ArticlePDF Available

A Real-Time MODIS Vegetation Product for Land Surface and Numerical Weather Prediction Models


Abstract and Figures

A technique is presented to produce real-time, daily vegetation composites at 0.01$^{circ}$ resolution $({sim}{rm 1}~{rm km})$ over the Conterminous United States (CONUS) for use in the NASA Land Information System (LIS) and weather prediction models. Green vegetation fraction (GVF) is derived from direct-broadcast swaths of normalized difference vegetation index from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the NASA Earth Observing System satellites. The real-time data and increased resolution compared to the 0.144$^{circ}$$({sim}{rm 16}~{rm km})$ resolution monthly GVF climatology in community models result in an improved representation of vegetation in high-resolution models, especially in complex terrain. The MODIS GVF fields show seasonal variations that are similar to the community model climatology, and respond realistically to temperature and precipitation anomalies. The wet spring and summer 2010 over the U.S. Plains led to higher regional GVF than in the climatology. The GVF substantially decreased over the U.S. Southern Plains from 2010 to 2011, consistent with the transition to extreme drought in summer 2011. LIS simulations depict substantial sensitivity to the MODIS GVF, with regional changes in heat fluxes around 100 ${rm Wm}^{-2}$ over the northern U.S. in June 2010. CONUS LIS simulations during the 2010 warm season indicate that the larger MODIS GVF in the western U.S. led to higher latent heat fluxes and initially lower sensible heat fluxes, with a net drying effect on the soil. With time, the drier soil eventually lead to higher mean sensible heat fluxes such that the total surface energy- output increased by late summer 2010 over the western U.S. A sensitivity simulation of a severe weather event using real-time MODIS GVF data results in systematic changes to low-level temperature, moisture, and instability fields, and improves the evolution of simulated precipitation.
Content may be subject to copyright.
A Real-Time MODIS Vegetation Product for Land
Surface and Numerical Weather Prediction Models
Jonathan L. Case, Frank J. LaFontaine, Jordan R. Bell, Gary J. Jedlovec,
Sujay V. Kumar, and Christa D. Peters-Lidard
Abstract— A technique is presented to produce real-time, daily
vegetation composites at 0.01° resolution (1 km) over the
Conterminous United States (CONUS) for use in the NASA
Land Information System (LIS) and weather prediction models.
Green vegetation fraction (GVF) is derived from direct-broadcast
swaths of normalized difference vegetation index from the Mod-
erate Resolution Imaging Spectroradiometer (MODIS) aboard
the NASA Earth Observing System satellites. The real-time data
and increased resolution compared to the 0.144° (16 km)
resolution monthly GVF climatology in community models result
in an improved representation of vegetation in high-resolution
models, especially in complex terrain. The MODIS GVF fields
show seasonal variations that are similar to the community
model climatology, and respond realistically to temperature and
precipitation anomalies. The wet spring and summer 2010 over
the U.S. Plains led to higher regional GVF than in the climatology.
The GVF substantially decreased over the U.S. Southern Plains
from 2010 to 2011, consistent with the transition to extreme
drought in summer 2011. LIS simulations depict substantial
sensitivity to the MODIS GVF, with regional changes in heat
fluxes around 100 Wm2over the northern U.S. in June 2010.
CONUS LIS simulations during the 2010 warm season indicate
that the larger MODIS GVF in the western U.S. led to higher
latent heat fluxes and initially lower sensible heat fluxes, with a
net drying effect on the soil. With time, the drier soil eventually
lead to higher mean sensible heat fluxes such that the total surface
energy output increased by late summer 2010 over the western
U.S. A sensitivity simulation of a severe weather event using
real-time MODIS GVF data results in systematic changes to low-
level temperature, moisture, and instability fields, and improves
the evolution of simulated precipitation.
Manuscript received June 1, 2012; revised November 26, 2012 and March 7,
2013; accepted March 11, 2013. Date of publication May 24, 2013; date
of current version December 17, 2013. This work was supported in part
by the National Aeronautics and Space Administration (NASA) Modeling
Analysis and Prediction Solicitation under Grant NNH08ZDA001N-MAP (PI:
Peters-Lidard/ Goddard Space Flight Center), and the NASA Science Mission
Directorate’s Earth Science Division in support of the Short-term Prediction
Research and Transition Center at the NASA Marshall Space Flight Center.
J. L. Case is with the NASA Short-term Prediction Research and Tran-
sition (SPoRT) Center, ENSCO, Inc., Huntsville, AL 35805 USA (e-mail:
F. J. LaFontaine is with the NASA Short-term Prediction Research and
Transition (SPoRT) Center, Raytheon, Huntsville, AL 35824 USA (e-mail:
J. R. Bell is with the Atmospheric Science Department, University of
Missouri, Columbia, MO 65211 USA (e-mail:
G. J. Jedlovec is with the NASA Marshall Space Flight Center/Short-term
Prediction Research and Transition (SPoRT) Center, Huntsville, AL 35805
USA (e-mail:
S. V. Kumar is with the NASA Goddard Space Flight Center, Science
Applications International Corporation, Greenbelt, MD 20771 USA (e-mail:
C. D. Peters-Lidard is with the Department of HSB, NASA Goddard Space
Flight Center, Greenbelt, MD 20771 USA (e-mail:
Color versions of one or more of the figures in this paper are available
online at
Digital Object Identifier 10.1109/TGRS.2013.2255059
Index Terms—Geoscience–Atmosphere–Atmospheric model-
ing, Geoscience–Land surface, Vegetation mapping.
EVAPOTRANSPIRATION from vegetative surfaces is an
important process that impacts the transport of moisture
into the atmosphere especially during the warm season. Vege-
tation coverage and health are commonly measured by polar-
orbiting satellites through the normalized difference vegetation
index (NDVI). The NDVI is based on the properties of healthy
vegetation, which has a high absorbance/low reflectance in
the visible portion (photosynthetically active region) of the
electromagnetic spectrum while having a high reflectance at
the near-IR (NIR) wavelengths. The NDVI is defined as the
combination of these reflectances
ρNIR +ρRED (1)
where ρNIR is the reflectance at NIR wavelengths
(0.75–1.5 μm) and ρRED is the reflectance at visible-
red wavelengths (0.6–0.7 μm). One of the first uses of this
channel combination as an index to denote green vegetation
was presented in [1]. NDVI ranges from 1to+1, where
values near 0 indicate little to no healthy vegetation while
negative values typically correspond to snow/ice cover under
clear skies, or to cloudy conditions. Depending on the
vegetation cover, type, and satellite footprint, NDVI values of
0.4–0.8 represent areas with near full coverage of healthy
Vegetation indices such as green vegetation fraction (GVF)
and leaf area index (LAI) are used by land surface models
(LSMs) to represent the horizontal and vertical density of plant
vegetation [2] in order to calculate transpiration, interception
and radiative shading. Latent heat flux from vegetation is
directly proportional to the value of GVF and strongly related
to LAI. Both of these indices are related to NDVI; however,
there is the problem of solving the two unknowns simultane-
ously from one NDVI measurement, as described in [2]. One
practice is to specify the LAI while allowing the GVF to vary
both spatially and temporally, as is done in the Noah LSM
[3], [4]. Operational versions of Noah within the National
Centers for EnvironmentalPrediction (NCEP) North American
Mesoscale model [5], [6], the Global Forecast System model
[7], [8], the Climate Forecast System version 2 (CFS; [9]), and
the Climate Forecast System Reanalysis (CFSR) [10] hold the
LAI fixed, while the GVF varies according to a global monthly
climatology. The same approach and GVF climatology are
0196-2892 © 2013 IEEE
also used for the North America Land Data Assimilation
System (NLDAS) [11], [12] and the Global LDAS coupled to
CFSR [10]. This GVF climatology was derived from NDVI
data on the National Oceanic and Atmospheric Administra-
tion (NOAA) Advanced Very High Resolution Radiometer
(AVHRR) polar orbiting satellite, using information from 1985
to 1991 [2], [13]. Representing data at the mid-point of
every month, the climatology is on a 0.144° (16 km) grid
and is distributed with the community weather research and
forecasting (WRF) model [4], [13], [14].
A limitation of the climatological dataset is that the annual
cycle of GVF is always represented the same in models from
one year to the next. In reality, the response of vegetation to
meteorological and climate conditions varies between seasons
and years based on anomalous weather and climate features.
For example, extreme events such as an unusual hard freeze,
late bloom due to colder than average temperatures, or drought
can lead to a vegetative response that is quite different than the
climatological representation. In addition, the dated informa-
tion (1985–1991) used to create the default GVF climatology,
coupled with the relatively coarse spatial resolution may not
be representative of current vegetative conditions in today’s
high-resolution numerical weather prediction (NWP) models.
Recent land use changes due to urban sprawl [15] may also
contribute to misrepresentations in the models.
To improve the representation of vegetation in LSMs and
NWP models, the NASA Short-term Prediction Research and
Transition (SPoRT) program created a high resolution, real-
time GVF composite that is updated on a daily basis with
near real-time, direct-broadcast NDVI swath data from the
Moderate Resolution Imaging Spectroradiometer (MODIS)
instruments aboard the NASA Earth Observing System Aqua
and Terra satellites. The SPoRT program specializes in transi-
tioning unique NASA, NOAA, and Department of Defense
satellite data and research capabilities to the operational
weather community to improve short-term weather forecasts
on regional and local scales. Several of these products have
been evaluated and transitioned for use in real-time local
and regional weather modeling applications, including a sea
surface temperature composite [16], [17] and the use of the
NASA Land Information System (LIS) [18], [19] for initializ-
ing land surface variables at high spatial resolution [20], [21].
The MODIS GVF dataset is another example of a real-time
SPoRT product designed to benefit short-term forecasting in
the operational weather community.
Previous studies have examined near real-time vegetation
datasets derived from the NOAA/AVHRR satellite [13] and
demonstrated their potential utility for real-time modeling
[22]–[24]. Other studies have examined vegetation datasets
derived from the NASA MODIS instruments for modeling
applications [25], [26]. This paper describes a Contermi-
nous U.S. (CONUS)-scale, 0.01° resolution (1 km), real-
time MODIS NDVI/GVF product to run in place of the
coarser-resolution climatological GVF currently implemented
in community models. The dataset is designed for regional,
convection-allowing resolution modeling applications over
the CONUS [27] using the LIS and coupled LIS/NASA
Unified-WRF (NU-WRF) modeling system. It has also been
transitioned to the community WRF model used for local mod-
eling applications at NOAA/National Weather Service forecast
offices [28]. The remainder of this paper is organized as fol-
lows: Section II describes the real-time MODIS vegetation and
product generation; Section III provides a brief background on
the LIS modeling system; Section IV presents comparisons
between the real-time MODIS and AVHRR climatology GVF,
and impacts on LIS/Noah LSM runs and NWP simulations
using NU-WRF. A summary is given in Section V, followed
by acknowledgements and references.
