Previous Page Table of Contents Next Page


Presenting the results of integrated surveys by computer mapping

Klaus VOLGER *

(*) Klaus Völger: UMWELT-DATA GMBH, 605 Offenbach, Ludwigstr. 33, West Germany.


1. Introduction - The need for planning-relevant information
2. The concept of cellular mapping
3. The concept of multi-source data
4. Extraction of planning-relevant information
5. Presentation of results
6. Computer-mapping and spectral pattern recognition


SUMMARY

Information relevant to regional planning must be collected from various sources, including remote sensing. The cellular mapping approach using one km² - cells is described by example. The degree of generalization is similar to that of automatic ERTS processing but many more sectors, e.g. socio-economic ones, can be covered. Also, there is compatibility between thematic maps from ERTS and from the computer.

1. Introduction - The need for planning-relevant information

Regional planning, like financial planning, requires a knowledge of assets and liabilities, of what one lays out and what is being taken in. The ecological balance sheet is normally in equilibrium except for human interaction and episodic natural disasters (like droughts, earthquakes) or blessings (extraordinary rainfalls, or volcanic ashrains as natural fertilizers).

Generally, there seems to be little doubt that an information base is required for planning at, e.g. regional scale. It can be compared to the foundation of a building. If the data base is too thin, the entire planning may collapse. On the other hand, if too much data is collected, a great deal of it may remain unused (in German this is called a "Datenfriedhof", i.e. a data-cemetery), information extraction becomes awkward, and the entire planning process is bound to be uneconomic. The collection of base data, therefore, is the middle route between Scylla and Charybdis in order to arrive safely at a sound set of planning-relevant information.

About ten years ago, a regional planning inventory of natural and socio-economic parameters in the western Sahara had specified the results very precisely. A large mass of questionnaires and land use maps had to be established. The method used at that time was a combination of photo-interpretation and field surveys. The planning of our integrated survey was done according to the critical path method, and the results were thematic maps, lists, histograms and a general report. The then available Gemini photographs were used to gather general information about the steppe areas not yet photographed from aircraft and particularly for little-known areas in the central Sahara. In the course of those ten years since, the concept of acquisition, processing and presentation of planning-relevant data has become a fascinating occupation for the present writer. Later experiences in Germany have led to the cellular approach, thus facilitating data handling by computers in a most simple way. (A later presentation by cellular computer-mapping on a line-printer proved a valuable decision aid in respect of judging the right amount of data to be surveyed, as I shall try to demonstrate in the next section.)

2. The concept of cellular mapping

In conventional maps, both topographic and thematic ones, the scale factor specifies the locational accuracy of details shown (e.g. in topographic maps, usually 0.2 mm at map scale for planimetry) and the graininess or resolution of information. In contrast to topographic maps, hardly any standards exist for the latter parameter mentioned in regard to thematic maps.

Needless to say, a topographic base is one prerequisite for any type of regional planning; however, it could be an elaborate contour map or just an aerial mosaic, depending upon the stage of development and the objectives of planning. The other information prerequisites or sectors may or may not be presented as maps - thematic maps in this context. In any case, the individual information sector has to be covered during the data acquisition phase. The data may be tabulated in other formats such as lists, histograms, etc.

For problems of regional scale, some map-like presentation is usually preferred, particularly if quantitative data can be shown in the geographic context. A group of curves or histograms laid out on a map is only a very poor substitute, because it is highly discontinuous. In this respect, the cellular approach to the mapping of statistical data has brought about a rather drastic change.

About ten years ago, at the Harvard School of Architecture, the town planners pioneered cellular computer mapping. It was based upon an approach developed by Howard T. Fisher of Northwestern University, Illinois, for constructing isopach maps from well logs. A land use and natural resources survey of New York State (ca. 150,000 km²) abbreviated "LUNR", was conducted in the late sixties for about 140 parameters. At about the same time and without knowledge of those endeavours in the USA, a detailed land use interpretation from aerial photos was executed for the city of Frankfurt under the guidance of the present author.

