AMAZON MODELING

1
AMAZON MODELING This document represents a
composite of multiple documents and
presentations on Amazon modeling. The intent is
to see where we are, then carry forward... ...
Much from Dani V, Vicky, Alex, Jeff......
2
Modeling surface hydrology across the Amazon
Basin using a Variable Infiltration Capacity
Approach
1LGTI, CENA, University of São Paulo 2School of
Ocean., University of Washington 3PPG FAA, ESALQ,
University of São Paulo 4LCE, ESALQ, University
of São Paulo 5Dep. Civil Engineering, University
of Washington
FAPESP 99/01159-4 03/13172-2 LBA-ECO CD06 and
ND09
3
Hydrology Model Variable Infiltration Capacity
or VIC
D. Victoria, et al
For each cell

• Meteorological drivers (max and min temperature,
  precipitation and wind)

• Vegetation distribution (forest, pasture, etc)

• Average soil properties soil texture

• Variable Infiltration curve empirical model that
  describes soil saturation related to soil
  moisture (calibration)

4
1. Vic Modeling 1.1 Forcing dataset The
precipitation, maximum and minimum temperature
and wind speed being used is from the CRU
dataset, a monthly climate dataset that has been
converted to daily time steps using the daily
variations in the COLA reanalysis dataset. This
means that we maintain the monthly average found
in the CRU dataset, together with the daily
variation provided by the climate models.
1.2 Soil texture. The soil texture map is the
basis for all other soil parameters used in VIC.
For the Amazon basin simulation we are using the
soil texture maps provided by Emilio, which I
believe are not documented anywhere. Basically,
what Emilio did was to group all common RADAM
soil profiles (based on soil type) and calculate
and average texture for each soil type. Then,
just assign the soil type polygons with the
averaged texture value. So far, the soil
properties used seems to be working
fine. 1.3 Routing network The routing network
being used was provided by Mariza. As far as I
can tell, it was based on the gtopo30 DEM (I do
have a lousy memory). I do not believe there is
much to be gained in the VIC scale (quarter of a
degree) by scaling up a finer flow direction map.
The important issue is how this scaling up is
performed. Mariza used a program that she wrote
to do this change in scale, based on some
published methodology developed for changing flow
direction scales. To sum it up, I dont think we
should worry about the routing maps for the VIC
application. DHSVM is a whole different problem
1.4 Sub basins. The Amazon basin was divided
into several sub-basins where the model is
calibrated separately. The sub-basins where
chosen based on the discharge measurements from
the CAMREX dataset and are Negro Madeira P
urus Juruá Japurá Santo Antonio do
Içá Itapeua (receives water from Japurá, Juruá
and Sto. A. Içá) Manacapuru (Itapeuá
Purus) Obidos (sum of all basins) The
calibration for each sub-basin is done comparing
the monthly modeled discharge against observed
discharge from CAMREX data. Three adjustment
coefficients are used to aid the calibration,
Nash, log Nash and difference in water volume
(modeled/measured). The parameters changed during
the calibration where b_inf VIC empirical
parameter for infiltration. Higher b, less inf,
more surf runoff. ds regulates baseflow. When
baseflow response will be linear or
exponential ds_max regulates max
baseflow depth depth of each of the 3 soil
layers. For instance, a thicker layer 2 will have
more water stored for ET. The model parameters
for each sub-basin where treated independently.
5
Flow direction and accumulation
Routing Model
DEM
6
Model calibration

• Soil depth 3 layers, down to 3 meters
• Variable Infiltration Capacity curve adjustment
• Regional vegetation parameters LAI, albedo, root
  depth and distribution, etc

Vegetation map spot images
Soil texture 0-30 and 30-100cm
Hueh et al., 2002
7
(No Transcript)
8
Challenging Issue Surface Forcing/Precip Fields
9
EARLY RESULTS 2004
10
Soil moisture
11
Evaporation (mm)
12
ET (mm)
13
Evaporation (mm)
14
Evaporation (mm)
15
146 110 73 36 0
Soil moisture
0 30 cm
349 272 196 119 42
30 100 cm
995 767 538 310 82
100 300 cm
16
Soil moisture
17
Monthly average values
Daily values
Shuttleworth, 1988 3 to 4 mm.day-1 Willmott,
1985 2 to 5 mm.day-1 GISS model 1.2 to 3.5
mm.day-1 GEOS-GCM 4 to 5.3 mm.day-1
18
Soil moisture
There is a lag phase between the soil moisture
maximum value of the 2nd and 3rd soil layer
19
Next steps from the III LBA Scientific
Conference (Brasília, July 2004) and beyond

