Babaeianetal 2018RSE

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/324756038
Mapping soil moisture with the OPtical TRApezoid Model (OPTRAM) based on
long-term MODIS observations
Article in Remote Sensing of Environment · June 2018

DOI: 10.1016/j.rse.2018.04.029
CITATIONS READS
110 1,125
5 authors, including:
Ebrahim Babaeian Morteza Sadeghi

University of Florida California Department of Water Reources
70 PUBLICATIONS 1,350 CITATIONS 86 PUBLICATIONS 1,975 CITATIONS
SEE PROFILE SEE PROFILE
T. Franz Scott B. Jones

University of Nebraska at Lincoln Utah State University
149 PUBLICATIONS 5,638 CITATIONS 272 PUBLICATIONS 8,257 CITATIONS
SEE PROFILE SEE PROFILE
All content following this page was uploaded by Ebrahim Babaeian on 14 May 2018.
The user has requested enhancement of the downloaded file.

1 Mapping Soil Moisture with the OPtical TRApezoid Model (OPTRAM) based
2 on Long-Term MODIS Observations
4 Ebrahim Babaeian1,*, Morteza Sadeghi2, Trenton E. Franz3, Scott Jones2, Markus Tuller1
5
1
6 Dept. Soil, Water and Environmental Science, The University of Arizona, Tucson, Arizona
2
7 Dept. Plants, Soils and Climate, Utah State University, Logan, Utah
3
8 School of Natural Resources, University of Nebraska-Lincoln, Lincoln, Nebraska
9
*
10 Corresponding Author at 1177 E. 4th Street, Tucson, Arizona 85721, USA
11 E-mail address: [email protected] (E. Babaeian), [email protected] (M. Sadeghi),
12 [email protected] (T.E. Franz), [email protected] (S.B. Jones), [email protected] (M. Tuller)
13
14
15 Abstract
16 The OPtical TRApezoid Model (OPTRAM) has recently been proposed for estimation of
17 soil moisture using only optical remote sensing data. The model relies on a physical linear
18 relationship between the soil moisture content and shortwave infrared transformed reflectance
19 (STR) and can be parameterized universally (i.e., a single calibration for a given area) based on
20 the pixel distribution within the STR-Normalized Difference Vegetation Index (NDVI)
21 trapezoidal space. The main motivation for this study is to evaluate how the universal
22 parameterization of OPTRAM works for long periods of time (e.g., several decades). This is
23 especially relevant with regard to uncovering the soil moisture and agricultural drought history in
1
24 response to climate change in different regions. In this study, MODIS satellite observations from
25 2001 to 2017 were acquired and used for the analysis. Cosmic-ray neutron (CRN) soil moisture
26 data, collected with the COsmic-ray Soil Moisture Observing System (COSMOS) at five
27 different sites in the U.S. covering diverse climates, soil types, and land covers, were applied for
28 evaluation of the MODIS-OPTRAM-based soil moisture estimates. The OPTRAM soil moisture
29 estimates were further compared to the Soil Moisture Active and Passive (SMAP) (L-band), the
30 Soil Moisture Ocean Salinity (SMOS) (L-band), and the Advanced AScatterometer (ASCAT)
31 (C-band) soil moisture retrievals. OPTRAM soil moisture data were also analyzed for potential
32 monitoring of agricultural drought through comparison of the OPTRAM-based Soil Water
33 Deficit Index (OPTRAM-SWDI) with the widely-applied Crop Moisture Index (CMI).
34 Evaluation results indicate that OPTRAM-based soil moisture estimates provide overall unbiased
35 RMSE and R between 0.050 to 0.085 cm3 cm-3 and 0.10 to 0.70, respectively, for all investigated
36 sites. The performance of OPTRAM is comparable with the ASCAT retrievals, but slighly less
37 accurate than SMAP and SMOS. OPTRAM and the three microvave satellites captured CRN soil
38 moisture temporal dynamics very well for all five investigated sites. A close agreement was
39 observed between the OPTRAM-SWDI and CMI drought indices for most selected sites. In
40 conclusion, OPTRAM can estimate temporal soil moisture dynamics with reasonable accuracy
41 for a range of climatic conditions (semi-arid to humid), soil types, and land covers, and can
42 potentially be applied for agricultural drought monitoring.
43
44 Keywords: Soil moisture mapping, The optical trapezoid model OPTRAM, Drought monitoring,
45 Cosmic-ray neutron soil moisture, MODIS, SMAP, SMOS, ASCAT
46
47 1. Introduction
2
48 Soil moisture is a highly dynamic state variable that controls fundamental hydrological
49 processes such as evaporation, infiltration, and runoff. It is also critical for management and
50 allocation of water resources, prediction and monitoring of drought, agricultural production,
51 mitigation of natural disasters and monitoring of ecosystem response to climate change
52 (Robinson et al., 2008; Vereecken et al., 2014; Babaeian et al., 2016).
53 Remote sensing techniques provide powerful means for characterizing and monitoring
54 the high spatiotemporal variability of soil moisture. During the past decade, several satellites
55 with various spatiotemporal resolutions have been launched for monitoring near surface (0-5 cm)
56 soil moisture, including the European Soil Moisture and Ocean Salinity (SMOS) Satellite,
57 launched in 2009 (Kerr et al., 2001), the Advanced Microwave Scanning Radiometer 2 (AMSR-
58 2) on the Global Change Observation Mission-Water (GCOM-W1) Satellite, launched in 2012,
59 the Advanced Scatterometer (ASCAT) launched in 2006 on the EUMETSAT MetOp-A and
60 MetOp-B satellites, the ESA Sentinel-1 Satellite launched in 2014 (Bartalis et al., 2007), and
61 NASA’s Soil Moisture Active-Passive (SMAP) Satellite, launched in 2015 (Entekhabi et al.,
62 2010). These microwave satellites yield the most accurate measurements of soil moisture
63 (Mattikalli et al., 1998) because of the high dielectric permittivity of soil water within the
64 microwave domain (Hallikainen et al., 1985) and the greater penetration of microwaves through
65 vegetation canopy and underlying soil (Escorihuela et al., 2010; Chapin et al., 2012;
66 Tabatabaeenejad et al., 2015).
67 Passive microwave satellites commonly provide high temporal resolution observations
68 (e.g., daily), but suffer from low spatial resolution. In this context, optical and thermal remote
69 sensing observations provide higher spatial resolution information and thus are often used to
70 enhance the passive microwave soil moisture maps through data fusion and downscaling
3
71 approaches (e.g., Piles et al., 2014; Merlin et al., 2012). Hence, development of robust optical
72 and thermal methods, in concert with microwave techniques, can improve remote sensing of soil
73 moisture.
74 The trapezoid (or triangle) model is a widely-applied approach to remote sensing of soil
75 moisture based on thermal (land surface temperature, LST) and optical data (Carlson et al. 1994;
76 Gillies and Carlson, 1995; Owen et al., 1998; Rahimzade-Bajgiran et al., 2013; Shafian and
77 Mass, 2015; Sun, 2016). Despite the triangle model’s obvious success discussed in Sadeghi et al.
78 (2017), it suffers from two inherent limitations. The first is the requirement of concurrent optical
79 and thermal data, which renders the model inapplicable to satellites that do not provide thermal
80 data, e.g. ESA Sentinel-2. The second limitation is that the land surface temperature is not only
81 affected by soil moisture content but also by ambient atmospheric conditions (e.g., wind speed,
82 air temperature, and air humidity). Hence, the conventional trapezoid model needs time
83 consuming and computationally demanding individual parameterization (calibration) for each
84 individual observation date. To overcome these two limitations, Sadeghi et al. (2017) proposed
85 the physically-based OPtical TRApezoid Model (OPTRAM) for estimation of surface soil
86 moisture. The OPTRAM trapezoid is formed by distribution of the normalized difference
87 vegetation index (NDVI) as a measure of vegetation fraction versus shortwave-infrared (SWIR)
88 transformed reflectance to obtain soil moisture content. This concept was introduced by Sadeghi
89 et al. (2015). The OPTRAM does not require a thermal band, hence, it is applicable to satellites
90 providing optical bands only (this resolves the first limitation of the conventional trapezoid
91 model). Because SWIR reflectance does not significantly change with ambient atmospheric
92 conditions, OPTRAM can be universally parameterized for a given location (this resolves the
93 second limitation of the conventional trapezoid model).
4
94 OPTRAM has been initially evaluated in Sadeghi et al. (2017) based on ESA Sentinel-2
95 and NASA Landsat-8 satellite observations for mapping of soil moisture in the Walnut Gulch
96 and Little Washita watersheds in Arizona and Oklahoma, respectively. Because Sentinel-2 was
97 only recently launched (i.e., in summer 2015), the time period covered in the Sadeghi et al.
98 (2017) study was limited to a few months in 2015 and 2016. Hence, the main motivation for this
99 current study has been to evaluate how the universal parameterization of OPTRAM (i.e., a single
100 calibration for a given area) works for long periods of time (e.g., several decades). This is
101 especially important with regard to expanding our ability to uncover the soil moisture history in
102 response to climate changes in different regions.
103 Long-term soil moisture data provide a useful measure for monitoring agricultural
104 drought (Chakrabarti et al., 2014), which is defined based on soil water deficit and its effects on
105 crop production. Recently, several studies have shown the potential of remotely sensed soil
106 moisture data for agricultural drought monitoring (Chakrabarti et al., 2014; Martinez-Fernandez
107 et al., 2016; Carrao et al., 2016; Mishra et al., 2017; Liu et al., 2017). Several soil moisture-based
108 drought indices have also been introduced for agricultural drought monitoring including the Soil
109 Moisture Index (SMI) (Sridhar et al., 2008), Soil Water Deficit (SWD) (Torres-Ruiz et al.,
110 2013), Plant Available Water (PAW) (McPherson et al., 2007), Drought Severity Index (DSI)
111 (Cammalleri and Vogt, 2016), Soil Moisture Drought Index (SMDI) (Sohrabi et al., 2015), and
112 Soil Water Deficit Index (SWDI) (Martinez-Fernandez et al., 2015). Among these drought
113 indices, SWDI is a simple agricultural drought index that is computed based on plant water
114 availability, which is assumed to not change over long time periods. The SWDI has been
115 recently used in conjunction with remotely sensed soil moisture estimates from SMAP and
116 SMOS for monitoring agricultural drought (Martinez-Fernandez et al., 2016; Mishra et al.,
5
117 2017). The application of microwave-based soil moisture retrievals (e.g., SMAP, SMOS,
118 Sentinel-1) for long-term monitoring of agricultural drought is limited due to the lack of soil
119 moisture data for long periods of time. The optical satellites such as Landsat and MODIS,
120 launched many years before SMAP, SMOS, and Sentinel-1, provide a unique opportunity for
121 such analysis in order to fill this critical observational gap.
122 The specific objectives of this study were to: (i) evaluate the universally-parameterized
123 OPTRAM for estimation of soil moisture with long-term MODIS data for four typical
124 watersheds in the United States, which exhibit diverse climates, soil types, and land covers, (ii)
125 compare the accuracy of OPTRAM soil moisture estimates with retrievals from microwave
126 satellites (i.e., SMAP, SMOS, ASCAT), and (iii) explore the feasibility of applying long-term
127 OPTRAM soil moisture data for agricultural drought monitoring.
128
129 2. Background: The Optical Trapezoid Model (OPTRAM)
130 Sadeghi et al. (2017) proposed a physically-based trapezoidal space termed the “optical
131 trapezoid model” (OPTRAM) for remote sensing of soil moisture based on optical data only. The
132 concept is based on the pixel distribution within the STR-NDVI space, where STR is the SWIR
133 transformed reflectance and NDVI is the normalized difference vegetation index, thereby
134 replacing LST in the conventional trapezoid model. Considering a linear relationship between
135 soil saturation degree, W (0 for fully dry and 1 for saturated soil) and STR (Sadeghi et al., 2015)
136 results in:
137 STR − STRd (1)