A. MODIS NDVI Compositing Algorithm
MODIS NDVI composites are created and updated daily
on a CONUS grid with 0.01° spacing (1 km), based on the
NDVI swath data obtained from the University of Wiscon-
sin/Space Science and Engineering Center’s MODIS direct
broadcast data stream [29]. The NDVI data are produced
with software based on the MODIS science team algorithms
and implemented for real-time data processing by the NASA
Direct Readout Laboratory at NASA Goddard Space Flight
Center as part of a suite of atmospheric and land prod-
ucts generated by the International MODIS/AIRS Process-
ing Package (IMAPP) ( The
NDVI swaths are derived using the corrected reflectance
algorithm developed by the MODIS rapid response sys-
tem (, but do not
account for atmospheric correction due to particulate aerosols
(e.g., [30]). To remove cloud contamination, MODIS cloud
mask information available from the suite of real-time products
is used to identify cloud-free regions in the compositing
The NDVI compositing technique follows a similar method-
ology to that used in the first-generation SPoRT sea surface
temperature (SST) compositing algorithm, which was solely
based on MODIS infrared and cloud-mask data over oceans
and large water bodies [16]. This technique generated SST
composites by obtaining a minimum of three cloud-free read-
ings at each pixel for a given collection period (typically 20
days), and then averaging the two warmest values to represent
SST at that pixel. To generate a single daily NDVI composite,
each cloud-masked MODIS swath is individually mapped to
the CONUS grid using utilities within the Man computer
Interactive Data Access System [31]. Negative NDVI values
are set to missing during the mapping step. In order to produce
a nearly continuous NDVI grid, an inverse time-weighting
algorithm is applied to the mapped swath data that queries
the previous 20 days for up to six pieces of NDVI data, using
the following formula:
where mis the number of NDVI values at a given pixel i,
NDVInis the data value at day n, and DaysAgonis the age of
day nprior to the current day. The influence of data near the
edge of the swath, which may have small uncorrected biases
because of atmospheric effects, is minimized through the use
of the multitemporal and multisatellite data at each point in
the compositing process. In situations where m=0, NDVIi
is set to missing.
The geographical extent of the NDVI grid ranges from
northern Mexico to southern Canada and includes all of the
CONUS. SPoRT has been producing these daily, real-time
composites since 1 June 2010. The daily NDVI grids are then
processed in the next step to derive GVF for use by LSMs
and NWP models.
B. Calculation of GVF for Use in Models
The GVF is calculated from the MODIS NDVI composites
on the identical 0.01° (1 km) grid following the procedures
outlined in [25] and [32]. For consistency with MODIS data,
the International Geosphere-Biosphere Programme (IGBP)
land-use classification [33] as applied to the MODIS instru-
ment [34] is used in the GVF calculation. The GVF is
computed by first determining the maximum NDVI at each
grid point (NDVImax)using the collection of daily NDVI
composites from 1 June 2010 to 1 Feb 2012 for both archived
and subsequent real-time GVF computations. The NDVImax
values are sorted as a function of the IGBP/MODIS land use
class by combining all grid points with the same land use
class into a single distribution, in order to identify the 90th
percentile of NDVImax for each land use class (representing
full vegetation coverage), and the fifth percentile for the
barren land use class (representing zero coverage). Results
are insensitive to the exact percentile selected to represent the
NDVImax value for full vegetation coverage [25], [32]. The
GVF at each grid point iis then computed as
where NDVIiis the NDVI composite value at grid point
ifrom (2), NDVISis a global constant given by the fifth
percentile of NDVImax for the barren land use class, and
NDVIV,iis the 90th percentile of NDVImax values for the land
use class Vat grid point i. As NDVI composites are collected
over a longer time period, the NDVImax can be periodically
updated to reflect subtle variations in the distributions of
NDVImax by land use class.
At this stage, any remaining pixels with missing NDVI
data are filled with time-interpolated monthly climatological
GVF data for the present day, thereby ensuring a continuous
GVF grid for use in models (mainly impacting snow-covered
pixels). The percentage of pixels in a given composite filled
with the GVF climatology ranged from 1% or less from June–
October to about 35% in February 2011 when snow cover was
at maximum extent over the domain. This data filling produced
no visual discontinuities in the warm season, but did result in
some noticeable discontinuities between the detailed MODIS
and the smooth climatology during winter months when snow
cover was most extensive. However, since evapotranspiration
is much less substantial in the surface energy budget during
the winter months (especially in snow-covered regions), the
impact of data filling should be minimal with the affected
grid points simply reverting back to the climatological GVF
To incorporate the real-time MODIS GVF into the LIS and
NU-WRF models, a new module was written within LIS to
read and process the daily MODIS GVF dataset in place of
the monthly climatological GVF database. The new module
enables the use of daily MODIS GVF in both retrospective
offline LIS and coupled LIS/NU-WRF simulations. The real-
time MODIS GVF is also available to the community WRF
model by re-formatting the data to be consistent with that of
the static input datasets of the WRF model. The much higher
spatial resolution (1kmversus16 km) and daily updates
based on real-time satellite observations have the capability
to improve the simulation of the surface energy budget as
demonstrated in Section IV, and in previous studies using
actual AVHRR GVF data [13], [22]–[24].
The LIS is a high-performance land surface modeling
and data assimilation system that integrates satellite-derived
datasets, ground-based observations, and model re-analyses to
force a variety of LSMs. By using scalable, high-performance
computing and data management technologies, LIS can run
LSMs offline globally with a grid spacing as fine as 1 km to
characterize land surface states and fluxes.
To demonstrate impact on land surface fluxes, control and
experimental versions of LIS were configured to run the Noah
LSM using the IGBP/MODIS land-use classification, with
all static and dynamic land surface fields masked based on
the IGBP/MODIS land-sea mask. The soil properties were
represented by the State Soil Geographic (STATSGO) [35]
database. Additional parameters for the LIS/Noah runs include
monthly climatologies of albedo [36], a 0.05° resolution
maximum snow surface albedo derived from MODIS [37],
the default AVHRR-based monthly GVF climatology in the
control LIS simulation [2], and a deep soil temperature
climatology (serving as a lower boundary condition for the
soil layers) at 3 m below ground, derived from 6 years of
NCEP Global Data Assimilation System (GDAS) 3-hourly
averaged 2-m air temperatures using the method described
in [3]. The motivation behind selecting the LIS parameters
described above is to use the same parameters as run in routine
operational land analyses at NCEP.
A. Comparison of Real-Time MODIS and Climatology GVF
Fig. 1 shows a sample comparison between the real-
time MODIS and AVHRR climatology GVF datasets from
17 July 2010. This day was selected to demonstrate the
increased resolution as well as how the real-time MODIS
GVF adjusts to regional precipitation anomalies. On this
day, the MODIS GVF values tended to be higher than the
time-interpolated AVHRR climatology in the western half of
the domain, especially over the High Plains from Texas to
Canada. The MODIS GVF provided much greater detail and
spatial variability in the complex terrain of the intermountain
Fig. 1. Depiction of the GVF on the CONUS domain for the (a) AVHRR-
based monthly climatology time-interpolated to 17 July, (b) SPoRT/MODIS
GVF from 17 July 2010, and (c) difference (MODIS – Climatology). Panel
(c) shows the quadrants of the domain used for the analysis in Section IV.
west. The improved resolution provided by MODIS and the
compositing approach helps to explain why many narrow strips
of positive GVF differences occurred in the west. The AVHRR
climatology dataset depicts GVF over much larger grid cells
than the MODIS product (200 times larger), thus averaging
out many of the horizontal variations related to the complex
terrain. As a result, the more heavily-vegetated corridors at
higher elevations are often too small to be resolved by the
16-km climatology product, and thereby take on a lower mean
GVF value largely based on the surrounding low-elevation
desert regions. Meanwhile, the real-time MODIS GVF is able
Fig. 2. Departure from average precipitation (in percent; available
from the National Weather Service River Forecast Center analyses at during 2010 for the months of (a) May,
(b) June, and (c) July. Regions of substantially above-average precipitation
are highlighted each month in the white outlines.
to resolve the heavily-vegetated foothills and mountainous
regions. A good example can be seen in the GVF differ-
ence in the Sierra Nevada range in east-central California
[Fig. 1(c)]. Elsewhere, slightly lower GVF occurred in regional
patches across the eastern U.S. and southeastern Canada, with
otherwise minor differences prevailing.
The higher GVF over the High Plains resulted from substan-
tial positive precipitation anomalies from May to July 2010.
Fig. 2(a) shows that much of the northern High Plains experi-
enced very substantial rainfall anomalies during May, ranging
from 150% to as much as 600% of normal rainfall, especially
from northeastern Montana to western South Dakota. The
month of June 2010 saw much of the same, but with a
southerly shift in the maximum positive rainfall anomaly into
the Central Plains [Fig. 2(b)]. In July [Fig. 2(c)], very large
precipitation anomalies shifted to the Southern Plains as a
result of early season tropical activity affecting Texas. The
Fig. 3. Correspondence between the SPoRT/MODIS one-year GVF change
and U.S. Drought Monitor product from summer 2010 to 2011 (a) one-year
GVFdifference(2Aug.2011–2Aug.2010), (b) U.S. Drought Monitor
product issued 3 August 2010, and (c) U.S. Drought Monitor product issued
2 August 2011.
higher real-time MODIS GVF are consistent with the positive
rainfall anomalies, as NDVI has been found to have a strong
correlation to antecedent precipitation in the preceding weeks
and months over the U.S. Central Plains [38].
The real-time MODIS GVF also responded in the opposite
sense during 2011, associated with the onset of intense heat
and severe drought during the summer months in the U.S.
Southern Plains. The one-year change in MODIS GVF valid
on 2 August 2011 shows a marked decrease across much of
Texas, Oklahoma, and southern Kansas [Fig. 3(a)]. At the
same time in the previous year, the U.S. Drought Monitor
product essentially depicted no abnormally dry or drought con-
ditions across the U.S. Southern Plains [Fig. 3(b), except for
Fig. 4. MODIS-based IGBP land use classification over the real-time GVF
CONUS domain, according to the scale provided at the bottom. Some low-
coverage categories are grouped together for simplicity sake.
Fig. 5. Time series of daily MODIS GVF and time-interpolated AVHRR
climatological GVF from 1 June 2010 to 1 January 2012 by land use
classification. Land use classes and the percent coverage in the CONUS
domain include (a) Grassland, (b) Cropland, (c) Open Shrubs, (d) Mixed
Forest, (e) Evergreen Needleleaf, (f) Deciduous Broadleaf, (g) Barren/Sparse,
and (h) Urban/Built-Up.
a small regional drought in parts of Louisiana and Arkansas].
The drought status changed dramatically by early August
2011, with the U.S. Drought Monitor product indicating
widespread D4 drought conditions from New Mexico eastward
to Louisiana [Fig. 3(c)]. The area of negative one-year GVF
Fig. 6. Departure from average temperature (°F, source: High Plains Regional Climate Center at for the time periods
of (a) September to November 2010, (b) December 2010, (c) March to May 2011, and (d) October to December 2011.
differences coincides with much of the 2 August 2011 D4
region (with the exception of sparsely-vegetated New Mexico).
This analysis demonstrates the need for a real-time GVF
product in models that can represent inter-annual vegetation
changes in response to regional climate anomalies.
Time series plots of the real-time MODIS versus AVHRR
climatological GVF are presented for several of the
IGBP/MODIS land use classes (Fig. 4) during 2010 and 2011
to compare the seasonal progressions of GVF. The time series
comparisons of the land use classes with the greatest coverage
on the CONUS domain of Fig. 4 are highlighted in Fig. 5. The
seasonal march of GVF compared similarly among most veg-
etation classes and dates, with a few noteworthy exceptions.
The real-time MODIS mixed forest and deciduous broadleaf
categories [Fig. 5(d), (f)] had slightly damped seasonal cycles
relative to the AVHRR climatology, whereas the cropland and
evergreen needleleaf experienced comparable magnitudes in
seasonal change [Fig. 5(b), (e)].
The shrubs category [Fig. 5(c)] had the largest systematic
change, with an increase in GVF over the AVHRR climatology
of 0.15–0.20 in the real-time MODIS product, especially
during summer 2010. This behavior is consistent with the
results of [39], who found that MODIS NDVI were con-
sistently higher than AVHRR NDVI over semi-arid regions.
