Home Articles Considerations about map-updating and images resolution

Considerations about map-updating and images resolution


Considerations about map-updating and images resolution

Serge Le Blanc
107 Chemin de Pechbusque
31 400 Toulouse

Methodology

Geographic objects of a topographic map
As it has been written in the introduction, there are three types of maps produced from satellite images: traditional line maps, new imagemaps and thematic land-cover / land-use maps. With or without image background, all these products shows updated geographic features like villages, roads or forests. Table 1 reviews all the geographic objects present on topographical maps at the most common scale.

Table 1 – geographic objects of a topographic map

Type
1/25 000 – 1/50 000
1/100 000 -1/200 000

Communications
major roads and motorways
secondary roads
tracks
footpaths
railways
Landing areas
major roads and motorways
secondary roads
major tracksrailways
Landing areas

Equipments
Power lines
Tunnels
Bridges
Sport fields
Power lines
Major tunnels
Major bridges

Settlements
High-density urban areas
Low-density urban areas
Villages
Isolated buildings
High-density urban areas
Low-density urban areas
Villages
Major churches

Relief
Contours at 5 to 20 m vertical interval
Height spots
Contours at 20 to 100 m vertical interval
Height spots

Hydrology
Rivers and channels
Streams
Lakes and dams
Springs
Wells
Rivers and channels
Lakes and dams

Vegetation / land cover
Cultivated area
Orchard, plantation
Grassland
Bush
Several types of forest
Rocky areas
Cultivated area
Bush
Forest
Rocky areas

Artificial limits
Administrative boundaries
Cadastral boundaries
Administrative boundaries

Writing
Toponymy
Toponymy

Tourist information
Sometime present
Frequently present

The main difference between each scale map is the density of details shown on the map. The purpose of this physical limitation is to leave the final paper product enough synthetic to be understandable at a glance.

Effective Resolution of images
This aspect is very important for the interpretability of the images.
The sampling interval is the most frequent parameter used to characterise remote sensing images: in a SPOT panchromatic image for example, one CCD element records the reflected light each 10m along the scan line (1,3 micron on the CCD) and a line each 1,504 nano-second along the orbit of the spacecraft. This parameter is usually given as the ‘pixel size’ (10x10m for SPOT P), and it is often misused as an indication of the resolution.

When a photographic film is scanned, the sampling interval is determined by the increments of the sensing element along a line of scan, and by the rotation of the drum between successive lines.

Unfortunately, this doesn’t mean that the signal recorded by the CCD element integrates only the reflected light of the objects located inside the pixel size area on the ground: the signal is also influenced by objects located outside this area. The area ‘seen’ by an individual sensing element is called the instantaneous field of view (IFOV). The objects located in the centre of the area have a stronger influence than those located near the edges. The reduction of influence on the signal of the objects according to their distance from the centre is called the modulation and is described by the MTF (modulation transfer function).

So the concept of “effective ground resolution” is more realistic: it indicates the diameter of a circular area where the modulation of signal is always higher than 50%. The corresponding area is called the effective instantaneous field of view (EIFOV). We have to note that if a digital image has an effective ground resolution much larger than the sampling interval, it looks blurred.

The MTF of a sensing device depends on many factors: the optical quality of the lens and the electronic characteristics (or film properties) are more or less stable in time, for a given device, but for remote sensing systems, the optical quality of the atmosphere is usually the most limiting factor. Two images of the same area, acquired with the same device at different dates can have very different MTF, and thereafter, very different potential for interpretation. Haze is an obvious explanation for images having a poor MTF, but urban pollution and forest fire smokes are not negligible, even when the latter are located very far away from the site (high altitude layers of smoke are frequent in tropical areas).

Measuring the effective resolution on an image is not an easy task if calibrated targets are not available, and it should be done separately on each image. Differences of MTF in different parts of an image can even be observed. The most commonly used method to evaluate MTF is based on the comparison of recorded standardised features (airfields, buildings, etc.) with reference images, having a known MTF (usually generated by a controlled degradation of a higher resolution image).

Based on the compilation of multiple scientific references, the effective resolution of SPOT panchromatic images is often put to between 15 and 25m and between 30 and 50m for multispectral mode, for Landsat 5 the effective resolution is in between 45 and 75m.

For aerial photographs, the resolution is usually expressed in line pairs/mm. Common values are 40 lp/mm on film, and 15 lp/mm for paper copies, and we can say that the resulting resolution of optical aerial film systems, including film and camera, is actually anticipated to be around 20 microns at the film scale. Like with satellite imagery, the principal cause of the degradation of the MTF is related to atmospheric conditions. The MTF of the film scanner is usually adapted to the sampling interval, in order to produce sharp images, that is to say with 20 microns resolution for the image the good scanning interval is 30 microns. For good 1:80 000 photos, this can be translated to an effective resolution (EIFOV) of approx. 2 to 2.4m, for 1:30 000 scale photos this gives a resolution a little bit better than 1m (90cm)!