W =
STRw − STRd
138 where:
6
(1 − R ) 2
139 STR = (2)
2R
140 where STRd and STRw are STR at dry (e.g., θ ~ 0 cm3 cm-3, where θ is volumetric moisture
141 content) and wet (e.g., θ = θs cm3 cm-3, where θs is saturated moisture content) states,
142 respectively, and R is surface reflectance for the SWIR electromagnetic domain (i.e., 2130 nm,
143 MODIS band 7). Assuming empirical linear relationships of STRd and STRw with vegetation
144 fraction, the dry and wet edges of the optical trapezoid are defined as follows (see Fig. 1):
145 STRd= id + sd NDVI (3)
146 STRW= iW + sW NDVI (4)
147 where id and sd are intercept and slope of the dry edge and iw and sw are intercept and slope of the
148 wet edge.
149 Based on Eqs. (1), (3) and (4), soil moisture within a given satellite image pixel can be
150 estimated from its STR and NDVI values:
151 id + s d NDVI − STR (5)

W =
id − i w + ( s d − s w ) NDVI
152 where the saturation degree W can be expressed as θ (cm3 cm-3) multiplied with the soil porosity
153 (assumed equal to θs).
154 As discussed in Sadeghi et al. (2017), unlike the conventional trapezoidal space (LST vs.
155 NDVI) that varies over time due to variation of the ambient atmospheric parameters, the STR-
156 NDVI space is nearly time-invariant because surface reflectance is only dependent on surface
157 properties and not on the ambient atmospheric parameters. Therefore, we hypothesize that a
158 universal parameterization of Eq. (5) is possible with prolonged time series of satellite
159 observations for any specific location.
7
160
161 Fig. 1. The optical trapezoid model (OPTRAM) relating STR [Eq. (2)] and NDVI.
162
163
164 3. Materials and Methods
165 3.1. Study Areas
166 The OPTRAM was evaluated for four watersheds in the United States with diverse
167 climates, soil types and land covers, where existing cosmic-ray neutron (CRN) soil moisture
168 observations were used as reference for soil moisture data (Fig. 2). Characteristics of the CRN
169 sites within the four watersheds are summarized in Table 2. A brief introduction of these
170 watersheds follows.
171
172
173
8
174
175 Fig. 2. Geographical locations of the watersheds considered for evaluation of OPTRAM. Landsat 8
176 shortwave infrared, near infrared, and green bands (September 2018) were used to create false color
177 composite images (from bands 7-5-3) in which land surfaces covered with vegetation appear deep green.
178 The locations of the cosmic-ray neutron soil moisture sites are marked with filled black circles.
179
180 Table 1. Characteristics of the CRN sites within the four selected watersheds.
9
Temporal Lattice Soil organic
Watershed’s name CRN site Latitude Longitude
coverage water (%) carbon (%)
Walnut Gulch Lucky 13-Mar-2012 to
31.744 -110.052 1.50 3.65
(WGW) Hills 31-Jan-2016
02-Jun-2010 to
Kendal 31.737 -109.942 2.40 0.80
7-Jan-2017
Willow Creek Tonzi 10-May-2011 to
38.432 -120.966 6.05 0.55
(WCW) Ranch 31-May-2017
Reynolds Creek Reynolds 10-Aug-2011 to
43.121 -116.723 5.70 2.62
(RCW) Creek 31-May-2017
Big Cypress Creek 19-Sep-2011 to
JERC 31.236 -84.462 1.08 0.66
(BCCW) 16-May-2017
181
182 3.1.1. Walnut Gulch, Arizona
183 The Walnut Gulch Watershed (WGW), located in southeastern Arizona (AZ), with an
184 area of 148 km2 is one of the most intensively instrumented semi-arid rangeland watersheds in
185 the world. It was developed as a research watershed for hydrologic and atmospheric sciences by
186 the United States Department of Agriculture (USDA) in 1953. The watershed has been
187 frequently used as a core validation site for aircraft- and satellite-based remote sensing of soil
188 moisture. The WGW includes two CRN sites: ‘Lucky Hills’ and ‘Kendal’. They are dominantly
189 covered with desert shrubs (two thirds) and desert grasses (one third). The soils are mostly sandy
190 loam and gravelly loam with little organic matter content. The climate is semi-arid with an
191 average annual temperature 17.7 ˚C and an annual average precipitation of 350 mm; about two-
192 thirds commonly falls during the summer monsoon season (July to September). The potential
193 evapotranspiration is approximately ten times the annual precipitation. The topography is gently
194 rolling hills incised by steep drainage channels (Renard et al., 1993; Keefer et al., 2008).
195
196 3.1.2. Willow Creek, California
10
197 The Willow Creek Watershed (WCW) with an area of 58 km2 is located East of
198 Sacramento, California (CA). The watershed encompasses the ‘Tonzi Ranch’ CRN site that has
199 been established in a fairly flat area and is covered with oak-grass savanna. Tonzi Ranch is one
200 of the SMAP validation sites. Sandy clay loam with low organic matter content is the dominant
201 soil texture. The land surface is covered with mixed forests, grasslands and shrublands. The
202 typical land forms consist of plains, shallow valleys, and hills. The elevation ranges from 50 to
203 450 m above sea level. The climate is Mediterranean with an average annual temperature of 16˚C
204 and an average annual precipitation of 560 mm mostly falling between November and March.
205
206 3.1.3. Reynolds Creek, Idaho
207 The USDA Reynolds Creek Experimental Watershed (RCW), located in Southwestern
208 Idaho (ID), is a key watershed that has been established to support research addressing issues of
209 water quantity and quality, seasonal snow, and rangeland hydrology of the interior Pacific
210 northwest. The RCW is equipped with one CRN station and very well instrumented with neutron
211 access tubes, weirs, fiber optic temperature cables, and snow survey stakes. It is one of the core
212 validation sites for remote sensing of soil moisture based on SMAP data. The climate in the
213 region varies from arid to temperate with an average annual precipitation ranging from less than
214 250 mm in the North part of the watershed to more than 1150 mm at the highest elevation in the
215 Southwest. Annual snow varies from approximately 20% of the annual precipitation in the lower
216 elevations to more than 70% at the higher elevations. The average annual temperature ranges
217 from 7.2 ˚C in low elevations to 3.9 ˚C in high elevations. The land use is dominated by rangland
218 that is covered with mixed sagebrush shrubs and grasses. The topography is fairly steep. The
11
219 Western part of the watershed is sparsly vegetated with very rocky soils, while the Eastern part
220 exhibits silty soils with dense vegetation cover.
221
222 3.1.4. Big Cypress Creek, Georgia
223 The Big Cypress Creek Watershed (BCCW) is located in the Southwest of Georgia (GA).
224 It covers the northern part of the Lower Flint River and the Lower Big Slough watersheds and
225 the western part of the Horseshoe Bend-flint River watershed. The BCCW encompasses the
226 ‘Jones Ecological Research Center’ (JERC) CRN site. The climate in the region is humid
227 subtropical. The average annual precipitation ranges from 1140 mm in the eastern part to 1400
228 mm in the southern part of the watershed. The average annual temperature ranges from 15.5 ˚C
229 in the North to 21 ˚C in the South (USGS, 1986). The region in entirely flat with mixed land
230 covers including evergreen forests with grass understory and agricultural lands. The predominant
231 soil type is very well drained deep sand with deep water table.
232
233 3.2. Reference Soil Moisture Data
234 As mentioned earlier, CRN observations were used as the reference soil moisture data.
235 The CRN method (Zreda et al., 2008) provides a unique opportunity to non-invasively measure
236 effective soil moisture within the topsoil and to fill the scale gap between point-scale in situ data
237 and large-scale remote sensing observations. Given a footprint area close to that of MODIS
238 (detailed below), the CRN data provide an excellent means for evaluation of OPTRAM soil
239 moisture estimates from MODIS observations.
240
12
241 Based on neutron transport simulations, Zreda et al. (2008) found that 86% of the neutron
242 signal occurs within a 335-m radius at sea level, which is nearly independent of the soil moisture
243 level. They also found that the vertical extent of the neutron signal depends on soil moisture
244 content and ranges from about 15 cm in saturated soils to about 70 cm in very dry soils. Recent
245 particle transport simulations show a similar footprint area in the order several hundred meters
246 albeit with more spatial sensitivity of soil moisture within the footprint (Desilets and Zreda,
247 2013; Kohli et al., 2015). The exact CRN footprint and sensitivity are still an active area of
248 research. The CRN data have been recently used for validation of spaceborne surface soil
249 moisture products such as SMAP, SMOS, and ASCAT for COSMOS sites in the United States,
250 Australia, Europe and Africa (Montzka et al., 2017; Akbar and Moghaddam, 2015; Fascetti et al.,
251 2016; van der Schalie et al., 2016). Currently, there are approximately 194 permanent CRN
252 stations worldwide (Andreasen et al. 2017), including 73 COSMOS stations in the U.S. (Zreda et
253 al., 2012), 20 TERENO stations in Germany (Baatz et al., 2015), 13 CosmOz stations in
254 Australia (Hawdon et al., 2014), and 32 COSMOS-UK stations in the United Kingdom (Evans et
255 al., 2016) all of which provide excellent data for remotely sensed soil moisture validation and
256 evaluation experiments.
257 The CRN detects the ambient low-energy neutron density in the air. The low-energy
258 neutrons are highly sensitive to the mass of hydrogen and thus soil water near the ground surface
259 (Zreda et al., 2012; Avery et al., 2016). A calibration function that relates the change in low-
260 energy neutrons to the change in hydrogen content is commonly developed for estimation of soil
261 moisture. The CRN is also affected by other hydrogen sources (e.g., atmospheric water vapor,
262 soil organic matter, and mineral lattice water), and thus may provide soil moisture values
263 exceeding the soil porosity, especially in very humid regions (Franz et al., 2012; Bogena et al.,
13
264 2013; Baatz et al. 2015). In such cases, the influence of additional sources of hydrogen should be
265 accounted for in the CRN soil moisture calibration. For this study, we corrected the CRN soil
266 moisture data based on the method proposed in Franz et al. (2013) and Bogena et al. (2013) as
267 follows:
 
