Mountains are areas of high relief having distinct changes in terrain slope and thus require a three-dimensional representation for spatial modeling. Maps and GIS in general treat the world as if it were flat (plain land); this two-dimensional view leads to incompatibility in appropriateness of GIS application between that for level land and mountains. Mountains have some very specific features that need heterogeneity. Unfortunately, it is not the practice, apart from few modifications made in GIS applications for mountain areas. The study illustrates integration of such factors in ecological zoning, land suitability classification and probability mapping, such as for land erosion. ‘Agricultural suitability/capability’ and ‘Vulnerability to soil erosion’ are two such examples.
Maps represent geographical area on the planar surface whereas due to slope differences in the mountains the actual surface area is greater, the discrepancy in area calculation leads to overestimation of population density (biotic). Watershed, having modal average slope in 20-25o ranges, has geographical area of 94 km2 as against the actual surface area of 105 km2. A large surface with very steep slope (e.g. cliff face) is reduced to negligible area on a map. Similarly, linear calculations and buffering become erroneous for mountains using simple GIS techniques. Shorter distances than actual are recorded from maps; such errors could lead to underestimation in the cost of road construction when using GIS methods. Pronounced slope and shadow effect cause much problem in interpretation of remote sensing image data. In the mountain areas, research is still needed on how to use shadow information (total shadow minus topographically caused shadow) for land cover classification. Thus, GIS has an important role in improving digital image processing in mountain areas.
Thermal differences due to aspect affects the limits of flora and fauna distribution. It is well known that the upper limit of any vegetation type is bound to be lower on the northerly aspects making them drier. Such factors were used in bio-climatic zoning in our study.
The Slope aspect can be used to our advantage in conjunction with geological structure to assess the trend of rock beds, which may be useful for planning of roads and other constructions. This approach was adopted in the finding causes of landslide zoning, which is rather recent, is being focused in various parts but none have been quite comprehensive. Inversion of temperature is another phenomenon dictating the land use in mountains. Settlements refer ridge tops and slopes than valley bottoms, as they are colder at night and foggy for longer period of the day. Accessibility is not easily defined in the hills as in the plains and this is an important factor for land use and infrastructure.
Spatial complexity of mountain regions makes extrapolation very difficult. The same locational theories of hierarchical distribution of settlements do not apply to these regions as in the plains. The complex interaction between various factors lead to this heterogeneity. Added to these are the irregularities of data and difficulties of data collection and fieldwork in these areas.
So far there are two approaches developed towards modeling the three-dimensional complexity of the mountains, these are the ‘DTM approach’ and the ‘Landscape approach’. The mountain specialists require a truly 3-GIS. However, incorporation of the approaches into knowledge based GIS is yet to be developed. Digital Terrain Models (DTM) helps in portraying the 3-dimensionsionality of the mountains, but overlaying procedures on DTM are yet not satisfactory. Better use of aerial photography is called for in this field. Global Positioning Systems (GPS) is used with GIS for mapping new (unmapped) features. With DTM, GPS and appropriate weightage attributed to slope, aspect and elevations in GIS it is possible to improvise the present inaccuracies of GIS applications for mountainous terrains.
In recent years, the growing concern over the environmental degradation of mountain ecosystems ahs gradually placed mountain issues in environmental and political agenda (Heywood et al., 1994). An example of this growing interest was the formulation of a Mountain Agenda for the UN in 1992. Several organisations are developing regional and national scale monitoring programmes in which GIS plays a central role. Hence it is very important to assess the applicability, or degree of accuracy in such GIS applications. Heywood et al., (1994) underestimated the uniqueness of GIS application in mountain environment, stating “that there is nothing unique about the character of GIS applications in mountain areas” although they add “nevertheless, the use of GIS in mountains require some special considerations”.
The primary criterion that distinguishes mountains from other land surfaces is its significant positive relief. Slope, aspect, complexity and heterogeneity of climate, vegetation, faunal and land use distribution patterns are all outcome of this primary factor, relief. The paramount effect of relief is nowhere more spectacular than in the Himalayas, and this is where our study is based. The physical characteristic that best defines mountains is their three-dimensionality. It is this three-dimensionality that poses the greatest challenge for modelling these regions using GIS, for the simple reason that most GIS and the data they incorporate still treat the world as if it were flat. GIS applications started in the West and gradually, through government or semi government organisational aid and private enterprises, spread to the underdeveloped countries. In India, particularly in the mountain areas the use of GIS has been mainly organisational. Therefore the fields of GIS application here have been land use analysis, hazard assessment, natural resource management, visualisation of terrain, ecological and hydrological modelling etc. Most of the work has been application of conventional GIS methods without much thought to the effect of relief and probable errors. The sphericity of the earth has long been recognised and assigned a role in geography but the topological nature of the surface has not received as much attention (Coffey, 1998).
