1500 0.1575 5cm [-]
(c) Surface soil moisture (left) and rainfall (right) soil moisture 0.00 0.25 0.50 0.75 Pr ec ip ita tio n [m m 15 m in 1 ] rain Average over 21 days (from May 23 - June 13)
Fig. 8. Mean daily cycles of (a) co- and cross-polarized backscatter, (b) surface canopy water and transpiration, and (c) soil moisture and rainfall, for the last 21-days of the season. Timing of sunrise and sunset are depicted with triangles in (b).
Although it is not a persistent contribution, double bounce can
716
still cause some sensitivity to soil moisture at HH polarization
717
until the end of the season. The seasonal increased sensitivity
718
to vegetation and reduced sensitivity to soil confirms previous
719
work on L-band (e.g. [20], [75], [80]).
720
Sub-daily backscatter variability has been attributed to
721
variations in VWC in several studies (e.g. [39], [41], [82]).
722
However, these satellite-based studies lacked ground validation
723
data. The unprecedented destructive sampling data presented
724
in this study confirm that sub-daily variations in VWC are
725
substantial (>0.5 kg m-2, Fig. 3) even though corn is an
726
isohydric species (i.e. water content is regulated through active
727
stomatal control). This motivates further research to include
728 7.0 6.5 6.0 5.5 5.0 4.5 0 [d B ] ( co -p ol ) (a) Backscatter HH VV 19.5 19.0 18.5 18.0 17.5 17.0 0 [d B ] ( cr os s-po l) Av cross-pol 0.00 0.05 0.10 0.15 0.20 0.25 SC W [k g m 2 ]
(b) SCW (left) and sap flow (right) SCW 0.00 0.02 0.04 0.06 0.08 0.10 Sa p fl ow [m m 15 m in 1 ] sap flow 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 0.1350 0.1425 0.1500 0.1575 5cm [-]
(c) Surface soil moisture (left) and rainfall (right) soil moisture 0.00 0.25 0.50 0.75 Pr ec ip ita tio n [m m 15 m in 1 ] rain Average over 4 days without rain (between May 23 and June 13)
Fig. 9. Mean daily cycles of (a) co- and cross-polarized backscatter, (b) surface canopy water and transpiration, and (c) soil moisture and rainfall, for only the 4 dry days within the last 21-days of the season. Timing of sunrise and sunset are depicted with triangles in (b).
other species. 729
The deepest drops in σ0were observed after the acquisition
730
at 9:30, when dew almost dissipated completely (Fig. 9). 731
This is also observed on dry days in Fig. 6 and Fig. 7. 732
Since transpiration rates become substantial after this time, 733
this suggests that diurnal VWC fluctuations govern the most 734
substantial part of the sub-daily σ0 cyles in mid- and late 735
season. 736
Several studies have attributed differences in diurnal 737
backscatter to the presence of dew, but did not account for 738
variations in internal VWC (e.g. [83]–[85]). The combina- 739
tion of intensive destructive vegetation sampling, continuous 740
unique insight into their combined influence on the dynamics
742
of sub-daily backscatter and how that varies throughout the
743
season. Gillespie et al. [17] provide one of the few studies in
744
which both internal and surface water are considered in the
745
context of dew detection. They found that the C-band
HH-746
pol backscatter on a night without dew had a peak 1.5 hours
747
after sunrise, corresponding to the peak in their observations
748
of internal VWC [17]. On the other hand, the backscatter
749
peak on a night with dew was at sunrise, which is the
750
moment beyond dew starts to disappear. Similarly, our L-band
751
observations show a sub-daily backscatter maximum around
752
sunrise, particularly in VV and cross-pol (Fig. 8 and Fig.
753
9). This is consistent with the peaks of dew accumulation
754
and suggests that dew can have a significant effect on the
755
timing and magnitude of the maximum of the sub-daily
756
backscatter cycle. While [17] provided the first indication that
757
dew formation and dissipation determines the peak of (C-band,
758
HH-pol) backscatter, the dataset was limited to two nights
759
and the use of visual inspection to confirm the presence of
760
dew. Our inclusion of continuous leaf wetness sensors allowed
761
us to capture the accumulation, peak and dissipation of dew
762
every night for the entire growing season, ensuring that our
763
conclusions are based on a diverse range of events.
764
The inclusion of continuous leaf wetness measurements also
765
provides unique, new insight into L-band backscatter
sensitiv-766
ity to rainfall interception. Light rain events, intercepted by the
767
vegetation, caused strong fluctuations in σ0 (3dB in cross-pol
768
and 2dB in co-pol), even though soil moisture was constant
769
(see Fig. 6a-c). The presence of surface canopy water is not
770
considered in current electromagnetic models (e.g. [75], [86],
771
[87]) or retrieval algorithms (e.g. [20], [79], [88]). The results
772
presented here demonstrate that SCW can have a significant
773
effect on σ0. Accounting for SCW in models and retrieval
774
algorithms can therefore be expected to lead to improved
775
retrievals of soil and vegetation variables.