It is worth noticing that the processing of the images to produce orthoimages damages to some extent the MTF, because of resampling.


Considerations about map-updating and images resolution

Basic principles of image analysis for mapping
Most of the work of map production and map updating is based on the recognition of objects or features present on the ground. After visual photo-interpretation of the images, fieldwork (‘completion’) is always necessary to fill the gaps in the interpretation and to collect information that is not available in the images (function of the buildings, village names, etc.).

Field completion is always necessary, but the amount of work needed to achieve it will vary to a very large extent according to the performance of the interpretation of the images: all undetected (or mis-interpreted) features have to be checked on the field.

Choosing one source of image can thus be seen as defining the best balance between the cost of acquiring better images and the reduction of cost if less field work is necessary. We based our analysis of image interpretability on concepts developed by NATO.

The interpretation of an object is organised in 4 hierarchical levels:

1/ Detection is the discovery of an object without recognition.
ex: there is a white linear feature in this corner of the image.

2/ Recognition is the ability to fix the identity of an object within a group type.
ex: this white linear feature is a road.

3/ Identification is the ability to place the identity of an object as a precise type.
ex: this road is a dual carriageway road.

4/ Technical analysis is the ability to describe precisely the attributes of the object.
ex: a low wall separates the two carriageways.

According to NATO, the resolution requested to achieve the interpretation of different objects is as indicated in table 2 (only objects interesting ‘civilian’ mapping are indicated, with the corresponding minimal scale of aerial photography):

This minimum scale is based on images having a good effective resolution, but the potential of interpretation depends also to a large extent on the contrast of the objects on the background.

  • Detection corresponds to the ability to count the number of objects.
  • Recognition is possible when the object is no more a dark spot but clearly a character
  • Identification is possible when one can read the character
  • Technical analysis corresponds to the
    analysis of the font used (italics, with or without serif, etc.)

As a rule of thumb, and for non-linear objects like tracks or footpaths, one can consider that well-contrasted objects of complex shape can be more or less detected only when their size is larger than 2 pixels, and identification possible for objects larger than 4 pixels. This is only true with a very good MTF (EIFOV=1.2 sampling interval). With an average image quality (EIFOV=1.5 S.I.), only objects larger than 5 pixels can be detected, and objects larger than 6 to 7 pixels can be identified in a reliable way with poor quality imagery.

Processing
To be used for mapping purpose all the satellite images must be geocoded and orthorectified, it means that digital processing put them in cartographic geometry. This is done using digital terrain model (DTM) and ground control points; both of them can be also derived from satellites systems.

The control points are generally provided through a ground survey conducted using satellite Global Positioning (GPS) techniques. In a very near future the use of GPS for precise positioning of the spacecraft on-orbit and stellar Attitude Recording System (ARS) would permit topographic mapping with a ground control point spacing of 500 to 1000 km (rather than 30 to 50km needed with systems like SPOT) so making considerable savings.

The DTM, used for the rectification, can be derived from existing contour lines or from satellite images. Indeed, thanks to their adjustable viewing angles, some optical satellites like SPOT or IRS can acquire stereographic pairs of images from which DTM can be calculated. Radar interferometry is another way to obtain DTM from images like ERS radar satellite for example. Contour lines and spot heights can be derived from those DTM if they are accurate enough.

Modern cartography use, today, two different work processes. The first one, “classical”, is based on photogrammetric tools and uses stereoplotters, fully digital in the best case, but more generally analytical plotters using always films. This solution needs very high educated and specialised operators, and also needs costly investment in hardware and software for the long term.

The second way of producing maps is also derived from the basic principals of photogrammetry, but it uses cheaper hardware and software based on powerful PCs, which have been recently introduced on the market. These new production chains are based on the use of orthoimages for map production and map updating.


Considerations about map-updating and images resolution

Relations between map scale and image resolution
To take full benefit of images at 1, 2.5, 5 and 10m resolution, they have to be displayed at least at full resolution, that is to say respectively at 1:2 000, 1:5 000, 1:10 000 and 1:20 000 approximately with a standard screen of 800×600 pixels.

The map as then of course to be displayed at the same scale in the updating process.

The following figures (5 to 10) are trying to illustrate the complexity of map production and map updating having regard to what is pertinent to be mapped, which detail if it is mapped is still readable compared to others neighbouring details and what is the good image resolution for 1:50 000 mapping.

One can say that the greater is the scale of the images (i.e.: the greater is the resolution) the best it is for visual photo interpretation. If this is especially true for production of new maps it is not so evident for map updating, where the image scale must be closely adjust to the final map scale and to the working scale.

In figure 5, above, four ortho-images, at resolutions of 1, 2.5, 5 and 10 metres and displayed at full resolution, are presented together with a 1:50 000 scale topographic map (centre) and the corresponding enlargements.

If for updating purpose of the 1:50 000 scale map sheet if it is interesting to work at 1:25 000 or 1:20 000 or even at 1:10 000 scale, it is no more reasonable to work at 1:5 000 scale or larger which are the scales for the higher resolution images.

