Previous Page Table of Contents Next Page

Terrain classification for the harvesting of tropical forests

D. Mazier, F. Baumgartner and C. Lepitre

D. MAZIER, F. BAUMGARTNER and C. LEPITRE were all associated with the Centre technique forestier tropical (CTFT), Nogent-sur-Marne. This article, originally written for UNASYLVA, is contained in a more lengthy and detailed version in the CTFT publication Bois et forêts des tropiques, No. 162.

Terrain classification is linked with evaluation of the accessibility of forest resources and is therefore of interest to research workers, planners and forest managers.

Numerous classification systems have been proposed, but most were conceived for application in temperate zones and serve essentially as a basis for deciding on the most suitable logging methods. They do not prove very useful in tropical forests in areas not under forest management, where the classification must take into account the difficulties not only of carrying out the harvesting operations as such (extraction), but also of creating the necessary infrastructure (roads).

A major forest inventory has been conducted in Gabon, covering an area of 3000000 hectares in a region situated to the north, east and south of Booué which will be served by the projected Trans-Gabonese railway (Figure 1). It was carried out by the Centre technique forestier tropical on behalf of FAO acting as executing agency for the United Nations Development Programme.


The region possesses a great variety of terrain types, ranging from fairly flat areas to others in which the topography makes harvesting operations extremely complicated. It was therefore considered essential to complete the inventory with a study of the terrain which would make it possible to characterize the different zones other than by mere qualitative appraisements. For this it was necessary to design quickly a practical method, adapted to conditions and able to provide a reliable, homogeneous classification of the terrain.

Several possibilities were considered, including the method based on structural geomorphology. In the end an original methodology was evolved, adapted to conditions in the Gabonese forests and to the data available. A concise description of this methodology is given in this article.

There are two essential differences between forest harvesting in Gabon and that conducted in temperate zones: yield per hectare and lack of infrastructure. The first task to be carried out in any harvesting operation is therefore to endow the area to be harvested with a network of roads and extraction routes. Hence the topography of the area influences two of the most important cost factors in timber production: establishment of the infrastructure, and extraction.

Some information about the physical features of the forest region studied are necessary for a better understanding of the study.

Climate. Although the climate does not really constitute a feature of the terrain, it is too important not to be taken into consideration.

Gabon, which straddles the Equator, has a moist, equatorial climate. In the zone with which we are concerned, there is little variation in climate and annual rainfall is between 1.6 and 2 m, distributed between four seasons. The geological formations (lower and middle pre-Cambrian) are relatively homogeneous and it is the tectonic movements that influence the contour of the land.

The undergrowth, which varies greatly in density, can hardly be counted an obstacle to harvesting operations. As for deadfalls, they are uniformly distributed and interfere but little with harvesting operations. The vegetation does not constitute an impediment, therefore. But the same cannot be said of the rocks that may encumber the ground. These consist sometimes of rocky slabs, but more often of large blocks which, if numerous and grouped together, make penetration for harvesting purposes impossible. This statement needs to be qualified, however. When it is a question of building forest roads, everything possible will be done to avoid traversing the rocky areas; and when an area of forest with some potential is so encumbered by rocks that crawler tractors cannot pass, this area will not be harvested. In Gabon unutilizable areas of this kind are usually scattered and not very extensive. The rocks are more numerous where the slope of the terrain is greater and are particularly frequent on the steep sides of mountains.

Soil quality. In road construction the quality of the soil is of great importance. The road-bed has to be laid on the surface layers of the soil which are often fairly uniform in a given area. What is of the greatest interest is the amount of laterite or other granulometric material available for constructing the bed. The number of borings made during the course of the normal inventory is too low to serve as the basis for a systematic location of gravel deposits: such a survey is beyond the means of the traditional type of inventory. Only traces which are clearly visible without any additional work will be noted. The presence of laterite can be detected in creek beds, breaks in slopes, the root system of uprooted trees, etc. It is advisable to include in any forest survey operation that is planned provision for the location and systematic recording of these traces.

STUDYING AERIAL PHOTOGRAPHS AS PART OF A SURVEY getting acquainted with the terrain

Most of the factors referred to above will not be examined in more detail, either because they are homogeneous throughout the area under study, or because of the impossibility of determining objectively how often and where they occur. In the following sections we are concerned only with a detailed study of the terrain. There are two ways of carrying out such a study:

· A microdescription of the terrain on the basis of the data collected by the field teams.

· A macrodescription of the terrain on the basis of existing photographic material and maps: of the 3000000 ha which have to be studied, about one half is covered by maps on a scale of l:50000 with an interval of 20 metres between contour lines, and the other half is almost entirely covered by aerial photographs on more or less the same scale.

