Ex7

advertisement
Estimating Evapotranspiration from Satellite
Prepared by Ayse Kilic
GIS in Water Resources, Fall 2014
PURPOSE
The purpose of this exercise is to utilize Landsat 8 data to calculate a vegetation index (NDVI) and
to estimate evapotranspiration (ET) from the NDVI in the ArcGIS environment. The Normalized
Difference Vegetation index (NDVI) is calculated from Landsat 8 TM.
COMPUTER AND DATA REQUIREMENTS
The data provided below have been prepared in a lab data folder under Exe7Data.Zip for the
North Platte River above Pathfinder Reservoir, Wyoming. It includes the following:
1. A toolbox, “DNtoRadiance_Nov_2014”, that has four models in it that will assist in the
calculation of reflectances, NDVI, surface temperature, and evapotranspiration. (i.e.
converts the “digital number” of the Landsat bands into actual reflectance units, estimates
LST from band 10, etc.).
DNtoRadiance_Nov_2014.tbx
2. NorthPlatte_1.gdb. This database contains
a. Reflectance bands for bands 4 and 5 of that you should have calculated for Exe 6, but
I am providing it here.
b. B10 digital number from Landsat 8 for June 20, 2013.
c. NDVI_toa: That you should have calculated for Exe 6, but I am providing it here.
d. Etrf24_ef_06202013_P35R31_L8_NP_Color.img . This is a raster file that contains
evapotranspiration in the form of Fraction of reference ET (or relative ET). It has
been derived from a land surface energy balance model (METRIC). It will be
compared with the relative ET map obtained from NDVI.
3. Sample_points_AK.shp. A shape file composed of a number of ‘points’ that will be
used for sampling data for the purpose of creating statistics for land use classes.
4. ETrF_frmNDVI.breaks2.xml. You can use this xml file to load the color for each class
of ETrF
SETTING UP YOUR DATA


First, make a directory (Exe7).
Copy the files above into this directory
1

The raster images for reflectance (ref_b4, ref_b53), and the DN b10 and NDVI should
each contain the two rows (31 and 32 of path 35) where you mosaicked these together
during exercise 6. Again, I have provided all of these to you in NorthPlatte_1.gdb
PART 1: MOSAICKING TWO LANDSAT ROWS FOR THE THERMAL BAND
As you have seen in Exercise 6, Landsat data come in ‘scenes’ (one path and one row) that are
about 160 km on a side (100 miles). One single scene (Path 35 Row 31) does not cover our
basin. Therefore, we need to also bring in Path 35, Row 32. We need to mosaic the two scenes
together for band 10 for path 35. We need to do this prior to calculating radiance and LST (Land
Surface Temperature).
However, to reduce your work load, I have done the mosaicking for you and am supplying you
with the two row mosaic named “b10.” The b10 is a digital number (tiff) image that originated
from the USGS Eros Data Center that represents thermal band 10 of Landsat 8 for the June 20,
2013 image.
PART 2: CALCULATION OF LAND SURFACE TEMPERATURE FROM LANDSAT 8
We have created a toolbox (tbx) that contains models from Model Builder to calculate land surface
temperature (LST) and evapotranspiration (ET) from vegetation index for Landsat 8 data. The
name of the toolbox is: DNtoRadiance_Nov_2014.tbx and it is included in your exercise folder.
You should ‘add’ this toolbox to your list of tools.
Under this toolbox, the “surface_temperature_model2” will calculate LST using band 10 of
Landsat 8. Right click on this tool and select ‘edit’ to see the toolbox displayed.
2
3
We will run this model in a few minutes, but first, let’s get acquainted with how we calculate
surface temperature from Landsat. You can click on the yellow boxes in the LST model above
to see the equations that we will cover below, as they have been implemented using the Model
Builder code. Note that in the tool, the inputs are in blue and include the b10 tif (DN image)
and the bands 4 and 5 reflectances along with four coefficients. The model computes both
narrow band thermal emissivity and the LST.
Background on Landsat surface temperature
Land surface temperature (LST) has been measured by Landsat since Landsat 4 was launched in
1982. Currently, Landsat 7 and 8 are in operation. The pixel size of thermal data is 120 m for
Landsat 4 and 5, 60 m for Landsat 7 and 100 m for Landsat 8. This compares to 30 m pixel size
for reflected short-wave data. Beginning in about 2009, all thermal data from Landsat has been
resampled to 30 m pixel size by the USGS EROS data center to conform to the 30 m pixel size of
short-wave data. Cubic convolution resampling is used. However, we should always keep in
mind the original ‘native’ resolution of the thermal pixels.