776
The significant effect of both dew and interception on σ0
777
illustrates the value of including continuous leaf wetness
778
sensors in microwave field campaigns and experiments. In
779
this study, SCW was estimated from leaf Wetness Sensor
780
data using a simple weighting based on LAI. While this
781
was sufficient to demonstrate the important influence that
782
SCW has on the canopy, rigorous validation of SCW
783
is essential in future experiments that seek to establish
784
quantitative relationships between SCW and σ0. Given that
785
microwaves penetrate the vegetation, future research should
786
also examine how the vertical distribution of surface canopy
787
water influences its effect on σ0.
788 789
V. CONCLUSIONS
790
Results from an intensive experimental campaign combining
791
sub-daily radar and vegetation water dynamics observations
792
were used to explore the sensitivity of L-band radar backscatter
793
to variations in surface and internal canopy water content of
794
corn. The daily cycle in radar backscatter was found to vary
795
in amplitude depending on the growth stage of the vegetation.
796
Though the strongest diurnal variations were observed during
797
the early vegetative stages, the limited vegetation scattering 798
and attenuation during this time suggests that these varia- 799
tions are attributed to surface soil moisture fluctuations and 800
heavy dew on the uppermost skin of the soil. As the canopy 801
approached full biomass, the sensitivity to the underlying 802
soil was strongly reduced, and the diurnal cycle in radar 803
backscatter was found to reflect temporal patterns in surface 804
and internal water content. 805
Radar backscatter, especially in cross-pol, was found to be 806
sensitive to surface canopy water, with temporal variations 807
in radar backscatter closely following the slow accumulation 808
and rapid dissipation of dew, and exhibiting transient but 809
significant increases in response to interception. In addition to 810
being a variable of interest in its own right, the prevalence of 811
dew during the night and early morning and its influence on the 812
radar backscatter highlights the potential influence of overpass 813
time on the interpretation of radar observations from sun- 814
synchronous satellites for vegetation monitoring. It also high- 815
lights the potential benefit of being able to choose sub-daily 816
SAR data at specific overpass times to avoid the confounding 817
influence of dew on the retrieval of biomass and internal water 818
content. Both the effects of surface and internal canopy water 819
on backscatter underscore the importance of including canopy 820
water dynamics in physical models, particularly those used to 821
simulate sub-daily radar observations. 822
One of the key challenges of exploiting sub-daily space- 823
borne SAR will be to disentangle surface and internal wa- 824
ter content. Continuous monitoring of surface canopy water 825
significantly improved the interpretation of sub-daily radar. 826
During daytime, interception events are often transient and 827
easily identifiable, and dew dissipation is often rapid. However, 828
the slower dynamics of dew accumulation and internal water 829
content variations are more difficult to separate. Developing a 830
reliable approach to monitor VWC continuously would ease 831
this separation of signals, and would improve the interpretation 832
of sub-daily radar significantly. The sensitivity to surface and 833
internal water content variations was found to be polarization 834
dependent. This suggests that sub-daily polarimetric SAR 835
(PolSAR) could be particularly useful to disentangle surface 836
from internal canopy water variations. 837
The results demonstrate a potentially valuable application 838
for sub-daily spaceborne SAR missions. However, the dataset 839
is limited to a single crop type and a single radar configuration. 840
There are many open questions to be addressed. Planned and 841
candidate missions have been proposed that could yield data 842
at different frequencies. Additional experimental research is 843
essential to explore the sensitivity of backscatter from L-, 844
to Ku-band to canopy water dynamics given the influence 845
that frequency will have on both the penetration depth in 846
the canopy and the sensitivity to the various vegetation con- 847
stituents. The influence of viewing geometry also warrants 848
investigation. The incidence angle of radar backscatter obser- 849
vations from geostationary satellites varies by latitude. Hence, 850
the suitability of sub-daily SAR data may be limited to certain 851
latitudinal bands. For constellations, a time series of data for 852
a given location on the ground will combine acquisitions that 853
may vary by incidence and azimuth angle. Both influence 854
on the relative sensitivity to surface and internal water content
856
and soil moisture and roughness should be characterized.
857
Moreover, given the importance of rainfall interception on
858
the radar signals and its complexity, more research should
859
be conducted on better estimating interception, under different
860
conditions, for different types and stages of vegetation, and the
861
effect of the distribution of intercepted water in the canopy on
862
backscatter.
863
The future availability of sub-daily fine resolution data on
864
surface and internal water content offers an extraordinary
865
opportunity to study plant water dynamics from a new
per-866
spective, and at the landscape scales most relevant for
under-867
standing water and carbon exchanges in the climate system.