Microdescription of the terrain

Microdescription of the terrain can be defined as description of the microtopography, that is to say, of the terrain as it appears to anyone moving about in the forest. It is limited to the local topography, but it is this that conditions road construction and the operation of the extraction machinery. This microdescription of the terrain is drawn up on the basis of the data obtained from questionnaires completed by the inventory teams. The sampling intensity is thus the same as for the evaluation of the forest potential.

For the inventory carried out in Gabon, the trees were counted and measured in strips 25 m wide lying across the transect line. The area of the record units was usually l ha (400 m long × 25 m wide). The recording form, in addition to data on the trees, included also information, at every 50 metres along the transect line, on the four slopes, i.e., the two longitudinal ones in the direction of the line, (average slope over 25 metres) and the two at right angles to the line on each side (average slope over 12.50 metres).

Although designed originally to assist in calculating the area of the record units in a horizontal projection, this information could also help to describe the microtopography.

It was thus possible to determine, every 50 metres along the transect line, the greatest slope in each of the four 25 m × 12.5 m quadrants of the record unit situated around this point, by the formula:

, P1 and P2

being the slopes of the two directions defining the quadrant. For a record unit of 1 ha, 400 m long, 32 values of P were thus obtained, grouped into the following classes:

0 to 20 %
20 to 30 %
30 to 40 %
40 to 50 %
50 to 60 %
more than 60%

The slopes thus classified were then regrouped for each of the 6 × 6 km squares to be inventoried (these were the systematically distributed "primary units" of a two-stage sampling design in which the second-stage sampling consisted, in each square, of two continuous parallel strips 6 km long by 25 m wide, each containing 15 record units of 1 ha.


Primary unit (6 km2 square)

Percentage classes

Less than 20

20 to 30

30 to 40

40 to 50

50 to 60

More than 60

























































In order to make possible quicker comparison between primary units, one can calculate the average weighted slope by weighing the mean values for each class of slope against the respective percentages as shown in the above table.

Figure 2 shows in diagram form the mean weighted slope gradients thus calculated per primary unit for a large part of the inventoried zone where a homogeneous inventory design was implemented. From this it will be seen that broadly speaking the slopes run from east to west.

To sum up, analysis of the field documents (recording forms) offers a certain number of advantages: it makes possible over-all comparisons between concessions and gives an idea of the real slope of the ground, which is very interesting as compared with the information obtained from aerial photographs and maps; since the forest cover hides the details of the relief of the ground, it is not possible to obtain an exact idea of the topography from an examination of photographs. However, it is advisable to complement field sampling by a study of photographic data and maps in order to obtain an over-all picture (macrodescription of the terrain).

Macrodescription of the terrain

The macrodescription of the terrain indicates only the topographic features visible on aerial photographs and does not take into account those on a small scale and the details concealed by vegetation. In addition to aerial photographs, use can also be made of contour maps on a scale of 1:50000, or of the map overlays which serve instead. Maps based on photographs facilitate analysis but eliminate certain details. They can be prepared or most parts of Gabon, because photographs at least are available.

A method for obtaining a macrodescription of the terrain was worked out in several stages, which are described below.

· First stage: assembling data. The first phase consisted of assembling a certain amount of factual data on maps and overlays on a scale of 1:50000 (with an interval of 2 m between contour lines), which covered half the inventoried zone. It was decided to start by using the map overlays.

On each map, covering an area of about 75000 ha, 9 sampling units were disposed in such a way as to obtain a homogeneous design with adjoining sheets. A few areas already being harvested were covered by sampling designs of varying, but much greater intensity.

The sampling unit was a square 2 × 2 km (400 ha), or 4 cm on the map. A square was chosen because it is the most regular geometric figure and therefore made it possible, by juxtaposition, to cover the whole area under study (the sides of the square were oriented north-south and east-west). The choice of the dimensions of the square represented a compromise between too small a unit, which would provide only very specific information for a restricted area, and too large a unit, which might cover several different types of relief but conceal the heterogeneity of the area by providing only information on averages.


The sampling design is shown at the bottom of the diagram. Each primary unit has an area of 6 km², and its zone of extension is 12 km². In the upper part of the diagram each primary unit is represented by its area of extension.

