NASAs Role in NIDIS

1
Drought Working Group analysis of model- produced
soil moisture as an index of agricultural
drought Randal D. Koster (GMAO,
NASA/GSFC) Zhichang Guo (COLA) Paul A. Dirmeyer
(COLA) Rongquian Yang (NCEP, NOAA) Ken Mitchell
(NCEP, NOAA) Cindy Wang (Chinese Academy of
Sciences) Dennis Lettenmaier (U.
Washington) Kingtse Mo (NCEP, NOAA) Wanru Wu
(NCEP, NOAA)
2
One of the goals of the U.S. CLIVAR drought
working group Develop a working definition of
drought (onset and demise) that is useful to both
the prediction/research and applications
communities. In this talk, we focus on
agricultural drought deficits in soil water
availability for vegetation (e.g., crop) growth.
What quantifiable index can we use to
characterize agricultural drought?
3
Some Potential Agricultural Drought Indices
Index Strengths Weaknesses
Available networks limited in most parts of the
world point measurements.
In-situ soil moisture measurements
Direct quantification of soil moisture.
4
Some Potential Agricultural Drought Indices
Index Strengths Weaknesses
Available networks limited in most parts of the
world point measurements.
In-situ soil moisture measurements
Direct quantification of soil moisture.
Measures only top several mm (cm) of soil, and
not under dense vegetation.
Satellite-based soil moisture measurements
Global estimates of areally-averaged soil
moisture.
5
Some Potential Agricultural Drought Indices
Index Strengths Weaknesses
Available networks limited in most parts of the
world point measurements.
In-situ soil moisture measurements
Direct quantification of soil moisture.
Measures only top several mm (cm) of soil, and
not under dense vegetation.
Satellite-based soil moisture measurements
Global estimates of areally-averaged soil
moisture.
An empirical estimate ignores some aspects of
antecedent meteorology.
Palmer drought index
Global estimates of drought state long history
of use.
6
Some Potential Agricultural Drought Indices
Index Strengths Weaknesses
Available networks limited in most parts of the
world point measurements.
In-situ soil moisture measurements
Direct quantification of soil moisture.
Measures only top several mm (cm) of soil, and
not under dense vegetation.
Satellite-based soil moisture measurements
Global estimates of areally-averaged soil
moisture.
An empirical estimate ignores some aspects of
antecedent meteorology.
Palmer drought index
Global estimates of drought state long history
of use.
Not a direct measurement soil moisture
estimates are model-dependent.
Model-derived soil moisture
Global estimates of areally-averaged soil
moisture reflects all prior meteorology.
7
Some Potential Agricultural Drought Indices
Index Strengths Weaknesses
Available networks limited in most parts of the
world point measurements.
In-situ soil moisture measurements
Direct quantification of soil moisture.
Measures only top several mm (cm) of soil, and
not under dense vegetation.
Satellite-based soil moisture measurements
Global estimates of areally-averaged soil
moisture.
Remainder of talk examine this weakness. Can
we get around it?
An empirical estimate ignores some aspects of
antecedent meteorology.
Palmer drought index
Global estimates of drought state long history
of use.
Not a direct measurement soil moisture
estimates are model-dependent.
Model-derived soil moisture
Global estimates of areally-averaged soil
moisture reflects all prior meteorology.
8
Study 1 Analysis of GSWP-2 Data
9
nd Global Soil Wetness Project

• This phase of the project takes advantage of
• The10-year ISLSCP Initiative 2 data set
• The ALMA data standards developed in GLASS
• The infrastructure developed in the pilot phase
  of GSWP
• GSWP-2 represents an evolution in multi-model
  large-scale land-surface modeling with the
  following goals
• Produce state-of-the-art global data sets of soil
  moisture, surface fluxes, and related hydrologic
  quantities.
• Develop and test in situ and remote sensing
  validation, calibration, and assimilation
  techniques over land.
• Provide a large-scale validation and quality
  check of the ISLSCP data sets.
• Compare LSSs, and conduct sensitivity analyses of
  specific parameterizations.