In both cases cited, the results were not only shown as line-drawn thematic maps but the inventory of land use was referenced to km²-cells in a geodetic coordinate network. These square cells were the basic units of reference of the survey. All information of a qualitative kind (e.g. kind of land use) or quantitative kind (e.g. absolute information, i.e. hectares of agricultural land use type x within km²-cell or relative information, i.e. dominant type > 50 hectares) are referenced to the coordinates of the cell, that is, its lower left corner, as is common practice in geodesy.

Cellular mapping: a location within the cell of, as in our example, say one km², is not possible and not even desirable. This in itself means a considerable generalization, at least as far as locational accuracy is concerned. The statistical accuracy remains untouched. Thus the first decision about locational accuracy and basically also about resolution is being forced.

The size of the cell may vary, depending upon the total area under consideration and the problems being investigated. Thus, in towns the size of the unit cell may be one hectare (= 0.01 km²), and for larger regions the cell may go up to 4, 9, 100 km² or any logical size in between.

One of the few formulas derived from our work seems to be: that for regions of any size, the number of unit cells should lie between 1,000 and 10,000, preferably around 5,000. By this the degree of generalization is indicated, if not already fixed.

The method of data handling is up to the planner and his facilities. It can be on large computers, minicomputers or, in our method, literally, by hand.

The first step is to draft a plan of data structure for each information sector (e.g. geology, hydrology, vegetation, agricultural land use, population, et al.). Hand in hand with it goes a concept of what data at what resolution are to be acquired. Simple considerations as to computer capacity and number of cards to be punched will force decisions rather early as to how many columns of a punch card are to be occupied. Multiplied with the number of cell units, this will result in the total number of punch cards or form sheets to be filled. This in itself is a very healthy, self-critical way to avoid the "data-cemetery".

Thus the cellular approach in itself, by sheer consideration of filling in punch cards of 80 columns, helps to decide the basic question of what is planning-relevant and what is not. The computer programs are available at nominal cost from the Harvard Laboratory for Computer Graphics and Spatial Analysis and from the US Department of Commerce. Modifications for minicomputers and even for manual updating and checking of certain questions have been developed by us recently for a German technical cooperation program in West Sumatra.

3. The concept of multi-source data

The input data for a regional information system can be derived from various sources. They could be grouped, according to the date of acquisition, as:

a) a priori data, i.e. knowledge about the region from literature, statistics, scientific papers, topographic maps, etc.

b) ad hoc data, i.e. recently acquired data, particularly for the acute problems.

According to the method of acquisition, they could be divided into two groups:

a) in situ data, i.e. data that are acquired on site, either by direct visual observation or by interviews,
b) remotely sensed data, i.e. an areal coverage from aircraft or satellites in multispectral bands.

There is a complementary interaction between the two latter groups. In the remote sensing community, in situ data are referred to as "ground truth", and they serve as a check on the results of an interpretation of aerial photographs or, in the case of multispectral computer analysis and land use classification, as training sets for the computer. The training sets are given into the computer to provide a known spectral signature of a certain class, e.g. wet grass" land, and also its width of variation.

A modification of the remote sensing/ground truth relation is provided by an additional lower flight level of remote sensing. At this lower altitude data of higher spectral and geometric resolution are acquired that are a great help in the manual as well as the automatic interpretation of high altitude, low resolution remote sensing data, e.g. from satellites like Landsat (ERTS) and Skylab.

These medium-level data are usually taken from aircraft, and it is sufficient if they cover a representative 5 or 10 percent of the total area. This concept is called "multi-stage sampling".