So very high resolution images (1 or 2.5 m) have to be sub-sampled as shown in figure 6 below, where we can notice that the visual aspect of all the images, which are displayed at the same scale of 1: 20 000, is identical whatever was the resolution of the original data set.

For the photo interpretation of the air field for example, the 1m image does not bring more information than the 10m one, for the interpretation of the building blocks in the town the 5m image which is displayed at half resolution, and even the 10m one at full resolution, are more informative than the more subsampled ones.

Geometric accuracy
The geocoding process has to reach the accuracy specifications of the topographic map. Planimetric accuracy is in general 0.2 mm at the publishing scale and elevation accuracy is suppose to be 1/3 of the contour interval. Table 4 gives these values for maps at the most common scales:

Table 4

Scale

Plani. accuracy (m)

Contour interval (m)

Elev. accuracy (m)

1/25 000

5

5 / 10

1.5 / 3

1/50 000

10

10 / 20

3 / 6

1/100 000

20

20 / 50

6 / 15

1/200 000

40

50 / 100

15 / 30

Accuracy of the geocoded image:
It means that if you want to update a 1:50 000 topographic map, you must be able to create a geocoded image with 10m accuracy.

The geometric quality of the geocoded image depends on:

  • the pixel size and the effective resolution of the sensor
  • the quality of the rectification model
  • the accuracy of the ground control points (GCP)
  • the accuracy of the DTM and the incidence
    angle of the satellite for the original image.

Let’s consider a 10m panchromatic SPOT image, taken with a quasi-vertical incidence angle, and corrected with sub-meter accuracy set of GCP. The best planimetric accuracy we can expect for the geocoded image is between 10 and 15 m, which is fully compatible with standards for maps between 1:50 000 and 1:100 000. In the same conditions, if the SPOT image is replaced by a one-metre resolution image, the final orthoimage will have a planimetric accuracy better than 2 to 3 m’s’, which is the standard for 1:25 000 maps. To conclude, we can say that today’s satellite images reach the geometric standards of 1:50 000 and 1:100 000 maps and tomorrow’s one-metre to five-metres resolution images will easily match the standards of 1:25 000 maps or larger.

Information extraction from satellite data

Let’s analyse for all the cartographic features (cf. table 1) their level of interpretability for four types of satellite images:

  • images with resolution around 10 m like SPOT images,
  • images with 5 and 2.5 m from satellites like IRS and the forthcoming SPOT 5 and ALOS,
  • and one meter images from IKONOS and from
    the future very high resolution satellites.

Communications:

Table 5

10 m images
5 m images
2.5 m images
1m images

major roads and motorways
Identification
Identification
Identification
Identification

Secondary roads
Recognition to Identification
Identification
Identification
Identification

Major tracks
Recognition to Identification
Identification
Identification
Identification

Other tracks
Detection to Recognition
Recognition
Recognition to Identification
Recognition to Identification

Railways
Recognition
Identification
Identification
Identification

Airports and Landing areas
Recognition
Identification
Identification
Technical analysis

Footpaths
No
No
No to Recognition
No to Recognition

Various image resolution for various steps of interpretation:

Air field and sports grounds


Considerations about map-updating and images resolution

Hydrology:

Table 6

10 m images

5 m images

2.5 m images

1m images

Rivers and channels

Recognition to Identification

Recognition to Identification

Recognition to Identification

Recognition to Identification

Streams

Detection to recognition

Recognition to Identification

Recognition to Identification

Recognition to identification

Lakes and dams

Recognition to Identification

Identification

Identification

Identification

Springs

No

No

No

No

Wells

No

No

No

No to detection

Equipments:

Table 7

10 m images
5 m images
2.5 m images
1m images

Power lines
Detection to Identification
Detection to Identification
Detection to Identification
Identification

Major tunnels
No to detection
No to detection
No to Identification
No to Identification

Other tunnels
No
No to detection
No to detection
No to detection

Major bridges
Detection to Identification
Detection to Identification
Recognition to Identification
Recognition to Identification

Other bridges
No
No to detection
No to Recognition
No to Recognition

Sport fields
Recognition to Identification
Identification
Identification
Identification to Technical Analysis

Settlements:

Table 8

10 m images
5 m images
2.5 m images
1m images

High-density urban areas
Recognition to Identification
Identification
Identification
Identification

Low-density urban areas
Recognition to Identification
Identification
Identification
Identification

Villages
Recognition to Identification
Identification
Identification
Identification

Isolated buildings
Detection to recognition
Recognition to Identification
Identification
Detection to identification

Vegetation / land cover:

Table 9

10 m images
5 m images
2.5 m images
1m images

Cultivated area
Identification
Identification
Identification
Identification

Orchard, plantation
Recognition to Identification
Recognition to Identification
Identification
Identification

Grassland
Identification
Identification
Identification
Identification

Bush
Identification
Identification
Identification
Identification

Forest
Identification
Identification
Identification
Identification

Rocky areas
Identification
Identification
Identification
Identification