The Influence of the Quality of Digital Elevation Data on the Modelling of Terrain Vehicle Movement

Abstract

This study investigated digital terrain models and options for their evaluation and effective usage. The most important result of this study was the introduction of the slope reduction method for low-detail elevation models. It enabled accurate results of passability analyses by performing adjustments of slopes. In addition, the goal was to determine the strengths and weaknesses of selected data for use in cross-country mobility analyses, followed by recommendations on how to use these databases efficiently to obtain accurate results. The selection of elevation databases (1 m, 5 m, 10 m, 30 m) was determined by the focus of data development projects of NATO and current scientific research projects of the Ministry of Defence of the Czech Republic. Key findings showed potential for use in practise for all tested elevation models. Efficient usage of low-detail models in CCM analyses is limited; nevertheless, they can be augmented with additional vector data or automated remote-sensing technologies.

1. Introduction

Terrain passability analysis as an integral part of battlefield intelligence plays a key role in military operations. Digital elevation models (DEMs) are the most detailed and up-to-date source of landscape information in comparison with other types of data (soil, forests, roads, settlements, etc.). The terrain relief is also the most stable part of the landscape. From the military perspective, DEMs are one of the only reliable sources of information for detailed, accurate and rapid terrain analyses. On the other hand, the quality of elevation models varies substantially depending on an area of the world. North Atlantic Treaty Organization (NATO) nations’ territory is covered with data with a high resolution and accuracy. The availability of high-quality data beyond European territory might be very limited (e.g., processed ready-to-use 1 m terrain models). Therefore, any output analysis from the data is constrained as well. It is essential to know what effect the detail of individual terrain models and slope models derived from them has on the results of terrain passability analyses. Besides the military perspective, there are other domains that have analysed DEMs for the extraction of landscape information. Study [1] referred to the impact on the protection of water resources, and photogrammetry-based mapping of microrelief forms was studied in [2,3], which used tri-stereo Pleiades images for the morphometric measurement.

The basis of this article was a study of the currently most used DEMs in the Czech Republic and NATO. To include different aspects of landscapes abroad, available global databases were included in the study as well. The results and recommendations of the article were aimed mainly at the NATO environment, i.e., NATO Command Structure, NATO Force Structure and NATO nations’ use cases. The analysis was performed by evaluating slopes from DEMs with the method of raster analysis, including implementing the influence of soils and the quality of terrain surfaces. Values of slopes and results of cross-country movement (CCM) analysis were compared for each terrain model in selected areas.

The key approach of the article, i.e., identifying the influence of slope accuracy on the results of CCM analyses, was based on both national and international studies. The basic methodology of analysis of elevation models was given in [4]. The study compared the accuracy of heights of terrain models with measured values in the field and assigned them specific output statistics. These statistics showed the specific advantage of 5 m × 5 m DEMs. It combined both assets in terrain detail and data availability across the world that are needed for practical use, such as visibility or passability analysis. Despite some problems with accuracy, especially in areas of significant microrelief shapes, DEMs comply with the declared accuracy and they are a major contribution to the field of geoinformatics and terrain analysis. The conducted tests confirmed the declared accuracy of all tested Czech models (used also in this article). Analysis of terrain measurements of the accuracy of elevation models, carried out by the University of Defence, Brno, and a comparison of the possibilities of using available elevation models with different levels of accuracy were the main prerequisites for a new, as yet unpublished methodology aimed at evaluating the influence of microrelief shapes generated from various elevation models on military mobility.

A stereo-vision-based terrain passability estimation method for off-road mobile robots is presented in [5]. The method models surrounding terrain using either sloped planes or a digital elevation model, based on the availability of suitable input data. This combination of two surface-modelling techniques increases the range and information content of the resulting terrain map. As defined in [6], the greatest asset to global elevation models are the TanDEM-X and ALOS projects. The TanDEM-X satellite mission has now mapped the Earth’s global topography with a spatial resolution of 0.4 arc-sec (about 12 m). While TanDEM-X is a commercial mission, a down-sampled elevation model (WorldDEM) with a 3 arc-sec resolution (commensurate to SRTM) and global coverage, the ALOS system is a new 3D model of the Earth’s surface of up to a 0.15 arc-sec (5 m) resolution. It will be generated from optical stereoscopy carried out aboard the ALOS satellite. The option of comparing various terrain models can be studied from [7]. It assesses uncertainties in a derived slope and aspects from a grid of DEMs. A quantitative methodology was developed for objective and data-independent assessment of errors generated from the algorithms that extract morphologic parameters from grid-based digital elevation model (DEM). The generic approach is to use artificial surfaces that can be described by mathematical models; therefore the ‘true’ output value can be pre-determined to avoid uncertainty caused by uncontrollable data errors. A prospective development and usage of a new surface model, namely, TanDEM-X High-Resolution Elevation Data Exchange Program (TREx), was introduced in [8]. Using this model to replace insufficient low detailed elevation models seems to be the only option to cover the information gap in some locations. One of the most important impacts of TREx is that it brings accurate results of the terrain analysis from a global perspective. The most useful analyses that can be gathered from it are visibility, propagation of radio signals, searching for areas suitable for air landing and route planning. Data can be procured in its raw form, but users may encounter issues with its extensive processing [9]. Global models can also be substituted for more detailed models in the case of the availability of local LIDAR data [10,11].