www.iges.org/gswp/ gswp_at_cola.iges.org
10
nd Global Soil Wetness Project

• This phase of the project will take advantage of
• The10-year ISLSCP Initiative 2 data set
• The ALMA data standards developed in GLASS
• The infrastructure developed in the pilot phase
  of GSWP
• GSWP-2 represents an evolution in multi-model
  large-scale land-surface modeling with the
  following goals
• Produce state-of-the-art global data sets of soil
  moisture, surface fluxes, and related hydrologic
  quantities.
• Develop and test in situ and remote sensing
  validation, calibration, and assimilation
  techniques over land.
• Provide a large-scale validation and quality
  check of the ISLSCP data sets.
• Compare LSSs, and conduct sensitivity analyses of
  specific parameterizations.

In GSWP-2, a number of land surface models were
driven with the same observations-based
meteorological forcing. What we will demonstrate
here is that the different models produce a
similar soil moisture product, when the product
is suitably scaled...
www.iges.org/gswp/ gswp_at_cola.iges.org
11
GSWP-2 Models (as of March 2005)
This page shows the international participation
in GSWP-2. The models analyzed here are circled.
Vertical structure shows soil layers for water
(W) and temperature (T), and the maximum number
of snow layers (S). Soil data sets are either
supplied by GSWP-2 (g) or the models default
(d). For vegetation distributions, GSWP-2
supplied datasets include IGBP (i) and SiB (s)
categories Sland has dynamic vegetation. Two
models have different time steps for energy (E)
and soil (S).
12
Ostensibly, the model-derived soil moistures
produced in GSWP (with the same atmospheric
forcing) are very different.
Southern U.S.
Europe
Sahara
Root zone soil moistures (degrees of saturation)
produced by the 7 land surface models at five
sites.
Sahel
Amazon
13
Such inter-model differences have long been
documented in the literature. They reflect a
simple and often overlooked fact For various
reasons, mostly related to model limitations, a
land models soil moisture variable is best
interpreted as an index of soil moisture state,
one that increases as the soil gets wetter and
decreases as it gets drier. In general, a
models soil moisture should not be considered an
absolute quantity that can be compared between
models or against direct observations. Its
MODEL DEPENDENT!
14
Scaling the data, to isolate temporal
variability Let w(j,n) models total soil
moisture for day j of year n. Define
w(j,n)
mw(j) WI(j,n)
----------------------------
sw(j) where
mw(j) Mean (over many years) of w on day j.
sw(j) Standard deviation of w on day j.
15
Note given the non-Gaussian nature of soil
moisture, there are better ways to scale the
data, particularly if a long data history is
available
CDF matching map percentiles.
For the GSWP2 analysis, with only 10 years of
data, we use the simpler standard normal
deviate approach. The use of the simpler
approach can only make things more difficult for
us, so if we still succeed
16
Raw model soil moistures
Scaled model soil moistures
Southern U.S.
Europe
Sahara
Sahel
Amazon
(31-day smoother applied)
17
Average r2 between models (degree to which the
models produce the same soil moisture
information, in terms of temporal variability,
with no smoothing)
Note When scaling the soil moisture, the
seasonal cycle is subtracted out before
statistics are computed, making it that much more
difficult to get a high r2.
18
Scaled model soil moistures
If an agricultural drought were defined as, say,
a soil moisture falling 0.5s below its
climatological mean for that time of year, then
all of the models would capture the 1988 Midwest
drought. Model dependence of soil moisture
values may not be such a big issue
19
Study 2 Study of North American Drought Lead
U. Washington. Slides adapted from originals by
Dennis Lettenmaier and Cindy Wang.
20
Models

• VIC Variable Infiltration Capacity Model
• (Liang et al. 1994)
• CLM3.5 Community Land Model version 3.5
• (Oleson et al. 2007)
• NOAH LSM NCEP, OSU, Air Force, Hydrol. research
  lab
• (Mitchell et al. 1994, Chen
  and Mitchell 1996)
• Catchment LSM NASA/GSFC Global Modeling and
  Assimilation Office
• (Koster et
  al. 2000 Ducharne et al. 2000)

21
Data

• All models driven with observations-based met
  forcing. Daily precipitation and temperature
  max-min, other land surface variables (downward
  solar and longwave radiation, near-surface
  humidity, and wind) derived via index methods.
  Methods as described in Maurer et al. (2002).
• Period of analysis 1920-2003 (after 5-year
  spinup).
• Spatial resolution 0.5? (3322 land grid cells)
• Domain conterminous United States.
• Soil and vegetation parameters differ for
  different models (generally NLDAS), as provided
  by model developers.

