In this lecture, you will study another example of

spatial data analytics problem in military applications.

In fact, military and intelligence sector

still create the largest demand of spatial data production.

So it is worth of tasting military application of spatial data science.

The problem is a defense research center is assigned to develop a framework

for designing optimal thermal observation device network against potential infiltration.

For the problem, they need

a spatial data analysis framework capable of optimal infiltration routing.

The framework should support dynamic generation of

network and minimum cost path finding of the network.

It can be used for a defense mode,

for which optimal distribution of TOD locations can be designed and for an attack mode

for which vulnerable infiltration routes can be generated

with a series of simulations with changing TOD locations.

The figure would explain the problem in a more friendly manner.

The map shows the concept of infiltration from beginning line to ending line.

The picture shows a real TOD,

which is the thermal imaging system,

which measures infrared energy emitted from an object with temperature.

In military, TOD is popularly used for

precise identification of long-range target during the night.

With respect to the given problem,

what do you think is the meaning of optimal infiltration route?

That would be the route which has the minimum sum of the TOD detection probability.

Assume that the raster data processing is applied,

and you could compute the detection probability of each cell.

The raster data converted to network,

and the detection probability will be the cost of each edge of the network.

For solving the problem,

we acquired the following spatial data of target area.

First of all, digital elevation model,

which will be used for visibility analysis,

also known as viewshed analysis,

the number of TODs and their locations,

and additional terrain features,

which could impact on

TOD detection probability such as vegetation, obstacles and hydrography.

The three datasets are illustrated in this slide.

The DEM of target area is given and,

hill-shading is applied for visualization.

Four TODs are assumed,

and they are located on the area of wide viewshed such as ridges or mountain tops.

Vegetation, hydrography and obstacles were presented as

terrain features which could impact on TOD detection probability and routing.

The flowchart summarizes a solution to the problem of finding optimal infiltration route.

Obstacle and hydrography can make non-passable areas.

Vegetation layer can produce concealment probability,

DEM and TOD locations could produce

TOD detection probability and then with consideration of concealment probability,

the final detection probability can be obtained.

With respect to the cost,

the final detection probability Dijkstra's algorithm can be applied,

after raster format is converted to network format.

Finally, you can get the optimal infiltration route at

the given setup of TOD locations and starting point of infiltration.

Note that it is just for the case of single starting point and fixed locations of TODs.

Later, we will repeat the same operation with different set-ups

of TOD locations and starting point of infiltration in red box.

To solve the problem,

an integrated framework of GIS and data analytics tool is

recommended, because it requires advanced network analysis and simulations,

which would not fit in GIS tools,

which only offers basic analysis capability.

Now, let's take step-by-step procedure to get the solution to the problem.

The first task is to use obstacles and hydrography layers to create non-passable areas,

where no route will be generated.

The table tells which features are passable and which features are non-passable.

The next task is to create a concealment probability map,

which means the covering likelihood of

infiltrating soldier or any military objects mainly due to vegetation type and density,

which could lower the TOD detection probability.

The tables shows density measures of tree and its corresponding concealment probability.

Based on the table and vegetation layer,

the concealment probability map could be produced.

Now, you are looking at it.

The figure illustrates TOD detection probability

based on TOD's location and ACQUIRE model,

which is dependent on the distance.

ACQUIRE model is proprietary and applicable to military TOD Model-TAS970K,

and it is confidential so that I cannot open

the detail of ACQUIRE model here, unfortunately.

The next task is to conduct a viewshed analysis,

which is to find visible areas from a specific location.

In the problem, viewshed analysis from TOD locations can delineate

areas where TOD can conduct a surveillance.

The figures are the outcome of four viewshed analysis from each TOD location.

The next step is to combine

TOD detection probability and outcome of a viewshed analysis layer.

Because the ACQUIRE model did not consider visibility,

the figure illustrates a TOD detection probability map

with consideration of visibility from viewshed analysis,

as well as the distance for ACQUIRE model.

Now, we have a TOD detection probability map.

However, it did not consider concealment probability as yet.

As mentioned, vegetation density can discount detection probability,

and the corresponding concealment probability map is already given.

The more concealment, the less detection probability.

So the factor, 1 minus concealment probability is multiplied to TOD probability,

as you can see in the equation,

and we can get the final detection probability map as shown in the figure.

The final detection probability map in a raster format is

converted to network format in which each cell is converted to a node,

each cell is connected to neighboring 8 cells.

With respect to the network data with cost of the final detection probability,

Dijkstra's algorithm was applied.

In this case, 7 arbitrary points were set up to starting points of infiltration,

and the optimal routes are presented,

which have the minimum sum of detection probability to ending line of infiltration.

While starting points of infiltration are changed,

you can find out vulnerable areas to the infiltration with

the present locations of TOD.

So far, you have a studied how to find the

optimal infiltration routes with respect to fixed TOD locations,

which can decide the spatial distribution of the final detection probability.

However, we did not consider the dynamic nature of infiltration problem.

TOD location can be changed,

in other words, spatial distribution of the detection probability can be changed.

For that, we should consider a simulation approach

with moving TOD locations around and apply the proposed solutions repetitively.

The figure shows the result of a simulation of 1,000 times,

where starting points are all the points on the starting lines of infiltration.

The white line segments represent

the vulnerable areas or vulnerable routes for infiltration in an attack mode.

At the same time,

the simulation approach can be used for optimal design of a TOD surveillance network

for a defense mode as well.