A certain number of parameters, selected to represent the maximum possible variations in terrain, were measured in these squares:

x1: slope with gradient of less than 20 percent
x2: slope with gradient of more than 40 percent

These limits of 20 to 40 percent were chosen for two main reasons: the first is that, where the gradient is less than 20 percent, harvesting is very easy, and where it is more than 40 percent, extraction by tractor is difficult; the second is of a practical kind: gradients of 20 percent and 40 percent correspond to the intervals between the 2 mm and 1 mm contour lines respectively. The area of these slopes is determined by a dot grid.

x3: number of changes in slope direction along the two median lines of the square;

x4: number of changes in slope direction along the sides of the square: it is assumed that each of the lines, median and side, is described by the observer who counts the number of times that the slope direction changes;

x5: number of rivers intersected by the median lines;

x6: number of rivers intersected by the sides of the square.

The number of changes in slope and the density of the drainage system determine the degree of fragmentation of the relief. The distinction between the sides and the medians of the square corresponded to different sampling designs, these lines having been selected because they were easy to plot. It should be noted that the parameters x3 and x4 are accurate and more sensitive than the parameters x5 and x6. Each crossing of a river indicated on the map corresponds to a change in slope, but the reverse is not true. In addition, the details of the drainage system depend on the accuracy of the recorders and on the type of relief.

x7 : number of contour lines intersected by the median lines;

x8 : number of contour lines intersected by the sides of the square (these two parameters give the total change in level);

x9: maximum variation in level within the square, in metres;

x10: total length of rivers within the square (determined by opisometer with an accuracy of ± 100 m).

These first ten parameters constitute a direct transcription of the information provided by the map. The following parameters result from further elaboration of these primary data.

x11: length of the road to be traversed, starting from the centre of the square in order to leave the circle inscribed around this centre without encountering any longitudinal slope exceeding 10 percent;

x12: sum of the transverse slopes expressed as a percentage, measured every 100 m along the road.

FOREST SURVEYOR looking for practical results

These last two parameters illustrate the degree of difficulty of the terrain. To these twelve parameters measured on the map overlays were added two other values formed by combining some of the previous parameters:

x13: x7/ x3, that is, the number of curves divided by the number of changes in slope, all measured along the median lines;

x14: x8/ x4 (the same as x13, but measured along the sides of the square).

Finally, each sampling square was coded from 1 to 5, according to its degree of difficulty:

Class 1: fairly even terrain,
Class 3: moderately uneven terrain,
Class 5: extremely uneven terrain,
Classes 2 and 4 being intermediate between these.

This first set of measurements was applied to 124 units in the inventoried area that were not yet under utilization, and some 30 units in the area that was already being harvested.

· Second stage: statistical analysis. Each sampling unit can be represented by a point in the 14-dimension space where the coordinates of the 14 axes would be the values assumed in this unit by the 14 preceding parameters. The geometric disposition of the points cannot be illustrated graphically. The statistical method best able to convey an idea of the cluster of points representing the sampling unit is "the principle component analysis", which consists, among other things, of projecting the cluster on those planes in the total space that are closest to the largest number of points, so as to arrive at the most representative diagrams possible.

(In what follows we will confine ourselves to the plane determined by the first main component [C1] - a straight line such that the sum of the squares of the distances of the points to their projections on that line are at a minimum - and by the second component [C2], perpendicular to C1 and such that the sum of the squares of the distances of the points to their projections on the plane of these two straight lines is at a minimum.)


The indication "points in the region of..." shows the location of a certain number of points drawn from different parts of Gabon. The "Monts de Cristal" points are very scattered: "i" varies enormously depending upon whether the observer is in a valley or on a hilly site.

This method of analysis was applied to the 156 sampling points studied and made it possible to obtain the following results:

Correlation between the 14 variables, in pairs.

Correlation between each of the variables and each of the first four main components.

A histogram of the frequency of the variables.

A graph plotting the position of the 156 points on the plane of the first two main components, C1 and C2.

· Third stage: The correlations between the 14 variables and the main components show that there is a good correlation between parameters x1, x2, x7, x8 and x9 and the component C1, and between parameters x3, x4, x5, x6 and x10 and the component C2. Of the remaining parameters, there is a certain degree of correlation between parameters x12 and x14 and C1, but no clear correlation for parameters x11 and x13.

Aims of the analysis

It would seem that the component C1 corresponds to the slope of the terrain: there is a strong correlation with x1 and x2. The component C2 would then correspond to the fragmentation of the relief.

One of the aims of the analysis undertaken was to see whether it was possible to class the types of terrain on a diagram by using only two variables selected as the best as a result of this analysis. These parameters must satisfy two criteria:

1. They must be well correlated with the results of the analysis;

2. They must be measurable not only on contour maps, but also on aerial photographs, so that it is possible to work even where no maps exist.