The CCM model used in NATO is called New NATO Reference Mobility Model (NRMM); see [12] for a detailed description. The analyses of the model and its gaps are in [13,14]. The NRMM was originally used to facilitate a comparison between vehicle design candidates by assessing the mobility of existing vehicles under specific terrain scenarios but has subsequently and most recently found expanded use in support of complex decision analyses associated with vehicle acquisition and operational planning support. A study [15] verified the usability of NRMM in new, so far untested conditions. It is based on a comparison of the empirically based NRMM with the physics-based Nepean Tracked Vehicle Performance Model (NTVPM) for assessing the cross-country performance of military tracked vehicles. The NRMM can be used strictly for the tactical level of planning. Such detailed data is not available across the whole Supreme Commanders of Allied Powers in Europe’s (SACEURs) area of interest (AOI). The AOI is an extended area beyond the territory of NATO nations where NATO can operate [16,17]. Global availability of high-detail elevation models is limited; therefore, other simplified CCM models must be considered. Future requirements of NATO towards CCM analyses are defined in [18], emphasising the ongoing important position of passability analyses in military planning.

The research of the CCM model in the Czech Republic at the University of Defence at the Department of Military Geography and Meteorology focuses on creating a system of coefficients representing various land features and conditions. Their analyses and implementations defining soil impact are in [19,20], refs. [21,22,23] focused on forest passability, refs. [24,25] describing problematics of microrelief obstacles and [26,27,28,29] consist of all environmental elements in one complex study. This approach was also partially included in the evaluation of data in the article. It is mainly in a form that considers which level of detail is required for data usability in the Czech military CCM model. Talhofer analysed the spatial database quality influence on the modelling of movement of vehicles in terrain in [30]. This general assessment of the quality of geospatial data can serve as a template for how to compare various datasets, including elevation models.

A similar approach as that found in the Czech Republic can be found among authors from other nations. Raster analysis is the main method used in the Polish CCM cartographic model [31,32]. This form was also assumed for the results of the article. The research introduced in [33] was specifically taken into consideration. It defines variances between elevation models with a focus on high-detail models (0.5 m, 1.0 m, 2.5 m, 5.0 m). It shows that the passable area of a smaller tested location is larger with higher detail terrain models. However, when studying large areas in a global aspect, where only lower-detail elevation models are available (5 m, 10 m, 30 m), the passable area is larger with lower-detail models. Processing and adjustment of these elevation models are necessary to obtain reliable CCM analysis results that correspond to the real terrain conditions. The other national models and studies are based on a vector line tactical level CCM analysis [34,35]. They do not consider the unavailability of detailed data; therefore, they do not fit within the focus of this article. Their focus was on automated navigation in the terrain. From the global point of view, this may be the only option before any global high-resolution elevation database is developed.

Elevation models are an important part of the complex analysis of the movement of vehicles in terrain; however, they cannot be considered in analyses without including other data and aspects. The data focus should be put on soil databases. The analysis of the Czech environment is introduced in [36]. The study considered the modelling of geographic and meteorological effects on vehicle movement, focusing on soil conditions and penetrometry; see also [37]. It consisted of an analysis of the characteristics of soil databases. The situation was similar to that of elevation models. National soil databases are relatively detailed enough to support digital terrain models with sufficient accuracy, but the global models are overly generalised and they cannot be used in detailed CCM analyses. Global data can be improved with methods for automatic refinement of soil databases [38] using an algorithm for refining soil data via comparison with relief models in a test grid with a cell of 100 m × 100 m.

Apart from soil databases, vector databases comprising features of the terrain should be taken into account for accurate passability analyses. One of the most used vector databases in NATO is the Multinational Geospatial Co-Production Program (MGCP) [39]. It is not completely a global database, though it has full coverage of the AOI. Since updating vector databases at the global scale can be substantially challenging, a new trend of using open-source data is rising. The OpenStreetMap is an example of a rapidly developing universal vector database [40]. NATO will most probably be using this kind of data more often in the near future. In general, contemporary vector databases do not consist of all features that are needed to enable accurate tactical-level CCM analyses.

Currently, there are no DEMs that would meet the requirement of high detail and global coverage at the same time in order to fulfil the needs of NATO and cooperating civilian organisations. The aim of this research was to verify the possibilities of usage of lower detail DEMs in practical tasks, predominantly in CCM analyses. The main asset of this study rested in a proposal for processing elevation models for their efficient evaluation in passability analyses done in larger areas.

2. Materials and Methods

2.1. Utilised Methods

The research in the article focused preferentially on elevation models to define the strengths and weaknesses of the models without any external influences. The comparison of the DEMs was based on preparing a suitable representative sample of data and determining a proper method of evaluation. Data from the Czech Republic was selected to test the accuracy of the slopes that were calculated from DEMs: Digital Terrain Model 3rd generation, 4th generation and 5th generation (DTM 3, DTM 4, DTM 5) and Digital Terrain Elevation Data 2 (DTED 2). The selection of the models was based on their occurrence in different parts of the globe. Whilst models with a resolution of up to 1 m predominate in Europe, areas of potential foreign operations are covered predominantly at most with elevation models with a maximum resolution of 30 m. Naturally, more accurate models might be locally available as well, e.g., laser scanning data. Nonetheless, from a military operation perspective, short notice ready-to-use terrain models with global coverage are the ultimate goal in the military environment. This excludes potential surface models that have other disadvantages for accurate CCM analyses. The challenge for elevation databases and their versatile usage is that any additional extensive processing of data prevents it from being operationally usable.

Although the key part of this research was based on the proper selection of elevation data, its usability in CCM analyses was guaranteed only with a suitable processing method. This article introduced its own approach to elevation data evaluation. This method was based on generating random points and general statistical comparisons of their slope values. The second part of the study was a comparison of the reliability of data, which was achieved by evaluating passability with a raster analysis for selected military vehicles. Five operationally representative areas were selected for the tests.

The analyses in the study were undertaken with the software ArcGIS 10.4.1 with its Toolbox functionalities; for points statistics, Extract values to points was used; and for areal statistics, Slope and Zonal statistics were used. Microsoft Excel 2019 was utilised for generating and evaluating statistical functions, e.g., mean slope and standard deviation. A major method used was the comparative method of map results or statistical results. The overview of the methodology and steps used in this research is in Figure 1.