The Landsats 5 and 7 had only a single thermal band, band number 6. Landsat 8 has two thermal
bands, bands 10 and 11. Post-launch testing of the two Landsat 8 bands has indicated that band
10 has higher accuracy than band 11, and is therefore utilized here. Band 10 covers the
wavelength region of approximately 10.5 to 11.2 micrometers and band 11 covers the
wavelength region of approximately 11.5 to 12.5 micrometers. Either band can be used to
estimate LST.
4
The following text describes the series of steps for calculating land surface temperature, LST,
from Landsat. Some of this text is based on the METRIC manual for determining
Evapotranspiration from Surface Energy Balance.
a) The spectral radiance (Lb) for thermal radiation (W/m2/sr/m)
Similar to the procedure used for short-wave data, the spectral radiance (Lb) for thermal radiation
is computed from the digital number of each individual 30 m pixel. Spectral radiance is the
outgoing radiation energy of the band as observed at the top of the atmosphere by the satellite. It
is calculated as:

LMAX  LMIN


(1)
Lb  
  (DN  QCALMIN )  LMIN
 QCALMAX  QCALMIN 
where
DN is the digital number of each pixel (recall that DN is used to efficiently “pack” data for
transmission from the satellite),
LMAX and LMIN are calibration constants found in the header file. These represent the
maximum and minimum radiance values used to scale the DN. LMAX and LMIN will have the
same dimensions as Lb and W/m2/sr/m.
The QCALMAX and QCALMIN are the highest and lowest range of values for rescaled
radiance in DN and are also found in the header file.
Eq. 1 essentially reverses the process used on the satellite to convert the Lb radiance into the DN
values for transmission to Earth.
The units for Lb are W/m2/sr/m. “sr” stands for ‘sterradian’ and “m” represents micrometers (1
millionth of a meter). Landsat 5 and 7 images are stored using one eight bit integer for each
pixel and band. Eight bit integers have the range of 0 to 28 –1, where 28 - 1 = 255. Landsat 8
images, being more modern, are stored as 16 bit data and have the range of 0 to 216 -1 = 65,535.
The values of LMAX and LMIN can vary from image date to image date. Therefore, it is
important to confirm, for each scene (path/row), the specific coefficients for each band from the
header file.
In the case of thermal data for Landsat 8, these LMAX and LMIN, QCALMAX and QCALMIN
values are found in the header file and are termed “Radiance_maximum_band_10” and
“Radiance_minimum_band_10”.
These following two values represent LMAX and LMIN for June 20, 2013 for path 35, row 31
for band 10:.
RADIANCE_MAXIMUM_BAND_10 = 22.00180
RADIANCE_MINIMUM_BAND_10 = 0.10033
Therefore, LMAX = 22.0018 W/m2/sr/m and LMIN = 0.10033 W/m2/sr/m.
5
The values of QCALMAX and QCALMIN are found in the header file of the June 20, 2013
image for path 35, row 31 for band 10
QUANTIZE_CAL_MAX_BAND_10 = 65535
QUANTIZE_CAL_MIN_BAND_10 = 1
Therefore QCALMAX = 65535 and QCALMIN = 1.
b)
The surface temperature (Ts, Kelvin)
The surface temperature is computed using the following modified Plank equation:
Ts 
K2


K
ln NB 1  1
 Rc

(2)
where: Ts is surface temperature (K), Rc is the corrected thermal radiance from the surface and
K1 and K2 are constants for Landsat 8 image. Units for Rc must be the same as those for K1
(W/m2/sr/μm).
For Landsat 8, constants may change from date to date, and therefore need to be read from the
header file for the specific image date. In the case of the June 20 image date for path 35, row 31
(over SE Wyoming):
K1_CONSTANT_BAND_10
K1_CONSTANT_BAND_11
K2_CONSTANT_BAND_10
K2_CONSTANT_BAND_11
=
=
=
=
774.89
480.89
1321.08
1201.14
Therefore, we will use K1 = 774.89 and K2 = 1321.08.
It is the K1 and K2 for band 10 that should be used for this exercise 7.
The K1 and K2 coefficients for Landsat 5 (L5) and Landsat 7 (L7) are also located in the header files, but
do not change for specific image date for L5 and L7. Table 1 provides the K1 and K2 for L5 and L7.
Table 1. Constants K1 and K2 for Landsat 5 (Markham and Barker, 1986; Chander et al., 2009) and
Landsat 7 (Landsat 7 Science User Data Handbook Chap.11, 2002; Chander et al., 2009) in W/m2/sr/μm
Landsat 5, Band 6
Landsat 7, Band 6
K1
607.76
666.09
K2
1260.56
1282.71
6
c) Surface emissivity (
Surface emissivity (is required to estimate LST. Surface emissivity is defined as the ratio of
the thermal energy radiated by the surface to the thermal energy radiated by a blackbody at the
same temperature. Two surface emissivitiesare used in energy balance work. The first is an
emissivity representing surface behavior for thermal emission in the relatively narrow band 6 (or
band 10) of Landsat (10.4 to 12.5 μm for band 6 of Landsat 5 and 7 and 10.5 to 11.2 μm for band
10 of Landsat 8). This narrow band emissivity is expressed as NB.