These results are further supported by [40], who compared
MODIS to AVHRR NDVI and found that during the grow-
ing season, MODIS NDVI exhibited greater sensitivity and
dynamic range than AVHRR NDVI due to the influence of
atmospheric water vapor on the AVHRR NIR band. The
narrower MODIS NIR band is more impervious to the water
vapor absorption portion of the spectrum, whereas the AVHRR
NIR band is strongly affected by water vapor absorption,
leading to lower NDVI values particularly during the humid
growing season [40].
The grassland class [Fig. 5(a), most prevalent over the High
Plains in Fig. 4] was also systematically higher during the
warm seasons, especially summer 2010 when high rainfall
anomalies occurred in this region. The cropland, barren, and
urban vegetation categories [Fig. 5(b), (g), and (h)] had the
most consistent seasonal comparison between the AVHRR
climatology and real-time MODIS GVF. This is particularly
true for the cropland category, likely attributed to consistent
farming practices and irrigation that yields similar signals in
NDVI from year to year.
A noteworthy feature in the time series charts is the apparent
phase shift in the MODIS GVF seasonal response relative to
the AVHRR climatology, which appears related to temperature
anomalies. First, the Autumn 2010 months were considerably
above average in temperature across much of the CONUS
[Fig. 6(a)], which likely contributed to a delayed decrease
in the MODIS GVF, most notable in the mixed forest and
needleleaf categories [Fig. 5(d) and (e)]. This was followed by
a precipitous drop in temperatures during December 2010 over
the northern and eastern U.S. [Fig. 6(b)] leading to the sharp
drop in MODIS GVF in the forest categories [Fig. 5(d)–(f)].
Second, cooler-than-average temperatures prevailed across the
northern and western U.S. in Spring 2011 [Fig. 6(c)], possibly
contributing to the slight lag seen in the evergreen needleleaf
category that predominates the northwest portion of the U.S.
[Figs. 4 and 5(e)]. Third, warmer-than-average temperatures
occurred once again during the Autumn 2011 months (espe-
cially in the northern U.S. [Fig. 6(d)] helping to slow the
decline of MODIS GVF relative to the climatology in the
forest categories [Fig. 5(d)–(f)].
B. Impact on LIS-Noah Offline Model Runs
To illustrate the impact of real-time MODIS GVF data,
LIS was configured to run version 2.7.1 of the Noah LSM
at 1-km grid spacing in an uncoupled, or offline mode using
climatology GVF versus real-time MODIS GVF during the
2010 warm season. First, a limited time-space simulation
comparison was made over an area of complex terrain centered
on Montana during June 2010 to show the impact of enhanced
MODIS resolution on the spatial distribution of fluxes. Second,
LIS was run for the entire 2010 warm season (June to October)
over the real-time MODIS GVF domain, decomposed into
the four quadrants shown in Fig. 1(c) to examine the mean
impact of the MODIS GVF in each region. For the full-
domain simulations, LIS was cold-started on 1 June 2008
with a uniform first-guess soil temperature and volumetric soil
moisture of 290 K and 25%, respectively, in all soil layers.
LIS-Noah was integrated for two years to 1 June 2010, using
a time step of 30 min. A sufficiently long integration time,
or “spin-up,” is necessary to ensure that the model states
can reach a fine-scale equilibrium with the forcing meteorol-
ogy [41], [42]. During the two-year spin-up integration, the
AVHRR-based GVF climatology was used. After 0000 UTC
1 June 2010, the spin-up run was re-started for two separate
offline integrations, a control run that continued using the
AVHRR climatological GVF, and an experimental run that
employed the daily MODIS GVF during the period of study
from 0000 UTC 1 June to 0000 UTC 1 November 2010.
In all LIS runs, atmospheric analyses from GDAS
[43]–[45] provided the required input fields to drive the LSM
integrations. The GDAS forcing fields of downward-directed
longwave radiation, surface pressure, 2-m air temperature,
and 2-m specific humidity are corrected topographically via
lapse-rate and hypsometric adjustments using the elevation
data differences between the LIS and native GDAS forcing
grids [41].
The Montana high-resolution run reveals detailed differ-
ences between the AVHRR climatology and the real-time
MODIS GVF related to the ability to resolve complex terrain
features. The climatology appears quite smooth on this scale
showing a minimum in GVF over the prairies of eastern
Montana, northern Wyoming, western South Dakota, and
central Idaho [Fig. 7(a)]. A broad maximum exists along the
Montana-Idaho border associated with the high terrain along
the Continental Divide. Meanwhile, the MODIS GVF field
shows much greater detail, able to resolve small river basins
with locally higher GVF and relatively lower GVF in the
higher mountain peaks in northwestern Wyoming and parts
of western Montana [Fig. 7(b) and (c)]. The difference field
shows that the real-time MODIS data depict much higher GVF
along the High Plains in the western Dakotas, and eastern
Montana and Wyoming [Fig. 7(c)].
Fig. 7. Depiction of the GVF on a domain with 1-km grid spacing centered
on Montana at 1800 UTC 27 June 2010 for (a) monthly GVF climatology
time-interpolated to 27 June, (b) real-time daily MODIS GVF valid on 27 June
2010, and (c) difference (MODIS – Climatology).
These variations in GVF directly impact the surface energy
budget through the partitioning of sensible and latent heat
fluxes. With the ability to better resolve the ridgetops, the
LIS-Noah run with the MODIS GVF depicted higher sensible
heat fluxes and substantially lower latent heat fluxes up to
100 Wm2over the high terrain where the MODIS GVF was
lower than the AVHRR climatology (Fig. 8). Meanwhile, the
sensible and latent heat fluxes were regionally altered by about
the same magnitude but the opposite sense over parts of the
Fig. 8. Differences in LIS-Noah simulated heat fluxes (Wm2, MODIS GVF
run minus AVHRR Climatology GVF run) valid at 1800 UTC 27 June 2010
for (a) sensible heat flux and (b) latent heat flux.
High Plains where the MODIS GVF was higher. Worth noting
in the difference field is the amount of local, detailed variations
consistent with the terrain elevation features of this region (not
A comparison of the real-time MODIS and AVHRR cli-
matology GVF on the full domain, broken up into the four
regions in Fig. 1(c), indicates that the MODIS GVF values
were on average higher than the AVHRR climatology in the
western half of the domain (Fig. 9). The northeast quad-
rant MODIS GVF means were slightly below the AVHRR
climatology GVF during July and August, then transitioned
to higher GVF by late summer and autumn. There were no
obvious weather anomalies in the northeast associated with
the lower summer GVF; however, the autumn featured well
above average temperatures across much of the eastern U.S.
[Fig. 6(a)], which likely caused the slower decline in the
real-time MODIS GVF relative to the AVHRR climatology.
The MODIS GVF remained closest to the climatology in the
southeast quadrant during most of the period of record. Like
the northeast quadrant, it was not until late in the warm season
when the MODIS GVF became consistently higher than the
AVHRR climatology.
The differences in the GVF datasets impacted the mean
surface fluxes and soil moisture output from the offline
Fig. 9. Comparison of the time-interpolated AVHRR monthly climatology
to the 2010 daily MODIS GVF, spatially-averaged over the four quadrants
depicted in Fig. 1(c). (a) Northwest. (b) Northeast. (c) Southwest. (d)
Fig. 10. Daily difference (MODIS – AVHRR Climatology) in the mean
sensible heat flux (solid lines), latent heat flux (dashed lines), and ground
heat flux (dotted lines) in LIS-Noah for the peak heating time period of
1800–2100 UTC, valid over the four quadrants depicted in Fig. 1(c).
(a) Northwest. (b) Northeast. (c) Southwest. (d) Southeast.
LIS-Noah integration. When the MODIS GVF was higher
than the AVHRR climatology, the mean 3-hourly latent heat
fluxes during peak daytime heating (i.e., 1800 to 2100 UTC)
increased by up to 30 Wm2, especially in the western half
of the domain (Fig. 10). The increase in the latent heat flux
resulted from the additional evapotranspiration due to the
higher coverage of healthy vegetation.
According to the surface energy balance equation
Rnet =Qh+Qle +Qg(4)
the net incoming radiation at the surface (Rnet ) is balanced
by the sensible heat flux (Qh), latent heat flux (Qle), and
heat flux into the ground (Qg). Assuming negligible changes
in surface albedo and all other factors being the same (i.e.,
identical atmospheric forcing for both model runs), an increase
Fig. 11. Daily difference in the mean volumetric soil moisture (MODIS
– AVHRR Climatology) in the top three LIS-Noah layers (0–10 cm,
10–40 cm, and 40–100 cm), valid over the four quadrants depicted in Fig. 1(c).
(a) Northwest. (b) Northeast. (c) Southwest. (d) Southeast.
in latent heat flux is offset by a corresponding decrease in
sensible heat flux and/or ground heat flux directed into the
soil, which is indeed the initial response during the peak
heating hours of 1800–2100 UTC over the western quadrants
[Fig. 10(a), (c)]. It should be noted that surface albedo was not
adjusted in these experiments due to changes in the input GVF,
nor was a dynamic surface albedo methodology implemented
into the offline simulations to modify albedo based on soil
moisture changes [46]. Newer versions of the Noah LSM
(v3.0+) enable adjustments to the surface albedo based on
the input GVF and enhanced look-up table attributes, resulting
in an albedo field consistent with the input GVF field. The
typical effect is to increase the surface albedo when the GVF is
smaller, resulting in a reduction of the net radiation. Although
not presented in this paper, these effects were seen in the U.S.
Upper Midwest associated with the severe drought of summer
2012 when using the real-time MODIS GVF compared to the
climatology GVF within LIS-Noah.
The increase in latent heat flux over the western half of
the domain had the most substantial impact on the volumetric
soil moisture. Fig. 11 shows the top three layers of mean
volumetric soil moisture, and how they responded to the
different GVF input. The soil dried more quickly in the
LIS-Noah run using MODIS GVF compared to the AVHRR
climatology, with volumetric soil moisture decreasing as much
as 0.015 (1.5%) or more. This more rapid soil drying was
a result of the increased mean latent heat flux caused by
the higher vegetation coverage, thereby extracting moisture
more quickly from the soil, particularly in the root zone
layers of 10–40 cm and 40–100 cm. The eastern half of the
domain experienced only marginal changes in volumetric soil
moisture, generally less than 0.5% by the end of the warm
An intriguing result of the offline LIS-Noah runs occurred
in the two western quadrants toward the end of the warm
season. While the latent heat flux remained higher in the
MODIS GVF simulations during the entire 2010 warm sea-
son [Fig. 10(a), (c)] because of the higher mean GVF
[Fig. 9(a), (c)], there was a distinct downward trend in the
difference fields beginning around 1 August in response to
the drier soil moisture in the MODIS GVF run. At the same
time, the mean sensible heat flux differences transitioned from
negative to positive, and remained slightly positive for the
remainder of the study period [Fig. 10(a), (c)]. The overall
drier soils over the western quadrants in the MODIS GVF run
enabled a slight increase in the mean skin temperature (not
shown) since less energy was required to evaporate moisture
from the soil medium, thereby producing the net increase
in mean sensible heat flux. The ground heat flux remained
consistently lower in the MODIS GVF run [Fig. 10(a), (c)]
resulting in cooler overall soil temperatures, especially in the
top 0–10-cm layer (not shown). Consequently, the overall
energy output directed to the atmosphere in the LIS-Noah
run using the real-time MODIS GVF increased relative to the
AVHRR climatology LIS run by late summer 2010 over the
western U.S., due to the increase in mean GVF.