2.2. Selected Digital Elevation Models

The digital terrain model (DTM) approximates the terrain surface or part of it by using a system of points in 2D space with specified height values. These points are predominantly organised into a regular grid or, in more detailed models, a triangular irregular network (TIN) [41]. Both these variants were included in the study. DEMs are the most important basis for evaluating the terrain relief. It varies according to spatial accuracy, where the height accuracy is lower than the position accuracy. In comparison with other types of data (e.g., vegetation, roads), the terrain does not change much in terms of extent and speed of changes, and thus, can be considered the most reliable. An overview of the elevation models considered in this study is in

Table 1. An overview of global elevation data and data from the territory of the Czech Republic and their altitude accuracy [4]. Unlike the rest of the models, DTM 5 is formed by an irregular grid of points (TIN).

Terrain Models	Resolution (m)	Vertical Accuracy (m)
Global Models
SRTM 1	30 m × 30 m	16–20 m
SRTM 3	90 m × 90 m	16–20 m
DTED 1	90 m × 90 m	3–20 m
DTED 2	30 m × 30 m	3–15 m
TREx	12 m × 12 m	2–10 m
Czech National Models
DTM 3	10 m × 10 m	1–7 m
DTM 4	5 m × 5 m	0.3–1 m
DTM 5	TIN—min. 1 m × 1 m	0.18–0.3 m
Local Models
LIDAR data	TIN—min. 1 m × 1 m	0.3 m

.

Global data is represented by Shuttle Radar Topography Mission (SRTM) radar data with an average detail of 30 m × 30 m [42]. Furthermore, some areas of the world are covered with DTED models with data levels 0, 1 and 2 [43], which are derived from other various methods of collecting altitude data (for example from SRTM). In 2015, a project of the new accurate global altitude model TREx was established. It is a joint project of NATO nations and non-member partners of the Alliance. It is based on the method of laser scanning. Because of the demanding character in terms of its time and capacity consumption, the project is particularly oriented towards specific locations of interest, primarily areas of deployment of troops of NATO nations.

This project was based on a similar principle as the creation of the global vector database MGCP [8]. TREx is a surface model and, therefore, also contains objects on the terrain (e.g., buildings, forests). Although a digital terrain model is not derived from the TREx data, a new initiative for the automatic generation of a 12 m model was scrutinised. TREx raw data in a combination with DTED 2 data can be used to interpolate a digital terrain model with a resolution of 12 m × 12 m, although the resulting accuracy is low. The main usage of such a model is currently for the MGCP topographic map concept. The contour lines are so far generated from DTED 2, which has limitations regarding use cases of terrain analyses.

The other option for detailed elevation data is local laser-scanning-based data. In areas of military operations, an accurate altitude model created from light detection and ranging (LIDAR) data is often the only detailed source [10]. It was the most accurate available source of altitude data in the International Security Assistance Force (ISAF) operation [11]. Its usability is versatile but obtaining this kind of data is time consuming. In the case of a crisis, the only option is ready-to-use lower-detail global elevation models.

From the elevation data perspective, a representative territory among NATO nations is the Czech Republic. Local sources in the Czech Republic are represented by various models: DTM 3 (10 m × 10 m), DTM 4, (5 m × 5 m) and DTM 5 (1 m × 1 m). Their creation and updating are based on a joint laser-scanning project of the Czech Office for Surveying, Mapping and Cadastre (ČÚZK) and the Office of Military Geography and Hydrometeorology (VGHMÚř). The most detailed model is the DTM 5. Since 2016, it has fully covered the territory of the Czech Republic and a local update of the model is subsequently underway [44].

DTM 5 was taken as a template and comparative database for analyses within this study. The detail of DTED 2 is not sufficient in comparison with DTM 4 and DTM 5; nonetheless, it is the only available detailed global terrain model for a large part of the foreign territory beyond the borders of NATO countries (in the AOI). DTED 1 is also the utilised global model; however, its basic 100 m × 100 m grid is completely insufficient for passability analyses. Large area CCM analyses use Global World Soil Database as a source of information about types of soils. Its scale is 1:500k and its detail is very insufficient. For this reason, small-scale CCM map outputs can use DTED 1 as an input elevation model for an overall picture.

Currently, a global elevation model created within the TREx project with a basic grid dimension of 12 m × 12 m is being developed [8]. This dimension is very close to DTM 3. Therefore, this model was selected for testing in order to determine its relationship with more detailed models and with DTED 2. TREx as a surface model is generally not appropriate for direct implementation in the passability analysis. On the other hand, it can be used without much loss of accuracy, for example, in the flat regions of Afghanistan or Mali.

2.3. Selection of Areas for Testing

To ensure the complexity of analyses, suitable areas representing the landscape of potential NATO foreign operations were selected. The selection represents possible types of reliefs across the AOI. Five different areas with a range of 3 km × 3 km are described in

Table 2. Characteristics of the 5 selected locations (3 km × 3 km) for testing the elevation models.

Area	Mean Slope	Mean Altitude (Above Sea Level)	Coordinates (WGS84)
Dobruška	6°	420 m	50.31° N, 16.22° E
Horní Cerekev	5°	645 m	49.32° N, 15.25° E
Znojmo	3°	211 m	48.83° N, 16.16° E
Kdyně	9°	576 m	49.39° N, 13.11° E
Dolní Morava	15°	670 m	50.11° N, 16.86° E

. The advantage was that each location represents a different type of terrain. This meant that areas with prevailing plains, hills and mountains were included. In terms of the morphometric division, the predominant plains and hills of selected locations represented the type of territory where military operations are conducted most often.

2.4. Selection of Vehicles for Testing

The impact of different elevation models must be tested not only in various territories but also with CCM parameters of different military vehicles. The aim of selecting suitable military vehicles was to create a balanced sample with a basic division into wheeled and tracked vehicles. The following vehicles were selected for data analysis:

• Off-road light vehicle—Land Rover DEFENDER 110 (LRD 110);
• Wheeled truck—TATRA T815 6 × 6 (T815);
• Infantry armoured tracked vehicle—Bojové vozidlo pěchoty 2 (BVP-2).

