Special Interests
Videography and River Monitoring

A methodological study for biotope and landscape mapping based on CIR aerial photographs
Landscape and Urban Planning Volume 41, Issues 3-4 Pages 183-192 6 July 1998
Sara A. O. Cousins* and Margareta Ihse Department of Physical Geography, Stockholm University S-10691 Stockholm Sweden
Abstract
In this paper, we present a method for base-line mapping of biotope and landscape elements in the rural Swedish agricultural landscape. The overall goal is to elaborate a classification system for a national landscape monitoring program, based on interpretation of existing colour infrared (CIR) aerial photographs at the scale 1:30000, and including a field control. The classification system developed was tested by mapping landscape elements in strategically selected test areas, and it is assessed with respect to interpretation accuracy. The landscape elements, mapped separately as patches, lines and points, are significant for the biodiversity on landscape level, and are susceptible to change. The classification system is based on a hierarchical approach in five levels, with regard to land use and management, nature type and succession stage, moisture, physiognomy, vegetation cover and plant species. By using the method and the suggested classification system, a base-line mapping can be done very quickly and accurately. The mapping rate is 1.4-2.8 km2/h and the interpretation accuracy is 95-99%.

Outline
1. Introduction
2. Physical settings of the study area
3. Materials and methods
4. Results
5. Discussion
6. Conclusion

1. Introduction

Developing rapid and accurate methods for monitoring landscapes and biodiversity is a high priority in Sweden and throughout Europe. The regional county councils in Sweden have got such new tasks on their agenda. However, so far there is no standardised national method. This study is a first step towards a National landscape monitoring system, based on biotopes and landscape elements. The study is restricted to methodological-aspects of base-line inventories for mapping the agricultural landscape, with special emphasis on biodiversity.
We assume, like O'Neill et al. (1988) and Forman (1995), that to ensure preservation of species and genetic diversity, there has to be a range and structure of biotopes that provide an ecologically functional landscape. The diversity of species can thus be mirrored in the diversity of biotopes.
In Sweden, a country dominated by coniferous forest, the rural agricultural landscape consists of a wide range of biotopes, habitats, and a rich diversity of species, including many threatened species (Ingelög et al., 1993aIngelög et al., 1993b). Out of 400 threatened plant species 300 belong to the agricultural landscape (Bernes, 1994). The important biotopes are mostly semi-natural, created by varying natural processes, combined with different methods of land management, often with long historical continuity (Ihse, 1989; Skånes, 1991Skånes, 1996; Sporrong et al., 1995).

Several authors have described biotopes in the agricultural landscape, which have a high diversity of plants, insects and birds (Ekstam et al., 1988; Ekstam and Forshed, 1988; Robertsson et al., 1990; Bengtsson-Lindsjö et al., 1991; Ihse and Norderhaug, 1995; Norderhaug et al., 1996). These biotopes are mainly found among semi-natural grasslands, deciduous woods and wetlands.
Small biotopes are also of importance for biodiversity, as they offer refuges and ways of dispersal, which is especially vital in a fragmented landscape. Such small biotopes include linear elements such as ditches, earth banks, stone walls, hedges, road verges, and point elements such as ponds, mounds of stones and solitary, old trees. Changes occurring in the landscape today constitute a threat to these biotopes and thus to biodiversity. Biotopes are fragmented and isolated, and the ecological infrastructure made up by the small biotopes is damaged (Ihse, 1995). Mechanisation and structural rationalisation, and in many cases afforestation are main causes of fragmentation and isolation, while changed species composition and species decrease are directly or indirectly caused by increased use of chemicals. The driving forces behind these changes are considered to be a combination of political decisions, technical and economical development (Ihse and Lewan, 1986). The causes and the driving forces will not be further discussed in this paper.

Studying biodiversity and the potential for biodiversity must be done in different scales, both in time and space (Forman, 1990; Zonneveld, 1990; Delcourt and Delcourt, 1992; van der Maarel and Sykes, 1993). This has been only marginally considered in most European countries (Ihse, 1996; Jongman, 1996). Biodiversity must be studied both on detailed level (genes and species) and general level (biotopes and landscapes), in order to fully understand the effects caused by rapid changes. This paper aims to test a method using existing colour infrared (CIR) aerial photographs (scale 1:30000) for detailed mapping of landscape elements that are important for biodiversity. Specific objectives are to develop a classification system which encompasses landscape elements of importance for preserving biodiversity and cultural-historical values, and to estimate the interpretation accuracy. The method and classification system are suggested for a base-line mapping in a landscape monitoring system.