Multi-source data should be formatted in such a manner that they are compatible with cellular storage and retrieval. This is usually no problem: on the contrary, it facilitates data-acquisition because the unit cell, usually one km², allows certain generalizations from the beginning, e.g. the percentage of a certain type of pasture within one cell need not be determined with exacting thoroughness; it is sufficient to estimate it, let us say, in six steps; none, 20 per cent, 40 percent, 60 percent, 80 percent, 100 percent (10 percent for each class). Logical thresholds of up to ten classes may be applied for each individual datum regardless of its source, e.g. from statistics, field work or remote sensing.

Input data differ in their dimensions. Besides undimensioned numbers (number of inhabitants of a certain hospital) which have the geometrical characteristic of the point (0-D), there are linear (1-D) data, areas (2-D), and spatial (3-D). The table below lists some examples:

Table 1

Dimension

0 - D point

1 - D linear

2 - D areal

3 - D spatial

examples of data input

hospital inhabitants

rivers, boundaries, houses per road km

land use, natural vegetation

relief

possible output

centrality, market place

necessary communication lines

potential for certain crops or grazing

catchment area for water quantities, agricultural or pasture potential due to slope

4. Extraction of planning-relevant information

In order to combine data obtained from the use of remote sensing techniques, mainly from aircraft and satellites, with information available from other sources, numerous different identification codes are being keyed to the 1 km² cells by administrative units, natural land units, land use units, watersheds, or in many other ways that may be needed during the planning process.

In our recent work for West Sumatra, a total of 98 different classes of base data have been identified, found relevant and stored for each of the 4,500 square km cells. The planners can now operate with a data bank of approximately 440,000 items, retrieving and combining information according to their needs, using the fundamental rules of mathematical logic. Almost any computer available on the market can print out all sets and subsets of the data stored.

Data can be computed and listed in any desired sequence, in the format of tables or of topographically fixed thematic maps. The more fascinating possibility is, however, the ability of the computer to combine any of the data classes with one or more other classes, by such logical operations as conjunction (AND), disjunction (OR), negation (NOT), implication (IF - THEN) and so forth. In other words, within the universe of the data stored, the planner can extract any set, subset, union of sets, intersection of sets, or complement sets which are relevant for his planning activities. For example, the intersection of the set A of all cells of unconsolidated volcanic rock with the set B of all cells of an altitude of more than 200 m will result in the subset C = A O B of cells with a certain land use potential.

Intersected again with the complement of the set D of medium slope cells, the resulting subset will contain all cells where the potential land use may be planned.

Figure

As the data bank for West Sumatra stores 98 categories of data, these data can be matched against each other in a tremendous number of combinations. It is obvious that the majority of possible matches will be meaningless, senseless, contradictory or logically impossible. The remainder, however, will still be more than any planning team can consider for practical processing. It follows, therefore, that this way of data storage and retrieval will provide the planner with all necessary information and eliminate to a large degree the need for further data collection at a later stage when the process of planning has changed the pattern originally envisaged.

The 98 categories of stored data for West Sumatra contain either absolute values like height above sea level and number of inhabitants, or relative values. Relative values are expressed as percentages within one cell, for instance 20 percent primary forest, 60 percent grassland with bushes, 20 percent cereals. Further categories identify up to 9 or 99 different properties of one cell like soil quality or the direction of the prevailing winds.

Two punched cards with 80 columns each were needed to register all data on one cell. Principally, punched card number one stores the natural data that are unlikely to change over a longer period of time except for major catastrophes. Punched card number two stores all data pertaining to land use and other socio-economic parameters that are subject to change both independently from or under the influence of project implementation. The data of the second punched card will have to be updated regularly if the data bank is to remain the most important planning tool for the local administration after the present planning project has been completed and implemented.

5. Presentation of results

There are various programs requiring various peripheral hardware for output. The simplest device is the normal line-printer using conventional characters like letters and figures. This is what we used. By repeated overprinting, a number of grey tones can be produced. Cellular maps produced in this way have the disadvantage of being hard to read if more than five classes are to be rendered. Therefore, we used the computer output as color separations and the resulting maps were printed in color. However, the normal procedure of representing one km² cell by 20 characters (5 per line, 4 in a column) does not produce satisfactory results, due to overprinting. Therefore, we applied a random distribution of fully blackened characters within the cell according to, e.g. the percentage of area covered at five or six classes, twenty dots being 100 percent, two dots 10 percent and so forth.