The selected vehicles are among the main currently used ones in the armed forces of the Czech Republic and partially also in NATO. Each of these vehicles has its own specifics, either in terms of technical parameters or driving characteristics; see the information given in [45]. Vehicle testing did not take place physically in the terrain and is based solely on theoretical research. This took the form of a site survey of local conditions and their possible impact on vehicles.

3. Results

The testing of the quality of digital terrain models was based on the utilisation of two methods. The first method was involved analysing the accuracy of derived slopes from the models. The second method was a comparison of the results of a CCM raster analysis for three selected vehicles.

3.1. Comparison of the Accuracy of Slopes Derived from Digital Terrain Models

The comparison of the accuracy of slopes was based on the usage of the tool Create random points in ArcMap. With this tool, 50,000 randomly placed points in the selected areas were created (five locations with 3 km × 3 km areas). A minimum distance between points of three meters was set (points were 3 m or more apart) to ensure a more even distribution. Values of slopes were assigned from the prepared slope maps of individual data sets to the generated points using the tool Extract values to points. The points were then divided into classes with a step of inclination of 5° and comparative statistics were calculated for each class. Mean values and standard deviations were selected as test statistics. The resulting set of values of the most important statistics is shown in

Table 3. A comparison of slopes in random points generated from elevation models and their statistics.

Slope Classes		Mean Slope (°)				Mean Slope Difference from DTM 5 (°)
Quantity	Slope	DTM 5	DTM 4	DTM 3	DTED 2	DTM 4	DTM 3	DTED 2
10,134	0–5°	3.14	4.09	5.21	5.17	0.95	2.07	2.02
12,389	5–10°	7.17	7.28	7.54	6.82	0.11	0.37	−0.35
6946	10–15°	12.37	12.07	11.74	9.94	−0.30	−0.62	−2.43
5933	15–20°	17.45	16.59	15.42	12.59	−0.87	−2.03	−4.86
5696	20–25°	22.49	21.41	19.11	15.85	−1.08	−3.39	−6.64
4987	25–30°	27.38	26.05	22.37	18.09	−1.33	−5.01	−9.29
2679	30–35°	32.10	29.81	23.37	18.33	−2.29	−8.72	−13.77
740	35–40°	36.95	31.24	22.80	15.75	−5.72	−14.15	−21.20
191	40–45°	42.00	33.36	22.64	14.23	−8.64	−19.36	−27.77
96	45–50°	47.41	37.25	26.36	10.48	−10.16	−21.05	−36.93
54	50–55°	52.53	40.48	28.51	8.73	−12.05	−24.02	−43.80
55	55–60°	57.48	46.31	30.92	6.90	−11.17	−26.56	−50.58
41	60–65°	62.82	48.48	26.30	5.42	−14.34	−36.51	−57.39
35	65–70°	67.43	51.05	31.97	5.80	−16.38	−35.46	−61.63
50,000	Total	14.27	13.77	12.64	10.54	−0.50	−1.63	−3.73

. The average slope was calculated as the mean value of the slope within one class. The average difference was given by subtracting slopes of a selected data set from the DTM 5. The standard deviation of the difference from the DTM 5 indicated the extent to which slope values were spread within one class.

Figure 2, Figure 3 and Figure 4 show histograms of all values of the standard deviations of slopes without division into classes, unlike in Table 3, where the values were divided into slope classes. The goal was to verify the trends in the accuracy of the data with two different methods. The histograms showed a deviation of the average value from the zero value for all elevation models, and at the same time, the range of standard deviations was compared. The lower the detail of the model, the higher the variance.

The comparison of the accuracy of slopes derived from elevation models produced the following results.

The number of points in each class did not represent an even distribution but fit more suitably to real conditions. Slopes up to 15° were 59% of the points and slopes above 30° were 8% of the points. The key threshold slopes for the assessment of passability, from 15° to 30°, represented 33% of the total number of determined slopes. This was sufficient due to the higher number of testing points. Graphs evaluating the characteristics of elevation models with individual classes of 5° inclination are shown in Figure 5, Figure 6 and Figure 7.

The assessment of the specific slope values and digital terrain models resulting from the analysis of Figure 2, Figure 3, Figure 4, Figure 5, Figure 6 and Figure 7 is given below.

Evaluation of specific slope values (slope classes):

• Low-detail elevation models did not differ from high-detail ones for lower slope values (slope values were very similar). The basic threshold was up to 10° for lower-detailed models (10–30 m) and up to 30° for medium-detailed models (5 m).
• Slopes up to 5°—The average slope value was higher in all tested models than in DTM 5. More detailed terrain models (DTM 5) had a more gradual slope in the lowest class. This inversed effect was specifically found in the flat areas only.
• Slopes 5–10°—The average values of slopes were very close for all models and the standard deviations were not very high either. It is possible to use less-detailed models in terrain analyses for slopes up to 10°.
• Slopes 10–30°—Almost a linear increase in slope could be observed depending on the resolution of a model. The average deviations with DTM 5 also had a linear increase. The more detailed the model, the greater the proportion of higher slopes. This slope class was still relatively reliable in all models (except DTED 2).
• Slopes above 30°—The comparative graphs deviated from the trend of the curves due to a smaller number of values, especially above 40° (see Figure 4).