868
By providing information on rapid surface and internal plant
869
water dynamics, sub-daily spaceborne SAR has the potential
870
to become a valuable source of data in the fields of hydrology,
871
land surface modeling, climate modeling, numerical weather
872
prediction and plant physiology.
873
APPENDIX
874
ELECTROMAGNETIC MODEL SIMULATIONS
875
A physical model for corn, developed at the Tor Vergata
876
University of Rome [86], [89], was used to illustrate
contribu-877
tions of soil and vegetation components to total backscatter,
878
and the changes during the season. This model is based on
879
radiative transfer theory and provides polarimetric backscatter
880
of agricultural fields. It is able to simulate both scattering
881
and extinction properties of vegetation elements and of the
882
underlying soil applying the most suitable electromagnetic
883
approximation, depending on the scatterer size and shape.
884
Furthermore, it is able to take into account multiple scattering
885
of any order and it can separate contributions of different
886
scatterers in the vegetation canopy.
887
The inputs for the model are listed in Table III. Soil root
888
mean square (RMS) height was estimated using the meshboard
889
approach described in [90], and was measured in the period
890
between sowing and crop emergence. Correlation length is
891
very difficult to measure accurately, because it is extremely
892
variable [91]. Therefore, we chose the correlation length which
893
gave the best fit between simulated and observed σ0during the
894
bare soil period. Plant density was averaged over 40 randomly
895
chosen samples. The model was run with a daily time step.
896
Because the model does not account for surface canopy water,
897
soil moisture values at 10:00 were used to ensure that dew had
898
dissipated from the canopy at the observation time. Since water
899
on leaves suppresses transpiration [10], the internal water
con-900
tent at 10:00 should be close to 6:00 observations. Time series
901
of vegetation parameters were linearly interpolated. Similar
902
to the observed σ0, cross-polarized backscatter represents the
903
average of VH and HV polarizations.
904
The model simulations (RMSE=3.91 dB) are presented in
905
Fig. 10. Note that the observed co-polarized backscatter is
906
underestimated by the model, while the cross-pol increase
907
due to vegetation growth is very well reproduced. Vegetation
908
scattering refers to the volume scattering by the vegetation
909
layer. Ground scattering refers to direct scattering solely from
910
the ground. Vegetation-ground scattering represents multiple
911
TABLE III
INPUT PARAMETERS FOR MODEL SIMULATION
Parameter Single value or time series Frequency 1.25 GHz
Incidence angle 40◦ Soil rms height 2.5 cm Soil correlation length 33 cm Surface soil moisture (10:00) Time series Crop height Time series Mg cylinders (stems) Time series
Mg discs (leaves) Time series
No. of leaves Time series LAI Time series Stem height Time series Stem radius Time series Leaf area Time series Plant density 8 plants m-2
scattering effects due to interactions between the vegetation 912
and ground. Double-bounce scattering represents the contri- 913
bution coming from specular reflection from the soil followed 914
by specular reflection by stems, and viceversa. 915
Co-polarized backscatter was dominated by the direct 916
ground contribution in the early season. Increasing VWC 917
during the vegetative stages (Fig. 2) results in attenuation of 918
the ground contribution, and an increase in the vegetation, 919
vegetation-ground and double-bounce terms. Double-bounce 920
scattering increases with stem growth and is most significant 921
in VV during the early vegetative stages. 922
Both co-polarized σ0 simulations are dominated by di- 923
rect vegetation scattering after May 16, when LAI>1, and 924
VWC>1.5 kg m-2. After May 23, when LAI>3.5 and
925
VWC>3.5 kg m-2, σ
V V simulations can be almost completely 926
explained by direct scattering from vegetation in mid- and 927
late season, and other scattering mechanisms are negligible. 928
However, indirect and direct scattering from the ground still 929
contribute to σHH to some degree in this period. These 930
results are comparable to those of [78], where a larger double 931
bounce effect at HH polarization is observed, due to a much 932
smoother soil surface. Cross-polarized backscatter (σcross) 933
was dominated by direct scattering from vegetation, even when 934
the plants were still small. 935
ACKNOWLEDGMENT 936
This project was supported by Vidi Grant 14126 from the 937
Dutch Technology Foundation STW, which is part of The 938
Netherlands Organisation for Scientific Research (NWO), and 939
which is partly funded by the Ministry of Economic Affairs. 940
The experiment was made possible by infrastructural and 941
technical support from the Agr. and Biol. Eng. Dept. and 942
PSREU at the University of Florida. The authors would like 943
to thank Daniel Preston and Patrick Rush for their technical 944
support, Eduardo Carrascal for data collection and processing, 945
James Boyer and his team for their on-farm logistical support, 946
and Roger DeRoo for useful insights about the UF-LARS 947
system. We also appreciate the efforts of the five anonymous 948