22
The challenge Different land schemes have
different soil moisture dynamics
Model simulated total soil moisture at
cell (40.25?N, 112.25?W)
23
Solution Normalized total column soil moisture
Recall there are more valid ways of scaling soil
moisture than using standard normal deviates

• For each model, total column soil moisture was
  expressed as percentiles.
• Percentiles were estimated for each model by
  month, using simulated total column soil moisture
  for the period 1920-2003.
• Percentiles were computed using the Weibull
  plotting position formula.

24
(No Transcript)
25
(No Transcript)
26
Averaged soil moisture percentiles 1932-38
27
Averaged soil moisture percentiles 1950-57
28
Spatial distribution of average (monthly)
between-model correlations of soil moisture
percentiles
29
Study 3 Objective Climate Drought Monitoring
over the United States Lead NCEP. Slides
adapted from originals by Kingtse Mo and Wanru
Wu.
30
Agricultural drought (SM percentiles, June
2008)
EMC/NCEP
All models capture the same basic features
Drought in SE, southern Texas and California.
Wetness in Great Plains.
But details differ.
31
Uncertainties of the NLDAS Compare VIC and Noah
over 1948-2003.
Soil moisture percentiles
Corr
RMS

• Differences are regionally dependent
• Over the areas east of 90W, differences are
  small.
• Over the areas west of 90W, differences are
  large.
• The RMS error is larger than 25 the difference
  between one drought class to another

Thanks Yun Fan and Andy Wood!!
32
Note similar result from these three studies
Between-model correlations are smallest in driest
areas.
33
Average (monthly) between-model correlations of
soil moisture percentiles U. Washington study
r2 values from GSWP2 study
correlation values from NCEP EMC study
34
Key Question Why is the model-dependence of a
soil moisture index larger (and thus the
potential usefulness of this index smaller) in
drier areas?
35
One major reason the potential for correlation
is tied to precipitation variance. A larger
year-to-year rainfall variability implies a
larger year-to-year soil moisture signal that all
models can more easily capture. If precipitation
variance is small, the model states arent
controlled as much by a large forcing signal, and
differences in model physics manifest themselves
more easily.
Correlation between models (GSWP2)
s2P
36
A key difference in model physics that can
manifest itself in the absence of strong
interannual precipitation forcing the models
water holding capacity.
e-folding time of soil moisture autocorrelation
(months) U. Washington study
Soil water holding capacity of six models (cm)
37
Differences between VIC and Noah (NCEP study)
Total SM anomaly percentile for selected River
Forecast Center areas Vic(Blue), Noah
(black) From 1950-2001 1. For RFCs east of
90-95W, VIC and Noah agree. e.g. the lower
Mississippi , Arkansas RFCs. 2. There are large
differences over the western region. e. g. the
Missouri , Colorado RFCs 3. VIC has more high
frequency components than the Noah.
3 month running mean
38
Another measure of agreement average standard
deviation of soil moisture values between models.
(GSWP2 study)
Before mapping
After mapping
39
Summary and Discussion
Land surface models use physically-based
formulations to integrate (over time) the effects
of meteorological forcing on soil moisture. The
models may provide information on soil moisture
state for evaluating agricultural drought.
But simulated soil moistures are
model-dependent. Nevertheless, we find that,
when interpreted in the context of their own
climatology, the seemingly different model
products are in fact consistent they provide
largely the same information on the time
variability of soil moisture at a point. The
model-dependence of a simulated soil moisture
product may not greatly limit its use in
characterizing agricultural drought.
40
Summary and Discussion (cont.)
This is particularly true over regions with
large interannual precipitation variance. The
use of a multi-model average of the scaled values
could help average out any model-specific
behavior that does remain after scaling
Scaled model soil moistures
Multi-model average
A particularly useful index for agricultural
drought? Something to consider!