Evaluation of digital terrain models (a comparison with DTM 5—1 m × 1 m):

• The less detailed the model, the bigger the deviation of the average slope and the bigger the variance in the standard deviation.
• The 5 m × 5 m models (DTM 4)—Achieved small deviations below 1° for slopes up to 20°. For bigger slopes, especially above 40°, the slope was almost less than one-third lower than in DTM 5. DTM 4 was suitable for slopes up to 20°, the use is not recommended for slopes above 40°.
• The 10 m × 10 m models (DTM 3)—Achieved small deviations below 1° for slopes up to 15°, and at 30°, the slope exceeded the limit of one-third of the slope difference compared to DTM 5. DTM 3 was suitable for slopes up to 15°; for slopes above 30°, it is not recommended.
• The 30 m × 30 m models (DTED 2)—They differed from DTM 5 by 1° already after 10° of slope inclination; from this value, the difference in inclination compared to DTM 5 was lower by more than one-third. Areas with inclinations above 50° in DTM 5 had less than 10° in DTED 2. The detail of the DTED 2 network of 30 m × 30 m points did not allow for identifying a more fragmented terrain relief. DTED 2 could only be used to determine slopes up to 10°.

3.2. Comparison of the Accuracy of Digital Terrain Models Using Raster Passability Analysis

Properties of digital terrain models can also be studied from the view of their use in the decision-making processes of commanders and staff. Data properties are evaluated from the area difference in the calculated passability. A difference was given by comparing selected models with the most detailed data (e.g., the difference between DTM 3 and DTM 5). Each pixel was assigned one of three passability values (GO, SLOW GO, NO GO) according to the parameters calculated from the traction curves of selected vehicles. For the calculation of the parameters, see [16]. This was based on the findings of the evaluation of dynamics of vehicle movement in the terrain of [46,47]. A total area of all pixels was then converted to percentages of passable, hardly passable, and impassable terrain. Map comparisons of the results of CCM analysis for the T815 vehicle and all tested terrain models are shown in Figure 8. A summary of passable area deviations between lower detail models and DTM 5 is displayed in

Table 4. Average deviations of the passable area calculated from elevation models (DTM 4, DTM 3, DTED 2) in comparison with DTM 5 with different parameters (vehicle type, humidity); results are mean values for five 3 km × 3 km areas, max–maximum available slope for the respective conditions, 1% deviation of the area means 9 hectares out of 900 hectares of one area and area deviations of elevation models with DTM 5 are always positive (larger passable areas).

CCM Parameters	Average Deviations of Passable Area in Comparison with DTM 5 (%)
	DTM 4	DTM 3	DTED 2
T815 GO moist (max slope = 6.25°)	1.32	2.71	4.39
T815 GO semi-moist (max slope = 12.08°)	1.31	2.54	3.32
LRD GO dry (max slope = 16.17°)	1.11	2.32	3.45
LRD GO + SLOW GO semi-moist (max slope = 21.82°)	0.77	1.83	2.37
BVP2 GO + SLOW GO semi-moist (max slope = 28.81°)	0.33	0.98	1.08

.

The raster analysis of passability produce the following results:

1.

Deviations in passable area (GO):

• The more detailed the model of terrain, the smaller the passable area.
• The passable area of 851 ha (out of a total of 900 ha of one 3 km × 3 km area) in DTM 5 represented

  ○

  860 ha in DTM 4 (1% larger passable area);

  ○

  870 ha in DTM 3 (2% larger passable area);

  ○

  880 ha in DTED 2 (3% larger passable area).

• These ratios may significantly vary in different types of landscape (mountains) or surface conditions (impassable soils) but have the same trend.

2.

Deviations in the hardly passable area (SLOW GO):

• The position and structure of hardly passable area remained unchanged in all models.

3.

Deviations in the impassable area (NO GO):

• The total area of impassable territory increased with the detail of a used model.
• The impassable area in DTM 5 (15 ha) represented

  ○

  75% of the area in DTM 4 (12 ha);

  ○

  33% of the area in DTM 3 (5 ha);

  ○

  20% of the area in DTED 2 (3 ha).

4.

Influence of other factors on the deviation of the passable area:

• The better the passability conditions, the smaller the area deviation when using less accurate relief models (see Table 4); this applied to the following conditions:

  ○

  lower soil moisture;

  ○

  more suitable soil types (clayey-sandy);

  ○

  more powerful vehicles (suitable for cross-country movement).

5.

Evaluation of digital terrain models:

• The 1 m × 1 m models (DTM 5):

  ○

  The accurate model was suitable for detailed CCM analysis.

• The 5 m × 5 m models (DTM 4):

  ○

  The deviation of the area of the passable terrain was small;

  ○

  DTM 5 was more suitable for fragmented terrain in detailed CCM analyses.

• The 10 m × 10 m models (DTM 3):

  ○

  The impassable area was not usually displayed in flat territories in these models (steeper slopes with shorter lengths);

  ○

  The area of the impassable territory was close to DTM 5 in mountainous areas with long slopes;

  ○

  The model could be used for less detailed CCM analyses.

• The 30 m × 30 m models (DTED 2):

  ○

  The impassable area was not usually displayed in flat areas in these models;

  ○

  The impassable area was close to DTM 5 in mountainous areas with long slopes;

  ○

  Reliable results of CCM analysis could not be achieved with DTED 2.

3.3. Efficiency Improvements of the Elevation Models

Before any wider global data improvements were introduced, the only option for more accurate CCM analyses was to implement a methodology regarding how to efficiently use low-detail elevation models. The usability of these models can be achieved by adjusting the maximum slope limits for a passable terrain. A hardly passable terrain can be considered impassable in low-detail models. To obtain similar results for the area of passability as high-detail elevation models, adjustments in

Table 5. Adjustments of maximum slope value reachable by a vehicle for passable (GO) and hardly passable terrain (SLOW GO) that were needed to achieve more accurate results for CCM analyses. Modifications calculated for DTM 4 (5 m × 5 m), DTM 3 (10 m × 10 m) and DTED 2 (30 m × 30 m).