A second emissivity term is an emissivity representing surface behavior for thermal emission in
the broad thermal spectrum (6 to 14 μm), and is expressed as . The is used during energy
balance work to calculate total longwave radiation emission from the surface and is not used
here.
The surface emissivities are computed using the following empirical equations developed by
Tasumi et al. (2003c) using leaf area index (LAI) based on soil and vegetation thermal spectral
emissivities from the MODIS UCSB Emissivity Library1. NDVI is used to filter soil and
vegetation (NDVI > 0) from water and snow (NDVI ≤ 0).
For NDVI > 0
NB = 0.97 + 0.0033 LAI;
0.98
for LAI  3
(3)
for LAI  3
For NDVI ≤ 0
Water, Albedo ( < 0.47,  = 0.99 and o = 0.985
Snow, Albedo ( ≥ 0.47,  = 0.99 and o = 0.985
The values for water and snow are set to the same values, but can be modified based on locally
calibrated emissivities and user preference.
d) The leaf area index (LAI)
The leaf area index, LAI, is the ratio of the total area of all leaves on a plant (one side of the
leaves) to the ground area represented by the plant and ranges from 0 to 6. LAI is dimensionless
(m2/m2) and is an indicator of biomass and canopy resistance to vapor flux.
The LAI is computed using the following empirical equation stemming from Trezza et al.
(unpublished):
LAI  11  SAVI 3 ;
LAI  6;
1
for SAVI  0.817
for SAVI > 0.817
http://www.icess.ucsb.edu/modis/EMIS/html/em.html
7
(4)
where SAVI is the SAVI calculated from Equation (5) using a value for L of 0.1. The maximum
value for LAI is 6.0, which corresponds to a maximum SAVI of 0.817. The relationship between
SAVI and LAI can vary with location and crop type.
e) Soil adjusted vegetation index (SAVI)
The SAVI is the soil adjusted vegetation index, and is similar to the NDVI. SAVI attempts to
“subtract” the effects of background soil from NDVI so that impacts of soil wetness are reduced
in the index. SAVI is computed as:
SAVI = (1 + L) (t,NIRt,RedL + t,NIRt,Red
(5)
where L is a constant for SAVI. If L is zero, SAVI becomes equal to NDVI. A value of 0.5
frequently appears in the literature for L. However, a value of 0.1 was found by Tasumi et al.
(2002) to better represent soils of southern Idaho. The value for L can be derived from analysis
of multiple images where vegetation does not change, but surface soil moisture does. ρt,NIR and
ρt,Red are the top of atmosphere reflectances for the NIR and red bands (bands 4 and 3 for
Landsats 5 and 7 and bands 5 and 4 for Landsat 8).
f) Thermal radiance from the surface (Rc, W/m2/sr/m)
Thermal radiance at the surface (Rc) is calculated from the thermal radiance sensed at the
satellite following Wukelic et al. (1989) as:
Rc 
Lb  R p
 NB
 1   NB  Rsky
(6)
where
Lb is the spectral radiance of the thermal band (6 or 10) (W/m2/sr/m),
Rp is the path radiance in the 10.4 – 12.5 or 10.5 – 11.2 μm band (W/m2/sr/m),
NB is the narrow band emissivity (unitless),
Rsky is the narrow band downward thermal radiation from a clear sky (W/m2/sr/m), and
τNB is the narrow band transmissivity of air (10.4 – 12.5 or 10.5 – 11.2 μm).
Eq. 6 essentially ‘corrects’ the Lb radiance for atmospheric attenuation. The units for Rc are
W/m2/sr/m. The corrected thermal radiance (Rc) represents the actual radiance emitted from
the surface whereas Lb is the radiance that the satellite “sees”. In between the surface and the
satellite two things occur. First, some of the emitted radiation is absorbed and reflected by the
atmosphere (transmissivity) and does not reach the satellite. Second, thermal radiation is emitted
by the atmosphere in the direction of the satellite (path radiance) and the satellite “thinks” that
this is from the surface.
For the surface_temperature_model2, we use Rp = 0.91, τNB = 0.866 and Rsky = 1.32 for low
aerosol conditions suggested by Allen et al. (2007). The values for Rp and τNB require the use of
an atmospheric radiation transfer simulation model such as MODTRAN and radiosonde profiles
representing the image and date.