 
=
268 θ v ρb  0.0808
− 0.115 − θlw − θ soceq  (6)
  N pih  
  − 0.372 
  N 0  
269 where θv is the volumetric moisture content (cm3 cm-3), θlw is lattice water content (g g-1), θsoceq is
270 soil organic carbon water content equivalent (g g-1), Npih is the corrected neutron intensity
271 (moderated neutron counts per time interval, cph), N0 is a specific calibrated parameter that
272 represents the count rate over dry soils (cph), ρb is the dry soil bulk density (g cm-3). The θlw and
273 θsoceq data were obtained from calibration datasets provided on the COSMOS data portal
274 (http://cosmos.hwr.arizona.edu). More details about CRN calibration and accounting for various
275 hydrogen sources are provided in Franz et al. (2012), Rosolem et al. (2013), Bogena et al.
276 (2013), and Lv et al. (2014).
277 The precipitation data were extracted from the closest SNOwpack TELemetry (SNOTEL) and
278 Soil Climate Analysis Network (SCAN) sites.
279
280 3.3. Satellite Data
281 3.3.1. MODIS
282 The Moderate Resolution Imaging Spectroradiometer (MODIS) is a key instrument on
283 board of NASA’s Terra (EOS AM-1) and Aqua (EOS PM-1) satellites that plays an important
284 role in studying land surface properties and processes and development of models to predict
14
285 global change. Terra and Aqua MODIS satellites capture the entire Earth surface every 1 to 2
286 days and acquire data in 36 individual spectral bands (400-14400 nm) at resolutions of 250 m
287 (bands 1-2), 500 m (bands 3-7), and 1000 m (bands 8-36). In this study, long-term cloud-free
288 MODIS land surface reflectance images (Terra MOD09A1 version 6) from 2001 to 2017 were
289 acquired and analyzed. This product provides spectral reflectance at bands 1-7 with 500 m
290 spatial resolution and 8-day temporal resolution as a gridded level-3 product in the Sinusoidal
291 projection. Each MOD09A1 pixel contains the best possible observation during an 8-day period
292 as selected on the basis of high observation coverage, low view angle, the absence of clouds and
293 cloud shadows, and aerosols. Reflectance values from bands 1 (Red) and 2 (Near Infrared, NIR)
294 were used to calculate NDVI = (NIR – Red)/(NIR + Red) and reflectance from band 7 (i.e.,
295 SWIR, 2130 nm) was used for calculating STR with Eq. (2). The main reason for selecting band
296 7 reflectance is its stronger linear correlation with soil moisture content when compared with
297 other SWIR wavelengths (e.g., 1240 and 1640 nm corresponding to MODIS bands 5 and 6), as
298 discussed in Sadeghi et al. (2015). Because MODIS SWIR band 7 used in Eq. (2) is sufficiently
299 far from the water vapor absorption bands in the SWIR region (i.e., 1400 and 1900 nm), the
300 effects of atmospheric water vapor can be neglected. The images were analyzed and projected
301 into WGS84 geographic coordinates using the HDF-EOS 2.14 tool (http://hdfeos.org/) and
302 MATLAB R2015b (Math Works Inc., Natick, MA) software.
303
304 3.3.2. SMAP
305 The L-band Soil Moisture Active and Passive (SMAP) satellite has been designed and
306 launched by NASA with the major aim of measuring and monitoring soil moisture changes
307 (Entekhabi et al., 2010). The SMAP released 7 advanced soil moisture products as levels 2, 3,
15
308 and 4 that provide surface and root zone soil moisture with spatial resolutions of 3, 9, and 36 km
309 and temporal resolutions of 3 and 24 hours. In this study, the level-3 product (L3_SM_P,
310 descending 6:00 a.m.), derived from the single channel algorithm V-Pol (SCA-V) and resampled
311 to a global and cylindrical 36 km resolution with Equal-Area Scalable Earth Grid (EASE-Grid
312 2.0) global projection, was applied and compared with CRN reference data to further evaluate
313 the MODIS-based estimates of soil moisture with OPTRAM. This level-3 product provides a
314 daily composite of half-orbit/swath estimates of global surface soil moisture (0-5 cm3 cm-3)
315 derived from the brightness temperature data measured by the passive microwave radiometer.
316
317 3.3.3. SMOS
318 Launched in 2009, the ESA Soil Moisture and Ocean Salinity (SMOS) satellite is the first
319 Earth observation mission dedicated to map and monitor surface soil moisture (Kerr et al., 2001).
320 To retrieve surface soil moisture (0-5 cm3 cm-3) the SMOS campaign uses the L-band microwave
321 emissions of the biosphere (L-MEB) radiative transfer model (Wigneron et al., 2007) and a
322 dielectric mixing model (Mironov et al., 2009). The SMOS level-2 and level-3 soil moisture
323 products have been successfully evaluated based on ground and satellite soil moisture datasets. A
324 recently released daily product – SMOS-INRA-CESBIO (SMOS-IC) – has been designed by the
325 French Institut National de la Recherche Agronomique (INRA) in collaboration with the Centre
326 d'Etudes Spatiales de la Biosphère (CESBIO) to provide global estimations of soil moisture.
327 The SMOS-IC algorithm is based on L-MEB, but simpler than the SMOS level-2 and
328 level-3 soil moisture retrieval algorithms. It is more independent from uncertainties of auxiliary
329 data, hence, provides more robust results. In contrast to the SMOS level-2 and level-3 algorithms
330 that correct for pixel heterogeneity, pixels are assumed homogeneous in SMOS-IC (Wigneron et
16
331 al., 2012; Fernandez-Moran et al., 2017). The SMOS-IC soil moisture maps are generated as
332 global products with a 25 km (EASE-Grid 2.0) grid size. In the current study, the daily SMOS-
333 IC soil moisture products (https://www.catds.fr/Products) from ascending orbit were used and
334 compared with the CRN reference data to further evaluate the MODIS-based estimates of soil
335 moisture with OPTRAM.
336
337 3.3.4. ASCAT
338 The Advanced Scatterometer (ASCAT) is an active C-band microwave sensor on the
339 METOP-A/B satellite platforms, launched in 2006. The sensor was initially designed for
340 monitoring wind speed and wind direction over the oceans, but it has been widely used for
341 monitoring soil moisture in terms of relative saturation (Wagner et al., 2013). The relative
342 saturation is estimated based on a simple linear change detection retrieval algorithm which is part
343 of the Water Retrieval Package (WARP) software (Naeimi et al., 2009). A shortcoming of this
344 algorithm is that it considers a linear relationship between the backscatter coefficient (dB) and
345 surface soil moisture, while assuming that surface roughness and land cover are constant in time.
346 While ASCAT provides accurate estimates of soil moisture for bare and sparsely vegetated soils,
347 retrievals for densely vegetated regions (e.g., tropical forests) are associated with significant
348 errors, since backscatter from soil and vegetation cannot be separated (Wagner et al., 2013). The
349 EUMETSAT has generated several ASCAT surface and root zone soil moisture products with
350 daily global coverage. In the current study, we used the H109 and H110 soil moisture products
351 from 2010 to 2016, with 12.5 km spatial sampling. The relative soil saturation was converted to
352 volumetric moisture content via multiplying with the soil porosity found in the Harmonized
353 World Soil Database (HWSD, Nachtergaele et al., 2012). Soil moisture data with noise
17
354 designation above 20% were excluded from our analyses, while all remaining data were
355 compared with the CRN reference data (Montzka et al., 2017).
356
357 3.4. Data Analyses
358 3.4.1. OPTRAM Parameterization
359 The dry and wet edges of OPTRAM (i.e., parameters of Eq. 5) were determined based on the
360 pixel distribution within the STR-NDVI space. The dry (id and sd) and wet (iw and sw) edge
361 parameters were determined via fitting of a straight line to the STR-NDVI point clouds. Because
362 of the feasibility of universal parameterization of the OPTRAM model (Sadeghi et al., 2017), an
363 integrated STR-NDVI trapezoid including several hundred images acquired in the period from
364 2001 to 2017 (see Table 2), was applied for each watershed. This approach aided in further
365 reduction of time-dependency of the dry and wet edge parameters and provided “effective”
366 model parameters for each watershed. The saturation degree (W) was mapped with Eq. (5) for
367 each date and watershed. The estimated W values for pixels containing the CRN sites were
368 converted to volumetric moisture contents by multiplying with the soil porosity values from the
369 HWSD.
370 In our analyses, pixels with negative NDVI values were filtered to omit pixels without vegetation
371 or soil (e.g., water surfaces, snow, rock). While Sadeghi et al. (2017) resampled Landsat-8 and
372 Sentinel-2 images to 120-m pixel size to remove effects of oversaturated pixels, here we
373 employed the original MODIS pixels, because oversaturated pixels are unlikely at 500-m spatial
374 resolution, unless they cover surface water bodies. The resultant optical trapezoids for the four
375 investigated watersheds (shown later in Fig. 3) verified this assumption, as the wet edge data
376 were obtained within the expected STR range for saturated/wet land.
18
377 As suggested by Sadeghi et al. (2017), fitting the edges of the optical trapezoids was done based
378 on visual inspection, rather than least-square regression, mainly to exclude the scattered points
379 around the main body of the trapezoids. The visual fitting, however, introduces user bias and
380 leads to some degree of uncertainty of model outputs. To investigate the effects of visual fitting
381 on soil moisture results we performed a sensitivity analysis, varying the initially visually-fitted
382 parameters by adding randomly-generated errors in the range of ±%5, ±%10, ±%15, and ±%20
383 of the initial values, where 4 replicates (R1 to R4 in Fig. 4) were considered for each level. To
384 quantify the uncertainties due to visual fitting soil moisture values calculated with perturbed
385 parameters were compared to those obtained with the initial parameters (see section 3.2).
386
387 3.4.2. Rescaling Soil Moisture Data
388 Soil moisture data from different sources (e.g., satellite vs. in situ, or varied satellites)
389 typically exhibit consistent temporal evolution, but different mean and standard deviation values.
390 Hence, rescaling data through matching the cumulative probability (CP) or cumulative
391 distribution function (CDF) has been recommended for reducing systematic biases between
392 different soil moisture datasets (Reichle and Koster, 2004; Koster et al., 2009; Brocca et al.,
393 2011; Mishra et al., 2017). The CDF matching removes impacts of time-invariant errors from
394 comparison between estimated and measured soil moisture and provides an objective basis for
395 intercomparisons of anomalies. In this paper, we followed the strategy of Reichle and Koster
396 (2004) and matched the CDF of remotely sensed soil moisture data to the CRN reference data
397 (Crow et al., 2005; Kumar et al., 2012). Accordingly, we calculated the bias-corrected (rescaled)
398 remotely sensed soil moisture, θRS, from original remotely sensed soil moisture, θ’RS, as follows:
σ CRN
399 ′ − µ RS )
θ RS = µCRN + (θ RS (7)
σ RS
19
400
401 where μ and σ represent the mean and standard deviation of soil moisture data, respectively, for a
402 given location for the study period, and subscripts RS and CRN denote the corresponding data
403 sources.
404
405 3.4.3. Performance Metrics
406 For each comparison of pairs of remotely sensed and reference (CRN) soil moisture
407 observations, the correlation coefficient (R), the root mean squared error (RMSE), bias, and
408 unbiased RMSE (ubRMSE) were applied as follows:
Cov(θ RS ,θ CRN )
409 R= (8)
σ RS σ CRN
1
∑ (θ RS − θ CRN ) i2
N
410 RMSE = i =1
(9)
N
1
∑ (θ RS − θ CRN )i
N
411 bias = i =1
(10)
N
412 ubRMSE = RMSE 2 − bias 2 (11)
413 where i denotes a given pair of estimates and measurements and N is the total number of paired
414 estimates and measurements. Because the CRN stations are not exactly aligned with the centers
415 of satellite RS pixels (MODIS, SMAP, SMOS, and ASCAT), all pixels covering at least 50% of
416 the CRN footprint were used for calculating these metrics.
417
418
419 3.4.4. Drought Analysis
20
420 Most of the proposed drought assessment indices are based on long-term atmospheric
421 variables (Palmer, 1965; McKee et al., 1993; Hogg, 1994), but they do not consider soil moisture
422 measurements, which are a direct measure of agricultural drought. One reason for this negligence
423 is probably due to a lack of soil moisture monitoring networks in the past. Today, satellite data
424 can provide soil moisture information with a wide range of spatial and temporal resolutions.
425 MODIS satellite data provide an excellent long-term database for drought monitoring
426 applications. With respect to its definition, agricultural drought is defined based on soil moisture
427 deficit and its effect on crop production, i.e., when evapotranspirative losses exceed the plant
428 available soil moisture defined as the difference between soil moisture at field capacity and that
429 at permanent wilting point (Kabat and Beekman, 1994). Martinez-Fernandez et al. (2015)
430 introduced a Soil Water Deficit Index (SWDI) for agricultural drought monitoring as follows:
θ − θ FC
431 SWDI = 10 (12)
θ FC − θ PWP
432 where θFC and θPWP (cm3 cm-3) are soil moisture content at field capacity and permanent wilting
433 point, respectively. Positive values of SWDI signify excess water in soil, while negative values
434 indicate some degree of soil drought. When SWDI is equal to zero, it means the soil is at field
435 capacity moisture content. The values equal to or smaller than -10 indicate that the moisture
436 content is below the lower limit of available water for plants, i.e., extreme water deficit in soil
437 (Martinez-Fernandez et al., 2015). As mentioned above, MODIS-OPTRAM soil moisture
438 estimates can provide a long-term dataset for agricultural drought monitoring based on
439 calculating SWDI (i.e., OPTRAM-SWDI). The θFC and θPWP were determined with the
440 ROSETTA pedotransfer model (Schaap et al., 2001) that uses clay content, sand content, silt
441 content, and bulk density data for parameterizing a soil water retention model (i.e., van
442 Genuchten, 1980) and thereby determining θFC and θPWP. These basic soil properties were
21
443 extracted for each site from the global ISRIC Soil database (https://soilgrids.org) with a spatial
444 resolution of 250 m.
445 The Crop Moisture Index (CMI) (Palmer, 1968) which is related to soil water storage was
446 used as reference to assess the accuracy of the OPTRAM-SWDI for agricultural drought
447 monitoring. The daily CMI is inversely calculated based on the subtraction of daily potential
448 evapotranspiration from daily precipitation (P − ET0), where the daily ET0 was calculated based
449 on the Thornthwaite method for each site (Thornthwaite, 1948; Rim, 2000).
450
451 4. Results and Discussion
452 4.1. OPTRAM Parameters
453 Figure 3 illustrates the pixel distribution within the STR-NDVI space for each watershed.
454 The black lines represent the optimized dry and wet edges and the colored lines represent the
455 randomly perturbed edges used for sensitivity analysis. As observed, a nearly trapezoidal shape
456 is formed by STR-NDVI point clouds for each watershed based on the integration of all MODIS
457 images listed in Table 2. Each of the four feature spaces exhibit a wide range of STR and NDVI
458 values that may be due to the variability of soil moisture, diverse climate, soil type, and land
459 cover. In general, STR and NDVI values are larger in regions with dense vegetation cover, e.g.,
460 Willow Creek and Big Cypress Creek watersheds (see also Fig. 2), which is due to high soil
461 moisture and/or vegetation water content. The larger STR values in densely vegetated soils
462 indicate higher moisture contents in the root zone. The visually optimized parameters for the
463 dry and wet edges for each watershed are listed in Table 2. Using these parameters, W in Eq. (5)
464 can be easily computed for any value of STR and NDVI.
465
22
466 Table 2. The dry and wet edge parameters [Eq. (5)] obtained for each watershed based on
467 long-term MODIS satellite data.
# MODIS Dry edge Wet edge
Watershed, State
images id sd iw sw
Walnut Gulch (WGW), AZ 543 0.10 2.60 2.50 1.70
Willow Creek (WCW), CA 750 0.00 2.80 1.50 10.0
Reynolds Creek (RCW), ID 450 0.23 1.10 3.00 5.50
Big Cypress Creek (BCCW), GA 560 0.00 0.70 0.00 19.0
468
469
470
471
23
472
473 Fig. 3. Pixel distribution within the STR-NDVI space formed with long-term MODIS data (2001-
474 2017) for the Walnut Gulch Watershed (WGW), Willow Creek Watershed (WCW), Reynolds
475 Creek Watershed (RCW), and Big Cypress Creek Watershed (BCCW). The solid and dashed
476 black lines represent the best visually fitted wet and dry edges, respectively. The colored lines,
477 used for sensitivity analysis, show randomly varied edges at levels of ±%5 (magenta with 4
478 replicates), ±%10 (red with 4 replicates), ±%15 (yellow with 4 replicates), and ±%20 (green with
479 4 replicates).
480
481 4.2. Sensitivity of OPTRAM Outputs to Model Parameters
482 Sensitivity of OPTRAM-based soil moisture estimates at CRN pixels to the model
483 parameters is shown in Fig. 4, where correlations between soil moistures estimates using the best
24
484 visually fitted parameters and the randomly perturbed parameters are shown. In general, strong
485 correlations are observed for all the uncertainty levels. The average RMSE values at levels ±%5,
486 ±%10, ±%15, and ±%20 for all the sites are 0.007, 0.010, 0.012, and 0.025 cm3 cm-3,
487 respectively, indicating that the OPTRAM outputs are not very sensitive to the model
488 parameters. Based on Fig. 4, uncertainties due to the visual fitting approach are anticipated to be
489 significantly less than the errors of optical or microwave remote sensing approaches (presented
490 later). In other words, the visual fitting approach results in acceptable soil moisture estimates,
491 despite its perceived user bias.
492
493
25
0.4 0.4 0.4 0.4
± %5 ± %10 ± %15 ± %20
0.3 R1 0.3 0.3 0.3
R2
R3
0.2 R4 0.2 0.2 0.2
WGW WGW
0.1 0.1 0.1 WGW 0.1 WGW
RMSE=0.0035 RMSE=0.0078 RMSE=0.0133 RMSE=0.0230