C. Impact on Simulation of a Severe Weather Event
To assess the potential impact in NWP models, a
convection-allowing configuration of NU-WRF was employed
for sample severe weather events in 2010 and 2011. The
NU-WRF configuration mimicked the WRF model as run in
real-time at the National Severe Storms Laboratory in support
of the NCEP Storm Prediction Center and NOAA/National
Weather Service forecast offices [27]. In this paper, results are
highlighted from the 17 July 2010 severe weather event over
the U.S. Upper Midwest.
The NU-WRF was run on a single domain at 4-km horizon-
tal grid spacing over a CONUS domain (model configuration
details summarized in Table I). Two separate NU-WRF sim-
ulations were made: a control run based on the climatology
GVF (hereafter “control”), and an experimental run based on
SPoRT-MODIS GVF (hereafter “MODIS”). Each simulation
was initialized at 0000 UTC 17 July and integrated out 36 h
covering the event of interest. To provide a proper land surface
initialization, a separate LIS-Noah spin-up simulation was
conducted over an identical 4-km CONUS domain as the
NU-WRF model configuration. The Noah LSM was integrated
for 2 years prior to 1 June 2010 in the same manner as
described in the previous section. The offline LIS runs were
then continued beyond 1 June 2010 using the AVHRR clima-
tology and MODIS GVF data to provide LSM initialization
data. The control simulation was initialized with the GVF
climatology and land surface fields from the control offline
LIS, whereas the MODIS simulation was initialized with
SPoRT/MODIS GVF data and land surface data from the
offline LIS run incorporating the MODIS GVF.
The 17 July 2010 event featured an optimal scenario for
examining the sensitivity of the model to the new GVF dataset
in that surface heating was minimally impacted by prevailing
cloud cover prior to convective initiation. In the early morning
hours of 17 July, a low pressure center was located in
western North Dakota with a warm front extending east to
Fig. 12. Meteorological observations from 17 July 2010 showing (a) surface weather map analysis at 1200 UTC and (b) Storm Prediction Center storm
reports for the 24-h period ending 1200 UTC 18 July.
WRF dynamical core Advanced Research WRF
Horizontal grid spacing 4.0 km
Horizontal grid
(west-east by
1200 ×800 points
Number of vertical
sigma levels 35
Integration time step 24 s
Initial and boundary
conditions NCEP NAM model
Planetary boundary
layer Mellor-Yamada-Janji´c scheme
Cloud microphysics WRF 6-class Single Moment (WSM6)
Dudhia shortwave
Rapid Radiative Transfer Model
parameterization none
central Minnesota [Fig. 12(a)]. The low moved east during
the day as the warm front lifted north. A dryline developed
in the western Dakotas and propagated eastward, leading
to the development of severe convection in an extremely
unstable environment over the eastern Dakotas by 2100 UTC
(surface-based convective available potential energy (CAPE)
of 40005000+Jkg
1). Several tornadoes and numerous
high wind and large hail reports occurred in the eastern
Dakotas, central Minnesota, Iowa, Nebraska, and northern
Missouri [Fig. 12(b)]. In particular, an exceptionally damaging
convective complex tracked southeastward from northeast
South Dakota, through Iowa, and into northeastern Missouri.
As will be seen, the use of the real-time MODIS GVF had a
positive impact on the evolution of this specific precipitation
Both the control and MODIS model runs produced very
little cloudiness during the several hours of peak solar heating.
Consequently, the differences in GVF over the sub-set region
of interest [Fig. 13(a)] translated almost directly into a change
Fig. 13. Difference fields from the 4-km NU-WRF model runs initialized
at 0000 UTC 17 July 2010 (MODIS – control) sub-set over the Upper
Midwest region, showing changes in (a) input GVF (%), (b) simulated sensible
2), (c) simulated latent heat flux at 19 h
2), (d) simulated 2-m temperature at 21 h (°C), (e) simulated
2-m dewpoint temperature at 21 h (°C), and (f) simulated surface convective
available potential energy at 21 h (CAPE, J kg1).
in the partitioning of the incoming shortwave radiation into
sensible and latent heat fluxes. At the approximate time of
peak surface heating (1900 UTC, 19-h forecast), a reduction
in the sensible heat flux of 50 Wm2or more [Fig. 13(b)]
in conjunction with an increase in latent heat flux up to
50–100 Wm2[Fig. 13(c)] occurred in the eastern parts of
Fig. 14. (a) NU-WRF 1-h accumulated precipitation (mm) for the 27-h
forecast of the control run, (b) MODIS run, (c) difference (MODIS – control),
and (d) stage IV precipitation analysis, valid for the hour ending 0300 UTC
18 July 2010.
North Dakota, South Dakota, and Nebraska corresponding to
the 5–30% increase in GVF [Fig. 13(a)]. Similar changes of
the opposite sense were found in central/northern Minnesota
associated with lower MODIS GVF; however, there was some
contamination from clouds and precipitation there.
These modifications in the heat fluxes due to GVF translated
into changes in the 2-m temperature and dewpoint temperature
[Fig. 13(d) and (e)] at 21 h just prior to convective initiation in
the warm, moist sector. The western part of the focus area with
higher MODIS GVF [Fig. 13(a)] experienced a net decrease
(increase) in the 2-m temperature (dewpoint temperature),
typically on the order of 1–2°, although local increases in
dewpoint temperature exceeded 4°C. The slightly lower 2-m
temperatures in the MODIS run over South Dakota were more
in-line with surface observations; however, both model runs
generally under-estimated the 2-m temperatures over eastern
Nebraska, and southern and western Iowa (not shown).
The net effect in the 21-h forecast was an overall increase in
the surface CAPE [Fig. 13(f)], especially over the western part
of the focus area. Portions of eastern Nebraska to southeastern
North Dakota had CAPE increases over 1000 J kg1.Inthis
instance, the higher GVF over the warm sector led to a greater
influx of moisture into a shallower boundary layer (not shown),
resulting in a net increase in moist static energy per unit
mass within the boundary layer, despite small decreases in
the 2-m temperature. Similar sensitivity studies demonstrated
the influence of real-time AVHRR vegetation data on the
partitioning of heat fluxes, and simulated 2-m temperature and
dewpoint temperature in NWP models [13], [24], as well as
statistically significant improvements to the simulated most
unstable CAPE [24].
Fig. 15. (a) NU-WRF 1-h accumulated precipitation (mm) for the 33-h
forecast of the control run, (b) MODIS run, (c) difference (MODIS – control),
and (d) stage IV precipitation analysis, valid for the hour ending 0900 UTC
18 July 2010.
The impact of the GVF differences on model simulated
precipitation was fairly subtle initially. Convective precipi-
tation initiated at 21 h in both simulations over extreme
southeastern North Dakota (not shown) and evolved into a
bow-shaped line in southern Minnesota by 27 h (Fig. 14).
The difference in the 1-h simulation precipitation at 27 h
[Fig. 14(c)] suggests that the MODIS run was slightly slower
than the control run over southernMinnesota. Both simulations
incorrectly produced a nearly continuous line of precipitation
while the observed precipitation resembled a more discrete
mode [Fig. 14(d)]. Over the next several hours, the control sim-
ulation quickly moved the precipitation into northern Missouri,
while the MODIS run regenerated/back-built convection more
similar to the observed evolution.By 33 h (0900 UTC 18 July),
the location and intensity of the MODIS 1-h accumulated
precipitation were more closely aligned with the Stage IV
precipitation analysis compared to the control run (Fig. 15).
Objective verification statistics of the forecast 2-m tem-
perature and 2-m dewpoint temperature at surface observing
stations illustrate the systematic changes due to the new
GVF dataset over parts of the focus area. The mean error in
2-m temperature [Fig. 16(a)] shows cooler MODIS simulated
temperatures by about 1.5 °C (mainly after 12 h) in the
Northern Plains region (NPL) [Fig. 16(c)], which had the
largest increases in the MODIS GVF over the climatology
GVF [Figs. 1(c) and 13(a)]. While the cooler MODIS-NPL
forecast 2-m temperatures introduced a slight negative bias
after 18 h, the 2-m dewpoint temperature mean errors tended
to improve on the control dry bias between 12–17 h and
again after 25 h [Fig. 16(b)] (MODIS-NPL became slightly
too moist between 18–25 h). The MODIS-NPL mean error
Fig. 16. Select verification statistics over the Northern Plains (NPL) and Midwest (MDW) NCEP verification regions for the 17 July 2010 control and
MODIS NU-WRF simulations. (a) Mean error for 2-m temperature (°C). (b) Mean error for 2-m dewpoint temperature (°C). (c) Map of NCEP verification
regions. (d) Critical success index (CSI) for the 5 mm h1accumulated precipitation between forecast hours 12 and 36 over the MDW verification region.
The mean sample sizes per forecast hour are: 287 and 253 for 2-m temperature and dewpoint temperature in NPL; 1492 and 1398 for 2-m temperature and
dewpoint temperature in MDW.
shows systematically higher 2-m dewpoint temperatures up to
2°C [Fig. 16(b)], especially beginning with the onset of the
diurnal heating cycle after 12 h. Further east over the Midwest
verification region (MDW) [Fig. 16(c)], very little systematic
change is seen in the simulated 2-m temperature and dewpoint
temperature, consistent with the relatively small GVF changes
there [Figs. 1(c) and 13(a)].
Precipitation verification statistics support the previous
subjective interpretations, as indicated by the higher values
of Critical Success Index scores in the MODIS run, espe-
cially associated with the nocturnal precipitation after 31 h
[Fig. 16(d)]. The improved simulated precipitation during this
time was likely due to the higher residual CAPE in the MODIS
run over eastern Nebraska and western Iowa during these
forecast hours (not shown). This event represents an ideal case
in which strong surface heating was unimpeded by antecedent
cloud cover, precipitation, or strong atmospheric forcing. Thus,
differences in the input GVF led to substantial changes in
simulated low-level thermal, moisture, and instability fields,
which ultimately affected the evolution of simulated precipi-
tation systems in a favorable manner.
This paper presented a methodology for producing daily
NDVI and GVF composites at 0.01° (1 km) resolution in
real time using direct broadcast swaths of MODIS NDVI data.
The direct broadcast NDVI data are produced by the MODIS
Rapid Response System and because of the real-time dis-
semination strategy, atmospheric correction due to particulate
aerosols are not accounted for in the NDVI calculations. The
calculation of GVF from NDVI is handled by vegetation class
following the previous documented methodologies of [25],
[32]. The GVF composites are updated daily in real time and
can be used in the LIS, NU-WRF modeling system, and the
community WRF model.
The real-time MODIS GVF product was compared to
the global AVHRR-based climatological GVF on a 0.144°
grid (16 km) currently available to community modeling
systems, such as LIS and WRF. Due to the much higher
spatial resolution, the MODIS GVF was better able to resolve
horizontal variations in GVF, especially in the complex terrain
of the western U.S. The real-time MODIS GVF was shown
to respond to regional climate anomalies in precipitation and
temperature during 2010 and 2011, with substantial regional
differences at times. The U.S. High Plains experienced above-
average precipitation during the spring and summer of 2010,
which resulted in substantially higher MODIS GVF compared
to the AVHRR climatology. In addition, the onset of extreme
drought and heat over the U.S. Southern Plains during summer
2011 yielded a large annual decrease in MODIS GVF of up
to 40%.
A sample offline LIS-Noah LSM run centered on Montana
quantified changes to the sensible and latent heat flux up to
100 Wm2or more using the real-time MODIS versus the
AVHRR climatology GVF, also illustrating the impacts of
improved resolution in an area of complex terrain. Composite
LIS-Noah simulation results for the 2010 warm season showed
that areas of higher mean MODIS GVF led to net increases
(decreases) in latent (sensible) heat flux. However, after a few
months of integration, the mean soil moisture in the MODIS
GVF simulation became drier over the western U.S. due to the
increased rate of soil moisture extraction in regions of higher
GVF. The overall drier soils eventually led to an increase in
the mean sensible heat flux and ultimately, an increase in the
net total energy directed to the atmosphere in the western U.S.