8
Fortunately, the effects of the three parameters on Rc are largely self-canceling and use of Lb
alone provides general atmospheric correction for most clear sky atmospheric conditions and
appears to be accurate for Ts of about 290-300 K. The result of no correction to Lb will be a
general underestimation of surface temperature (Ts) by up to about 5o C for warmer portions of
an image Allen et al. (2007).
Now let’s run the temperature model and see just how hot things are in Wyoming . If you
double click on the surface_temperature_ model2, you will see the followings inputs and the
outputs:
Inputs are b4 reflectance (red band), band 5 reflectance (NIR) that are used to calculate SAVI and
LAI, and thermal band (b10) that we mosaicked above. Note that the b4 and b5 images must be
mosaicked also. You should have done that for exercise 6. Outputs are LST and Narrow band
emissivity.
9
Remember that the Lmax, Lmin, K1 and K2 are coefficients that come from the header file for the specific
Landsat image date. These will almost always be the same value for two rows that appear in the same path
on the same date.
Note that at the end of the running of the model, we are getting some warnings about the
difference in projections between the “WGS84” used by the Landsat L1T TIFF images and the
Albers that is used by the Data Frame:
Now we should be able to see both the original ‘tif’ files of Digital numbers for band 10 (b10)
and the new “LST” (land surface temperature) layers. Your viewer may look like this:
10
Notice that LST varies from 268 to 333 Kelvin, and the emissivity is 0.98 and did not vary across the
scene (the emissivity model is not quite working. If anyone would like to ‘fix’ it, they can have 3 pts
extra credit ).
We should colorize the LST map to help us visualize areas of high and low surface temperature and to
improve its appearance. Let’s select a color scheme that shows hottest as red and coolest as blue.
The resulting two screenshots below show the Pathfinder Reservoir where the temperature of the water
surface is much cooler than the surrounding rangeland. The mountains in the south part of the screenshot
are also cool.
Question: Why are the mountains cool? Give two reasons. (turn in your answer)
11
We can zoom in on some of the irrigated areas in the center part of the basin and see individual
fields (below). We can see that the center pivot fields are somewhat out of shape because of the
use of WGS84 projection with the Landsat L1T images from the USGS, whereas the native
projection system for the ‘basin’ layer is Albers Equal Area. For a ‘serious’ product, we would
want to reproject the Landsat and associated products into the Albers Equal Area, or vice-versa.
Note that the fields are a little ‘fuzzy’ due to the 100 m native pixel size of the thermal band of
Landsat 8. The same fields presented later in NDVI are more clear because of the 30 m pixel
size of the short wave bands.
12
Turn in: Colorized surface temperature map and legend for the entire basin area and for a close up
showing some irrigated agriculture.
PART 3. ESTIMATING EVAPOTRANSPIRATION from NDVI
The normalized difference vegetation index, NDVI, is a quantity used to assess the presence of live green
(𝑁𝐼𝑅−𝑅𝐸𝐷)
vegetation. The NDVI was computed in Exercise 6 using the formula: 𝑁𝐷𝑉𝐼 = (𝑁𝐼𝑅+𝑅𝐸𝐷) and will be
used for this exercise.
The RED and NIR stand for the spectral reflectance measurements acquired in the red and near-infrared
regions of electromagnetic spectrum, respectively. NDVI takes values from -1 to 1. The higher the NDVI,
higher the fraction of live green vegetation present. Band 5 (0.77-0.90 µm) of Landsat 8 measures the
reflectance in NIR region and Band 4 (0.63-0.69 µm) measures the reflectance in the Red region.
When we estimate ET from NDVI, we are relying on a proportional relationship between the amount of
vegetation and the amount of transpiration (ET = evaporation from soil and transpiration from the
vegetation). We need to appreciate that this is only an approximate relationship. If the soil is dry in the
root zone, as is common in rangeland and forested areas during parts of the year, then transpiration, T,
will be less than expected, due to water stress. Conversely, if a recent rain event has wetted the upper soil
layer, then there will be some evaporation, E, in addition to T, so that total actual ET may exceed the ET
estimated from NDVI.
13
Therefore, we need to appreciate that this estimate for ET is an approximation, however, it is generally in
the ‘ballpark’ and can be useful as a first and rapid estimate. We will compare our estimate of ET with a
more accurate ET derived from surface energy balance in a later section.
ET not only varies with vegetation, it also varies with weather. ET will be greater under dry, windy
conditions than under humid, calm conditions, and under sunny skies as compared to cloudy skies.
Therefore, we will couple our estimate of relative ET from NDVI with the reference ET that takes into
account the effects of weather. We will call the relative ET from NDVI the “fraction of reference ET”
(ETrF). ETrF is, as described above, generally proportional to NDVI. ETrF generally ranges from 0 to
about 1.0. A value of 1.0 means that the fraction of reference ET is 1.0, so that the ET for that pixel
equals the reference ET value. The reference ET in this case is the “tall” or alfalfa reference ET that is
usually calculated using the ASCE Penman-Monteith equation (ASCE 2005).