This paper deals with the source of errors encountered in GIS application to mountain environment using examples from some case studies and references. The paper also goes further to evaluate some of the options for improvising GIS applications for mountain areas.
Errors in GIS applications for mountain environment
“Error and uncertainty are common features of cartographic information, so it is hardly surprising that these aspects are also present in digital version of analogue maps” (Openshaw et al. 1991). GIS is a powerful tool in spatial analysis and its power is obvious in that it has the potential to dramatically increase both the magnitude and importance of errors in spatial databases. Burrough (9186) identifies three main groups of factors that govern errors that may be associated with spatial data processing. These are: (a) obvious sources of error (human), (b) errors arising from natural variations or from ordinal measurements, and (c) errors arising through processing.
The errors associated with GIS applications specific to mountain regions, our current topic of discussion, are from the second group. No map is entirely error free, but errors due to natural variations in mountainous terrain are significant. Positional error, aerial interpolation error and linear measurement error increases with slope.
Error in area calculation
Calculation of population density (e.g. bio-density) is exaggerated by underestimation of area for the sloping terrain. The actual surface area on a sloping terrain is greater than the geographical area depicted on the map that represents surface of the earth as a flat surface. The actual area is the product of the geographical area and Cosine of the average slope angle of the place. The smaller the unit size the greater the accuracy (Fig. 1).
Fig. 1: Descrepancy between actual area and geographical area depending on slope of land
Error in linear calculation: Similarly the linear distances over sloping land are greater in reality than depicted on the maps. The actual distance in this case can also be calculated by finding the product of the mapped distance and the Cosine of the average slope angle traversed. Thus distances derived from GIS give wrong information, such as shorter road distances and lower drainage density. As a result cost deduced for transmission line or road construction would be an underestimated value, even if the difference is not too large.
Positional error: As a result of shorter distance derived from GIS for a mountainous terrain, the placement of points according to linear measurement becomes erroneous when applied on sloping terrain. For example, if sampling points are to be located 1 kilometre from a given point and their positions are worked out with GIS considering a two-dimensional surface, the resultant position would be further away than a kilometre due to slope involved. For that matter how would the measurement of point patterns be carried out using nearest-neighbour analysis or quadrant sampling for mountain areas? Such questions become more relevant as detailed GIS analyses are more frequently being carried out on larger scale maps.
Buffer zone error: Since linear distances are underestimated a buffer zone generated by GIS that ignores the land slope would enclose a wider area than actually intended. The proportion of additional distance brought into the buffer zone would be directly proportional to the slope angle of the land.
Ineffectiveness of straight-line accessibility derivation: In plain land is derived simply by multiple buffer generation from the road (line) for an area and from a settlement centre (point) for evaluating accessibility of a village to different service centres. This simple method does not hold true for mountain regions, where slope and other physical impediments must be taken into account when working in scales larger than 1:50,000. Some accessibility maps have been made taking the contours into consideration (Bournay & Pradhan, 1994; Trapp, 1995) but existence of un-traversable slopes do not seem to have been considered. Accessibility passages only come below 40o slopes for foot-tracks and within 10o for motorable roads. Therefore the algorithm for accessibility mapping in mountain areas needs to be far more intelligent.
Veregin (1995) very well defines the types and sources of error. “Errors result from inadequate data acquisition methods that do not truthfully capture the real world phenomena. This conceptualization has much in common with statistical treatment of error in terms of bias and precision… That is, encoded values represent approximations… Errors can therefore be reduced through the use of more refined data acquisition techniques …and methods of repeated sampling.” Apart from this is inherent variability of geographical phenomena. Veregin states that in case thematic attributes, methods for measuring and documenting error can be differentiated in terms of the scale of measurement as approximate, whereas, for interval and ratio data, error can be measured in terms of the mean deviation between actual and observed values at a sample of locations. This provides an error index analogues to the root mean squared error (RMSE) for elevation data.
Effect of Slope factor
Although for most purposes, the earth may be regarded as a uniform surface, it does have variability in its relief.. Similarly, any conceptual surface may be examined in terms of its degree of relief. Relief is commonly expressed by of gradient, which identifies the change in the vertical dimension as the horizontal dimension changes (Coffey, 1981).