Thus a screening effect was achieved for color printing or for constructing a small number of color-composites on transparent foil (color-diazomaterial such as Foldex, 3M-Colorkey, etc.).

The table below shows the various single-theme topics and the mode of presentation that was used in our recent work for the West Sumatra region:

Combinations of these parameters are easily possible, such as forests and grassland (i.e. natural vegetation) alone or with existing agriculture.

The true capability of the computer-methods shows up when multiple-theme topics are printed, i.e. by a combination of various inputs. It is obvious that the eliminating parameters for, e.g. agricultural potential depend upon local particularities. Therefore, they must be defined for each region and, as shown by our experience in Sumatra, the parameters will be changed, modified and amended as the planning goes on.

The following eliminating parameters were chosen to print out the gross agricultural potential:

a) Government land (= forests),
b) present agricultural land use more than 50 percent in each cell,
c) swamps,
d) slope over 40 percent,
e) height above 1.500 m.

For the net potential, a further narrowing down, in addition to the factors a) through e) was done by the following factors:

f) high valley density,
g) depth of valleys over 50 m.

In both cases the population distribution was printed in grey shades over the color map.

It should be noted that these remarks are intended as a simple example to demonstrate the possibilities of multiple-theme computer mapping.

At present, an adaption of the programs for midicomputers is being developed with the object of placing the planning information system into the hands of local authorities. It is obvious that many socio-economic factors and even political intentions of the local authorities may be merged into the system. Total areas in hectares can be listed, and also market situations may be simulated.

6. Computer-mapping and spectral pattern recognition

The degree of generalization produced by line printers and subsequently color-coded maps depends upon the cell size chosen. With the random dot method the apparent geometric resolution is better than the nominal size of the cell. The resulting maps show considerable similarity to thematic maps, especially land use maps, from multispectral scanners using spectral pattern recognition techniques on a computer. Here the size of the instantaneous field of view (IFOV) or picture element (pixel) determines the geometric resolution.

By gridding, i.e. introducing a geodetic reference net into the pattern recognition output, the results become directly comparable with each other and one output may serve as check on the other. What seems more practical and appears to us an important future tool for regional planning is the possibility of combining the results of both methods. The approach used by those authors who are entirely systems-oriented towards spectral pattern recognition will in all probability never supply a hundred percent of all thematic map information needed by regional planning. This can be stated at least for present satellite data. Multispectral scanners and high-resolution photographic systems aboard aircraft, however, can fill the gap until more sophisticated satellite systems become available, such as EOS and others expected to be operational around 1980. A mission definition study for a hydrological observation satellite (Hydrosat) and an agricultural one (Agrosat) have recently been completed by a German firm for the Federal German Government (Ministry of Research and Technology). These studies will be handed over to the Committee on Peaceful Applications of Outer Space of the United Nations, and are hoped to be a stimulant for further work.

Table 2

Topic

Number of classes

Full tone per cell or random dots

B/W or color

Source of data

mean elevation

10

f.t.

B/W

maps

valley density

7

f.t.

B/W

maps

relative altitude

7

f.t.

B /W

maps

difference

8




natural vegetation


r.d.

color

aerial photos + ERTS

forest types

3

r.d.

color

aerial photos + ERTS

grass and bushland

4

r.d.

color

aerial photos + ERTS

agriculture and pasture

8

r.d.

color

aerial photos + ERTS

population

(1 character = 100 persons)

r.d.

B/W

total number from census, distribution by houses in aerial photos

ethnic distribution

3

r.d.

color

similar to population distribution


Previous Page Top of Page Next Page