We use the following conditional statement to estimate ETrF:
If NDVI>0, ETrF = 1.25 NDVI
Else, ETrF = 0.7
The 0.7 estimates ETrF for any water bodies that may exist in the image. Water usually has a negative
NDVI. The 0.7 value assumes that evaporation from water is about 70% of the tall reference ET value.
The statement translates into following in the raster calculator used in Model Builder.
Con("%ndvi_toa%" > 0.0, 1.25*"%ndvi_toa%", 0.7)
Double click on ETrF_from_NDVI tool, specify the input by locating NDVI_toa under NorthPlatte.gdb.
Save your output into the same geodatabase:
14
The result should look like that below.
15
We should colorize the ETrF map to help us visualize areas of high and low ET and to improve its
appearance.
We have made a color scheme for ETrF that can be imported and used. That scheme, named
ETrF_frmNDVI.breaks2.xml can be imported in the properties / symbology menu as shown below.
You will want to select “Classified” in the “Show:” window on the left. Then right click on the “Sym..
Range” list in the center of the window and select “Load Class Breaks”. You can direct to the
ETrF_frmNDVI.breaks2.xml scheme. You may want to left click on the “Label” tab and select “Format
Labels” to reduce the number of digits displayed in the label.
The ETrF_frmNDVI.breaks2.xml scheme will assign a different color each 0.1 ETrF. You can select the
color scheme as shown that goes from yellow as low to green and blue as high.
The resulting map should look like:
16
Differences in ETrF (relative ET) are clearly visible between various land uses, and various
fields. If we zoom in on the irrigated areas in the center of the basin, we can see the higher ETrF
associated with the irrigated parcels:
17
Again, we can see that the center pivot fields are somewhat out of shape because of the use of
WGS84 projection with the Landsat L1T images from the USGS, whereas the native projection
system for the ‘basin’ layer is Albers Equal Area. For a ‘serious’ product, we would want to
reproject the Landsat and associated products into the Albers Equal Area, or vice-versa.
Turn in: Colorized ET from NDVI map and legend for the entire basin area and for a close up showing
some irrigated agriculture (see above).
Part 4. Comparing ETrF from NDVI with ETrF derived by Surface Energy Balance
A growing area for remote sensing is the determination of ET using what is called a surface
energy balance (SEB). Some details on SEB are given in the lecture entitled “Lecture on Energy
Balance”. The advantage of SEB is that actual ET is determined, including when ET is reduced
by lack of soil water availability and when ET is increased due to recent rain events that wet the
soil surface and by irrigation. One such SEB method that is used by a number of western and
18
midwest states including Nebraska is the METRIC process (Mapping ET at high Resolution
using Internalized Calibration). The disadvantage of METRIC is that it takes some expertise in
ET physics to review and calibrate images. This is becoming more and more automated over
time, however. The other disadvantage of SEB methods is that they require a thermal band on
the satellite. Therefore, many SEB methods use Landsat images to get ET at the field scale using
thermal and reflected imagery.
We have applied METRIC to this June 20, 2013 image and have derived an ETrF map. We will
now compare the ETrF from METRIC with the ETrF using some spatial statistics and sampling.
Let’s first open the METRIC ETrF map and colorize it. The name of the ETrF image from
METRIC is: etrf24_ef_06202013_P35R31_L8_NP_Color.img. That image is in ERDAS
Imagine format and is located in the Exercise 7 data folder provided by the instructor. ArcMap
is able to read the img format.
Add the etrf24_ef_06202013_P35R31_L8_NP_Color.img image to the ArcMap and colorize it
using the same ETrF_frmNDVI.breaks2.xml scheme to colorize it. The result should look like the
following:
Notice the similarity of the METRIC ETrF map (on left) to the ETrF from NDVI map (on right)
created earlier. However, there are differences, as we will explore below.
The yellow and blue colors outside the Landsat path represent values of -2 and 2 that were
intentionally embedded by METRIC. Let’s get rid of these by excluding them in the symbology
menu. Go to symbology for the ETrF from METRIC and select the “Classify” button. Then
19
select “Exclusion…” and then specify the values or ranges of values to exclude. In our case, we
exclude -2 and 1.9 – 2.1. These are separated by a semicolon: “-2; 1.9-2.1”. The result should
look like the following side-by-side. Note that you may have to reload the
ETrF_frmNDVI.breaks2.xml scheme a second time.
20
Turn in: A screenshot of the side-by-side ETrF from the two methods, similar to that shown
above.