GO or SLOW GO Slope Value (°)	Reduced Slope Values (°)
	DTM 4	DTM 3	DTED 2
10	9.9	9.6	9.3
15	14.5	13.4	12.8
20	19.0	17.3	16.4
25	23.5	21.2	19.8
30	27.8	25.1	23.6
35	32.0	29.0	27.0
40	36.0	32.8	30.5
45	40.0	36.5	34.0

had to be performed. These adjustments need to be applied to elevation models based on the traction parameters of a vehicle (maximum reachable slope), which were calculated from traction curves or the DMA (Defence Mapping Agency) model; for details, see [16]. The values in Table 5 were calculated from the mean slope values of elevation models for each slope class (see Table 3). Afterwards, these values had to be modified according to the results of CCM map analyses, displayed in Table 4 and

Table 6. The comparison of the extent of passable, hardly passable, and impassable areas with different elevation models (DTM 5, DTM 4, DTM 3, DTED 2) without slope corrections and with slope corrections. Location: Kdyně, 6 km × 6 km, moist conditions, GO max slope = 10°, SLOW GO max slope = 20°. The main improvements in the results are displayed as bold values.

		DTM 5 Area (%)	DTM 4 Area (%)	DTM 3 Area (%)	DTED 2 Area (%)
No slope corrections	GO	72.01	73.17	74.44	73.04
	SLOW GO	24.71	24.70	24.40	26.07
	NO GO	3.28	2.13	1.15	0.90
With slope corrections	GO	72.01	72.65	72.06	69.19
	SLOW GO	24.71	24.35	24.77	26.59
	NO GO	3.28	3.00	3.17	4.23

and as map results in Figure 8, Figure 9 and Figure 10.

If a calculated maximum reachable slope for a hardly passable terrain (SLOW GO) for a vehicle was, for example, 25°, then by using the 10 m × 10 m elevation model, the maximum reachable slope was only 21.2° and the CCM analysis map should be adjusted. DTM 5 served as the most accurate model database to calculate the slope deviations of other elevation models from the real terrain. DTM 5 is a very accurate model (1 m × 1 m) but still deviates from the real terrain. Given the trends of lower-detail models displayed in Table 5, the estimated maximum reachable slope value reductions for DTM 5 would be 0.2° for 20° slopes, 0.5° for 30° slopes and 1.5° for 45° slopes. The accuracy of slope reductions was tested for a different extent and conditions. The tests were performed in the form of CCM map analysis in a 3 km × 3 km area in Dolní Morava (arid and moist conditions), a 6 km × 6 km area in Kdyně (moist conditions) and five combined selected areas of 5 km × 5 km (moist conditions). The main comparative characteristics were variations of passable/impassable area between corrected models and models not corrected. The other characteristic was a comparison of the position of a hardly passable and impassable terrain. The results with the calculated slope reductions (Table 5) were more accurate and were markedly closer in position to the most detailed elevation model, namely, DTM 5, than results without slope reductions (see Table 6,

Table 7. The comparison of the extent of passable, hardly passable, and impassable areas with different elevation models without slope corrections and with slope corrections. Location: Dolní Morava, 3 km × 3 km, moist conditions, GO max slope = 10°, SLOW GO max slope = 20°. The main improvements in the results are displayed as bold values.

		DTM 5 Area (%)	DTM 4 Area (%)	DTM 3 Area (%)	DTED 2 Area (%)
No slope corrections	GO	43.67	44.25	43.77	43.72
	SLOW GO	36.36	37.37	40.79	42.53
	NO GO	19.97	18.38	15.44	13.76
With slope corrections	GO	43.67	43.61	41.12	38.89
	SLOW GO	36.36	35.49	35.16	37.03
	NO GO	19.97	20.90	23.72	24.08

,

Table 8. The comparison of the extent of passable, hardly passable, and impassable areas with different elevation models without slope corrections and with slope corrections. Location: Dolní Morava, 3 km × 3 km, arid conditions, GO max slope = 20°, SLOW GO max slope = 30°. The main improvements in the results are displayed as bold values.

		DTM 5 Area (%)	DTM 4 Area (%)	DTM 3 Area (%)	DTED 2 Area (%)
No slope corrections	GO	80.03	81.62	84.56	86.24
	SLOW GO	16.12	15.29	14.38	12.89
	NO GO	3.85	3.09	1.06	0.87
With slope corrections	GO	80.03	79.10	76.28	75.92
	SLOW GO	16.12	15.50	18.70	17.81
	NO GO	3.85	5.40	5.02	6.27

and

Table 9. The comparison of the extent of passable, hardly passable, and impassable areas with different elevation models without slope corrections and with slope corrections. Combined 5 areas, 3 km × 3 km, moist conditions, GO max slope = 10°, SLOW GO max slope = 20°. The main improvements in the results are displayed as bold values.

		DTM 5 Area (%)	DTM 4 Area (%)	DTM 3 Area (%)	DTED 2 Area (%)
No slope corrections	GO	77.34	78.55	79.28	80.54
	SLOW GO	16.64	16.35	16.72	16.27
	NO GO	6.02	5.09	4.00	3.19
With slope corrections	GO	77.34	78.24	77.91	76.82
	SLOW GO	16.64	15.86	15.61	15.55
	NO GO	6.02	5.90	6.48	7.63

).

The accuracy of the selected method of slope corrections could be observed from an extent of impassable terrain calculated from the selected elevation models. The impassable area was significantly more accurate with implemented slope corrections. More important is the fact that these slope adjustments allowed for distinguishing hardly passable and impassable terrain with noticeably higher precision.

On the right side of the map (Figure 9) is a valley that represents an impassable obstacle. Without corrections, the valley was evaluated as hardly passable when using DTM 3 and DTED 2. With implemented corrections, the valley was correctly displayed as impassable also in these lower detail models. Furthermore, shapes of other major terrain features representing hardly passable terrain were displayed more accurately (e.g., a small valley in the lower-left corner of the map). The detail of DTM 5 allowed for displaying a fragmented terrain (e.g., top left corner).