An NWP case study of a warm-season severe weather event
illustrated how GVF differences affected the low-level thermal,
moisture, and instability fields that ultimately impacted the
precipitation evolution in a positive manner. Cases such as the
17 July 2010 event, where strong surface heating maximizes
the differences in heat fluxes, have the greatest potential to
benefit from the inclusion of real-time vegetation information.
Future work shall involve extending the NDVI/GVF com-
positing algorithm for use in future satellite platforms. Because
of the simple compositing algorithm, such a regional real-time
vegetation product could easily be applied to other current
and future sensors aboard both polar orbiting satellites (e.g.,
Visible Infrared Imager Radiometer Suite on Suomi-NPP)
and the Geostationary Operational Environmental Satellite-R
The authors would like to thank the constructive suggestions
given by three anonymous reviewers, which strengthened
the manuscript substantially. Computational resources for the
simulation work were provided by the NASA Center for
Climate Simulation at the NASA Goddard Space Flight Center.
Mention of a copyrighted, trademarked or proprietary product,
service, or document does not constitute endorsement thereof
by the authors, ENSCO Inc., Raytheon, the National Aero-
nautics and Space Administration, the SPoRT Center, or the
U.S. Government. Any such mention is solely for the purpose
of fully informing the reader of the resources used to conduct
the work reported herein.
[1] C. J. Tucker, J. A. Gatlin, and S. R. Schneider, “Monitoring vegetation
in the Nile delta with NOAA-6 and NOAA-7 AVHRR imagery,
Photogram. Eng. Remote Sens., vol. 50, no. 1, pp. 53–61, Jan. 1984.
[2] G. Gutman and A. Ignatov, “Derivation of green vegetation fraction
from NOAA/AVHRR for use in numerical weather prediction models,”
Int. J. Remote Sens., vol. 19, no. 8, pp. 1533–1543, Aug. 1998.
[3] F. Chen and J. Dudhia, “Coupling an advanced land-surface/hydrology
model with the Penn State/NCAR MM5 modeling system. Part I: Model
description and implementation,” Mon. Weather Rev., vol. 129, no. 4, pp.
569–585, Apr. 2001.
[4] M. B. Ek, K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren,
G. Gayno, and J. D. Tarpley, “Implementation of Noah land surface
model advances in the national centers for environmental prediction
operational mesoscale Eta model,” J. Geophys. Res., vol. 108, no. D22,
pp. 1–16, Nov. 2003.
[5] Z. I. Janjic, J. P. Gerrity, Jr., and S. Nickovic, “An alternative approach to
nonhydrostatic modeling,” Mon. Weather Rev., vol. 129, no. 5, pp. 1164–
1178, May 2001.
[6] Z. I. Janjic, A nonhydrostatic model based on a new approach,”
Meteorol. Atmos. Phys., vol. 82, nos. 1–4, pp. 271–285, Jan. 2003.
[7] J. Han and H.-L. Pan, “Revision of convection and vertical diffusion
schemes in the NCEP global forecast system,Weather Forecast.,
vol. 26, no. 4, pp. 520–533, Aug. 2011.
[8] F. Yang, H.-L. Pan, S. K. Krueger, S. Moorthi, and S. J. Lord,
“Evaluation of the NCEP global forecast system at the ARM SGP site,
Mon. Weather Rev., vol. 134, no. 12, pp. 3668–3690, Dec. 2006.
[9] S. Saha, S. Nadiga, C. Thiaw, and J. Wang, “The NCEP climate forecast
system,J. Climate, vol. 19, no. 15, pp. 3483–3517, Aug. 2006.
[10] S. Saha, S. Moorthi, H.-L. Pan, X. Wu, J. Wang, S. Nadiga, P. Tripp,
R. Kistler, J. Woollen, D. Behringer, H. Liu, D. Stokes, R. Grumbine,
G. Gayno, J. Wang, Y.-T. Hou, H.-Y. Chuang, H.-M. H. Juang, J. Sela,
M. Iredell, R. Treadon, D. Kleist, P. Van Delst, D. Keyser, J. Derber,
M. Ek, J. Meng, H. Wei, R. Yang, S. Lord, H. Van Den Dool, A. Kumar,
W. Wang, C. Long, M. Chelliah, Y. Xue, B. Huang, J.-K. Schemm,
W. Ebisuzaki, R. Lin, P. Xie, M. Chen, S. Zhou, W. Higgins, C.-Z. Zou,
Q. Liu, Y. Chen, Y. Han, L. Cucurull, R. W. Reynolds, G. Rutledge,
and M. Goldberg, “The NCEP climate forecast system reanalysis,” Bull
Amer. Meteorol. Soc., vol. 91, no. 8, pp. 1015–1057, Aug. 2010.
[11] K. E. Mitchell, D. Lohmann, P. R. Houser, E. F. Wood, J. C. Schaake,
A. Robock, B. A. Cosgrove, J. Sheffield, Q. Duan, L. Luo, R. W. Hig-
gins, R. T. Pinker, J. D. Tarpley, D. P. Lettenmaier, C. H. Marshall, J. K.
“The multi-institution north american land data assimilation system
(NLDAS): Utilizing multiple GCIP product and partners in a continental
distributed hydrological modeling system,” J. Geophys. Res., vol. 109,
no. 7, pp. 1–117, Apr. 2004.
[12] Y. Xia, M. Ek, H. Wei, and J. Meng, “Comparative analysis of relation-
ships between NLDAS-2 forcings and model outputs,” Hydrol. Process.,
vol. 26, no. 3, pp. 467–474, Jan. 2012.
[13] L. Jiang, F. N. Kogan, W. Guo, J. D. Tarpley, K. E. Mitchell,
M. B. Ek, Y. Tian, W. Zheng, C.-Z. Zou, and B. H. Ramsay, “Real-
time weekly global green vegetation fraction derived from advanced
very high resolution radiometer-based NOAA operational global veg-
etation index (GVI) system,” J. Geophys. Res., vol. 115, no. D11,
pp. D11114-1–D11114-22, Jun. 2010.
[14] W. C. Skamarock, J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker,
M. G. Duda, X.-Y. Huang, W. Wang, and J. G. Powers, “A description
of the advanced research WRF version 3,” NCAR, Boulder, CO, USA,
Tech. Rep. NCAR/TN-475+STR, Jun. 2008.
[15] T. N. Carlson and S. T. Arthur, “The impact of land use—Land cover
changes due to urbanization on surface microclimate and hydrology:
A satellite perspective,” Global Planet. Change, vol. 25, nos. 1–2,
pp. 49–65, Jul. 2000.
[16] S. L. Haines, G. J. Jedlovec, and S. M. Lazarus, “A MODIS sea surface
temperature composite for regional applications,” IEEE Trans. Geosci.
Rem. Sens., vol. 45, no. 9, pp. 2919–2927, Sep. 2007.
[17] L. Schiferl, K. K. Fuell, J. L. Case, and G. J. Jedlovec, “Evalu-
ation of enhanced high resolution MODIS/AMSR-E SSTs and the
impact on regional weather forecasts,” in Proc. 14th Symp. Integr.
Observ. Assimilat. Syst. Atmosph., Oceans, Land Surf., Jan. 2010,
pp. 1–11.
[18] S. V. Kumar, C. D. Peters-Lidard, Y. Tian, P. R. Houser, J. Geiger,
S. Olden, L. Lighty, J. L. Eastman, B. Doty, P. Dirmeyer, J. Adams,
K. Mitchell, E. F. Wood, and J. Sheffield, “Land information system—
An interoperable framework for high resolution land surface modeling,”
Environ. Model. Softw., vol. 21, no. 10, pp. 1402–1415, Oct. 2006.
[19] C.D. Peters-Lidard, P. R. Houser, Y. Tian, S. V. Kumar, J. Geiger,
S. Olden, L. Lighty, B. Doty, P. Dirmeyer, J. Adams, K. Mitchell,
E. F. Wood, J. Sheffield, “High-performance earth system modeling with
NASA/GSFC’s land information system,” Innov. Syst. Softw. Eng.,vol.3,
no. 3, pp. 157–165, Jul. 2007.
[20] J. L. Case, W. L. Crosson, S. V. Kumar, W. M. Lapenta, and
C. D. Peters-Lidard, “Impacts of high-resolution land surface initial-
ization on regional sensible weather forecasts from the WRF model,” J.
Hydrometeorol., vol. 9, no. 6, pp. 1249–1266, Dec. 2008.
[21] J. L. Case, S. V. Kumar, J. Srikishen, and G. J. Jedlovec, “Improving
numerical weather predictions of summertime precipitation over the
southeastern United States through a high-resolution initialization of the
surface state,” Weather Forecast., vol. 26, no. 6, pp. 785–807, Dec. 2011.
[22] T. M. Crawford, D. J. Stensrud, F. Mora, J. W. Merchant, and
P. J. Wetzel, “Value of incorporating satellite-derived land cover data in
MM5/PLACE for simulating surface temperatures,” J. Hydrometeorol.,
vol. 2, no. 10, pp. 453–468, Oct. 2001.
[23] N. P. Kurkowski, D. J. Stensrud, and M. E. Baldwin, “Assessment
of implementing satellite-derived land cover data in the Eta model,”
Weather Forecast., vol. 18, no. 3, pp. 404–416, Jun. 2003.
[24] K. A. James, D. J. Stensrud, and N. Yussouf, “Value of real-time
vegetation fraction to forecasts of severe convection in high-resolution
models,” Weather Forecast., vol. 24, no. 1, pp. 187–210, Feb. 2009.
[25] J. M. Miller, M. Barlage, X. Zeng, H. Wei, K. Mitchell, and D. Tarpley,
“Sensitivity of the NCEP/Noah land surface model to the MODIS
green vegetation fraction data set,Geophys. Res. Lett., vol. 33, no. 13,
pp. L13404-1–L13404-5, Jul. 2006.
[26] R. L. Ruhge and M. Barlage, “Integrating a real-time green vegetation
fraction (GVF) product into the land information system (LIS),” in Proc.
15th Symp. Integr. Observ. Assimilat. Syst. Atmosph., Oceans, Land Surf.,
Jan. 2011, pp. 1–12.
[27] J. S. Kain, S. R. Dembek, S. J. Weiss, J. L. Case, J. J. Levitt, and
R. A. Sobash, “Extracting unique information from high-resolution
forecast models: Monitoring selected fields and phenomena every time
step,Weather Forecast., vol. 25, no. 5, pp. 1536–1542, Oct. 2010.
[28] J. L. Case, F. J. LaFontaine, A. L. Molthan, B. T. Zavodsky, and
R. A. Rozumalski, “Recent upgrades to NASA SPoRT initialization
datasets for the environmental modeling system,” in Proc. 37th Nat.
Weather Assoc. Annu. Meeting, Oct. 2012, pp. 1–9.
[29] H.-L. Huang, L. E. Gumley, K. Strabala, J. Li, E. Weisz, T. Rink,
K. C. Baggett, J. E. Davies, W. L. Smith, and J. C. Dodge, “International
MODIS and AIRS processing package (IMAPP): A direct broadcast
software for the NASA earth observing system,” Bull. Amer. Meteorol.
Soc., vol. 85, no. 2, pp. 159–161, 2004.