OPTRAM Soil Moisture using Randomly Perturbed Parameters (cm3 cm-3)
0.0 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
0.4 0.4 0.4 0.4
± %5 ± %10 ± %15 ± %20
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2
WCW WCW WCW

0.1 0.1
WCW
0.1 0.1
RMSE=0.0032 RMSE=0.0080 RMSE=0.0070

0.0 0.0
RMSE=0.0208
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
0.4 0.4 0.4 0.4
± %5 ± %10 ± %15 ± %20
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2
RCW RCW RCW RCW

0.1 0.1 0.1 0.1

0.0 0.0 0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
0.4 0.4 0.4 0.4
± %5 ± %10 ± %15 ± %20
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2
BCCW BCCW
0.1 0.1 0.1 BCCW 0.1 BCCW

0.0 0.0 0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
494 OPTRAM Soil Moisture using the Best Visually Fitted Parameters (cm3 cm-3)
495 Fig. 4. OPTRAM soil moisture estimates at CRN sites using the best visually fitted parameters
496 (Table 2) against randomly perturbed parameters used for sensitivity analysis for ranges of ±%5,
497 ±%10, ±%15, and ±%20. Each row of plots shows data from a different site: Kendal, AZ
498 (WGW), Tonzi Ranch, CA (WCW), Reynolds Creek, ID (RCW), and JERC, GA (BCCW). The
499 root mean squared error (RMSE) denotes averaged values of the RMSE from 4 replicate data
500 sets shown as different colors.
501
502
503
504
26
505 4.3. Rescaled CDFs
506 Comparisons between the original and rescaled CDFs of CRN reference soil moisture
507 data and soil moisture estimates from OPTRAM, SMAP, SMOS, and ASCAT for each site and
508 for all dates are illustrated in Fig. 5. A large mismatch is shown between the original CDFs,
509 while the rescaled CDFs coalesced well. This result indicates that the different sources resulted
510 in different, but highly-correlated, soil moisture values. One probable reason for the observed
511 mismatch could be the spatial scale mismatch, i.e., different image pixel size and sensing
512 volume. As seen in Fig. 5, ASCAT tends to produce lower soil moisture values for most sites.
513 For example, at the Kendal and Lucky Hills sites, the ASCAT soil moisture estimates indicate
514 very dry conditions (<0.05 cm3 cm-3) during most of the dry season, while OPTRAM never
515 reached values that low. The highest soil moisture values in most sites were reported by either
516 OPTRAM or SMAP, which could be due to their deeper sensing depth, i.e., deeper penetration of
517 SMAP’s L-band and OPTRAM’s responsiveness to plant canopy-derived root zone soil
518 moisture. As mentioned, the rescaled CDFs for the remotely sensed soil moisture data were well
519 matched to CRN soil moisture data at all four sites, leading to reduced error metrics such as
520 RMSE and bias. However, it should be noted that the correlation coefficient (R) will not change
521 due to CDF matching, as CDF matching is basically a linear transformation of the original data
522 and has nothing to do with random variations that determine R.
523
524
27
1.0 1.0
0.8 CRN
0.8
CRN
OPTRAM (original) OPTRAM (rescaled)

0.6 0.6
SMAP (original) SMAP (rescaled)
SMOS (original) 0.4 SMOS (rescaled)

0.4
ASCAT (original) ASCATP (rescaled)
0.2 0.2
Kendal, AZ
Kendal, AZ
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
Lucky Hills, AZ
Lucky Hills, AZ
1.0 1.0
Cumulative Probability
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
Tonzi Ranch, CA Tonzi Ranch, CA
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
Reynolds Creek, ID Reynolds Creek, ID
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
JERC, GA JERC, GA
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
Soil Moisture (cm3 cm-3)

525
526 Fig. 5. Cumulative Probability for matching the CDFs based on CRN data, original (left column) and
527 rescaled (right column) remote sensing-based estimates of soil moisture from OPTRAM, SMAP, SMOS,
528 and ASCAT for sites Kendal, AZ; Lucky Hills, AZ; Tonzi Ranch, CA; Reynolds Creek, ID; and JERC,
529 GA.
530
28
531 4.4. Soil Moisture Estimates
532 Figure 6 depicts a comparison of OPTRAM-derived soil moisture daily temporal
533 dynamics with those of CRN, SMAP, SMOS and ASCAT from 2010 to 2017. The OPTRAM
534 estimates exhibit distinct dry and wet cycles and capture the dynamics of CRN soil moisture at
535 the five selected sites, especially at Tonzi Ranch and Reynolds Creek, indicating good
536 performance of OPTRAM (mean ubRMSE value of 0.0675 and mean R value of 0.3940). Data
537 shown in Fig. 6 clearly demonstrate distinct differences in climate, land cover and soil variability
538 among the four watersheds. Compared to the other three sites, the range of OPTRAM and CRN
539 soil moisture is limited at Kendal and Lucky Hills (Walnut Gulch) sites because these two sites
540 have a semi-arid climate with well-drained coarse-textured soils and limited precipitation. A
541 wider range of soil moisture content is evident for Tonzi Ranch and Reynolds Creek, which is
542 associated with more frequent precipitation events during each season and also the existence of
543 clay loam and silty soils across these two sites. In addition, a wider range of OPTRAM soil
544 moisture compared to CRN observations is also evident at each site during the studied period,
545 especially for humid climates. This is partly due to the higher noise of the satellite-based
546 estimates compared with the noise of ground (CRN) measurements. Another physical reason,
547 especially for bare soils, is that OPTRAM-based estimates represent the surface (skin) soil
548 moisture for a thin layer with a few mm depth, whereas CRN data provide effective soil moisture
549 down to about 50 cm for nearly dry soils (Zreda et al., 2012). This fact results in remotely sensed
550 soil moisture values to be more responsive to atmospheric factors (i.e., precipitation, radiation,
551 and evaporative demand), explaining the high temporal variability and dynamic range of surface
552 soil moisture. The CRN soil moisture values are generally higher during summer and dry
29
553 periods. This result is mainly due to the neglected effect of high biomass growth rate affecting
554 CRN-based soil moisture (Andreasen et al., 2017).
555
OPTRAM CRNP SMAP SMOS ASCAT Precipitation
0
0.4
0.3
50
0.2
0.1 100
0.0
Arizona (Walnut Gulch, Kendal)
150
01-Jan-10 01-Jan-11 01-Jan-12 01-Jan-13 01-Jan-14 01-Jan-15 01-Jan-16 01-Jan-17
0
0.4
0.3
50
0.2
0.1 100
0.0
Arizona (Walnut Gulch, Lucky Hills)
150
0.5 0
Soil moisture (cm cm )
-3
0.4
Precip (mm d-1)

50
3
0.3
100
0.2
0.1 150
0.0 California (Willow Creek, Tonzi Ranch)

200
0
0.5
0.4
50
0.3
0.2
100
0.1
0.0 Idaho (Reynolds Creek)

