One of the GIS areas where much remains to be done is
the domain of time analysis. Indeed, if almost all the
available tools focus on the spatial evolution of a phenomenon, one finds
oneself quite helpless when it comes to visualizing or analysing a phenomenon
that, also, evolves as a function of time.
We have had, for quite a long time, animation tools that allow us to see a
sequence of maps and perceive changes as a function of time. But,
although useful for communication, these tools do not allow any serious data
analysis.
We will discuss, in a series of articles, the ArcMap and
ArcGis Pro new available tools as Time Pattern Mining Toolbox. Firstly,
we will follow an example, with ArcGis Pro and then, with ArcMap.
What is a Spatial-Temporal
Cube?
The following image
summarizes the answer to this question.
On the left we have our points. They
are distributed on the XY space and it is this distribution we, usually, study
and analyse with the ArcGis tools. But
for a same location, we have several points that correspond to a different
moment. That is why the points are distributed, also, on the Z axis
which corresponds to the time.
The spatial-temporal cube will allow us to analyse the
distribution of our points on these three axes.
First important thing to remember: The tools we will see in
relation with the space-time cube do not allow analysing a given phenomenon
from its attributes. They are limited to the
analysis of the occurrence or not of a phenomenon (the point exists or does not
exist).
The process is simple; we will determine a box size (bin), such
as the pixels or the cells of the rasters, but 3 dimensional. And
we will determine the number of points contained in each box. It is
this number that we will visualize and analyze.
Let’s take an example, which
will serve us to follow the process in detail. We have weather forecasts
that contain wind gust forecasts. These forecasts for 8 days,
every 3 hours for an area encompassing Brittany.
An
overview of the attribute table allows viewing the “gust” attribute
which corresponds to the speed (in knots) of wind gusts, as well as a
“DateTime” field which indicates the Date and the hour of each
forecast.
Maybe, one day we will be able to analyse the bursts
values distribution as a function of time. For
the time being, since we are limited to the study of the occurrence of points,
we can, for example, define a speed threshold that seems risky and analyse the
distribution of points that exceed this threshold value.
If we set the threshold value of 20 knots, we will create a
new data layer with the points for which this value is exceeded. The points
where this value is not exceeded will not be present. To
this end, we select by attributes, gust> = 20, and export the selected
points.
Now, we have a layer to build our cube.
Creation of the spatial-temporal
cube
To build the cube, which will
be stored in a NetCDF (.nc) format, we will use the Create
Spatio-Temporal Cube tool in the Spatiotemporal Models Explorer
toolbox .
Once
launched the command, the parameter window appears.
The input entities correspond to our burst
selection layer> 20 knots.
The output spatial-temporal cube will be our cube.
The time field will be used to define the Z axis. It
is absolutely mandatory to set a Date field. For
those who are still in shapefiles, remember their Date fields can only contain
the date but not the time. If you are going to work with
time steps less than the day long, you must have a geodatabase entity class.
For now, we do not have a template cube on which to retrieve
the other information. If we make a series of cubes,
the first one will serve as model, we must verify that the steps and the boxes locations
will be the same for all cubes produced.
The time interval is the size of the cube boxes on the
Z axis. In other words, the time step that we will apply to our analysis.
In our example, the forecasts time step is 3 hours.
A box size of 12 hours will produce 16 time steps in Z. A box
size of 24 hours, will produce 8.
One of the constraints of the cube is that there must be at
least 10 time steps. We will choose the time step of
12 hours.
The distance interval is the cell size in the XY plane.
If you do not fill one or both intervals, the program
calculates it by default.
Visualization of the spatial-temporal
cube
To visualize the cube you have to download a toolbox
containing two tools: one that allows visualizing the cube in 2D and another
that allows visualizing it in 3D.
The
toolkit can be downloaded from http://esriurl.com/SpaceTimeCubeUtilities
.
Once downloaded and
uncompressed, you can add it to the geoprocessor by clicking on the Ribbon
Insert -> Toolbox
We will see the visualization in 3D.
You must have a 3D window
(Scene) in your project. If
you do not have one, click Insert-> New Scene
You must remove the reference to an elevation layer.
Indeed, if you do not do this, the columns of the cube will
be displayed relative to the ground level and you will lose visibility of the
correspondence of the different time steps.
In the Content panel, click on your scene -> properties
-> Elevation surface, deploy Ground and delete all services present.
Execute the visualization
command of the spatial-temporal cube in 3D:
Check the cube you have just created. Give
a name to your visualization layer.
At this stage, in the variables to display, you will only
have Count. In the next article we will discuss how to use
other variables after using the Hot Spot Analysis command .
Afterwards, depending on your
graphics map, you will have the following result:
You can navigate in 3D on the screen to visualize
where are the time steps that exceed the most, the burst threshold value. But this remains a visual
evaluation.
To statistically determine the number of overshoots, the Emerging
Hot Spot Analysis command provides the standard deviation of the
results.
We will discuss this topic in the next article.