The difference in area calculation does not affect land use since land revenue measurements only consider horizontal land surface, and for all practical uses (construction of house, of tanks, agriculture) a quasi horizontal land is put to use. Vertical component of the land is not utilized. However, for forest cover the actual area perimeter and land surface is kept in record that does not match (is greater than) the geographical area depicted on the map. Therefore when a forestis delineated in GIS its area is underestimated. The problem becomes acute in the case of cliffes or very steeply sloping land that may be the hanitat of particular plant species.
In our study area, while analysing the land use changes, it was observed that the new agricultural extensions during 1963 and 1993 were predominantly (26.5%) on the 20o – 30o average slope areas since the lower slopes were already under cultivation and higher slopes are not preferable.
Effect of topographic aspect
South facing aspects in the northern hemisphere are sunnier (receiving longer period of solar radiation), and therefore warmer and drier. As a result the upper limits of occurrence of any fauna or flora are higher on the southern aspects than the northern aspects. Even the snow-line is lower on the northern aspects and the snowmelt regime too is different. The southern aspects being sunnier and warmer have less forest cover and are preferred for agricultural purpose, whereas northern aspects are more forest clad. Southern aspects are more prone to forest fires too. Such is the impact of aspect on the local moisture regime. Therefore, aspect must be considered along with edaphic, macro-climatic and infrastructural factors in land use planning for such areas. In our study area the agro-climatic regions were defined taking into consideration both elevation and aspect, and based on that the land suitability classification was carried out. Agricultural extension during 1963 to 1993 were mainly (28%) on the south-western slopes and eastern slopes of the watershed, and marginal on the northerly slopes. There was less extension on the south facing slopes due to non-availability of land. In soil fertility evaluation study slope and aspect derived from DTM have provided two of the extrapolation factors in soil carbon and nitrogen contents (Schmidt, 1991) mapping.
The same geological structure has different facet according to the aspect. A north-east dipping rock strata has dip slope on the northeast aspect and anti-dip slope on the south-west aspect. This fact has significant implication on slope-cut road construction. Consequently the anti-dip slope with inward sloping rock beds provides a safer side for road construction as dip slopes would be prone to slope failure. In the Himalayas, where due to thrusting the general orientation of the rock strata is more or less uniclinal, this method is applicable. This rule was applied in our study in combination with lithology and structure, in the Garhwal Himalayas for (i) finding the major causes of landslides; and (ii) for assessment of the appropriateness of the proposed road route in a watershed. In the former study it was found that lithology and construction aggravate landslides more than the dip direction of the slopes. Whereas in the latter study where the general dip is 30o E of N at 45o to 60o slopes, and the proposed road is mostly planned on the right (west) bank of the river traversing unsafe dip slopes where slope-cut for road construction will result in rock slip and landslides. Only 20% of the proposed route lies on comparatively safe slopes (with rocks dipping away from the road cut) for road construction. Moreover the route runs across a major active landslide. Re-routing of the road on the anti-dip slopes as much as feasible and using dip slopes only where the slope is below 25o is recommended.
Heterogeneity of mountain landscape: Mountain ecosystems are very heterogenous in the spatial distribution of any feature, therefore interpolation, or extrapolation of data leads to largely erroneous results. Extrapolation can be very risky without in depth understanding of the inter-relationships of the variables involved in the extrapolation, and the risks of inaccurate or improbable extrapolation increase with the number of variables and the complexity of the environment. The data ffor almost all aspects of mountain ecosystems are very heterogeneous in their length and frequency of record, spatial coverage and availability. The three-dimensional complexity of the mountain can exacerbate these problems to an extreme extent (Heywood, et al., 1994).
Data sources and related problems: The main data sources for mountain regions are similar to that of the lowlands but due to its inherent heterogeneity higher resolution of remotely sensed data and greater sampling density for ground truthing is necessary. One new and reliable source of digital locational data is Global Positioning System (GPS) which is particularly helpful in representing break-line features characteristic of mountain areas. However, locking-on at least 4 satellites even in mountainous terrain is often problematic when working in confined areas, such as gorges (stocks and Heywood, 1994) so GPS may not prove to be effective in certain areas.
There are some inherent problems related to mapping from remote sensing and aerial photographs. Geometric distortion arises when either an aerial photo or satellite imagery is used to record a mountainous region. The displacement between observed and true map locations of ground feature has been estimated as + 9 pixels, i.e.270m for 30m resolution (Hill & Kohl, 1988). Fukushima (1988) cites root mean square (RMS) errors of 101.6 m in mountain areas, compared to errors of 3.4 m in areas of low relief using SPOT data. DTM have been used to assist in the correction of data in such cases (Haefner & Hugentobler, 1988). One of the methods of error rectification in satellite images is shadow matching.