150
0
0.3
0.2 100
0.1
200
0.0
Georgia (Big Cypress Creek, JERC)
300
556 Date (dd-m-yy)
30
557 Fig. 6. Time series of remotely sensed (OPTRAM, SMAP, SMOS, ASCAT) and CRN-based soil
558 moisture and precipitation at Kendal, AZ; Lucky Hills, AZ; Tonzi Ranch, CA; Reynolds Creek, ID,
559 and JERC, GA. The precipitation data were extracted from SNOTEL (SNOwpack TELemetry) and
560 SCAN (Soil Climate Analysis Network) sites closest to each CRN station. Note that for the Walnut
561 Gulch sites, no precipitation was recorded by the SCAN rain gauges from August 2012 to
562 November 2013.
563
564 In Fig. 6, rescaled OPTRAM soil moisture estimates are also compared with rescaled soil
565 moisture retrievals from SMAP, SMOS, and ASCAT. As seen, OPTRAM estimates follow the
566 seasonal variations of SMAP, SMOS, and ASCAT retrievals. Although a similar overall pattern
567 is observed, the OPTRAM values are generally lower, especially during warmer periods. During
568 or shortly after a precipitation event, OPTRAM data are quite noisy which may be due to low
569 quality MODIS optical images resulting from cloud and atmospheric water vapor effects.
570 The original ASCAT, SMOS, and SMAP soil moisture retrievals (not presented here)
571 were much more variable showing a wider range of values than those of OPTRAM and CRN for
572 most sites. This greater variability may primarily be due to the different sensing depths of
573 microwaves, which are known to be shallower for shorter wavelengths. For instance, C-band
574 ASCAT is sensitive down to a 2 cm depth and can be even less than 1 cm during a strong rainfall
575 event, whereas L-band SMOS penetrates to about 5 cm (Al-Yaari et al., 2014). Therefore, during
576 and shortly after a precipitation event, microwave soil moisture retrievals associated with a very
577 wet surface layer show higher soil moisture values than CRN data, whose penetration depth
578 exceeds 15 cm. This discrepancy between microwave and CRN data is reduced with time after
579 precipitation because the soil profile tends toward a more uniform moisture distribution. Another
580 reason for the observed discrepancies could be due to high spatial soil moisture variability after
581 intense precipitation events (also spatially variable), magnifying effects of the scale mismatch
582 between lower resolution microwave and higher resolution CRN observations. For some cases,
583 for example the Arizona sites, a time lag is observed for ASCAT soil moisture data, when
31
584 moisture values fall below 0.20 cm3 cm-3. This time lag can be attributed to the temporal
585 variations of surface roughness and land cover, which are assumed to be constant in the ASCAT
586 retrieval algorithm (WARP) (Wagner et al., 2013). The lag over summer can be related to the
587 increase in vegetation cover and thereby decrease in soil moisture content due to plant water
588 uptake.
589 Scatterplots between CRN-derived and remotely-sensed soil moisture estimates for each
590 site are presented in Fig. 7. Correlations between the CRN estimates and rescaled OPTRAM and
591 microwave satellite soil moisture estimates range between 0.100 and 0.927. Based on RMSE
592 estimates, SMAP and SMOS performed better than OPTRAM and ASCAT in most sites.. This
593 was generally expected since microwave signals at lower frequencies (L-band) are much more
594 sensitive to soil water (i.e., dielectric constant) and microwave-based retrievals provide more
595 accurate soil moisture estimates than optical methods. In our study, comparison of OPTRAM
596 estimates with ASCAT retrievals (C-band) over all sites indicate that OPTRAM performed
597 similar to or better than ASCAT for all sites, except for the JERC site in GA.
598
599
600
32
OPTRAM SMAP SMOS ASCAT
0.5 0.5 0.5 0.5
RMSE: 0.0719 RMSE: 0.0437 RMSE: 0.0693 RMSE: 0.0705
0.4 0.4 0.4 0.4
Kendal, AZ
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2
0.1 0.1 0.1 0.1
0.0 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
0.4 0.4 0.4 0.4
Lucky Hills, AZ
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2

Remotely Sensed Soil Moisture (cm cm )
-3
0.1 0.1 0.1 0.1

3
0.0 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
0.5 0.5 0.5 0.5
RMSE: 0.0860 RMSE; 0.0735 RMSE: 0.0795 RMSE: 0.1127
Tonzi Ranch, CA
0.4 0.4 0.4 0.4
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2
0.1 0.1 0.1 0.1
0.0 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
0.5 0.5 0.5 0.5
Reynolds Creek, ID
0.4 0.4 0.4 0.4
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2
0.1 0.1 0.1 0.1
0.0 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
0.3 0.3 0.3 0.3
0.2 0.2 0.2 0.2
JERC, GA
0.1 0.1 0.1 0.1
0.0 0.0 0.0 0.0

0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3
3 -3
601 Measured Soil Moisture (cm cm )
602 Fig. 7. Scatterplots comparing remotely sensed (OPTRAM, SMAP, SMOS, and ASCAT) with CRN-based soil
603 moisture for Kendal, AZ; Lucky Hills, AZ; Tonzi Ranch, CA; Reynolds Creek, ID, and JERC, GA.
604
33
605 Table 3 compares soil moisture estimation performance of OPTRAM and microwave
606 satellites with CRN estimates in terms of R, RMSE, unRMSE, and bias for each site. These error
607 metrics indicate a generally good agreement between the OPTRAM and CRN estimates.
608 OPTRAM yields slightly larger RMSE values than SMAP and SMOS, but equal to or better than
609 ASCAT for most sites. At Kendal, Lucky Hills and JERC, OPTRAM performed well with RMSE
610 and ubRMSE close to ASCAT. OPTRAM yielded ubRMSE values between 0.0501 to 0.0853 cm3
611 cm-3 and R values between 0.100 and 0.746 when compared with CRN. At the JERC site in GA,
612 OPTRAM shows the best agreement with CRN, where ubRMSE was 0.0501 and bias was 0.0066
613 cm3 cm-3. At all sites except Tonzi Ranch, OPTRAM overestimated soil moisture with an
614 average bias of 0.019 cm3 cm-3. This overestimation was referred to as an ‘oversaturated issue’
615 in the original OPTRAM model development (Sadeghi et al., 2017), explained as the result of
616 temporary shallow surface ponding and/or canopy interception of water, which increase STR
617 values for a given NDVI and thereby overestimate soil moisture content. At Tonzi Ranch,
618 OPTRAM slightly underestimated soil moisture with a bias of -0.011 cm3 cm-3.
619
620
621 Table 3. Performance metrics for evaluation of OPTRAM, SMAP,
622 SMOS, and ASCAT soil moisture estimates based on CRN observations
623 at each of the investigated sites.
COSMOS
Metric OPTRAM SMAP SMOS ASCAT
site
bias 0.0186 -0.0216 0.0452 0.0015
Kendal,
ubRMSE 0.0695 0.0380 0.0525 0.0705
AZ
RMSE 0.0719 0.0437 0.0693 0.0705
R 0.3533 0.8522 0.5629 0.2629
N 201 63 67 172
bias 0.0399 0.0069 0.0102 0.0015
Lucky
ubRMSE 0.0504 0.0339 0.0561 0.0680
Hills, AZ
RMSE 0.0643 0.0346 0.0571 0.0681
R 0.3098 0.7892 0.4243 0.1609
N 108 28 37 101
bias -0.0109 0.0065 0.0327 -0.0076
Tonzi
ubRMSE 0.0853 0.0732 0.0725 0.1125
Ranch, CA
RMSE 0.0860 0.0735 0.0795 0.1127
34
R 0.461 0.927 0.7918 0.752
N 244 74 90 206
bias 0.0109 -0.0070 0.0424 -0.0094
Reynolds
ubRMSE 0.0822 0.1284 0.0582 0.0921
Creek, ID
RMSE 0.0823 0.1286 0.0720 0.0925
R 0.7460 0.5041 0.6202 0.6104
N 177 64 43 73
bias 0.0066 -0.0015 -0.0792 0.0015
JERC, GA ubRMSE 0.0501 0.0310 0.0517 0.0365
RMSE 0.0505 0.0311 0.0946 0.0366
R 0.1000 0.7829 0.7907 0.4977
N 139 36 41 109
N: the number of data pairs used to calculate metrics.
624
625 In theory, the ASCAT sensor should provide more accurate retrievals of soil moisture
626 owing to the greater sensitivity of active microwave reflections to dielectric permittivity of soil
627 water. The relatively large errors observed for ASCAT (compared to SMAP and SMOS) is
628 consistent with several studies. Peng et al. (2015) found RMSE value 0.096 and bias value -0.053
629 cm3 cm-3 for ASCAT soil moisture retrievals for southwestern China. Rotzer et al. (2014)
630 obtained RMSE values between 0.07 and 0.09 cm3 cm-3 and an R value 0.25 for ASCAT for
631 agricultural lands within the Ruhr catchment in Germany, and Djamai et al. (2015) reported
632 RMSE values ranging from 0.07 to 0.11 and R values from 0.10 to 0.37 for ASCAT for
633 agricultural land and forests in Canada. In conformity with Montzka et al. (2017), the lowest
634 error metrics are found for the SMAP and SMOS retrievals with very similar R values but
635 somewhat larger ubRMSE values, likely due to the number of data points, study period, and use
636 of corrected CRN data [see Eq. (6)]. Considering five site averages, the SMAP retrievals provide
637 the highest R (0.77) and lowest ubRMSE (0.061 cm3 cm-3) values followed by SMOS retrievals
638 (R = 0.64 and ubRMSE = 0.058 cm3 cm-3), and ASCAT products (R = 0.40 and ubRMSE = 0.076
639 cm3 cm-3). At the Kendal, Lucky Hills and JERC sites, SMAP soil moisture retrievals fall within
640 the mission target accuracy of 0.04 cm3 cm-3.
35
641 Considering data bias, SMAP tends to slightly underestimate soil moisture at Kendal,
642 Reynolds Creek, and JERC and the ASCAT underestimates soil moisture at Tonzi Ranch and
643 Reynolds Creek. Underestimations for SMAP may arise from the smaller sensing depth and
644 larger radiometer footprint compared with CRN measurements, as discussed above. For ASCAT
645 the underestimation could additionally be due to the uncertainty in soil porosity when the relative
646 saturation (values between 0 and 1) is converted to volumetric moisture content. The
647 underestimation could also be due to bias in CRN measurements. For instance, the Tonzi Ranch
648 site is located in a flat area where the CRN footprint is subject to water ponding after heavy
649 precipitation events, resulting in higher CRN soil moisture values. At the Reynolds Creek site
650 located in a mountainous area, the CRN footprint includes a surface layer of snow during the
651 winter. This can result in soil moisture estimates exceeding the soil porosity, as indicated in Fig.
652 6.
653 At the JERC site, remotely sensed data exhibit the smallest bias scores, which may arise
654 from the narrow range of soil moisture dynamics. For instance, OPTRAM shows R equal to 0.10
655 but ubRMSE equal to 0.0501 cm3 cm-3. This result indicates that the use of these standard metrics
656 is not always suitable for in-depth evaluation of soil moisture accuracy. The correlation between
657 in situ soil moisture and satellite observations is commonly not very large and a typical R range
658 between 0.40 to 0.80 is reported, but even a complete lack of correlation does not necessarily
659 mean that the satellite based estimations are wrong (Wagner et al., 2013). In such a case, as we
660 have done in this study, it is more appropriate to interpret remote sensing results in a relative
661 context for example by comparing with performance of other satellite data against the same
662 reference data rather than attributing absolute meaning to the results (Wagner et al., 2013).
663
36
664 4.5. Drought Indices
665 As mentioned earlier, long-term MODIS data can provide a historical database of soil
666 moisture estimates preceding the launch of microwave satellites. This long-term database is a
667 valuable asset for an array of applications such as agricultural drought assessment and climate
668 studies. To check the performance of OPTRAM for this application, SWDI was computed based
669 on OPTRAM estimates and compared with an atmospheric drough index, CMI. Weekly SWDI
670 temporal dynamics based on OPTRAM soil moisture estimates for our five selected sites are
671 compared with CMI values in Fig. 8 (left). In this study, the OPTRAM-SWDI approach is
672 similar to Mishra et al. (2017) and Martinez-Fernandez et al. (2016), who used SMAP and
673 SMOS soil moisture retrievals to compute SWDI to monitor agricultural drought, respectively. A
674 similar pattern is observed between OPTRAM-SWDI and CMI at all sites except the Big
675 Cypress Creek, which shows large discripancies over the whole period in terms of temporal
676 evolution and identification of drought onsets. This mismatch may arise from the narrow range
677 of soil moisture in what is a predominantly sandy textured soil of the Big Cypress Creek (see
678 Fig. 6). The relatively low water storage capacity (θFC and θPWP equal to 0.13 and 0.05
679 respectively) limits plant soil moisture availability, producing larger differences in calculated soil
680 moisture based on CMI (i.e., P − ET0). The OPTRAM-SWDI is an indicator of water availability
681 in soil, which can be used to estimate agricultural drought.
682
37
30 10 1.0
0.4
Arizona (Walnut Gulch, Kendal) Arizona (Walnut Gulch, Kendal)
Soil moisture (cm3 cm-3)

