DTED is dead.

DTED is dead.

 

 

The Map won’t tell you everything you need to know

 

 

Throughout my career as a professional soldier I was reminded that the map only tells so much about the terrain, and that in ground operations the terrain would be a prime point of consideration for all tactical decision-making. This went straight into design principles of Steel Beasts, of course; we’re deliberately omitting some elements from the student’s view as to reward a direct study of the virtual terrain rather than its abstracted map representation. Maybe we could do more here – but it’s a start.
Simulations are used for training; they are used in research and experimentation. And yet, the vast majority of experimentation and training is taking place in synthetic terrain that is usually based on elevation data with a mesh width of 30m (DTED2), if not worse. There are a number of reasons for that – availability and budget restraints are often named – but in addition, often enough the underlying simulation can’t handle better quality.
Even though the terrain may appear quite realistic at first sight, anyone working in the business of delivering training to the warfighter must be aware that the human eye and brain are complicit in the business of making the human observer believe “being there”, willing to suspend disbelief, willing to yield to the overwhelming firepower of visualization. We must therefore make conscious efforts to take a step back and analyze what is being presented to us. What is “real data”? What is embellishment by the simulation’s terrain and render engine – like adding geotypical details, or even those that just look pretty? What is simply our own brain filling in the blanks?
The latter part is hardest to bypass – but as it is also a common issue for every single virtual simulation. My focus in this piece is on the source data, what we simulation engineers make of it, and what effects this has on simulated outcomes.

Sweet poison

Low resolution terrain databases are cheap to come by and they usually perform with high frame rates – what’s not to love about that? Those minor details that are missing, can’t we do without them?
In some places we can. In others, it is somewhat dangerous to do so. Even 10m grids create few places to hide for combat vehicles positioned in the synthetic environment. This results in longer lines of sight, longer times of exposure, earlier target detection, and it shifts the balance of the exercises in favor of weapon systems with stand-off. The enormous illusion-firepower of simulations then helps to elevate the concept nearly to dogma: That we will have stand-off because our simulations predict that we do.

We are confronted with simulation-assisted bias.

Procedural embellishment

Likewise, low resolution databases cannot adequately recreate narrow valleys with a road network that competes with rivers and buildings for scarce flat ground. Our team is currently working on a database covering the area of operation of the Battle of the Bulge, east of and around the town of Bastogne, Belgium.

As it turned out, DTED2 30m grid data were entirely useless without procedural modifications to restore some of the detail lost in coarse quantization of the underlying terrain. Consequently we developed tools in the last months to restore some of the lost detail in low resolution databases, but this can help only up to a degree. Lost information can only be “re-invented” (=made up in a plausible manner), but not restored.

Creating synthetic worlds is a collaborative effort. At each step people give their best to make the result presentable. Artists want to make buildings look pretty, and terrain to look less sterile. So details get added to make the terrain more believable, simply because everybody is giving their best to create the best value for those who are supposed to be trained with the simulation. However, each of these well-meaning attempts to overcome deficits in computer hardware or availability of data further blurs the line between what is based on hard data, and what is made up in pursuit of immersion. Add to that exaggerated marketing claims like “15cm resolution” when that resolution is actually based on procedural modification – in other words, making things up. Add artificial ants crawling through synthetic grass when zooming up, and the illusion is perfect – except, it remains an illusion, as impressive as the underlying technology may be.
LIDAR scan data do offer a way to reduce the simulation-assisted bias. At eSim Games we are proud to help our customers to utilize this new source to create better terrain databases. High resolution terrain data used to be a scarce and precious treasure. With rapidly falling prices due to semi-automated data collection (aerial LIDAR scans) and more sophisticated utilization of satellite data telemetry better raw data are increasingly becoming a commodity; all of Denmark, for example, has already been scanned at 2m resolution.

LIDAR, the remedy

Our customers’ experimentation with LIDAR scan based terrain databases clearly show that simulated outcomes are often shockingly different when comparing them with conventional databases. Armored vehicles and anti-tank missiles designed to maximize their stand-off advantage suddenly find themselves outnumbered and in duel situations of under 500m range in terrain where the old database predicted long lines of fire.

Where terrain details allow to effectively mask vehicles in a tactically sound manner, AAR analysis shows that where in 10m grid based databases impact locations from incoming fire are nearly evenly distributed between hull and turret, in high-resolution terrain the impact locations almost completely shift to the turret areas – which happens to be in line with observable reality since WWII days.
Even where the terrain is mostly flat, an elevated railway embankment may create considerable dead space in which a lot of bad surprises can be hidden. Even minor depressions create opportunities for ambushes. In fact, ambush becomes the default tactic for defenders in almost every scenario where 10m grid databases rarely offer suitable places to hide effectively:

High time for simulation engines to catch up!