These locations are usually joined as one area in lower detail models, especially with details that were 10 m × 10 m and above. This was the reason for the higher ratio of impassable terrain with corrected slopes. Due to the fragmented terrain, the mentioned location was, in the end, completely impassable. This meant that the lower-detailed models were displaying the correct situation. The results in Figure 9 were confirmed in Figure 10 for a different location.

The extent of the impassable area is given with the corrections adjusted to a more real coverage and position. It shows that the determination of impassable and hardly passable areas when using a lower-detail elevation model can be very variable. The method of slope correction was the right way to improve a current state, which was verified by better coverage results in comparison with DTM 5.

4. Discussion

The results of analyses of the digital terrain models DTM 5, DTM 4, DTM 3 and DTED 2 showed different qualities based on the resolution of the network of points. Each elevation model had its possible usage in practise: from the most detailed 1 m × 1 m model, namely, DTM 5, with a universal usage to the low detail model 30 m × 30 m with limited use cases, mostly for small-scale overview maps. Currently, the lower detail elevation models are not used efficiently. Efficient use can be achieved by adjusting the accuracy to high-detail elevation models. These adjustments would increase the usability of all elevation models. Before implementing any modifications to digital terrain models, other types of data need to be considered as well.

Individual parts of the landscape, i.e., terrain relief and soil cover, are represented by a different quality of data. Terrain relief is covered with detailed and accurate elevation models in the territory of European states. They are suitable for planning military operations down to the tactical level. Soil and surface conditions data is less reliable for detailed CCM analyses. This data has little detail or inaccurate determination of soil type boundaries. It is necessary to consider the shortcomings of the data used to properly interpret the results of passability analyses.

4.1. Evaluation of Digital Terrain Models

Digital terrain models are not globally available in adequate detail (minimum 10 m × 10 m) and area coverage to support any possible location for a foreign operation at short notice. Currently, it is not possible to perform accurate and detailed CCM analyses in missions abroad without the use of mobile mapping devices (satellite systems, drones), which use various types of sensors and technologies (photogrammetry, laser scanning, radar data, etc.). The biggest disadvantage of this mobile sensors approach is the time delay (months) in collecting and processing data to cover the entire area of a recently established operation. The solution lies in creating a universal database with complete coverage and sufficient detail. This coverage should include the entire NATO area of interest (AOI), which reflects NATO nations’ territory and the NATO area of responsibility (AOR) covering adjacent territories in Europe, the western part of Asia and the northern part of Africa. NATO does not have sufficient resources to create or purchase such data. Nevertheless, it could utilise the capabilities of national geographic services, the Multinational Geospatial Support Group (MN GSG), the NATO Communication and Information Agency (NCIA) or a type of direct support of foreign operations, which was done using a Resolute Support Reach-Back Afghanistan Cell (RS RAC). After some time, a military operation is supplied with detailed data, which corresponds to the production in the territories of NATO nations [48]. Nevertheless, sufficient data that would cover immediate needs in the case of a crisis operation are not available, either in the required extent or quality.

Models with a basic resolution of at least 1 m × 1 m (e.g., DTM 5) are the most detailed terrain models of NATO member countries. However, models with a resolution of 5 m × 5 m (e.g., DTM 4) are also sufficient in their detail to achieve accurate results of passability analyses. This type of model can be less demanding in terms of data flow for calculations via a network. DTM 3 corresponds in detail with the new worldwide terrain surface model TREx. At this level of detail (10 m × 10 m), the model cannot be used for detailed CCM analyses. Nonetheless, data can be used for less fragmented terrain without vegetation cover. This is the case for some foreign operations, such as in Mali. These areas are already mostly covered by the TREx model. DTED 2 is a low-detail elevation model that allows for only general terrain analyses. The optimal terrain model with a balanced ratio of detail and accuracy of the results of passability analyses when considering the complexity of its acquisition is a model with a resolution of 5 m × 5 m. The goal of geographical development within NATO should be to ensure comprehensive coverage with enhanced data in the NATO AOI.

4.2. Usability of Soil Databases in CCM Analyses

Soil databases with sufficient quality are not available in the areas of foreign military operations. Despite a lower accuracy in comparison with elevation models, soil databases can be used for terrain analyses. However, to achieve the required detail of data, a more accurate mapping of soil type boundaries would be necessary. For example, in the territory of the Czech Republic, the Digital Soil Map 1:50,000 need to be updated [20]. Since DTM 4 was identified as a sufficiently accurate terrain model, the optimal density of measurement points for soil characteristics is 5 m × 5 m at the boundaries between soil types. Apart from geometric accuracy improvement, other information needs to be added to the assessment of soil types, such as an actual soil moisture level and an assessment of the quality of the surface itself. Vector databases that contain surface roughness and vegetation coverage parameters with sufficient detail and area coverage do not exist. Accurate soil databases would not be much use without this additional surface information.

From the global perspective, universal soil data is represented by the Harmonized World Soil Database (HWSD), which combines various national soil databases. Its detail is very low and represents a scale of 1:500,000. It can only be used for lower-accuracy estimates of passability analysis for the operational and up to the strategic level of military planning. A detailed study of improvements of the HWSD and its usability in national systems was given in [49]. One of the possible options to replace low-detailed surface data is the Destination Earth initiative. It is a project of the European Commission with the aim of collecting and evaluating satellite images. One of the important use cases of this project for the CCM analysis domain is the evaluation of surface temperatures to determine soil and surface types. Even though this method is not accurate enough, it can improve data availability in territories of potential conflicts across the world [50].

A study [51] showed that if the terrain has arid or semi-moist conditions, then the influence of soil types in the CCM can be disregarded. The only significant exception was the alluvial soil type, which occurs in some humid valleys. Except for arid conditions, these valleys are impassable for most types of vehicles (track or wheeled). The other exception is a sandy surface, which substantially limits the movement of wheeled vehicles. A vast majority of soil types are different from the alluvial and sandy types. Therefore, the focus should be put on distinguishing correct humidity conditions of the surface and properly processing a utilised elevation model.