20 0.8
SWDI (OPTRAM) Soil moisture content (OPTRAM)
5 0.3
CMI
CMI (mm d-1)

10
NDVI
NDVI
0.6
SWDI
0.2
0 0
0.4
-10 0.1
-5
0.2
-20
0.0
-30 -10 0.0

01-Jan-01 01-Jan-04 01-Jan-07 01-Jan-10 01-Jan-13 01-Jan-16 01-Jan-01 01-Jan-04 01-Jan-07 01-Jan-10 01-Jan-13 01-Jan-16
30 10 1.0
0.4

20 0.8
5 0.3
CMI (mm d-1)

10
0.6
NDVI
SWDI
0.2
0 0
0.4
-10 0.1
-5
0.2
-20
0.0
-30 -10 0.0

30 10 1.0
0.4
California (Willow Creek, Tonzi Ranch)
California (Willow Creek, Tonzi Ranch) Soil moisture (cm3 cm-3)
20 0.8
0.3
CMI (mm d-1)
5
10
SWDI
0.6
NDVI
0.2
0 0
0.4
-10 0.1
-5
0.2
-20
0.0
-30 -10 0.0

30 15 1.0
Idaho (Reynolds Creek) Idaho (Reynolds Creek)
0.5
20 10 0.8
CMI (mm d-1)
0.4
10 5
0.6
NDVI
SWDI
0.3
0 0
0.2 0.4
-10 -5
0.1
0.2
-20 -10
0.0
-30 -15 0.0
0.20 1.0
30 20
Georgia (Big Cypress Creek, JERC) Georgia (Big Cypress Creek, JERC)
20 0.15 0.8
10
CMI (mm d-1)
10
0.10 0.6
NDVI
SWDI
0 0
0.05 0.4
-10
-10 0.00 0.2
-20
-0.05 0.0
-30 -20
01-Jan-01 01-Jan-04 01-Jan-07 01-Jan-10 01-Jan-13 01-Jan-16
01-Jan-01 01-Jan-04 01-Jan-07 01-Jan-10 01-Jan-13 01-Jan-16
683 Date (dd-m-yy) Date (dd-m-yy)
684 Fig. 8. Temporal dynamics of weekly OPTRAM-SWDI compared with CMI (left), and OPTRAM-based soil
685 moisture and NDVI values (right) for five sites.
686
38
687 Figure 8 (right) provides a comparison betweeen OPTRAM soil moisture and NDVI time
688 series for each site for the 2001 to 2017 period. It is obvious that OPTRAM soil moisture and
689 NDVI show similar temporal evolution for the dry and wet periods (with a small time lag), which
690 is related to the response of vegetation to soil moisture content. The OPTRAM soil moisture and
691 NDVI also show similar patterns with the corresponding OPTRAM-SWDI and CMI values at
692 each site. The NDVI and other similar vegetation indices such as Soil Adjusted Vegetation Index
693 (SAVI by Huete, 1988) and Enhanced Vegetation Index (EVI by Liu and Huete, 1995) have
694 been widely used for drought monitoring through correlation with other climatic drought
695 indicators. However, NDVI anomalies may show time lags of several weeks with atmospheric
696 variables (e.g., precipitation as discussed by Zhang et al., 2013) that restrict the application of
697 NDVI for drought monitoring, as is also evident in Fig. 8 (right).
698 Based on OPTRAM-SWDI values, all sites have experienced a wide range of drought
699 severity from no drought (SWDI ≥ 0) to extreme drought (SWDI ≤ -10) during the 2001 to 2017
700 period. In general, drought events are clearly captured by OPTRAM-SWDI based on higher
701 negative values. For example in Arizona, a significant deficit of precipitation from May 2002 to
702 October 2004 and February 2006 to November 2007 led to severe and extreme drought
703 conditions evidenced by OPTRAM-SWDI values that reduced to around -17 and daily CMI
704 values that reduced to around -5. This drought pattern coincides with the U.S. Drought Monitor
705 data (https://droughtmonitor.unl.edu) in Arizona where more than 50% of the state area
706 experienced exceptional and extreme drought conditions during this period (data are not shown).
707 These results are promising and indicate the importance of long-term MODIS data (since 2000)
708 for generating historical long-term soil moisture dynamics with the OPTRAM approach to show
709 agricultural drought periods, particularly in regions where soil moisture information is lacking.
39
710 This could also potentially provide means for long-term (decadal-scale) drought monitoring
711 based on OPTRAM-MODIS soil moisture information, which is key to developing early warning
712 systems for agricultural drought.
713 A critical issue of soil moisture estimation from optical satellite data is that estimates are limited
714 to the surface soil layer (i.e., a few mm). Agricultural drought is related to soil water availability
715 for plants, water which is mainly stored in the deeper root zone layer. However, strong
716 correlations exist between surface (or near-surface) and root zone soil moisture (Albergel et al.,
717 2008; Hirschi et al., 2014). With OPTRAM, it is assumed that θ (or W) in Eq. (5) is correlated to
718 root zone soil moisture through the vegetation response to soil water deficit in the root zone layer
719 (Sadeghi et al., 2017). Many other studies have tracked the change in root zone soil moisture via
720 remotely sensed vegetation indices because soil moisture status in the root zone influences the
721 vegetation water content and thereby changes the spectral characteristics of the vegetation
722 (Santos et al., 2014). These findings suggest that OPTRAM soil moisture estimates can be
723 directly used for agricultural drought monitoring and would be a suitable alternative for coarse
724 resolution microvawe-based soil moisture retrievals, which are currently too coarse for drought
725 monitoring over agricultural fields.
726
727 5. Summary and Conclusions
728 This paper presents a comprehensive evaluation of the optical trapezoid model
729 (OPTRAM) for estimation of soil moisture. The evaluation includes four distinct watersheds in
730 the U.S., which are diverse in climate and land cover. Long-term MODIS images from 2001-
731 2017 were employed to obtain a universal parameterization of the OPTRAM model for
732 estimation of soil moisture. Cosmic-ray neutron (CRN) soil moisture data from the U.S.
40
733 COSMOS network, as well as three other existing satellite-based surface soil moisture products,
734 including SMAP, SMOS, and ASCAT retrievals, were used to evaluate OPTRAM’s accuracy.
735 The reference CRN soil moisture exhibits an area of influence similar to the size of a MODIS
736 pixel. This similarity helps to reduce uncertainties in the comparisons. The CDF matching
737 method was applied to eliminate bias (time-invariant errors) and to rescale OPTRAM and
738 microwave-based soil moisture estimates. Long-term MODIS data (since 2001) was used to map
739 long-term soil moisture dynamics with OPTRAM and to track environmental changes. For
740 example, a soil water deficit index (OPTRAM-SWDI) was calculated based on OPTRAM soil
741 moisture estimates and a soil water availability concept to explore the potential of OPTRAM-
742 MODIS to monitor agricultural drought. The OPTRAM-SWDI was compared with an
743 atmospheric drought index (CMI) to identify and capture drought dynamics over diverse
744 climates, soil types and land covers.
745 Matching the CDFs significantly decreased the estimation errors (bias, RMSE, ubRMSE),
746 while having no impact on R values. It was found that the original OPTRAM and microwave soil
747 moisture estimates were highly correlated with CRN soil moisture at each site. OPTRAM
748 successfully captured the temporal dynamics of the SMAP, SMOS, and ASCAT soil moisture
749 retrievals for all sites. OPTRAM performed well for all test sites with ubRMSE ranging between
750 0.0501 and 0.0853 cm3 cm-3 and R ranging between 0.10 and 0.70. The SMAP and SMOS
751 retrievals were more accurate than the OPTRAM estimates for most CRN sites, verifying higher
752 sensitivity of passive microwaves to soil moisture than optical reflectance observations. The
753 OPTRAM, however, performed similar to or slightly better than the ASCAT sensor.
754 A universal parameterization of OPTRAM can be obtained for a given region based on
755 long-term MODIS data. In this study we used a visual inspection approach for delineating the
41
756 dry and wet edges. A performed sensitivity analysis indicates that OPTRAM outputs are not
757 highly sensitive to the dry and wet edge model parameters. Nonetheless, in our ongoing work we
758 strive to develop a physical edge detection approach to eliminate any potential user bias.
759 Due to moderate spatial and high temporal resolution of MODIS images, OPTRAM
760 estimates could be helpful for monitoring soil moisture for agricultural fields and the STR-NDVI
761 space could be used for downscaling of microwave satellite soil moisture products.
762 A similar pattern was observed between OPTRAM-SWDI and CMI at most sites. The
763 drought events were clearly captured by OPTRAM-SWDI. These results demonstrate that the
764 OPTRAM-SWDI is able to adequatly capture drought dynamics for a wide range of climate, soil
765 type and land cover conditions, portending OPTRAM’s feasibility as a tool for agricultural
766 drought monitoring. The calculation of SWDI is based on soil water retention parameters such as
767 the field capacity and permanent wilting point. The international soil datasets such as the
768 Harmonized World Soil Database (FAO/ISRIC/ISSCAS/JRC, 2012) in conjuction with a
769 pedotransfer function approach can provide good estimates of soil water retention parameters.
770 Thus, where soil property databases are available, SWDI can be estimated from the OPTRAM.
771
772 6. Acknowledgments
773 The authors gratefully acknowledge funding from the National Science Foundation
774 (NSF) grant no. 1521469 and acknowledge use of data from the COsmic-ray Soil Moisture
775 Observing System (COSMOS), funded by the US National Science Foundation (ATM-0838491).
776
777 7. References
42
778 Akbar, R., Moghaddam, M., 2015. A combined active passive soil moisture estimation algorithm
779 with adaptive regularization in support of SMAP. IEEE Transactions on Geoscience and
780 Remote Sensing. 53(6), 3312–3324.
781 Albergel, C., Rudiger, C., Pellarin, T., Calvet, J. C., Fritz, N., Froissard, F., Martin, E., 2008.
782 From near-surface to root-zone soil moisture using an exponential filter: An assessment of
783 the method based on in-situ observations and model simulations. Hydrology and Earth
784 System Sciences. 12, 1323–1337.
785 Al-Yaari, A., Wigneron, J.-P., Ducharne, A., Keer, Y.H., Wagner, W., De Lannoy, G., Reichle,
786 R., Al Bitar, A., Dorigo, W., Richaume, P., Mialon, A., 2014. Global scale comparison of
787 passive (SMOS) and active (ASCAT) satellite based microwave soil moisture retrievals with
788 soil moisture simulations (MERRA-Land). Remote Sensing of Environment. 152, 614-626.
789 Andreasen, M., Jensen, K.H., Desilets, D., Franz, T.E., Zreda, M., Bogena, H.R., Looms, M.C.,
790 2017. Status and perspectives on the cosmic-ray neutron method for soil moisture estimation
791 and other environmental science applications. Journal of Hydrology. 16(8).
792 Avery, W.A., Finkenbiner, C., Franz, T.E., Wang, T., Nguy-Robertson, A.L., Suyker, A.,
793 Arkebauer, T., Arriola, F.M., 2016. Incorporation of globally available datasets into the
794 roving cosmic-ray neutron probe method for estimating field-scale soil water content.
795 Hydrological Earth System Science. 20, 3859–3872.
796 Baatz, R., Bogena, H.R., Franssen, H.-J.H., Huisman, J.A., Montzka, C., Vereecken, H., 2015.
797 An empirical vegetation correction for soil moisture content quantification using cosmic ray
798 probes. Water Resources Research. 51, 2030–2046.
799 Babaeian, E., Homaee, M., Montzka, C., Vereecken, H., Norouzi, A.A. and van Genuchten,
800 M.T., 2016. Soil moisture prediction of bare soil profiles using diffuse spectral reflectance
801 information and vadose zone flow modeling. Remote Sensing of Environment. 187, 218-229.
802 Bartalis, Z., Wagner, W., Naeimi, V., Hasenauer, S., Scipal, K., Bonekamp, H., Figa, J.,
803 Anderson, C., 2007. Initial soil moisture retrieval from the METOP-A Advanced
804 Scatterometer (ASCAT). Geographical Research Letters. 34, L20401.
805 Bogena, H.R., Huisman, J.A., Baatz, R., Franssen, H.-J.H., Vereecken, H., 2013. Accuracy of the
806 cosmic-ray soil water content probe in humid forest ecosystems: The worst case scenario.
807 Water Resources Research. 49, 5778–5791.
808 Brocca, L., Hasenauer, S., Lacava, T., Melone, F., Moramacrco, T., Wagner, W., Dorigo, W.,
809 Matgen, P., Martinez-Fernandez, J., Llorens, P., Latron, J., Martin, C., Bittelli, M., 2011. Soil
810 moisture estimation through ASCAT and AMSR-E sensors: An intercomparison and
811 validation study across Europe. Remote Sensing of Environment. 115, 3390–3408.
812 Cammalleri, C., Micale, F., Vogt, J., 2016. A novel soil moisture‐based drought severity index
813 (DSI) combining water deficit magnitude and frequency. Hydrological Processes. 30(2), 289-
814 301.
815 Carlson, T.N., Gillies, R.R., Perry, E.M., 1994. A methoCd to make use of thermal infrared
816 temperature and NDVI measurements to infer surface soil water content and fractional
817 vegetation cover. Remote Sensing Reviews. 9(1-2), 161-173.
43
818 Carrao, H., Russo, S., Sepulcre-Canto, G., Barbosa, P., 2016. An empirical standardized soil
819 moisture index for agricultural drought assessment from remotely sensed data. International
820 Journal of Applied Earth Observation and Geoinformation. 48, 74-84.
821 Chakrabarti, S., Bongiovanni, T., Judge, J., Zotarelli, L., Bayer, C., 2014. Assimilation of SMOS
822 soil moisture for quantifying drought impacts on crop yield in agricultural regions. IEEE
823 Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 7(9), 3867-
824 3879.
825 Chapin, E., Chau, A., Chen, J., Heavey, B., Hensley, S., Lou, Y., Machuzak, R. and
826 Moghaddam, M., 2012. AirMOSS: An airborne P-band SAR to measure root-zone soil
827 moisture. Proc. IEEE Radar Conference. Atlanta, GA, 693–698.
828 Crow, W., Koster, R., Reichle, R., Sharif, H.,2005. Relevance of time-varying and time-invariant
829 retrieval error sources on the utility of spaceborne soil moisture products. Geophysical
830 Research Letters. 32, L24405. doi: 10.1029/2005GL024889.
831 Desilets, D., Zreda, M., 2013. Footprint diameter for a cosmic-ray soil moisture probe: Theory
832 and Monte Carlo simulations. Water Resources Research. doi:10.1002/wrcr.20187.
833 Djamai, N., Magagi, R., Goita, K., Hosseini, M., Cosh, M.H., Berg, A., Toth, B., 2015.
834 Evaluation of SMOS soil moisture products over the CanEx-SM10 area. Journal of
835 Hydrology. 520, 254–267.
836 Entekhabi, D., Njoku, E.G., O'Neill, P.E., Kellogg, K.H., Crow, W.T., Edelstein, W.N., Entin,
837 J.K., Goodman, S.D., Jackson, T.J., Johnson, J. and Kimball, J., 2010. The soil moisture
838 active passive (SMAP) mission. Proceedings of the IEEE. 98(5), 704-716.
839 Escorihuela, M.J., Chanzy, A., Wigneron, J.P., Kerr, Y.H., 2010. Effective soil moisture
840 sampling depth of L-band radiometry: A case study. Remote Sensing of Environment. 114,
841 995–1001.
842 Evans, J.G., Ward, H.C., Blake, J.R., Hewitt, E.J., Morrison, R., Fry, M., et al., 2016. Soil water
843 content in southern England derived from a cosmic-ray soil moisture observing system:
844 COSMOS-UK. Hydrological Processes. 30, 4987-4999.
845 Fascetti, F., Pierdicca, N., Pulvirenti, L., Crapolicchio, R., Munoz-Sabater, J., 2016. A
846 comparison of ASCAT and SMOS soil moisture retrievals over Europe and Northern Africa
847 from 2010 to 2013. International Journal of Applied Earth Observation and Geoformation.
848 45, 135-142.
849 Fernandez-Moran, R., Al-Yaari, A., Mialon, A., Mahmoodi, A., Al Bitar, A., De Lannoy, G.,
850 Rodriguez-Fernandez, N., Lopez-Baeza, E., Kerr, Y., Wigneron, J.P. 2017. SMOS-IC: An
851 alternative SMOS soil moisture and vegetation optical depth product. Remote Sensing. 9,
852 457.
853 Franz, T.E., Zreda, M., Rosolem, R., Ferré, T.P.A., 2013. A universal calibration function for
854 determination of soil moisture with cosmic-ray neutrons. Hydrological Earth System Science.
855 17, 453–460.
856 Franz, T.E., Zreda, M., Rosolem, R., Ferre, T.P.A., 2012. Field validation of a cosmic-ray
857 neutron sensor using a distributed sensor network. Vadose Zone Journal. 11(4).
44
858 Gillies, R.R., Carlson, T.N., 1995. Thermal Remote Sensing of Surface Soil Water Content with
859 Partial Vegetation Cover for Incorporation into Climate Models. Journal of Applied
860 Meteorology. 34, 745-756.
861 Hallikainen, M.T., Ulaby, F.T., Dobson, M.C., El-Rayes, M.A., 1985. Microwave dielectric
862 behaviorofwetsoil-part1: empirical models and experimental observations. IEEE
863 Transactions on Geoscience and Remote Sensing. 23(1), 25–34.
864 Hawdon, A., McJannet, D., Wallace, J., 2014. Calibration and correction procedures for cosmic-
865 ray neutron soil moisture probes located across Australia. Water Resources Research. 50,
866 5029-5043.
867 Hirschi, M., Mueller, B., Dorigo, W., Seneviratne, S. I., 2014. Using remotely sensed soil
868 moisture for land–atmosphere coupling diagnostics: The role of surface vs. root-zone soil
869 moisture variability. Remote SensIng of Environment. 154, 246–252.
870 Hogg, E.H., Hurdle, P.A., 1995. The aspen parkland in western Canada: a dry-climate analogue
871 for the future boreal forest? Water, Air, and Soil Pollution. 82, 391–400.
872 Huete, A., 1988. A soil adjusted vegetation index (SAVI). Remote Sensing of Environment. 25
873 (3), 295–309. doi:10.1016/0034-4257(88)90106-X.
874 Kabat, P., Beekma, J., 1994. Water in the unsaturated zone. In: Ritzema, H.P. (Ed.), Drainage
875 Principles and Applications. ILRI Publication 16, second ed. International Institute for Land
876 Reclamation and Improvement, Wageningen, the Netherlands. 383-434.
877 Keefer, T.O., Moran, M.S. Paige, G.B., 2008. Long ‐term meteoro
878 database, Walnut Gulch Experimental Watershed, Arizona, United States. Water Resources
879 Research. 44(5), W05S07.
880 Kerr, H., Waldteufel, P., Wigneron, J.P., Martinuzzi, J., Font, J., Berger, M., 2001. Soil moisture
881 retrieval from space: The Soil Moisture and Ocean Salinity (SMOS) mission. IEEE
882 Transactions on Geoscience and Remote Sensing. 39, 1729–1735.
883 Kohli, M., Schron, M., Zreda, M., Schmidt, U., Dietrich, P., Zacharias, S., 2015. Footprint
884 characteristics revised for field-scale soil moisture monitoring with cosmic-ray neutrons.
885 Water Resources Research. 51(7), 5772-5790.
886 Koster, R.D., Guo, Z., Yang, R., Dirmeyer, P.A., Mitchell, K., Puma, M., 2009. On the nature of
887 soil moisture in land surface models. Journal of Climate. 22, 4322-4335.
888 Kumar, S.V., Reichle, R.H., Harrison, K.W., Peters-Lidard, C.D., Yatheendradas, S., Santanello,
889 J.A., 2012. A comparison of methods for a priori bias correction in soil moisture data
890 assimilation. Water Resources Research. 48, W03515.
891 Liu, D., Mishra, A.K., Yu, Z.B., Yang, C.G., Konapala, G., Vu, T., 2017. Performance of SMAP,
892 AMSR-E and LAI for weekly agricultural drought forecasting over continental United States.
893 Journal of hydrology. 555, 88-104.
894 Liu, H.Q., Huete, A., 1995. A feedback based modification of the NDVI to minimize canopy
895 background and atmospheric noise. IEEE Transactions on Geoscience and Remote Sensing.
896 33 (2), 457–465.
45
897 Lv, L., Franz, T.E., Robinson, D.A., Jones, S.B., 2014. Measured and Modeled Soil Moisture
898 Compared with Cosmic-Ray Neutron Probe Estimates in a Mixed Forest. Vadose Zone
899 Journal. 13(12). doi:10.2136/vzj2014.06.0077
900 Martinez-Fernandez, J., Gonzalez-Zamora, A., Sanchez, N., Gumuzzio, A., 2015. A soil water
901 based index as a suitable agricultural drought indicator. Journal of Hydrology. 522, 265–273.
902 Martinez-Fernandez, J., Gonzalez-Zamora, A., Sanchez, N., Gumuzzio, A., Herrero-Jimenez,
903 C.M., 2016. Satellite soil moisture for agricultural drought monitoring: Assessment of the
904 SMOS derived Soil Water Deficit Index. Remote Sensing of Environment. 177, 277–286.
905 Mattikalli, N.M., Engman, E.T, Jackson, T.J., Ahuja, L.R., 1998. Microwave remote sensing of
906 temporal variations of brightness temperature and near surface soil water content during a
907 watershed-scale field experiment, and its application to the estimation of soil physical
908 properties. Water Resources Research. 34, 2289–2299.
909 McKee, T.B., Doesken, N.J., Kleist, J., 1993.The Relationship of Drought Frequency and
910 Duration to Time Scales. Proceedings of the Eighth Conference on Applied Climatology.
911 American Meteorological Society: Boston; 179–184.
912 McPherson, R.A., Fiebrich, C.A., Crawford, K.C., Kilby, J.R., Grimsley, D.L., Martinez, J.E.,
913 Basara, J.B., Illston, B.G., Morris, D.A., Kloesel, K.A. and Melvin, A.D., 2007. Statewide
914 monitoring of the mesoscale environment: A technical update on the Oklahoma Mesonet.
915 Journal of Atmospheric Oceanic Technology. 24(3), 301–321.
916 Merlin, O., Rüdiger, C., Al Bitar, A., Richaume, P., Walker, J. Kerr, Y., 2012. Disaggregation of
917 SMOS soil moisture in southeastern Australia. IEEE Transactions on Geoscience and
918 Remote Sensing. 99, 1-16.
919 Mironov, V.L., Kosolapova, L.G., Fomin, S.V., 2009. Physically and mineralogically based
920 spectroscopic dielectric model for moist soils. IEEE Transactions on Geoscience and Remote
921 Sensing. 47, 2059–2070.
922 Mishra, A., Vu, T., Veetti, A.V., Entekhabi, D., 2017. Drought Monitoring with Soil Moisture
923 Active Passive (SMAP) Measurements. Journal of Hydrology. 552.
924 Montzka, C., Bogena, H.R, Zreda, M., Monerris, A., Morrison, R., Muddu, S., Verrecken, H.,
925 2017. Validation of spaceborne and modelled surface soil moisture products with cosmic-ray
926 neutron probes. Remote Sensing. 9, 103.
927 Nachtergaele, F.O., van Velthuizen, H.T., Verelst, L., Wiberg, D., Batjes, N.H., Dijkshoorn, J.A.,
928 van Engelen, V.W.P., Fischer, G., Jones, A., Montanarella, L., et al., 2012. Harmonized
929 World Soil Data base (Version 1.2), FAO, Rome, Italy, IIASA, Laxenburg, Austria.
930 Naeimi, V., Scipal, K., Bartalis, Z., Hasenauer, S., Wagner, W., 2009. An improved soil moisture
931 retrieval algorithm for ERS and METOP scatterometer observations. IEEE Transactions on
932 Geoscience and Remote Sensing. 47, 1999-2013.
933 Owen, T.W., Carlson, T.N., Gillies, R.R., 1998. An Assessment of satellite remotely-sensed
934 Land Cover Parameters in Quantitatively Describing the Climatic Effect of Urbanization.
935 International Journal of Remote Sensing. 19, 1663-1681.Palmer, W. C., 1965.
936 Meteorological drought. U.S. Weather Research Paper 45. Washington D.C.: U.S. Weather
937 Bureau.
46
938 Palmer, W. C., 1968. Keeping track of crop moisture conditions, nationwide: The new crop
939 moisture index. Weatherwise. 21(4), 156–161
940 Peng, J., Niesel, J., Loew, A., Zhang, S., Wang, J., 2015. Evaluation of Satellite and Reanalysis
941 Soil Moisture Products over Southwest China Using Ground-Based Measurements. Remote
942 Sensing. 7, 15729-15747.
943 Piles, M., Sánchez, N., Vall-llossera, M., Camps, A., Martínez-Fernández, J., Martinez, J. and
944 Gonzalez-Gambau, V., 2014. A downscaling approach for SMOS land observations:
945 Evaluation of high-resolution soil moisture maps over the Iberian Peninsula. IEEE Journal of
946 Selected Topics in Applied Earth Observations and Remote Sensing. 7(9), 3845-3857.
947 Rahimzadeh-Bajgiran, P., Berg, A.A., Champagne, C. and Omasa, K., 2013. Estimation of soil
948 moisture using optical/thermal infrared remote sensing in the Canadian Prairies. ISPRS
949 Journal of Photogrammetry and Remote Sensing. 83, 94-103.
950 Reichle, R. H., Koster, R.D., 2004. Bias reduction in short records of satellite soil moisture.
951 Geographical Research Letters. 31, L19501.
952 Renard, K.G., Lane, L.J., Simanton, J.R., Emmerich, W.E., Stone, J.J., Weltz, M.A., Goodrich,
953 D.C. and Yakowitz, D.S., 1993. Agricultural impacts in an arid environment: Walnut Gulch
954 studies. Hydrological Science and Technology. 9(1-4), 145-190.
955 Rim, C.S., 2000. A comparison of approaches for evapotranspiration estimation. KSCE Journal
956 of Civil Engineering. 4(1), 47-52.
957 Robinson, D.A., Campbell, C.S., Hopmans, J.W., Hornbuckle, B.K., Jones, S.B., Knight, R.,
958 Ogden, F., Selker, J., Wendroth, O., 2008. Soil moisture measurement for ecological and
959 hydrological watershed-scale observatories: A Review. Vadose Zone Journal. 7(1), 358-389.
960 doi:10.2136/vzj2007.0143.
961 Rosolem, R., Shuttleworth, W.J., Zreda, M., Franz, T.E., Zeng, X., Kurc, S.A., 2013. The effect
962 of atmospheric water vapor on neutron count in the cosmic-ray soil moisture observing
963 system. Journal of Hydrometeorology. 14, 1659–1671.
964 Shafian, S., Maas, S.J., 2015. Improvement of the Trapezoid method using raw Landsat image
965 digital count data for soil moisture estimation in the Texas (USA) High
966 Plains. Sensors. 15(1), 1925-1944.
967 Sadeghi, M., Babaeian, E., Tuller, M., Jones, S.B., 2017. The optical trapezoid model: a novel
968 approach to remote sensing of soil moisture applied to Sentinel-2 and Landsat-8
969 observations. Remote Sensing of Environment. 198, 52-68.
970 Sadeghi, M., Jones, S.B., Philpot, W.D., 2015. A linear physically-based model for remote
971 sensing of soil moisture using short wave infrared bands. Remote Sensing of
972 Environment. 164, 66-76.
973 Santos, W.J.R., Silva, B.M., Oliveira, G.C., Volpato, M.M.L., Lima, J.M., Curi, N., Marques,
974 J.J., 2014. Soil moisture in the root zone and its relation to plant vigor assessed by remote
975 sensing at management scale. Geoderma. 221, 91–95.
976 Schaap, M.G., Leij, F.J., van Genuchten, M.T., 2001. ROSETTA: a computer program for
977 estimating soil hydraulic parameters with hierarchical pedotransfer functions. Journal of
978 Hydrology. 251(3-4), 163-176.
47
979 Sohrabi, M.M., Ryu, J.H., Abatzoglou, J., Tracy, J., 2015. Development of soil moisture drought
980 index to characterize droughts. Journal of Hydrologic Engineering. 20(11), 04015025.
981 Sridhar, V., Hubbard, K.G., You, J., Hunt, E.D., 2008. Development of the soil moisture index to
982 quantify agricultural drought and its “user friendliness” in severity-area-duration assessment.
983 Journal of Hydrometeorology. 9(4), 660-676.
984 Sun, H., 2016. Two-Stage Trapezoid: A New Interpretation of the Land Surface Temperature
985 and Fractional Vegetation Coverage Space. IEEE Journal of Selected Topics in Applied
986 Earth Observations and Remote Sensing. 9(1), 336-346.
987 Tabatabaeenejad, A., Burgin, M., Duan, X., Moghaddam, M., 2015. P-band radar retrieval of
988 subsurface soil moisture profile as a second-order polynomial: First AirMOSS results. IEEE
989 Transactions on Geoscience and Remote Sensing. 53(2), 645–658.
990 Thornthwaite, C.W., 1948. An approach towards a rational classification of climate.
991 Geographical Review. 38, 55-94.
992 Torres-Ruiz, J.M., Diaz-Espejo, A., Morales-Sillero, A., Martín-Palomo, M. J., Mayr, S.,
993 Beikircher, B., Fernández, J.E., 2013. Shoot hydraulic characteristics, plant water status and
994 stomatal response in olive trees under different soil water conditions. Plant and Soil. 373(1-
995 2), 77-87.
996 U.S. Geological Survey, 1986. Hydrologic events and surface-water resources in National Water
997 Summary: U.S. Geological Survey Water-Supply Paper 2300.
998 Van der Schalie, R., Kerr, Y. H., Wigneron, J. P., Rodríguez-Fernández, N. J., Al-Yaari, A., de
999 Jeu, R.A.M., 2016. Global SMOS soil moisture retrievals from the land parameter retrieval
1000 model. International Journal of Applied Earth Observation and Geoinformation. 45, 125-134.
1001 van Genuchten, M. Th., 1980. A closed-form equation for predicting the hydraulic conductivity
1002 of unsaturated soils. Soil Science Society of America Journal. 44, 892–898.
1003 Vereecken, H., Huisman, J.A., Pachepsky, Y., Montzka, C., van der Kruk, J., Bogena, H.,
1004 Weihermüller, L., Herbst, M., Martinez, G., Vanderborght, J., 2014. On the spatio-temporal
1005 dynamics of soil moisture at the field scale. Journal of Hydrology. 516, 76-96.
1006 Wagner, W., Hahn, S., Kidd, R., Melzer, T., Bartalis, Z., Hasenauer, S., Figa-Saldana, J., de
1007 Rosnay, P., Jann, A., Schneider, S., Komma, J., 2013. The ASCAT soil moisture product: a
1008 review of its specifications, validation results and emerging applications. Meteorologische
1009 Zeitschrift. 22, 5-23.
1010 Wigneron, J.P., Schwank, M., Baeza, E.L., Keer, Y., Novello, N., Millan, C., Moisy, C.,
1011 Richaume, P., Mialon, A., Al Bitar, A., et al., 2012. First evaluation of the simultaneous
1012 SMOS and ELBARA-II observations in the Mediterranean region. Remote Sensing of
1013 Environment. 124, 26-37.
1014 Wigneron, J.P., Kerr, Y., Waldteufel, P., Saleh, K., Escorihuela, M.J., Richaume, P., Ferrazzoli,
1015 P., de Rosnay, P., Gurney, R., Calvet, J.C., et al., 2007. L-band microwave emission of the
1016 biosphere (L-MEB) model: Description and calibration against experimental data sets over
1017 crop ﬁelds. Remote Sensing of Environment. 107, 639–655.
1018 Zhang, F., Zhang, L., Wang, X., Hung, J., 2013. Detecting agro-droughts in Southwest of China
1019 using MODIS satellite data. Journal of Integrative Agriculture. 12 (1), 159–168.
48
1020 Zreda, M., Shuttleworth, W.J., Zeng, X., Zweck, C., Desilets, D., Franz, T., Rosolem, R., 2012.
1021 COSMOS: the Cosmic-ray Soil Moisture Observation System. Hydrological Earth System
1022 Science. 16, 4079-4099.
1023 Zreda, M., Desilets, D., Ferre, T.P.A., Scott, R.L., 2008. Measuring soil moisture content non-
1024 invasively at intermediate spatial scale using cosmic-ray neutrons. Geographical Research
1025 Letters. 35, L21402.
49
View publication stats