Classification of the terrain by means of two parameters would make it possible to work rapidly, while the use of 12 parameters cannot be adopted for the study of large regions.

(a) According to the first component (C1): x1 and x2 satisfy the above conditions; x2 has the disadvantage that it gives many zero values, which greatly diminishes its interest; x1 is much more widely applicable, but in the case of very uneven terrain it would lack refinement, since it, too, would provide many zero values.


It was therefore decided to try a combination of these two parameters, which would make possible a complete description of the terrain, both in the intermediate zones where both parameters give valid results, and in the zones at either extreme, where the use of x1 for very uneven terrain and of x2 in flat terrain lacks precision. This combination is expressed in the following equation by the index "i":

(x1 and x2 expressed as percentages in a given square)
"i" is expressed in degrees:
i = 0 on flat terrain
1 = 100 if all the terrain has a slope gradient of over 40%.

The angle "i", expressed in degrees, will hereafter be referred to as the slope index.

(b) According to the second component (C2): x3, x4, x5, x6 and x10 all meet the criterion of correlation with the second component. In addition, they are largely independent of x1 and x2; x4 (number of changes in slope along the sides of the square) presents certain advantages over x10 (total length of rivers). In fact, the density of the drainage network shown on the map depends on whoever prepared the map. This drawback attached to x10 is even more inconvenient when it comes to photographs, in which it is not clear where rivers end; x4 was therefore retained. To improve the quality of the parameter linked to the second component, x3 can be added. The resulting variable, number of changes in slope along the sides and medians of the square, will be called f (fragmentation index).

Figure 3 shows the position of each of the 156 sampling units on a graph whose abscissa is slope index "i" and whose ordinate is fragmentation index "f", the class of difficulty being shown for each point. The graph indicates that the subjectively estimated difficulty level corresponds, despite some overlap, with variations in the slope index. The values (i < 1, f < 15) at the lower left of the graph represent units in the region of Daloa, Ivory Coast, where forest exploitation is considered easy, the Gabonese area that comes nearest to this optimum (i = 2.5, f = 18) represents a flat, swampy area north of Koumaneyong. The graph also demon strafes the great variability of fragmentation index "f".

· Fourth stage: practical application. The classes of difficulty (Figure 4) were worked out bearing in mind on the one hand the preceding grades of difficulty, and on the other hand the necessity of keeping approximately the same number of classes. In addition, examination of the diagram and of the 1:50000 contour maps of areas with very uneven terrain showed the necessity of providing for more categories for difficult terrain.

The final classification adopted for the slope index (i) was as follows:

0 to 12:

easy terrain

13 to 24:

average terrain

25 to 38:

moderately difficult terrain

39 to 54:

difficult terrain

55 to 69:

very difficult terrain

70 +:

extremely difficult terrain

For the fragmentation index, only two categories were employed: terrain with little fragmentation, "f" up to and including 40, and very fragmented terrain, "f" of 41 and over. This classification corresponds to conditions in Gabon, except for the coastal plain, where the main value of "f" is around 4041. For a study of other areas, it might be useful to distinguish a greater number of categories within the fragmentation index.

Within a sampling unit, "i" is measured in the following way: the 4-cm square, taken from a map on a scale of 1: 50000 with 20-m intervals between the contour lines, is covered by a 64-dot grid; the number of points at which the slope is less than 20% - interval greater than 2 mm and those where it is above 40% interval less than 1 mm - are ascertained with the help of a transparent, double decimetre graduated on the lower side placed against the grid, and using where necessary a magnifying glass - a relatively slight enlargement suffices. If y1 is the first figure obtained and y2 is the second, the slope index, "i", is equal to the value of i in grades such that .

Then the number of changes in slope are counted along the sides of the square and along the two medians shown on the grid. Different measurements are made on the same units in order to ascertain the accuracy: the relative error between measurements may reach 15 %, but cannot exceed 5 % if a lens is used.

Various trials were made on a section covering 60000 ha, always on a scale of 1:50000, in order to perfect the practical sampling design. Using the Universal Transverse Mercator grid, a systematic sampling design was experimented using progressively higher sampling intensities. For a given intensity, the contours of areas of equal difficulty (according to the slope index) were traced on the map; then the limits of the slope index categories were identified by a thorough study of all the squares on the map. This showed that excellent results could be obtained by using a sampling design in which the distance between two counted squares equals two squares in both a latitudinal and a longitudinal direction, that is, a minimum sampling intensity of 1/9 = 11.1%.