4.3. Recommendations for Improvements of Elevation Models

The introduced method of slope reduction for the usability of low detail elevation models in CCM analyses was affirmed as correct. All results showed an improvement in the accuracy of passable areas, both in extent and position (as shown by the Figure 10 map comparisons and Table 6, Table 7, Table 8 and Table 9 numbers in bold). However, the size and number of testing areas were limited. Verifying this concept at a global scale would require extensive data testing in large areas across Europe, Asia and Africa for various elevation data sources. The key element is also a focus on specific restricted areas for passability. These are hilly and mountain areas, where a key slope limit of approximately 30° occurs more often than in other types of terrain. Additional testing sites should make the slope reduction values presented in Table 5 more accurate. Further slope adjustments to the values in Table 5 would probably only be minor.

Global detailed data readiness is something that NATO currently lacks. The usability of digital terrain models is not limited to only high-detail models. NATO can use low-resolution models in foreign operations in a time frame before any detailed elevation data is procured, e.g., LIDAR data. Low-detail elevation models can fulfil this aim via the additional processing of data introduced in this chapter using the method of slope reduction. Accurate CCM analyses cannot be achieved without including microrelief forms in the analysis. DTM 5 (1 m × 1 m) consists of all microrelief objects, for example, embankments. The other elevation models mostly do not consist of these objects. Remedy can be achieved with vector data that consists of roads and other line or point objects that are important for passability. These objects could be incorporated into analyses with low-detail elevation models as impassable obstacles. The current ongoing project of the global vector database MGCP can serve as a source of information for objects on the terrain. Nowadays, expectations for improvements in global elevation models are focused on the TREx project. TREx is a surface model; therefore, its usage for accurate CCM analyses is limited.

Nevertheless, new methods for processing TREx data into the form of a digital terrain model are currently in development. Although this might make a new global digital elevation model with a resolution of 12 m × 12 m available, the accuracy of altitude of areas formerly consisting of forest and built-up areas is low. The reason for this is that data from these locations are interpolated from lower detail models, e.g., DTED 2. TREx is also available in its raw unprocessed form as a cloud of points. Using this data can allow for rapid reaction analyses for any crisis region across the world. The disadvantage is in more complicated processing, mostly because of the overly large data volume.

So far, no global elevation model has met all requirements needed for accurate CCM analyses. Global coverage with ready-to-use data is the highest priority. The second priority is the detail of an elevation model, which ideally should be 5 m × 5 m. The surface model, such as TREx, is not a sufficient replacement for a regular digital terrain model. For the moment, the only possible solution for this would be the use of automated systems fitted on military vehicles that evaluate passability parameters in real time. Even though the capability of automated navigation of military vehicles is not widespread in NATO, the systems will become the future of CCM modelling.

To compare the influence of the use of different elevation models on the determination of terrain passability, different statistical methods can be used. The use of the parametric ANOVA method would be possible under the assumption of a normal distribution of data. However, this distribution does not occur in the event of extreme changes in the terrain, e.g., changes in adhesion due to precipitation. For non-parametric analysis of the variance of GO, SLOW GO and NO GO areas, it is possible to use, for example, the Kruskal–Wallis test or multiple comparisons with the adjustment of the significance level using the Bonferonni method. However, the validity of these methods must be verified via physical testing of off-road vehicles in different types of terrain.

The main conclusion of the study is that under the current conditions, the usability of the selected elevation databases for accurate and detailed comprehensive CCM analyses is limited. A future focus should be to find a clear solution that allows for utilising existing databases efficiently; see also [52,53,54,55]. The essential part is also adopting selective measures for data collection for CCM analyses in real time using means of remote detection of passability parameters. That includes monitoring necessary methods and technologies of an automated vehicle driving in space and making active steps to implement them into practise.

5. Conclusions

This research studied the possibilities for raster analyses of elevation models and other options for their evaluation and effective usage. One of the most important results of this study was the introduction of the method of slope reduction for low-detail elevation models. Furthermore, the goal was to determine the strengths and weaknesses of selected data for use in cross-country mobility analyses, supplemented by recommendations on how to use these databases efficiently. The selection of databases was determined by the focus of data development projects of NATO and current scientific research projects of the Ministry of Defence of the Czech Republic. The presented methodology of data assessment for the purposes of military geographical analysis of the terrain can be further used in military practice, e.g., in foreign missions where high-detail data may not be present. Key findings of the elevation models analysis show the potential of their usage in practise. DTM 5, which is a model with a minimum resolution of a grid of points of 1 m × 1 m, is the most detailed model in terms of the evaluation of slopes and passability. It can be replaced by using DTM 4 (5 m × 5 m) in terrain analyses, for example, the slope reduction for 30° slopes is 2.2°. The elevation model with lower detail DTM 3 (10 m × 10 m) had a 4.9° slope reduction for 30° slopes and DTED 2 (30 m × 30 m) had a 6.4° reduction. These models can be used for general analyses of less fragmented terrain.

The selected areas, where analyses were performed, were a balanced sample of the landscape. It allowed for the assessment of the most important terrain parameters that have a fundamental influence on the conduct of military operations. Nevertheless, it is necessary to perform a complete validation of the slope reduction method from other various representative regions across NATO AOR (territory of NATO nations) and NATO AOI (territory beyond NATO nations). A larger number of locations would enable a more accurate determination of the reliability of databases when used in the cross-country mobility analysis. The challenge NATO has is a limited global short-notice availability of detailed digital terrain models. Without these models, no detailed and accurate CCM analysis can be performed. The only option is to process available elevation models into a form that allows for acceptable results, such as the slope reduction method.