Exercise Title: Gridding Point
Shapes in Saga
Abstract: Using a
very straightforward gridding method, the previously created point shapes (of
World Ocean Database (WOD) ocean temperatures) are gridded using Liberia area dummy grids as
target geometries. Some time is spent working on the problem of
color palettes for the grids, resulting in some general recommendations.
Finally, a method to extract samples of data grids as ASCII for close
inspection is described.
Preliminary Reading (in
OceanTeacher, unless otherwise indicated):
Murray Brown; final
panels based on a recent short discussion on the Saga wiki pages between
user Ellen Gonzalez and Volker Wichman about visualizing grid cell
|1. Now, we're
going to grid the point shapes we created in the previous exercise.
This is not a course in gridding -- a very complex subject in its own
right -- but just an introduction to the concept. For this reason,
we're going to use a relatively simple algorithm, INVERSE DISTANCE WEIGHTED. You
can come back to the algorithm selection steps and try other methods on your own time.
|2. Load these
- Use FILE >
SHAPES > LOAD to load all of the above shapes and the World Borders
- Use FILE > GRID > LOAD to load the dummy grids. You should
have at least the 1-degree version and the 0.1-degree version.
|3. Here is the full
complement of data files, for students who have created all the above files,
including the dummy grids.
Each grid has a "system" line
that identifies it by these items (left to right). These examples refer to the uppermost
- Resolution in Degrees: 0.01º
- Number of Columns: 2000
- Number of Rows: 1500
- Center Longitude (lower left cell): -22.995º
- Center Latitude (lower left cell): -5.995º
Below that is listed the name of the actual grid. All of these
are dummy grids, with structure but no meaningful contents.
|4. Take a minute to browse
the many "gridding" tools available in Saga at TOOLS > GRID >
GRID-GRIDDING. This should give you some idea of the
complexity of the subject.
This is not an intensive course in gridding, so
we won't take much time here. You can explore other methods on your
own. In the panels to follow we'll use a favorite method that seems to
work well with marine station data.
|5. Select TOOLS >
GRID > GRID-GRIDDING > INVERSE DISTANCE WEIGHTED. This window opens with a
typical set of options for you to select/provide. An article above
gives more background information about this gridding method.
As you work on this
window, and all Saga tools, small helpful hints about the different items
will appear at the bottom left, as you see here.
|6. Make these selections:
- POINTS - Use drop-down menu and select the jfm_0m point
- ATTRIBUTE - Select TEMPERATURE.
- TARGET GRID - Select GRID (which means we will name one later)
- SEARCH RANGE - Select LOCAL
- MAXIMUM SEARCH DISTANCE - Select 4 degrees (you can experiment with this later)
- NUMBER OF POINTS - Select ALL POINTS WITHIN SEARCH DISTANCE
- DISTANCE WEIGHTING - Select INVERSE DISTANCE TO A POWER
- POWER - Select 2
Of course, you are encouraged to experiment with all of these selections,
when you have your own data to work on. The best approach is to vary
each one separately to see the effects on the resulting grid.
Now click OK.
|7. WORDS OF
WISDOM. What are these values for? Imagine a formula that
will take the scattered data points and calculate a regular grid (i.e. the
target grid) of values. Each new value to go in the grid is calculated
from the nearby attribute measurements. The values to be used are
limited both by the search distance (i.e. the maximum distance away from the
grid point to look for measurements) and by the number of points (which
could also be a limiting factor). But when the points are being
considered in the calculations, then distant points are given much less
importance in the calculation than nearby points. The importance is
calculated from the inverse of distance, i.e. the further away, the less
important. But the algorithm goes even further and adjusts the inverse
of the distance to be exponential (the power). If the power is 2, for
example, then the importance is proportional to 1/distance squared; if 3,
for another example, then the importance is proportional to 1/distance
cubed. So the importance of the values drops off very fast with higher
powers, less so with smaller ones. Larger values would not lead to
very smooth grids, because the calculations would reflect very small areas
quite close to the grid point; smaller values would lead to smoother grids,
for the opposite reason. This very short explanation just refers to
this algorithm, INVERSE DISTANCE WEIGHTED. There are dozens in the
literature, and each has its own characteristics and concepts.
|8. For GRID SYSTEM, select
the 1º dummy grid for Liberia.
Make sure that GRID is
set to CREATE. Then click OK.
will see a new object appearing under the 1º system.
Notice that it is identified by the name of the variable, plus [INVERSE
DISTANCE WEIGHTED] a very generic name.
PRACTICAL TIP: To avoid
problems later on with grid identification, you should change the name of
this grid, to something easily recognized, like temp_jfm_0m.
This tool uses the same name for all the products here, so do not leave the
automatic name unchanged.
|10. To see the new grid,
double-click on the name, and select NEW for the map.
|11. Here's the new grid.
Notice that the grid is complete, and has cells in the northeast corner
which are on land. We'll learn in a later lesson how to deal
with this common result of gridding marine data.
NOTE: Yes, there is
an area of obviously poor quality data, the narrow vertical "stripe" in the
center. This happens all the time with archive data. Just live
with it for now, and later we might address it during quality control work.
|12. Let's fix up the colors.