PREPARING TO FEEL A TREE WITH A POWER-SAW forest surveyors should take him into account, too

LOADING TIMBER IN AN AFRICAN FOREST they knew the poor was solid

This systematic design was supplemented by the study of a few additional squares, for which the slope index category war, not clear. In regions with very heterogeneous relief, these additional sampling points will be more numerous than in cases where the relief is relatively homogeneous. Finally, the over-all sampling intensity adopted was about 14%. The practical design, carried out by a sampling study, fixed the limit of the zones of equal difficulty solely on the basis of the slope index. The fragmentation index, with fewer categories and much easier to discern, was kept in second place.

The simultaneous but rapid examination of aerial photographs proved interesting, and even essential where the terrain was easy - "i" between 1 and 12. In fact, there are two typical cases that may arise:

- The documents on a scale of 1:50000 are not complete maps, but map overlays which do not indicate swampy formations;

- The contour lines shown on the maps are 20 m apart, so that slight variations in relief are not visible; in aerial photographs, on the other hand, short but steep slopes are clearly visible ("orange peel" relief).

APPLICATION TO AERIAL PHOTOGRAPHS. Toward the end of 1973, about half the inventoried zone in Gabon was covered by map overlays on a scale of 1:50000. The other half was covered only by aerial photographs. In this latter half one way of classifying the terrain would be to transfer to the aerial photographs the method used for the map overlays. The slope index could be determined by laying a dot grid of the area selected on one of a stereographic pair of photographs and ascertaining by means of the stereoscope the points where the slope is less than 20% and those where it is more than 40 % This procedure presupposes that the observer is experienced in estimating gradients from aerial photographs and that he periodically controls the gauge of his instrument. In theory, the only difficulty lies in the variation in scale of the photographs. In order to be accurate, therefore, it is advisable to determine the exact scale of the photographs and to bear this in mind in fixing the size of the sampling unit.

Another method would be to proceed by analogy. In the part covered by map overlays an aerial coverage exists on a scale approximately the same as for the rest of the area to be inventoried. One could, therefore, as a first stage, prepare a check sample of stereograms representative of the different types of terrain and subsequently class the area to be studied by referring to this sample.

The aerial photographs are examined under the stereoscope; using the check stereograms, the limits of the different slope index classes are plotted on the photos of a single strip. The photos from this strip, and then from other strips, are compared and the homogeneity of the limits traced are verified. Often the limits do not correspond exactly as between one strip and another; this is due to the poorer quality of photographs along their edges, and it is necessary to adjust this by stereoscopic examination from strip to strip - with a slight overlap of 10% to 20% - or even by a fresh examination under the stereoscope of the photos from a single strip.

STUCK IN THE MUD learning about the quality of the soil after the road is built

Since the fragmentation index "f" is divided into much less refined categories, in the case of Gabon its different values were not shown in detail on the maps, particularly since there is little variation over fairly extensive areas; some regions show little fragmentation, while others are uniformly broken up.


In general the results obtained from both methods of description correlate fairly well. But macrodescription of the terrain using maps does not give a really accurate picture of the slopes. It has the advantage, however, of providing a comprehensive view of the terrain as a whole, whereas the indications provided by the microdescription, based on field observations, are limited in scope by the sampling intensity used for the inventory.

Apart from this, it is necessary to set the results of the macrodescription against the over-all nature of the relief (uniformity or variability). Thus, a hill with difficult slopes, but situated in the middle of an easy area, will not present any great obstacle to harvesting operations; it will always be possible to construct the essential access roads. But if hills of this type are frequent, without wide valleys between, this will render harvesting operations difficult, if only by reason of the cost of penetrating the mountainous area. In the peninsula of Azuero, in Panama, where this method of macrodescription was employed, the difficulties of the terrain are undoubtedly indicated by the high slope indexes, but they also derive from the fact that the relief is everywhere the same, without any valleys of penetration. The difficulties noted in Gabon, in the eastern region of Mouila, also have to be evaluated bearing in mind the fact that the relief remains the same over vast areas (there is only one large valley).

Finally, the study carried out, which does not pretend to be applicable in all cases, makes it possible to draw a certain number of conclusions. First, it is important that, in addition to the usual data (on trees, slopes, etc.), the field survey team record also other information useful for future harvesting operations, such as, for example, indexes of material suitable for use in road-building, or the presence of rocky slabs or blocks. Also, the macrodescription of the terrain made on the basis of maps and aerial photographs should be effected in conjunction with the photo-interpretation studies necessary for forest inventory work as such. Apart from the fact that the data would be used at the same time by the same interpreters, this procedure would also make it possible to demonstrate interesting correlations between types of forest and types of relief.

Previous Page Top of Page Next Page