• Improve calibration needs work
• Extend time period
• Couple with Biogeochemical Model
• Update vegetation library (MODIS LAI and Albedo)
  - partially
• Improve visualization and analysis tools for the
  results

20
Ji-Paraná Physical template
21
Precipitation
Evapotranspiration
22
Soil Moisture
Layer 1
Layer 3
23
Ji-Paraná Modeled discharge
24
(No Transcript)
25
(No Transcript)
26
Dani Reports 1, Madeira
Preliminary results Results for each of the
already calibrated sub basins are presented.
Since all sub-basins where calibrated
independently, a new calibration effort should be
carried out, trying to maintain a higher
consistency in the parameters across basins, for
example, start the calibration in one basin using
the parameters from an already calibrated basin
so they are better related.
27
Santo A. do Içá
N model 108 e N obs 108 r2 -3.507 rl
-9.508 V -0.582
28
As everybody knows, precipitation is
underestimated. The CRU monthly precipitation is
equal to the measured discharge, as shown bellow.
For this reason, the modeled discharge is no
good
29
n
m
n
m
n
m
30
(No Transcript)
31
Modeled (VIC) x Measured (ANA) discharge
CV () Underestimated by VIC () Superestimated by VIC () 11 VIC ANA()
Óbidos 34 86 8 6
Manacapuru 40 92 7 1
Itapeua 33 98 1 2
Purus 77 60 36 4
Japurá 36 63 28 9
Negro 23 69 22 9
Juruá 36 26 67 8
Madeira 48 69 23 8
Sto Antonio Içá 28 100 0 0
() of the total months
32
Madeira simulation results
1 - VIC modeling synthesis. A brief description
of the modeling environment. A better description
of the modeling parameters are given in
progress_report1.doc. The VIC simulation was done
using a spatial resolution of 0.25 degrees,
dividing the Madeira river basin in 1629
individual cells. Necessary model parameters
are 1.1 Soil texture A soil texture map was
obtained from Emilio. Its a RADAM based soil
properties map for the brazilian amazon and, for
the area outside of Brazil, its based on the FAO
soil map. 1.2 Other soil properties necessary
(Ksat, bulk density, porosity, field capacity)
are derived from a table that is used for all VIC
applications (http//www.hydro.washington.edu/Lett
enmaier/Models/VIC/Documentation/Info/soiltext.htm
l). 1.3 Vegetation cover derived from the 1km
Spot vegetation map for the Amazon (Vicky sent me
that one) and grouped so for each VIC cell we
have the percent covered by each vegetation
class, that is, for each VIC cell we could have
75 forest, 10 pasture and the rest is bare
soil. 1.4 Vegetation parameters These were
mainly based on the parameters estimated by
Santiago, derived from Amazonian deforestation
and Climate, and also LDAS and SIB2 parameters
used by Rafael Rosolem and Humberto Rocha. 1.5
Forcing dataset time period. Daily simulation
from Jan-1-1981 to Dec-31-1991 was carried out.
This period was chosen because of data
availability. Some new datasets are being
evaluated but Im reluctant in mixing different
datasets. I believe that could lead to more
errors. Also, Im using a different approach from
Coe et al.. They use a monthly dataset along with
a weather generator to get daily values. Im
using the same monthly dataset (CRU) but,
inserting the daily variation observed in a
climate model to the monthly observations (COLA
dataset). Recently, Erich Collicchio got hold of
a new daily precipitation dataset from CPTEC that
apparently in very good BUT has only data from
2004 to present. Could be used to see the big
drought effects on the model BUT, we still need
temperature and wind data for the same period.
1.6 Comparing results Along the text some
statistical comparison parameters will be shown
for the modeled vs. observed discharge. Some
parameter names are in Portuguese because they
are generated automatically (sorry about that).
The parameters are r2 Nash-Shutclif parameter
(goodness of fit) rl Log(Nash) parameter
(goodness of fit but more influenced by low
flow) V Difference in water volume, negative
means lower discharge modeled. Modeled and
measured discharge average and standard deviation
(media e std) and its ratio.
33
Rio Madeira
After calibration, the Madeira basin modeled
discharge agree very well with the CAMREX
discharge data for either monthly and daily
discharge (Figure 1 and Figure 2). N model 100
e N obs 100 r2 0.837 rl 0.784 V
-0.041 Media modelada 28737.053 Media obs
29955.627 Razao media model/obs 0.959 STD model
16919.086 STD obs 16793.993 Razao std
model/obs 1.007 For the monthly results, both
ratios (average discharge and standard deviation)
are close to 1. Also, the fit parametrs (r2 and
rl) show good agreement. Note that the first year
of simulation (1981) is left out of the
comparisons due to model spin-up.
34
Rio Madeira
N model 3026 e N obs 3026 r2 0.817 rl