[30] X. Xiao, B. Braswell, Q. Zhang, S. Boles, S. Frolking, and B. Moore, III,
“Sensitivity of vegetation indices to atmospheric aerosols: Continental-
scale observations in Northern Asia,” Remote Sens. Environ., vol. 84,
no. 3, pp. 385–392, Mar. 2003.
[31] M. A. Lazzara, J. M. Benson, R. J. Fox, D. J. Laitsch, J. P. Rueden,
D. A. Santek, D. M. Wade, T. M. Whittaker, and J. T. Young, “The
man computer interactive data access system: 25 years of interactive
processing,” Bull. Amer. Meteorol. Soc., vol. 80, no. 2, pp. 271–284,
Feb. 1999.
[32] X. Zeng, R. E. Dickinson, A. Walker, M. Shaikh, R. S. DeFries, and J.
Qi, “Derivation and evaluation of global 1-km fractional vegetation cover
data for land modeling,” J. Appl. Meteorol., vol. 39, no. 6, pp. 826–839,
Jun. 2000.
[33] T. R. Loveland, B. C. Reed, J. F. Brown, D. O. Ohlen, Z. Zhu, L. Yang,
and J. W. Merchant, “Development of a global land cover characteristics
database and IGBP DISCover from 1 km AVHRR data,Int. J. Remote
Sens., vol. 21, nos. 6–7, pp. 1303–1330, 2000.
[34] M. A. Friedl, D. Sulla-Menashe, B. Tan, A. Schneider, N. Ramankutty,
A. Sibley, and X. Huang, “‘MODIS collection 5 global land cover:
Algorithm refinements and characterization of new datasets,” Remote
Sens. Environ., vol. 114, no. 1, pp. 168–182, Jan. 2010.
[35] D. A. Miller and R. A. White, “A conterminous United States multi-
layer soil characteristics data set for regional climate and hydrology
modeling,” Earth Interact., vol. 2, no. 2, pp. 1–26, Jan. 1998.
[36] B. P. Briegleb, P. Minnis, V. Ramanathan, and E. Harrison, “Compar-
ison of regional clear-sky albedos inferred from satellite observations
and model computations,” J. Climate Appl. Meteorol., vol. 25, no. 2,
pp. 214–226, Feb. 1986.
[37] M. Barlage, X. Zeng, H. Wei, and K. E. Mitchell, “A global 0.05° max-
imum albedo dataset of snow-covered land based on MODIS observa-
tions,” Geophys. Res. Lett., vol. 32, no. 17, pp. L17405-1–L17405-5,
Sep. 2005.
[38] J. Wang, P. M. Rich, and K. P. Price, “Temporal responses of NDVI
to precipitation and temperature in the central Great Plains, USA,”
Int. J. Remote Sens., vol. 24, no. 11, pp. 2345–2364, Nov. 2003.
[39] R. Fensholt and I. Sandholt, “Evaluation of MODIS and NOAA AVHRR
vegetation indices with in situ measurements in a semi-arid environ-
ment,Int. J. Remote Sens., vol. 26, no. 12, pp. 2561–2594, Dec. 2005.
[40] A. Huete, K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Fer-
reira, “Overview of the radiometric and biophysical performance of the
MODIS vegetation indices,” Remote Sens. Environ., vol. 83, nos. 1–2,
pp. 195–213, Nov. 2002.
[41] B. A. Cosgrove, “Real-time and retrospective forcing in the North
American land data assimilation system (NLDAS) project,” J. Geophys.
Res., vol. 108, no. D22, pp. 1–12, Oct. 2003.
[42] M. Rodell, P. R. Houser, A. A. Berg, and J. S. Famiglietti, “Evaluation
of 10 methods for initializing a land surface model,” J. Hydrometeorol.,
vol. 6, no. 2, pp. 146–155, Apr. 2005.
[43] D. F. Parrish and J. C. Derber, “The national meteorological center’s
spectral statistical-interpolation analysis system,” Mon. Weather Rev.,
vol. 120, no. 8, pp. 1747–1763, Aug. 1992.
[44] J. C. Derber and W.-S. Wu, “The use of TOVS cloud-cleared radiances
in the NCEP SST Analysis System,” Mon. Weather Rev., vol. 126, no. 8,
pp. 2287–2299, Aug. 1998.
[45] SSI Analysis System, NCEP Environmental Modeling Center, College
Park, MD, USA, Apr. 2004.
[46] B. F. Zaitchik, J. A. Santanello, S. V. Kumar, and C. D. Peters-
Lidard, “Representation of soil moisture feedbacks during drought in
NASA unified WRF (NU-WRF),” J. Hydrometeorol., vol. 14, no. 1,
pp. 360–367, Jan. 2013.
Jonathan L. Case has performed land surface
and numerical weather prediction sensitivity studies
using NASA datasets for the Short-term Prediction
Research and Transition (SPoRT) team since 2006.
He has transitioned several unique real-time NASA
datasets into operations within the National Weather
Service forecast offices by configuring local mod-
eling platforms to incorporate the datasets. He runs
and manages a high-resolution, real-time configura-
tion of the NASA Land Information System running
the Noah land surface model, which provides input
land surface initial conditions to local modeling systems over the eastern U.S.
Prior to joining SPoRT, he served under the Applied Meteorology Unit at
Cape Canaveral, FL, USA, for nearly eight years, supporting the U.S. Space
Program by transitioning technology into operations for the U.S. Air Force
45th Weather Squadron, the National Weather Service (NWS) Spaceflight
Meteorology Group, and the NWS forecast office at Melbourne, FL. He has
authored more than 50 conference proceedings, extended abstracts, reports,
and refereed journal articles since 1998.
Frank J. LaFontaine has over 20 years of experi-
ence in meteorology, research, and software develop-
ment. His work includes software/algorithm devel-
opment and analyses of NASA and NOAA satel-
lite data and products for Short-term Prediction
Research and Transition (SPoRT). He also serves as
a Scientist and Software Developer for the NASA
Marshall Space Flight Center airborne research
instruments AMPR and C-STAR. He has partici-
pated in 17 NASA field experiments. Software skills
include FORTRAN-77, McIDAS, CSH, Perl, and
some experience with IDL, C, HDF, and HTML. He has co-authored numerous
peer-reviewed papers, conference preprints, and posters.
Mr. LaFontaine is a member of the American Meteorological Society.
Jordan R. Bell is a senior atmospheric science
student at the University of Missouri, Columbia,
He was an intern at the NASA Short-term Pre-
diction Research and Transition (SPoRT) Center
in 2011, during which he worked with SPoRT
scientists to evaluate the impacts of the real-time
SPoRT/MODIS Green Vegetation Fraction on land
surface and numerical weather prediction models.
He has presented results of the intern project in
New Orleans, LA, USA, at the 92nd American
Meteorological Society annual meeting in January 2012.
Gary J. Jedlovec has spent most of the last 25 years
developing and evaluating algorithms to retrieve
geophysical parameters from remotely sensed air-
craft and satellite measurements for regional climate
studies and weather forecasting applications. He
has authored papers on these approaches including
the retrieval of land and sea surface temperatures,
and atmospheric water vapor, approaches to detect
clouds in polar and geostationary satellite imagery,
and in the assimilation of AIRS radiance and profile
data to improve short-term weather forecasts. He is
currently leading an effort to transition the use of unique NASA EOS satellite
data (from MODIS, AMSR-E, and AIRS instruments) into various National
Weather Service forecast offices and other operational end users to improve
short-term weather forecasts on a regional and local scale.
Sujay V. Kumar received the Ph.D. degree in civil
engineering with emphasis on computer-aided high-
performance computing applications from North
Carolina State University, Raleigh, NC, USA, in
He is currently a Senior Scientist with Science
Applications International Corporation and conducts
research in the Hydrological Sciences Laboratory,
NASA Goddard Space Flight Center. He is the Chief
Architect of the NASA Land Information System, a
high-performance platform for high-resolution land
surface modeling and data assimilation. His current research interests include
land surface modeling and data assimilation, land-atmosphere interaction stud-
ies, computational modeling, optimization, and high-performance computing.
Christa D. Peters-Lidard received the B.S. (summa
cum laude) degree in geophysics and a minor in
mathematics from Virginia Polytechnic Institute and
State University (Virginia Tech), Blacksburg, VA,
USA, in 1991, and the M.A. and Ph.D. degrees
in the Water Resources Program from Princeton
University, Princeton, NJ, USA, in 1993 and 1997,
She is currently the Chief of the Hydrological
Sciences Laboratory, NASA’s Goddard Space Flight
Center, Greenbelt, MD, USA, where she has been a
Physical Scientist since 2001. She was an Assistant Professor with the School
of Civil and Environmental Engineering, Georgia Institute of Technology,
Atlanta, GA, USA, from 1997 to 2001. Her current research interests include
land-atmosphere interactions, soil moisture measurement and modeling, and
the application of high-performance computing and communications technolo-
gies in Earth system modeling.
Dr. Peters-Lidard is currently the Chief Editor for the American Meteoro-
logical Society (AMS) Journal of Hydrometeorology, and an elected member
of the AMS Council. She has also served as an Associate Editor for the
Journal of Hydrology and Water Resources Research. Her Land Information
System Team received the 2005 NASA Software of the Year Award. She is
a member of Phi Beta Kappa, and was awarded the Committee on Space
Research Scientific Commission A Zeldovich Medal in 2004 and the Arthur
S. Flemming Award in 2007. She was elected as an AMS Fellow in 2012.
... Large-scale LSMs without dynamic vegetation modeling are strongly limited by the assumption that vegetation has a recurring annual cycle, i.e., using climatological LAI and GVF input. In reality, the vegetation's response to meteorological and climate conditions varies due to interand intra-annual weather and climate anomalies (Case et al., 2013). ...
... In addition, they can also be assimilated for state updating in LSMs with dynamic vegetation simulation (Sabater et al., 2008;Barbu et al., 2011Barbu et al., , 2014Albergel et al., 2017;Kumar et al., 2019). Earlier studies indicated that replacing climatological vegetation by interannually varying satellite-derived indices can improve modeled energy fluxes as well as surface temperature and moisture in both offline LSM simulations (Miller et al., 2006;Case et al., 2013;Yin et al., 2016) and atmosphere-coupled LSMs (Crawford et al., 2001;James et al., 2009;Boussetta et al., 2013;Ge et al., 2014;Kumar et al., 2014). The largest improvements are obtained during extreme meteorological anomalies (Case et al., 2013). ...
... Earlier studies indicated that replacing climatological vegetation by interannually varying satellite-derived indices can improve modeled energy fluxes as well as surface temperature and moisture in both offline LSM simulations (Miller et al., 2006;Case et al., 2013;Yin et al., 2016) and atmosphere-coupled LSMs (Crawford et al., 2001;James et al., 2009;Boussetta et al., 2013;Ge et al., 2014;Kumar et al., 2014). The largest improvements are obtained during extreme meteorological anomalies (Case et al., 2013). In this study, it is expected that, besides meteorological anomalies, large-scale land cover conversions, such as deforestation, also alter the vegetation strongly from its climatological representation. ...