First, in the properties panel, select COLORS > TYPE > GRADUATED COLORS.
|13. Then just below it, find
the tiny ellipsis (...) to the right side of 10 COLORS.
|14. This opens the COLORS
window. Click on COUNT and change 10 to 50 or 100.
Then click OK.
|15. Then select PRESETS to
find a long list of "pre-packaged" color palettes (some with appropriate
names). Select RAINBOW and click OK. Click OK again to leave the
At the bottom of the properties panel, you can click
SETTINGS > APPLY to see your choices in action.
|16. This is our 0-m
temperature grid, with a good color palette. Including the "stripe".
|17. But what about the value
range that goes with the palette? Right-click on the grid name, and
|18. This histogram is very
similar to the ones we created in
5.1 Converting a Data
Table to a Point Shape in Saga, in so we'll use the same value range of 21-30º.
|19. Back in the properties
panel for this grid, you can see here where to enter these values.
Then click SETTINGS > APPLY (at the bottom of the panel).
|13. Now repeat the gridding
process, as you did above, but this time select the next smaller-resolution "dummy
grid." In this case there is a 0.5º grid available, so it is
Make sure GRID is set to CREATE.
Then click OK.
|14. Repeat this process with
the same point shape, but with each of the dummy grids you have. This
shows you what the different RESOLUTIONS provide.
resolutions range from 1º (top
left) to 0.01º (lower right).
[The last one will take much longer than the others, so just be patient.]
The images look "smoother" as you get a higher resolution, but you can
see that false features begin to appear, such as the many small "bull's
eyes" (as they are called).
|15. All of these
have been set to the same 100-color rainbow palette and 21-30º
value range, so they are visually comparable. There is nothing
"sacred" about this range, but you recall we obtained it by visual
inspection during the exercise 5.1 Converting a Data
Table to a Point Shape in Saga.
|16. Also, recall that
during the process of making the surface analyses in ODV we remarked that
the data probably should be interpreted/visualized at resolutions not much
smaller than 1-2º. So, all things
considered, our target resolution for these specific data from ODV will be the 1º dummy grid. You should spend time
with your own data to see what resolution, palette and value range gives results that satisfy your
research or data management needs.
|17. You can delete the grids
with resolution smaller than 1º (just right-click on the "grid system"
and select CLOSE on each one). Leave only the 1º system in place.
|18. Now, you can
perform the gridding on the remaining 3 point shapes, using the 1º
dummy grid as the target geometry. Be very careful to set the
selection window to TEMPERATURE for the variable to grid, and CREATE for the
target (so you don't over-write an existing grid).
|19. Here, all four 1º
|0-5 m, JFM
||0-5 m, JAS
|3500-4500 m, JFM
||3500-4500 m, JAS
The top row looks fine, but the application of the same palette to the
deeper analyses completely obscures any details. We need to abandon previous
MDL recommendations to use only one palette, and be realistic about
cases like this.
|20. Here we used a different
palette for the deeper grids,
based on a value range of 0-3º, and it really
makes the data more understandable. There may even be a seasonal
signal, but that would require much more study to confirm.
|21. So, based on
this new insight, here are the MDL Rules for Gridded Data
- Climatological Rasters.
- When multiple grids/rasters are similar, in average and range, then
use the same color palette for all, based on good representative minimum
"low" and maximum "high" values.
- The selected
be clearly stated, (often) whole-number values.
- But, when natural value fields are very dissimilar, such as the sea
surface versus the deep sea below the permanent thermocline, for good
scientific reasons, then select
separate palettes, and carefully distinguish
between the separate palettes in your reports
- In certain cases, moreover, temporal (often "seasonal") patterns
are highly significant but might be obscured by a
single "annual" palette. Be sensitive to such situations, and
consider adopting appropriate seasonal palettes if this clarifies such
- Operational Rasters. The emphasis is on pattern or
feature recognition in synoptic rasters, and much less so on
comparisons. So don't worry about common palettes, and use the
actual data range that the display program automatically applies.
|22. In the setup for
the gridding algorithm you used here you could probably get much
smoother fields with a larger search radius and other changes. You can
explore these effects on your own. For open ocean data, collected over
many decades, it is not surprising to find this level of "noise" in the
data. Much more smoothing is needed to prepare the figures you'3w12ll see
in various Atlases.
right-click on each of the new grids, and save them in the folder PRODUCTS >
SAGA > GRIDS with these respective filenames:
And don't forget that each Saga grid file (*.sgrd) is accompanied by *.mgrd
and *.sdat auxiliary files.
24. Now we'll explore a little-known data inspection
tool in Saga, to see that actual contents of a new grid. Here we've
deleted everything except the 0-5m grid for the JAS temperatures.
|25. To choose an area for
inspection, click the ACTION TOOL in the row of Saga commands.
|26. Now use your cursor to
drag a rectangle around an area of special interest. It can be up to
100 cells by 100 cells, but not larger.
27. Now select the ATTRIBUTES tab in the SETTINGS
panel. You'll immediately see a table of the data values in the
selected grid area. The empty cells in the graphic are also empty in
|28. You can think of lots of
ways to use this capability when you're inspecting new grids for problems or
questions of accuracy. Also you can block areas of this matrix and
paste them into other applications, if you don't need anything larger than
100 X 100 cells.