2. Physical settings of the study area
The study area is the county of Malmöhus, the regional administrative unit situated in the southernmost part of Sweden. It is mostly an agricultural landscape, where four different landscape regions can be distinguished, differing in bedrock, soil, topography, climate, hydrology and vegetation, as well as present and past land use and management. We will give a description of the physical settings, to enable the understanding of actual and potential biodiversity, as this will enable comparisons with other regions and thus provide possibilities to adopt the method outside the study area.
The plains is dominated by sedimentary rocks, rich in nutrients. They have highly fertile soils, mostly clayey till and in some places silty or sandy soils. Large horsts of granite and gneiss, running in NW-SE direction, have soils that are poor in plant-accessible nutrients and minerals.
There are few lakes, ponds and watercourses remaining, both due to permeable bedrock and to artificial drainage during the last 100 years. The topography in the plains is gently undulating with a relative relief of 50 m. The relative relief on the horsts reaches about 100 m. Vegetation in the study area belong to the nemoral zone, thus deciduous broad-leaved trees, mainly beech and oak, dominate the forests.
The mean temperature in the coldest month, February, is between -1°C and 0°C, and in the warmest month, July, between 16 and 18°C. Mean annual precipitation lies in the interval 550-800 mm (National Atlas of Sweden, 1995). The land use on the plains is mainly food crop production. Forestry and animal husbandry are common on the horsts and in their vicinity (Sporrong et al., 1995). Based on the above landscape, four areas were distinguished (Fig. 1):
1. Intensively farmed large scale agricultural plains, with few small biotopes (Äspö area)
2. Intensively farmed medium to large scale agricultural plains, with some lakes and watercourses (Igelösa area)
3. Hummocky landscape with a mixture of medium scale agriculture and forestry (Häckeberga area) 4. Hummocky extensively farmed, small scale agricultural landscape, with many semi-natural grasslands and a large number of small biotopes (Bessinge area)

3. Materials and methods
Diapositive CIR aerial photographs were chosen for data capture. CIR aerial photographs cover the whole of Sweden (scale 1:30000 in the south, 1:60 000 in the north), and were photographed by the Swedish Land Survey during the 1980s. A second coverage of photographs has been initiated, and will be in the future available for follow-up studies in a monitoring program. Four test areas measuring 5 km×5 km were chosen (a-d), each area covered by a Swedish Cadastral Map (scale 1:10000). The areas interpreted in Äspö, Igelösa and Häckeberga were 5 km×5 km, i.e. whole map sheets. Bessinge has a very heterogeneous landscape with a vast number of landscape elements. The aerial photograph interpretation of this area had to be restricted. Using a random choice analysis 50 squares (250 m×250 m) were selected for the interpretation. Linear and point elements were interpreted in each square. For the data collection of patch elements, seven 1 km×1 km areas were picked out strategically where the greatest chance of catching the largest number of landscape elements, covering all the variation, was at hand.
The CIR aerial photographs were interpreted by visual interpretation in a zoom stereoscope, with continuous magnification from 2-15 times. The data were collected according to the PLP-method, e.g. by dividing the landscape features into patches (areas), lines and points, and interpreting them on separate transparencies.
Classification codes of each landscape feature were also interpreted on a separate transparency, to facilitate forthcoming scanning of the transparencies to a digital data base. The classification system is based on what was supposed to be detectable in CIR aerial photographs, and what was regarded to be of importance for the biodiversity. It has a hierarchical structure. The variables are further defined and described in Cousins and Ihse (1996).

4. Results
The classification system is presented in Table 1, Table 2 and Table 3.
The classification system for patches is divided into two main groups, open and forested land, each with four levels for interpretation. An additional fifth level incorporates the field inventory, i.e. ground truth and data on vegetation composition that are not detectable in the aerial photographs. The elements are described according to land use and management, nature type and moisture, physiognomy, vegetation cover and selected species in the tree and bush layer.
The system is not truly hierarchical, meaning that all `boxes' in the system are systematically filled. This is because all sublevels are not relevant for all higher classification levels (e.g. the moisture level is of little relevance for the nature type `fen'). The attributes in level 3 could be combined with all those in level 2, those in level 4 with level 3, etc. This makes the system very flexible.
An example will be given from the land use class grassland, where 68 patches of grassland were subdivided into 404 observations, considering all four levels and taking into account about 30 different variables (Table 4). The level land use/management is not regarded relevant for linear and point elements. Today small biotopes are not generally managed in the agricultural landscape. Thus, the classification system for linear elements and points consists of three levels only, and one additional for field control. The interpretation accuracy is presented in Table 4 and Table 5 and a confusion matrix in Table 6.
The interpretation accuracy is based on about 500 patch elements (50% of all interpreted) and about 600 linear and point elements (90% of all interpreted), that field were checked. The interpretation accuracy for both patch elements and point and linear elements is calculated for each variable and each level. The first level of patches has an interpretation accuracy of 97%, the second 95%, the third 99% and the fourth level 98%. The interpretation accuracy for small biotopes are calculated together for linear and point objects and are for the first and third levels 97%, for the second 96%.