0.749 V -0.042 Media modelada 28576.845 Media
obs 29839.585 Razao media model/obs 0.958 STD
model 17192.263 STD obs 17075.800 Razao std
model/obs 1.007 For the daily comparison, the
parameters r2 and rl are a bit lower but still,
represent a good fit (excluded first year of
data).
35
Rio Madeira ET v Precip
Considering that, by getting the modeled
discharge correct, the other modeled parameters
(ET and soil moisture) are also correct (and that
is a BIG CONSIDERING), the following analyses
can be made.
Figure 3. Daily ETPrecip for Madeira. Blue line
shows 11 correspondance
Analyzing the same graph but now showing the
different months (Jan, Apr, Jul, Oct, Figure 4),
we see that these are mainly days from the dry
season (Jul) and from the beginning of the rainy
period (Oct). The higher July ET is explained by
the fact that the values plotted are daily basin
average so, average basin precipitation could be
very low for a given day but, the forests located
near the rivers (riparian zones) would still have
access to water and maintain higher ET then the
rest of the basin, resulting in higher average
basin ET. Unfortunatelly, this sort of dynamic,
where water is maintained in the riparian zones,
is not explicitly represented by VIC since there
is no lateral movement of water during the
simulation.
The Madeira basin and other Amazon basins (except
NONONONO) show a distinct and expected pattern
when ET vs. Precip is analysed. That is, with low
precipitation, evapotranspiration is limited by
the amount of water available but, as
precipitation increases, the limiting factor for
ET becomes available energy. This shows that
water availability is a limiting factor for ET in
the Madeira Basin and that there is no ONE SINGLE
FACTOR limiting ET (Figure 3). Also, its
interesting to note that, not a lot of points are
above the 11 line (blue line). These points
represent days when ET was higher then
precipitation. Analyzing the same graph but now
showing the different months (Jan, Apr, Jul, Oct,
Figure 4), we see that these are mainly days from
the dry season (Jul) and from the beginning of
the rainy period (Oct).
36
Rio Madeira ET v Precip July
Nonetheless, average July ET has a clear gradient
along the basin, with higher ET close to the
Amazon main channel (Figure 5), obviously related
to precipitation (Figure 6). This gradient
stresses the finding that there is a clear water
limitation to ET over the Madeira basin. Also, on
Figure 5, apparently there is higher ET where the
Madeira main channel is located. January does not
show such gradient (Figure 8 and Figure 9) (sorry
I could not get both graphs with the same color
scale).
Figure 6. Monthly July precipitation (average
82-91)
Figure 5. Monthly July ET (average for 1982 to
1991)
Figure 8. Monthly January ET (average for 1982 to
1991)
Figure 9. Monthly January precipitation (ave 82
91)
37
Another explanation for ET gt Precip during July
is water stored in the soil (Figure 7)
Figure 7. Juyl soil moisture (second soil layer,
ave 82-91)
38
As expected, soil moisture for the first soil
layer varied accordingly to precipitation (Figure
12) and a clear separation for each month can be
seen (Figure 13).
Figure 13. Daily soil moisture (1st layer) and
precipitation, separated by months
Figure 12. Daily soil moisture (1st layer) and
precipitation
39
Separating the data by months its clear that, by
May, when the rain starts to stop, the soil
moisture is gradually depleted to maintain forest
ET (Figure 15).
For the second soil layer a higher variation in
soil moisture is seen for the low precipitation
period (Figure 14).
Figure 14. Modeled daily soil moisture for the
second soil layer and precipitation
Figure 15. Modeled daily soil moisture for the
second soil layer and precipitation for different
months
40
For the third soil layer, a circular pattern is
seen, where soil moisture is depleted from April
to October, when it reaches them minimum and then
starts to be recharged in November to March
(Figure 16 and Figure 17)
Figure 16. Modeled daily soil moisture for the
third soil layer and precipitation
Figure 17. Modeled daily soil moisture for the
third soil layer and precipitation for separate
months
41
Soil moisture dependence on precipitation varies
across depth, that is, for the first layer, there
is a strong sync between precipitation and
moisture (Figure 18). For the second layer
(Figure 19) the variations are not aligned, a lag
close to two months, while for the third (Figure
20) layer, there is a clear lag between the
precipitation cycle and soil moisture variation
of almost 6 months.
Figure 19. Seasonal variation of monthly
precipitation and modeled soil moisture of the
second layer
Figure 20. Seasonal variation of monthly