Part 5. Statistical Analyses and Comparisons of ETrF and Surface Temperature
Generally, and engineer or scientist needs to use statistics and quantitative comparisons to
demonstrate differences between two products or estimates, rather than just showing ‘pretty
pictures.’ In the following, we will conduct some sampling of about 180 points in our basin.
Those points were selected by the instructor and are distributed in agricultural (NLCD classes 81
and 82), rangeland (NLCD classes 52 and 71), forest (NLCD classes 41 and 42), wetlands (90
and 95) and water (NLCD class 11).
I will first show you how to make a shape file composed of a number of ‘points’ that will be used
later for sampling. I will provide you with a shape file of about 182 points, of which 150 points
lie within the basin. That shape file is named “Sample Points.” You are welcome to use this
shape file for sampling, but I encourage you to make and use your own for practice.
We create a shape file to enter our sampling point into. Go to ArcCatalog and to the
NorthPlatte.gdb. Right click on the name of your exe7 folder and select “new” and “shapefile.”
21
Give the shapefile the name “Sample_points” and make sure the type is ‘point.’ Also, we need
to define a coordinate system. We will use the WGS1984 coordinate system of Landsat since we
will be sampling the Landsat based image products. Our zone is UTM Zone 13N:
Once the Sample_points.shp is created, add it to the ArcMap project.
Now we will add points to the shape file using the ‘editor.’ Go to ‘editor’ and select “start
editing”. If the “editor” is not present at the top, you can load it by right-clicking the ‘grey’ area
on the menu bar and selecting ‘editor’ from the tool list.
22
Before you start the editor, you may want to zoom in on the ETrF or LST image (or even the
color composite image) so that you can see the features, and then start the editor.
When you select ‘start editing’, choose ‘sample points’ from the menu:
Hit “OK”.
Then, from the upper menu bar, select the “Create Features” tool (see below):
23
Then, in the “Create Features” Window, select the name of your point file (i.e., Highlight it):
This should make the “point” tool ‘active’ on the menu bar:
Now, everywhere we click on the map, ArcMap and Editor should create a point. Go crazy
clicking on points in ag fields, rangeland, water and forests in the mountains. Collect about 200
points. You can move around the map using the scroll buttons.
When you are finished, you need to save your points by first select “Stop Editing” from the
“Editor” menu. It will ask you if you wish to save the edits. Say “yes.” You should now have a
“sample points” file that contains those points. The result should look something like (you can
make the symbols larger and colored using symbology):
24
The total number of sample points in the Sample_points_AK shp file is shown below. Note that
about 30 of the points lie outside of the basin. Therefore, when this shp file is used later for
sampling of the NLCD feature, which has been clipped to the North Platte basin, the 30 points
that lie outside the basin will not have any values for NLCD and therefore will not be retained in
the resulting table of sampled values that come from the zonal statistics. This is a good learning
exercise on the behavior of sampling and tables.
25
The total number of sample points in the Sample_points_AK shp file is shown above.
Turn in: Location map of your sample points (over lay onto the NLCD). Note: you can chose to use my
sample_points_AK shape file. However, still please turn in the location map.
26
Zonal Statistics on the Sample points
We will sample the two sets of ETrF (from NDVI and from METRIC), the surface temperature,
the NDVI, and the Land Use (NLCD) for all of our sample points. We will use Zonal Statistics
to do this.
Go to the “zonal statistics as a table” tool:
We will need to sample each of the five images one at a time. Let’s start with the NLCD:
27
Note that we are using the “FID” field from the Sample_points feature, since these are numbered
from 0 to 180-something. We are sampling the NLCD_2011 feature and placing the results into
a table named sampled_NLCD. You will need to browse to find the NLCD_2011 from exercise
6. When the sampling is completed, we can open the sampled_NLCD table and we should see
statistics for all of the points:
28
As mentioned above, about 30 of the 182 locations are omitted from the above table because
they lie outside the North Platte River basin (above Pathfinder Dam) and therefore, they do not
have entries in the table.
Note that the values for min, max, mean are all the same because we are sampling points.
Therefore, to make this more efficient, let’s run again, and select ONLY “Mean”:
29
.
30
Now, let’s repeat this process for the two ETrF images, the NDVI and the LST. For consistency,
let’s call the tables as sampled_NDVI, sampled_LST, sampled_ETrF_from_NDVI and
sampled_ETrF_from_METRIC.
31
We should now have five TABLES in our data frame TOC:
32
If we open the sampled_NLCD, we should see about 8 to 10 classes. The NLCD classes for our
image are listed in the table below that is taken from NLCD attribute table:
33
Ultimately, we are going to create Excel plots to evaluate our data. To do this, we have to:
1.
2.
3.
4.