Afforested arable land or grassland are most difficult to interpret, especially if they are afforested by deciduous trees. Solitary coniferous trees are most underestimated, due to low contrast, low reflectance and small perimeter and area, while the interpretation accuracy of old solitary deciduous trees is high even in this scale (100%). There is very little confusion between the classes on generalised level, 6-9% between arable land and cultivated grasslands and ley pastures. Time estimated for interpretation is 1.4-2.8 km2 per hour.

5. Discussion
The method described here can be used for mapping landscape features and biotopes in an agricultural landscape, and it can also be used as a base for estimating the potential biodiversity.
The generalised classification level can be used for planning purposes at scale 1:50000, the more detailed levels for base-line inventories for planning at scale 1:10000 or 1:20000, the scales commonly used in Sweden. We have assumed that there is a link between landscape and biotope diversity and the diversity of species, and that landscapes with many biotopes and landscape elements with a dense ecological infrastructure provide a good base for sustained biological diversity.

We will give some comments on the chosen variables regarding potential high biodiversity: The age of forest is distinguished, as old forest have higher biodiversity than young or clear-felled. The `hagmarksartad' deciduous or mixed forest is a forest succession of abandoned semi-wooded meadows and pastures. It has been shown to be important for preserving the biodiversity as it contains species of both the forest flora, old stagheaded trees and especially species from old continuously managed grasslands (cf. Skånes, 1996).
Riparian zones, shown to have high biodiversity values, are valuable for the ecosystem function, and are severely affected by changes. Afforested arable land may increase the biodiversity when planted with deciduous trees, while afforestation on unfertilised grasslands will always have a negative effect on the biodiversity.
The urban land of houses and gardens, often with old hedges and old deciduous trees, are important for the dispersal of species, mainly as stepping stones, as they may contain the only features of semi-natural vegetation in the large-scale intensively farmed agricultural plains.
Earth banks, stone walls, farm tracks, ditches, heaps of stones and ponds have more or less broad zones of grassland vegetation (dry, mesic or moist), which have high potential for biodiversity, compared to the surrounding arable fields.

Colour infrared aerial photographs give the best information on vegetation. Earlier investigations have shown that CIR aerial photographs are superior to black and white photographs (Ihse, 1978), and superior to visual interpretation in satellite images (Cousins, 1997). Old black and white photographs have earlier been used in change detection studies in Sweden (Ihse and Lewan, 1986; Ihse, 1989; Skånes, 1996), as the Swedish regular aerial photography coverage stretches back to the 1930s, and provides a unique archive for studies of landscape change. The knowledge from these studies was used in the choice of change-susceptible variables in the classification system.
The quality of the mapping is highly dependent on the quality of the photographs; good contrast in colours, optimal date for photography and the time lap between interpretation and photography dates. For a base-line mapping the time between photography and mapping ought to be as short as possible. However, we found that even ten years old photographs could be used. Only about 20 elements, of more than 1000 patches, points and linear elements controlled in field, have been removed or radically changed. This gives an uncertainty in the assessment of accuracy, but there will not be any important differences in the overall interpretation accuracy. We suggest that also comparatively old photographs can be used for a base-line inventory, as data capture by interpreting CIR aerial photographs is still much better than by interpreting more recent black and white photographs. The combination of variables for landscape and biotope mapping used in this study, gives a very high interpretation accuracy, 95-99%, compared to earlier investigations of vegetation mapping, where 81-87% were considered satisfactory (Ihse, 1978). The estimated time for interpretation of landscape elements (1.4-2.8 km2/h) is comparable with mapping vegetation types by CIR aerial photographs (1-2.5 km2/h) (Ihse, 1978). It is four to five times quicker than field based inventories. Borg (1977) estimated field based vegetation mapping to be about 2 km2 per day. 6.

Conclusion
Colour infrared aerial photographs are well adopted for base-line inventories of landscape, with respect to biodiversity. The scale 1:30000 ensures that many landscape details and vegetation types can be interpreted, and management continuity and natural-cultural aspects can be taken into consideration. The classification system is flexible and encompasses landscape elements considered important for biological diversity in the agricultural landscape, such as patches of grasslands, deciduous forests, wetlands and small linear and point biotopes. The assessment of accuracy (95-99%) shows that this method is very reliable. The method requires a certain amount of field inventories. The interpretation can pinpoint the areas with a potential for high biodiversity.