precipitation and modeled soil moisture of the
third layer
Figure 18. Seasonal variation of monthly
precipitation and modeled soil moisture of the
first layer
42
Humidade de Solo v Precip
43
Monthly precipitation and ET both follow the same
pattern, once again showing the water limitation
to ET (Figure 10).
Figure 10. Precipitation and modeled ET time
series for the Madeira basin
44
A regression between precipitation and ET was
carried out using the following mathematical
formulation (formula suggested by Perinho, its
the same formulation for maximum photosynthesis
production. I put the ETmin parameter)
Figure 11. Daily ETPrecip for the Madeira Basin,
along with regression curve and parameters
45
The regression gives a minimal ET of 0.86 mm/day
and max of 4.22 mm/day. These values are not
actual max and min ET values but its interesting
to use them as a cross basin comparison (Table
1). Table 1 Regression parameters for each
modeled Basin. Regression could not be
established for Jurua basin
Basin et max et min k
Madeira 4.22 0.858 0.9
Negro 3.38 0.789 1.06
Purus 4.06 0.798 1.06
Jurua 4.559 3837
From Table 1 we see that Madeira apparently has
higher ET then the rest of the Basins. Another
noticeable fact is that a regression could not be
established for the Jurua basin, showing that the
relation between precipitation and ET for that
basin is not very strong. No statistical analysis
was carried out to see if regression parameters
do, in fact, differ.
46
Rio Negro
The calibration parameters changed for the basin
where b 0.45 ds 0.45 ds_max ksat3 (ksat
layer 3) d20.5 d10.1 Adjustment
coefficients r2 0.7 rl 0.644 V 0.279
47
NEGRO
Here we see daily NEGRO ET vs. Precip. Its clear
that ET has an upper limit (around 4 mm/day) and
that limit is only reached when precip is above 5
mm. Also, ET will rarely be higher then precip
(above blue line). Could this means that soil
water is not as important? Because, even if we
have low precip, ET could be maintained by soil
water. The probable explanation is shown on the
graph bellow.
48
NEGRO
Here we have normalized soil water storage (mean
monthly soil water / max (mean monthly soil
water) for each soil layer against precip. Both
layer 1 and 3 points are packed, that is, no
big dispersion. Also, layer 1 has a good relation
with precip while layer 3 still is influenced by
precip but on a smaller scale. For layer 2, the
one that influences ET the most, we observe more
scattered points but yeat, the same level of
influence presented in layer 1. This explains why
we dont see too many days with ET above precip.
Because days with low precip will not have high
water storage in the soil. Looking in the soil
water time series, we see how all 3 layers follow
the same pattern found in the precip.
49
NEGRO
50
NEGRO
Another interesting find is that, even though
precip for the Negro basin is similar for April
and July, ET for April is higher .How did I
arrive at this conclusion? Both April and July
points (red and blue, respectively) are
concentrated around the same precipitation range,
from 2 to 20 mm/day but, ET in April is
apparently higher, as seen above.
51
Rio Juruá (Purus, Japura, Içá)
Calibration for the Juruá is still troublesome.
It could be due to the simplicity of the routing
model presented in VIC or some other factor.
Still more calibration needs to be done. Also, I
just finished a new cell-to-cell routing program
and Ill see if there are any differences between
the VIC routing and the cell-to-cell routing.
Ill present 2 model runs for this basin. The
calibration coefficients for the last model run
have been misplaced (happens after 61
runs) b0.02 r2 0.585 rl 0.788 V -0.061
52
Rio Juruá
r2 0.52 rl 0.717 V -0.235 The last run does
look better but there is a larger error in water
discharge volume (V)
53
Rio Purus
Calibration is still under way so there are no
good results. The best result so far isRodada
17, ds0.1 dsmax20 b.15 N model 92 e N obs
92 r2 -0.142 rl 0.137 V -0.447
54
(No Transcript)
55
(No Transcript)
56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
(No Transcript)
60
(No Transcript)
61
Effects of deforestation on hydrologic
response from scale reduction to large basins
INPA/INPE Group Daniel Andrés Rodrigues, Javier
Tomasella, Luz Adriana Cuartas, Ralph Trancoso da
Silva, Rita de Cassia da Silva.
62
Water Balance Comparison of cumulative changes of
storage
63
Water balance for the Ji-Paraná river basin,
using a simple method through GIS and Remote
Sensin (D. Victoria et al in press) Preliminary
application of VIC, S. Santiago et al in prep.
PRELIMARY MODELS LANDUSE CHANGE SHOULD HAVE
EFFECT
EARLY- better now
64
Do we see such signals, at (Ji-Parana) Mesoscale?
65
E AGORA?