Join the tables
Export to a dbf
Import to Excel
Create the Graphics and statistics
Step 1. Joining the five tables:
We will join the five tables using sample NLCD as a target table. The FID will be used as
common field that the join will be based on.
34
35
If you open the sampled_NLCD table, you will see that the table has been expanded to include
the five columns from NDVI table. However, the column heading for NDVI is mean just like the
NLCD heading. Lets rename the NDVI heading by going to its properties as shown below.
Lets change Alias to NDVI.
Also, if the “create index” window pops up, you should select “YES”:
36
Lets join the ETrF from METRIC to the NLCD table also. Repeat the same joining process for
the NLCD table. However, this time you need to select the second “FID” column from the list. It
will have an ‘*’ next to it in the table above. In the join list below, it will be the second FID
entry.
37
Change the column header ‘Mean’ to ETrF_METRIC. Repeat this joining process for the
sampled_ETrF_METRIC and sampled_LST tables. Each time, select the ‘last’ “FID” in the
drop down list for the common field, and each time, rename “Mean” to a unique name.
When we are finished with the joining of the five tables, the final table for Sample_NLCD
should look like the following:
38
Step 2. Exporting the Table to Excel
Now we can export the sampled_NLCD table to a dbf so that we can open it in Excel.
From the open sampled_NLCD table, click in the upper lefthand corner and select “export” and
specify a ‘.dbf’ file type and name. Place the exported file into your project folder so that you
can find it. In the second menu that shows up, you will want to export ‘all records.’
I named my dbf file “Sampled_NLCD_NDVI_ETrF_LST.dbf”.
39
Now, open Excel, and go to the folder where the dbf is stored: (make sure that the “All files
(*.*)” is selected so that the ‘.dbf’ file shows up 
The table will be opened in Excel, and it will look like a mess:
40
So, the first thing we should do is to format columns so that they are easier to look at (NLCD
should have no decimals, NDVI and ETrF should have three decimals and LST one decimal).
Secondly, we have columns that we don’t need, namely the RowID, COUNT, AREA columns
that appear five times, plus the extra four FID columns. Therefore let’s delete those columns so
that the result looks like (I renamed the “MEAN” column “NLCD”):
Also, note that in the ArcMAP table, we renamed the “MEAN” headings to NDVI, etc. These
have reverted to “Mean_1”, “Mean_12”, etc. when we exported it. We need to rename these
again. (sorry).
41
Let’s save this file using “Save as” an Excel file.
Step 3: Now we can go to work creating some scatter plots to see relationships.
We are interested in the following questions, for example:
a.
b.
c.
d.
How does ETrF from NDVI compare to ETrF from METRIC energy balance?
How does the ETrF comparison vary with landuse type?
How does ETrF from METRIC relate to NDVI? (how does that vary by land use type?)
How does ETrF from METRIC relate to LST? (and how does that vary by land use
type?)
To answer the “by land use type” questions above, we will use different symbol types for each
land use. First, we need to sort the data by land use type to get all land uses into groups.
42
To sort the rows in Excel by NLCD, highlight all rows, including the header row. Select “Data”
from the menu bar and “Sort”. Select “NLCD” from the “Sort by” dropdown box. We will sort
by Values and from Smallest to Largest:
Recall from the table of NLCD values below the various definitions for NLCD classes. To keep
things simple, let’s group the data into the following general classes to plot with different
symbols:
Water
Forest
Rangeland
Agriculture
Wetlands
11
41 and 42 (deciduous and evergreen forest)
52 (shrub) and 71 (herbaceous) (Shrubs include sagebrush and
herbaceous is used in Wyoming for range grasses like crested wheat
grass, buffalo grass, bluestem, etc.)
81 and 82
90 and 95
43
Note that if you used different sampling points, you may have additional classes that you will
want to include in those groups. We are going to leave out out the 21 and 31 points (developed
open space and barren land) because there are only two of them.
We can colorize the “groups” of NLCD in the NLCD column for our benefit (see screen shot
below). Then we can create our first scatter plot to evaluate questions 1 and 2 above, by:
1. Plotting ETrF_from_NDVI (as Y) vs. ETrF_from_METRIC (as X)
2. Making five series in the graph, for the five groups in our list above.
The result is the graphic shown below, where we have done the following to improve its
appearance. Please make these ‘improvements’ also. It is good practice in making presentation
and report-quality graphics, and is a good reflection on UNL:
a. We made the plotting area square so that the data plot 1:1 as a 45 degree angle.
b. We made the scales of the Y and X axis the same (0 to 1.4)
c. We added a 1:1 line by adding it as a sixth series (just indicate values of 0,1.4 for each
entry) and we made it a solid black line with no symbols
d. We made each series symbol larger and with black outline and with different color and
different symbol shape so that they can be readily distinguished.
e. We made the axis labels have one decimal place and we made them 12 pt font.
f. We added titles for the axes and made them 14 pt
g. We added a title and made it 12 point and we made the title box have white fill so that the
grid lines did not show through
h. We made the total graph outline go away and we made the plotting area outlines black,
not grey.