Full-text available
In this study, we tested the impact of a revised set of soil, vegetation and land cover parameters on the performance of three different state-of-the-art land surface models (LSMs) within the NASA Land Information System (LIS). The impact of this revision was tested over the South American Dry Chaco, an ecoregion characterized by deforestation and forest degradation since the 1980s. Most large-scale LSMs may lack the ability to correctly represent the ongoing deforestation processes in this region, because most LSMs use climatological vegetation indices and static land cover information. The default LIS parameters were revised with (i) improved soil parameters, (ii) satellite-based interannually varying vegetation indices (leaf area index and green vegetation fraction) instead of climatological vegetation indices, and (iii) yearly land cover information instead of static land cover. A relative comparison in terms of water budget components and “efficiency space” for various baseline and revised experiments showed that large regional and long-term differences in the simulated water budget partitioning relate to different LSM structures, whereas smaller local differences resulted from updated soil, vegetation and land cover parameters. Furthermore, the different LSM structures redistributed water differently in response to these parameter updates. A time-series comparison of the simulations to independent satellite-based estimates of evapotranspiration and brightness temperature (Tb) showed that no LSM setup significantly outperformed another for the entire region and that not all LSM simulations improved with updated parameter values. However, the revised soil parameters generally reduced the bias between simulated surface soil moisture and pixel-scale in situ observations and the bias between simulated Tb and regional Soil Moisture Ocean Salinity (SMOS) observations. Our results suggest that the different hydrological responses of various LSMs to vegetation changes may need further attention to gain benefits from vegetation data assimilation.
... The variables taken or derived from satellite data were elevation (RCMRD Geoportal, 2015); precipitation (Funk et al., 2015); surface water availability (Senay et al., 2013); baseflow, runoff, soil moisture, and actual evapotranspiration (Case et al., 2014) (Vargas et al., 2015); and population density (Stevens et al., 2015). More detail about the actual nature of the features is given below in Section 2.4. ...
... The Noah-Multiparameterization (Noah-MP) LSM relies on water and energy balances to physically describe land-atmosphere interaction processes under multiple specifications (Niu et al., 2011). Mean daily soil moisture (m 3 /m 3 ) at depths of 0-10 cm, 10-40 cm, 40-100 cm, and 100-200 cm and total monthly actual evapotranspiration (mm), baseflow (mm), and runoff (mm) were derived from local instances of the Noah-MP LSM run per month (Case et al., 2014). ...
Full-text available
Groundwater is an important source of water for people, livestock, and agriculture during drought in the Horn of Africa. In this work, areas of high groundwater use and demand in drought-prone Kenya were identified and forecasted prior to the dry season. Estimates of groundwater use were extended from a sentinel network of 69 in-situ sensored mechanical boreholes to the region with satellite data and a machine learning model. The sensors contributed 756 site-month observations from June 2017 to September 2021 for model building and validation at a density of approximately one sensor per 3700 km². An ensemble of 19 parameterized algorithms was informed by features including satellite-derived precipitation, surface water availability, vegetation indices, hydrologic land surface modeling, and site characteristics to dichotomize high groundwater pump utilization. Three operational definitions of high demand on groundwater infrastructure were considered: 1) mechanical runtime of pumps greater than a quarter of a day (6+ hrs) and daily per capita volume extractions indicative of 2) domestic water needs (35+ L), and 3) intermediate needs including livestock (75+ L). Gridded interpolation of localized groundwater use and demand was provided from 2017 to 2020 and forecasted for the 2021 dry season, June–September 2021. Cross-validated skill for contemporary estimates of daily pump runtime and daily volume extraction to meet domestic and intermediate water needs was 68%, 69%, and 75%, respectively. Forecasts were externally validated with an accuracy of at least 56%, 70%, 72% for each groundwater use definition. The groundwater maps are accessible to stakeholders including the Kenya National Drought Management Authority (NDMA) and the Famine Early Warning Systems Network (FEWS NET). These maps represent the first operational spatially-explicit sub-seasonal to seasonal (S2S) estimates of groundwater use and demand in the literature. Knowledge of historical and forecasted groundwater use is anticipated to improve decision-making and resource allocation for a range of early warning early action applications. Supporting local groundwater management would improve resilience to drought. Kenya groundwater use estimated with sensors, satellite data, and machine learning. Historical use was modeled with up to 75% cross-validated accuracy. Forecasts for the 2021 dry season indicated up to 80% external validation accuracy.
... Passive microwave measurements are affected by satellite incidence angles and are attenuated by vegetation canopies that dampen soil emissions from the ground (Dobson et al., 1985;Pellarin et al., 2016;Su et al., 2020). LSM also has limitations originating from uncertainties associated with model parameterization, such as climatology-based vegetation parameterization (Case et al., 2014;Fang et al., 2018b;Liu et al., 2011;Ullah et al., 2018). The simplified formulation of physical processes, such as the 1-d formulation of vertical water flow, can also cause errors in LSM-based SM data (Balsamo et al., 2009;Bi et al., 2016;Chen et al., 2016). ...
... For example, the backscatter coefficient of ASCAT SM is largely dependent on the local incidence angle and vegetation canopy that can cause multiple interactions, volume scattering, and double bounce (Dobson and Ulaby, 1986;Ulaby et al., 1974). The fixed climatologybased vegetation parametrization (Case et al., 2014;Fang et al., 2018a;Jiang et al., 2010) and simple 1-d vertical formulation of the water balance in Noah LSM may also cause a biased SM estimate (Bi et al., 2016;Chen et al., 2016;Pitman, 2003). With the exception of bias, NCA-LDAS generally yielded better performance metrics when compared to NLDAS-2, which shared the same Noah LSM without DA, even when TC and VOD values were high. ...
Three widely used primary soil moisture (SM) data sources, namely, in-situ measurements, satellite observations, and land surface models (LSM), possess different characteristics. This study combined three SM data sources using machine learning (ML): random forest, artificial neural networks, and support vector regression, and simple averaging ensemble approaches to produce improved daily SM data over the contiguous United States (CONUS). For each ML model, three schemes were tested using different independent variables, namely, satellite-derived, LSM-derived, and both. Triple collocation analysis (TCA) was adopted to address the scale mismatch problem between in-situ and coarse gridded SM data. The proposed approach was evaluated using the International Soil Moisture Network (ISMN), Soil Moisture Active Passive Core Validation Sites (SMAP CVS), and TCA. In the ISMN-based evaluation, the proposed ML-based ensemble generally produced better evaluation metrics and showed robust skills over topographically complex and densely vegetated regions where existing SM products showed poor skills. The SMAP CVS-based evaluation demonstrated that the ML ensemble approach yielded a better performance than the existing SM datasets, resulting in a correlation coefficient of 0.78, unbiased root mean squared difference of 0.035 m³/m³, and bias of 0.006 m³/m³. In addition, the TCA results additionally confirmed that the ML-based ensemble had better spatiotemporal quality than the other SM products. The data-driven approach proposed in this study has three major novelties: (1) the proposed ML-based method synergistically merges various data sources to improve SM; (2) the performance of the proposed ML-based SM was robust to topography and vegetation; and (3) the average ensemble of three ML models additionally improves performances. The SM time-series data generated by the proposed approach are expectedly suitable variables for environmental and climate applications over CONUS. The research findings suggest that ML algorithms can be effectively used for modeling dynamic soil moisture.
... The physical model method is to establish a model about the physical relationship between vegetation spectral information and FVC by studying the interaction between light and vegetation [13], [14]. Due to the problem of mixed pixels in satellite imagery, the dimidiate pixel model is proposed to distinguish vegetation and background information [10]. Several studies estimate global FVC based on VIIRS surface reflectance data using machine learning methods, such as back propagation neural networks and general regression networks [12].Some studies use deep learning regression models to estimate FVC in savanna ecosystems [11]. ...
... In natural scenes, the dynamic brightness range refers to the ratio of the highest light intensity to the lowest light intensity [39]. The dynamic range of an image can be defined as the logarithmic ratio between the largest and smallest readable signal in a given scene [44] Dynamic Range (dB) = 20 × log 10 Max Signal Min Signal . ...
Full-text available
Measured fractional vegetation cover (FVC) on the ground is very important for validation of the remote sensing products and algorithms. However, because of the influence of some factors such as the angle of illumination and vegetation density, the existence of vegetation shadows limits the accuracy of FVC estimation. This article proposes a deep learning method to reduce the FVC estimation error based on high dynamic range (HDR) images with vegetation shadows (HDR REC-DL method). The HDR REC-DL method can accurately extract FVC from HDR images with complex texture information on vegetation shadows. This method is based on the U-Net convolutional network structure for semantic segmentation of images containing vegetation shadows, and the segmentation results are less affected by vegetation types. Results from the HDR REC-DL method were highly similar to the vegetation segmentation results from visual interpretation. Values of the kappa coefficient, F1 score (F1), recall, and mean intersection over union of the HDR REC-DL method were 0.926, 0.942, 0.924, 0.916 for sunny weather and 0.903, 0.974, 0.983, and 0.895 for cloudy weather, respectively. Compared with the vegetation segmentation accuracy of the shadow-resistant algorithm, the HDR REC-DL method increases the kappa coefficient, F1, and mIOU by 21%, 16%, and 29% for sunny weather, and by 11.1%, 3.6%, and 10.3% for cloudy weather, respectively. The HDR REC-DL method provides a novel method for accurately estimating FVC from images containing vegetation shadows.
... The real-time daily GVF analyses were used to overwrite the default monthly climatological vegetation fraction data used by the WRF model at 00:00 UTC on each day. Using real-time, satellite-derived GVF instead of a monthly GVF climatology has been shown to improve the representation of the surface energy budget and subsequent model forecasts during the warm season (Case et al., 2014). In Fig. 2f, it is evident that use of the real-time GVF led to lower leaf area index ( Fig. 2e; computed internally by the WRF model) across most of the domain compared to the climatological vegetation data (Fig. 2d), with the exception of some forested regions in the northern portion of the domain and bands of enhanced leaf area index surrounding metropolitan areas such as Chicago. ...
Full-text available
High-resolution simulations were performed to assess the impact of different parameterization schemes, surface datasets, and analysis nudging on lower-tropospheric conditions near Lake Michigan. Simulations were performed where climatological or coarse-resolution surface datasets were replaced by high-resolution, real-time datasets depicting the lake surface temperatures (SSTs), green vegetation fraction (GVF), and soil moisture and temperature (SOIL). Comparison of two baseline simulations employing different parameterization schemes (referred to as AP-XM and YNT, respectively) showed that the AP-XM simulation produced more accurate analyses on the outermost 12 km resolution domain but that the YNT simulation was superior for higher-resolution nests. The diurnal evolution of the surface energy fluxes was similar in both simulations on the 12 km grid but differed greatly on the 1.3 km grid where the AP-XM simulation had a much smaller sensible heat flux during the daytime and a physically unrealistic ground heat flux. Switching to the YNT configuration led to more accurate 2 m temperature and 2 m water vapor mixing ratio analyses on the 1.3 km grid. Additional improvements occurred when satellite-derived surface datasets were incorporated into the modeling platform, with the SOIL dataset having the largest positive impact on temperature and water vapor. The GVF and SST datasets also produced more accurate temperature and water vapor analyses but had degradations in wind speed, especially when using the GVF dataset. The most accurate simulations were obtained when using the high-resolution SST and SOIL datasets and analysis nudging above 2 km a.g.l. (above ground level). These results demonstrate the value of using high-resolution satellite-derived surface datasets in model simulations.
Land surface models (LSMs) rely on vegetation parameters for use in hydrological and energy balance analysis, monitoring and forecasting. This study examines the influence that vegetation representation in the Noah-Multiparameterization (Noah-MP) LSM has on hydrological simulations across the diverse climate zones of western tropical South America (WTSA), with specific consideration of hydrological variability associated with the El Niño Southern Oscillation (ENSO). The influence of model representation of vegetation on simulated hydrology is evaluated through three simulation experiments that use: (1) satellite-derived constant MODIS; (2) satellite-derived time-varying MODIS; (3) the Noah-MP dynamic leaf model. We find substantial differences in vegetation fields between these simulations, with the Noah-MP dynamic leaf model diverging significantly from satellite-derived vegetation fields in many ecoregions. Impacts on simulated hydrology were, however, found to be modest across climate zones, except for select extreme events. Also, although impacts on hydrology under ENSO-induced variability were small, we find that the Noah-MP dynamic leaf model simulates a positive relationship between rainfall and vegetation in humid ecoregions of WTSA, where satellite observations may indicate the opposite. The relatively small sensitivity of simulated hydrology to vegetation scheme suggests that the performance of hydrological monitoring and forecasting in WTSA that use Noah-MP is largely unaffected by the choice of vegetation scheme, such that using a simple climatological default is generally no worse than adopting more complicated options. The presence of some differences between the time-varying and constant MODIS simulations for hydrologic extremes, however, indicates that time-varying MODIS configuration might be more suitable for hydrological hazards applications.