Babaeianetal 2018RSE

Uploaded by

Copyright:

Available Formats

Babaeianetal 2018RSE

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Babaeianetal 2018RSE

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Article in Remote Sensing of Environment · June 2018

Ebrahim Babaeian Morteza Sadeghi

SEE PROFILE SEE PROFILE

T. Franz Scott B. Jones

SEE PROFILE SEE PROFILE

The user has requested enhancement of the downloaded file.

2 on Long-Term MODIS Observations

11 E-mail address: [email protected] (E. Babaeian), [email protected] (M. Sadeghi),

12 [email protected] (T.E. Franz), [email protected] (S.B. Jones), [email protected] (M. Tuller)

32 monitoring of agricultural drought through comparison of the OPTRAM-based Soil Water

42 potentially be applied for agricultural drought monitoring.

45 Cosmic-ray neutron soil moisture, MODIS, SMAP, SMOS, ASCAT

50 allocation of water resources, prediction and monitoring of drought, agricultural production,

51 mitigation of natural disasters and monitoring of ecosystem response to climate change

52 (Robinson et al., 2008; Vereecken et al., 2014; Babaeian et al., 2016).

58 2) on the Global Change Observation Mission-Water (GCOM-W1) Satellite, launched in 2012,

66 Tabatabaeenejad et al., 2015).

67 Passive microwave satellites commonly provide high temporal resolution observations

83 consuming and computationally demanding individual parameterization (calibration) for each

86 moisture. The OPTRAM trapezoid is formed by distribution of the normalized difference

87 vegetation index (NDVI) as a measure of vegetation fraction versus shortwave-infrared (SWIR)

93 second limitation of the conventional trapezoid model).

102 response to climate changes in different regions.

121 such analysis in order to fill this critical observational gap.

127 OPTRAM soil moisture data for agricultural drought monitoring.

129 2. Background: The Optical Trapezoid Model (OPTRAM)

136 results in:

137 STR − STRd (1)

145 STRd= id + sd NDVI (3)

146 STRW= iW + sW NDVI (4)

148 wet edge.

150 estimated from its STR and NDVI values:

151 id + s d NDVI − STR (5)

153 (assumed equal to θs).

159 observations for any specific location.

164 3. Materials and Methods

165 3.1. Study Areas

170 watersheds follows.

182 3.1.1. Walnut Gulch, Arizona

196 3.1.2. Willow Creek, California

206 3.1.3. Reynolds Creek, Idaho

220 exhibits silty soils with dense vegetation cover.

222 3.1.4. Big Cypress Creek, Georgia

233 3.2. Reference Soil Moisture Data

239 moisture estimates from MODIS observations.

256 evaluation experiments.

276 (2013), and Lv et al. (2014).

278 Soil Climate Analysis Network (SCAN) sites.

280 3.3. Satellite Data

281 3.3.1. MODIS

282 The Moderate Resolution Imaging Spectroradiometer (MODIS) is a key instrument on

302 MATLAB R2015b (Math Works Inc., Natick, MA) software.

304 3.3.2. SMAP

317 3.3.3. SMOS

335 moisture with OPTRAM.

337 3.3.4. ASCAT

357 3.4. Data Analyses

358 3.4.1. OPTRAM Parameterization