44
1.4
Sampled Locations in North Platte River
Basin above Pathfinder Reservoir
ETrF from NDVI
1.2
1.0
Water
0.8
Forest
Rangeland
0.6
Agriculture
Wetlands
0.4
1:1
0.2
0.0
0.0
0.2
0.4
0.6
0.8
1.0
ETrF from METRIC
45
1.2
1.4
Let’s take time out and answer some questions about the above graph:
Question: Please comment on how well the ETrF from NDVI agrees with the more accurate
ETrF from METRIC for each class. Comment on why the agreement is good for each class or
bad for the class. For example, why is the agreement good for irrigated agriculture, but not very
good for rangeland. (turn in)
Now, let’s make two more scatter graphs to see how ETrF from METRIC behaves vs. NDVI and
vs. LST for all classes. In Excel, let’s copy the graphic that we have made and then we can
modify it to plot ETrF from METRIC (as Y) vs. NDVI (as X):
1.4
ETrF from METRIC
1.2
Sampled Locations in North Platte River
Basin above Pathfinder Reservoir
y = 1.15x + 0.04
R² = 0.594
1.0
Water
0.8
Forest
Rangeland
0.6
Agriculture
Wetland
0.4
Linear (Agriculture)
0.2
0.0
-0.6 -0.4 -0.2
0.0
0.2
0.4
0.6
0.8
1.0
NDVI
Note that the 1:1 line does not fit the data anymore because we are plotting ETrF vs NDVI.
NDVI generally ranges from 0.1 for bare soil to 0.7 or 0.8 for full vegetation, and NDVI is
usually negative for water. Negative NDVI occurs for water because the red band reflects more
than the NIR band for water. In the case of soil and vegetation, however, the NIR band reflects
more than the red band.
Instead of the 1:1 line, we have added a ‘trend line’ to the AGRICULTURE points. You can do
this by right clicking on the ag. Data on the plot and selecting trend line. If you highlight the
boxes for ‘display equation’ and ‘display R2’ . We can see a relatively strong linear relationship
between ETrF from energy balance and NDVI.
46
47
Let’s take time out and answer some questions about the above graph of ETrF from METRIC vs.
NDVI:
Question: Please comment on how well the ETrF agrees with NDVI. Which classes show a
strong relationship between ETrF and NDVI? Which ones do not? Give some reasons why they
do or do not. (turn in)
Our last plot is ETrF from METRIC vs. Land Surface Temperature (LST):
You can copy the ETrF vs. NDVI plot and turn it into ETrF vs. LST. Add a trend line for
agriculture like you did for ETrF vs. NDVI:
1.4
ETrF from METRIC
1.2
Sampled Locations in North Platte River
Basin above Pathfinder Reservoir
y = -0.035x + 11.57
R² = 0.859
1.0
Water
0.8
Forest
Rangeland
0.6
Agriculture
Wetland
0.4
Linear (Agriculture)
0.2
0.0
280
290
300
310
320
330
Land Surface Temperature
Turn in the three scattered plot figures created above.
Question: Do you think you can estimate ET from LST alone for specific land use classes, for
example, agriculture, and forest? Explain your reasoning (turn in).
48
5. REFERENCE and ACTUAL EVAPOTRANSPIRATION
The reference evapotranspiration we are going to use is the ‘alfalfa reference’ ET, generally abbreviated
as ETr. The alfalfa reference ETr contrasts with the other type of reference ET, which is for clipped grass
vegetation. Grass reference ET is usually abbreviated as ETo. Because alfalfa is taller and leafier than
clipped grass, ETr is typically about 20 to 30% greater than ETo. ETr represents the upper limit of ET as
constrained by environmental energy to convert liquid water to vapor. This energy comes mostly from
solar radiation and air temperature.
We will estimate the spatial distribution of the Actual Evapotranspiration (ET) by multiplying
reference ET with the ETrF rasters:
ET = ETr X ETrF
Question: ETr for June 20, 2013 at the HPRCC Encampment weather station is 8.68 mm/day,
then what is the daily estimate for ET for the five LU classes sampled? You can use the
sampled data that are in the Excel spreadsheet to make this estimate.
The ETr for the month of June is 270 mm, then what is the monthly ET for the five LU classes
sampled? (turn in)
The final raster map to produce is one of monthly ET for the Two Row Landsat Image for the Month of
June, where alfalfa reference ETr is 270 mm for the month. Turn in.
SUMMARY OF ITEMS TO BE TURNED IN
49
Download