Quantifying the influence of factors on changes in fractional vegetation cover (FVC) is critical for assessing regional environmental changes and consequent ecological protection. However, accurately identifying the factors responsible for vegetation changes remains a challenge. This study focuses on the Wumeng Mountain Area, China (WM), where the ecological environment is extremely fragile and the social economy underdeveloped. Using the enhanced vegetation index to calculate FVC, Sen's slope trend analysis, Mann–Kendall test with the trend-free prewhitening procedure, Pettitt change-point test, and Hurst exponent, we analyzed the spatiotemporal variations in vegetation from 2000 to 2019 and projected future variations. The geographical detector model was used to analyze the spatial differentiation driving mechanism of changes in vegetation cover in the WM. We observed that the spatiotemporal variation of vegetation in the WM was significant between 2000 and 2019. The areas of the WM with extremely significant growth and significant growth accounted for 32.57% (p < 0.01) and 15.28% (0.01 < p < 0.05), respectively. The mutation years of the significantly changed vegetation were concentrated between 2007 and 2011. However, 36.09% of vegetation growth exhibited strong unsustainable characteristics and based on the past 20 years, a potential decreasing trend that has great uncertainty in the future. The geographical detector model indicated that temperature and soil type were the primary driving forces for spatial differentiation of vegetation changes in the WM, with q values of 0.131 and 0.101, respectively. Interactions between climate, topography, and human activities promote vegetation growth in a nonlinear fashion
Vegetation is parameterized in the operational National Oceanic and Atmospheric Administration National Water Model (NWM) using climatologies of leaf area index and green vegetation fraction (GVF) which do not capture interannual variability or seasonal anomalies. This study investigates the impact of assimilating real-time Visible Infrared Imaging Radiometer Suite (VIIRS) GVF on NWM streamflow and land surface energy and moisture fluxes. Using a subset of the NWM over three humid watersheds in Alabama, VIIRS GVF was assimilated into an experimental NWM configuration to replace the default GVF climatology to provide a more accurate representation of the land surface vegetation greenness. Although the use of real-time VIIRS GVF in place of the climatological GVF slightly improved NWM stream-flow prediction, results were not statistically significant. GVF assimilation also improved the interannual representation of land surface fluxes with increases in GVF resulting in increased evapotranspiration and latent heat.
Full-text available
The rapid development of remote sensing technology has brought abundant data support for deep learning based temperature forecasting research. However, recently proposed methods usually focus on the temporal relationship among temperature observation information, whereas ignore the spatial positions of different regions. Motivated by the observation that adjacent regions usually present similar temperature trends, in this paper we consider the temperature forecasting as a spatiotemporal sequence prediction problem, and propose a new deep learning model for temperature forecasting, Self-Attention Joint Spatiotemporal Network (SA-JSTN), which simultaneously captures the spatiotemporal interdependency information. The kernel component of the SA-JSTN is a newly developed Spatiotemporal Memory (STM) unit, which describes the temporal and spatial models via a unified memory cell. STM is constructed based on the units of the convolutional LSTM (ConvLSTM). Instead of using simple convolutions for spatial information extraction, in STM we improve ConvLSTM by a self-attention module, which has significantly enhanced the global spatial information representation ability of our proposed network. Compared with other deep learning based temperature forecasting methods, SA-JSTN is able to integrate the global spatial correlation into the temperature series prediction problem, and thus present better performance especially in short-term prediction. We have conducted comparison experiments on two typical temperature data sets to validate the effectiveness of our proposed method.
Full-text available
Fraction of green vegetation, fg, and green leaf area index, L g , are needed as a regular space-time gridded input to evapotranspiration schemes in the two National Weather Service (NWS) numerical prediction modelsÐ regional Eta and global medium range forecast. This study explores the potential of deriving these two variables from the NOAA Advanced Very High Resolution Radiometer (AVHRR) normalized di erence vegetation index (NDVI) data. Obviously, one NDVI measurement does not allow simultaneous derivation of both vegetation variables. Simple models of a satellite pixel are used to illustrate the ambiguity resulting from a combination of the unknown horizontal (f g) and vertical (L g) densities. We argue that for NOAA AVHRR data sets based on observations with a spatial resolution of a few kilometres the most appropriate way to resolve this ambiguity is to assume that the vegetated part of a pixel is covered by dense vegetation (i.e., its leaf area index is high), and to calculate f g = (NDVI-NDVIo)/(NDVI 2 -NDVIo), where NDVIo (bare soil) and NDVI 2 (dense vegetation) are speci® ed as global constants independent of vegetation/soil type. Global (0´15ß) 2 spatial resolution monthly maps of f g were produced from a 5-year NDVI climatology and incorporated in the NWS models. As a result, the model surface ¯ uxes were improved.
Full-text available
This study reports results from an analysis of the relationship between atmospheric forcing and model‐simulated water and energy fluxes for the North American Land Data Assimilation System Project Phase 2 (NLDAS‐2). The relationships between mean monthly precipitation and total runoff are stronger in the Sacramento (SAC) and variable infiltration capacity (VIC) models, which grew out of the hydrological community, than in the Noah and Mosaic models, which grew out of the soil‐vegetation‐atmosphere transfer (SVAT) community. The reverse is true for the relationship between mean monthly precipitation and evapotranspiration. In addition, surface energy fluxes in VIC are less sensitive to model forcing (except for air temperature) than those in the Noah and Mosaic model. Notwithstanding these general conclusions, the relationships between forcings and model‐simulated water and energy fluxes for all models vary for different seasons, variables, and regions. These findings will ultimately inspire a combination of SVAT‐type model energy components with hydrological model water components to develop a SVAT‐hydrology model to improve both evapotranspiration and total runoff simulations. Copyright © 2011 John Wiley & Sons, Ltd.
Full-text available
The reanalysis at National Centers for Environmental Prediction (NCEP) focuses on atmospheric states reports generated by a constant model and a constant data assimilation system. The datasets have been exchanged among national and international partners and used in several more reanalyses. The new data assimilation techniques have been introduced including three-dimensional variational data assimilation (3DVAR), 4DVAR, and ensembles of analyses such as ensemble Kalman filter (EnKF), which produce not only an ensemble mean analysis but also a measure of the uncertainty. The new climate forecast system reanalysis (CFSR) was executed to create initial states for the atmosphere, ocean, land, and sea ice that are consistent as possible with the next version of the climate forecast system (CFS) version 2, which is to be implemented operationally at NCEP in 2010. Several graphical plots were generated automatically at the end of each reanalyzed month and were displayed on the CFSR Web site in real time.
This analysis system is called the spectral statistical-interpolation (SSI) analysis system because the spectral coefficients used in the NMC spectral model are analyzed directly using the same basic equations as statistical (optimal) interpolation. Favourable features include smoother analysis increments, greatly reduced changes from initialization, and significant improvement of 1-5 day forecasts. The objective function is a combination of forecast and observation deviations from the desired analysis, weighted by the inverses of the corresponding forecast- and observation-error covariance matrices. There are two principal differences in how the SSI implements the minimization of this functional as compared to the current OI used at NMC. First, the analysis variables are spectral coefficients instead of gridpoint values. Second, all observations are used at once to solve a single global problem. No local approximations are made, and there is no special data selection. -from Authors
A new physics package containing revised convection and planetary boundary layer (PBL) schemes in the National Centers for Environmental Prediction's Global Forecast System is described. The shallow convection (SC) scheme in the revision employs a mass flux parameterization replacing the old turbulent diffusion-based approach. For deep convection, the scheme is revised to make cumulus convection stronger and deeper to deplete more instability in the atmospheric column and result in the suppression of the excessive grid-scale precipitation. The PBL model was revised to enhance turbulence diffusion in stratocumulus regions. A remarkable difference between the new and old SC schemes is seen in the heating or cooling behavior in lower-atmospheric layers above the PBL. While the old SC scheme using the diffusion approach produces a pair of layers in the lower atmosphere with cooling above and heating below, the new SC scheme using the mass-flux approach produces heating throughout the convection layers. In particular, the new SC scheme does not destroy stratocumulus clouds off the west coasts of South America and Africa as the old scheme does. On the other hand, the revised deep convection scheme, having a larger cloud-base mass flux and higher cloud tops, appears to effectively eliminate the remaining instability in the atmospheric column that is responsible for the excessive grid-scale precipitation in the old scheme. The revised PBL scheme, having an enhanced turbulence mixing in stratocumulus regions, helps prevent too much low cloud from forming. An overall improvement was found in the forecasts of the global 500-hPa height, vector wind, and continental U. S. precipitation with the revised model. Consistent with the improvement in vector wind forecast errors, hurricane track forecasts are also improved with the revised model for both Atlantic and eastern Pacific hurricanes in 2008.
One of the challenges in land surface modeling involves specifying accurately the initial state of the land surface. Most efforts have focused upon using a multiyear climatology to specify the fractional coverage of vegetation. For example, the National Centers for Environmental Prediction (NCEP) Eta Model uses a 5-yr satellite climatology of monthly normalized difference vegetation index (NDVI) values to define the fractional vegetation coverage, or greenness, at 1/8° (approximately 14 km) resolution. These data are valid on the 15th of every month and are interpolated temporally for daily runs. Yet vegetation characteristics change from year to year and are influenced by short-lived events such as fires, crop harvesting, droughts, floods, and hailstorms that are missed using a climatological database. To explore the importance of the initial state vegetation characteristics on operational numerical weather forecasts, the response of the Eta Model to initializing fractional vegetation coverage directly from the National Oceanic and Atmospheric Administration's Advanced Very High Resolution Radiometer (AVHRR) data is investigated. Numerical forecasts of the Eta Model, using both climatological and near-real-time values of fractional vegetation coverage, are compared with observations to examine the potential importance of variations in vegetation to forecasts of 2-m temperatures and dewpoint temperatures from 0 to 48 h for selected days during the 2001 growing season. Results show that use of the near-real-time vegetation fraction data improves the forecasts of both the 2-m temperature and dewpoint temperature for much of the growing season, highlighting the need for this type of information to be included in operational forecast models.
Positive soil moisture-precipitation feedbacks can intensify heat and prolong drought under conditions of precipitation deficit. Adequate representation of these processes in regional climate models is, therefore, important for extended weather forecasts, seasonal drought analysis, and downscaled climate change projections. This paper presents the first application of the NASA Unified Weather Research and Forecasting Model (NU-WRF) to simulation of seasonal drought. Simulations of the 2006 southern Great Plains drought performed with and without soil moisture memory indicate that local soil moisture feedbacks had the potential to concentrate precipitation in wet areas relative to dry areas in summer drought months. Introduction of a simple dynamic surface albedo scheme that models albedo as a function of soil moisture intensified the simulated feedback pattern at local scale-dry, brighter areas received even less precipitation while wet, whereas darker areas received more-but did not significantly change the total amount of precipitation simulated across the drought-affected region. This soil-moisture-mediated albedo land-atmosphere coupling pathway is structurally excluded from standard versions of WRF.