Data models for mountain environment: The proliferation of GIS is explained by its unique ability to assimilate data from widely divergent sources, to analyze trends over time, and to spatially evaluate potential environmental impact caused by development (ICIMOD, 1995). Therefore, GIS should be capable of assimilating the unique factors controlling the land use pattern in mountains.
To develop an understanding and appreciation of the optimal locations for settlements and other land uses, it is also essential that the GIS should be able to identify areas affected by the likelihood of natural hazards, whose distribution is influenced by complex interactions between local climates, human activities, soil, bedrock and vegetation characteristics (Hewitt, 1992).
A widely researched GIS application in mountain environments is in landslide hazard zoning (Rengers et al., 1992) van Westan, 1992, 1993, 1994) where either qualitative or quantitative analysis is performed. The analyses have taken various factors of mass movement like geological structure, lithology, hydrological conditions, vegetation, angle of slope and aspect of slope into account. A deterministic model of landslide hazard using GIS has become rather popular.
To capture the three-dimensionality complexity of the mountain areas two broad groups of approaches have developed in GIS, one is the DTM approach and the other is the landscape approach. The first involves the use of digital terrain models to provide categorization of zones of elements of a mountain area according to slope, elevation and aspect. The latter uses landscape units, constructed from a systhesis of environmental data, which reflect the character (structure, sustainability and responsiveness) of an area rather than its physical form (height, shape and exposure) alone (Heywood et al 1994) .
The DTM approach: Digital Terrain Models (DTMs) are mathematical models to graphically represent the elevation of the terrain derived from elevation data (z values at x,y). Information on intermediate height, aspect, slope, shape, radiation incidence, hill shadows, visibility and cut-and-fill estimates can be derived from these models. Data for DTM comes from the following sources viz., ground surveys, topographic maps, satellite imageries, aerial photographs and GPS. There are two main structuring approaches in DTM; (a) grid-based and (b) triangular irregular network (TIN) method. The former is less computationally intensive than the TIN method. First developed by Peuker (1978), the TIN method is better capable of working on randomly located height data, incorporating break lines, and reduces the data volume. The use of DTMs to help in modeling dynamic mountain ecosystems (Walsh et al., 1994) involve probability mapping for pedictable natural hazards using multi-criteria analysis. Weight is attributed to various factors, some of which like slope angle and aspect are derived from DTM. In satellite images of mountains, land cover should be derived from total shadow comprises minus the topographic shadow (Paracchini & Folving, 1994) and the integration of DTM with GIS will help in developing accurate semi-automatic surface cover mapping from satellite imagery.
The landscape approach: The landscape approach incorporates behavioral approach of the landscape is based on the principles of landscape ecology (Forman & Gordon, 1986). Permanent and dynamic conditions of the land are differentiated. The method of delineating landscape units requires prior knowledge and use of Knowledge-based systems. One basic problem associated with this approach is that of assessment of stability. Stability is generated by the overlay of landscape units and is not easily interpreted by non-experts in the field. This is posing problems in using this approach in policy making.
Fig. 3: Erosion impact of proposed motor road in Pranmeti watershed
Even if total representation of all the complexities of mountain area is not possible, the major and relevant elements characterising the terrain such as slope, aspect,elevation, macroclimate and actual accessibility should be taken into consideration in GIS applications for mountain areas if realistic results are desired. The importance of field data and ground thruthing should not be underplayed while using advanced techniques like remote sensing or GPS. The user must not be complecent with a two-dimensional GIS when dealing with mountain areas. The active use of DTM, with overlay and other spatial analyses being carried out on the three-dimensional model is strongly suggested, and appropriate weightings should be given to different topography-related factors for the mountain regions. Ideally the mountain scientists require a ‘truly 3D GIS’ (Stocks & Heywood, 1994), and possibly ERDAS Imagine is a step in this direction. Understanding the role and inter-relationship of various factors in ecological setup and land use is essential for spatial analysis. The TIN DTM using statistical techniques (spatial moving averages, kriging and other interpolation methods) is more advisable for mountain areas unless a very high-resolution grid DTM is used (which creates heavy data files). Draping and ‘cut and fill’ methods should be correct and visually satisfactory. The landscape approach is more advisable in agro-economic studies otherwise the DTM approach is preferred fore spatial analysis of mountain regions. Finally greater resolution and accuracy of data is required in mountain regions due to their spatial complexity.