4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4):
Problems, Prospects and Research Needs. Banff, Alberta, Canada, September 2 - 8, 2000.

Automated landform classification using DEMs:

a conceptual framework for a multi-level, hierarchy of hydrologically and geomorphologicaly oriented physiographic mapping units

GIS/EM4 No. 198

Robert A. MacMillan
David H. McNabb
R. Keith Jones

Abstract

Resource inventory systems are at the heart of virtually all efforts to manage natural resources data. They provide the foundation for responsible management and planning and the basis for applying management knowledge and experience over space and time. In many jurisdictions, resource inventories are undergoing change in response to improvements in technology and concerns about costs and effectiveness of current methods. A new generation of classification and mapping systems is emerging that makes use of improved mapping and modeling technologies, while continuing to draw upon the existing foundation of classification and mapping protocols. A critical feature of most resource inventory systems is the definition of fundamental classes of spatial entities that can be recognized both conceptually and physically as discrete spatial entities. Most mapping systems rely on recognition of landform or landscape based spatial entities to define this basic structural framework. Landforms provide a physical framework for describing the environment and the ecological processes and interactions that occur within the environment (Rowe, 1979). Numerous investigators have examined ways in which digital elevation data can help define landform-based units to act as basic spatial and structural entities for soil, terrain or ecological maps. These investigations have demonstrated many useful concepts. To the best of our knowledge no proposed system of automated terrain or ecological mapping has advanced beyond research investigation to achieve routine, operational use. This paper describes a conceptual design for creating landform-based spatial entities from digital elevation data to support multi-level, hierarchical integrated natural resource inventory. The geomorphic/hydrologic spatial entities defined by our procedures are intended to provide a framework for mapping more specific ecological entities of interest.

Keywords

Automated landform classification, ecological modeling, multi-level, hierarchical ecological classification, integrated hydrological, ecological and geomorphological spatial entities

Introduction

This paper describes the initial stages of a project to define a tool kit to process digital elevation data to define a multi-level hierarchy of hydrologically and geomorphologically oriented spatial entities. These landform-based spatial entities are intended to act as a basic structural framework for different kinds of natural resource inventories and their interpretations (e.g. maps of soils, terrestrial eco-systems, wildlife habitat, forest productivity, carbon sinks). The paper outlines an initial conceptual design and provides examples of its application to different scales of digital elevation data.

Problem statement

The problem addressed by this paper is how to define landform (and landscape) based spatial entities at a variety of scales to provide a meaningful spatial framework for describing and interpreting ecological conditions and modeling ecological processes. Since most ecological processes of interest involve interactions between spatial entities, it is necessary to define how the spatial entities interact and otherwise relate to one another. Since both ecological processes and management activities take place over a variety of scales, it is desirable to define a multi-level hierarchy of spatial entities, with explicit procedures for nesting and aggregating within the hierarchy. And finally, since the entities, and the procedures used to define them, are designed to be applicable to a wide range of locales and conditions, the entities must be simple and must be defined using flexible and robust classification procedures. We have elected to adopt a heuristic, rule based approach implemented using fuzzy logic.

Background

Situation analysis

Natural resource managers increasingly must address many complex and inter-related management and planning challenges at local to global scales. These include a myriad of inter-twined social, economic and ecological drivers. Examples include growing globalization, increased competition, greater market-awareness and demands around sustainability issues, the necessity for cost-effective operations, more accurate forecasting capability; compliance with standards and environmental certification, an expanding number of international and national monitoring obligations, and various commitments related to sustainable resource development. Resource inventory and interpretation systems are at the heart of virtually all of these contemporary resource issues by providing the foundation information for responsible management and planning. They provide the basis to apply and extend management knowledge and experience over space and time, increasingly with the aid of spatial planning and decision support models.

Against an ever-growing list of requirements, resource inventories in many jurisdictions are undergoing significant change. Many of the changing expectations noted above drive the need for a new generation of classification and mapping systems. The composition, structure, and arrangement of these multi-scale ecosystem complexes influence many business objectives including resource allocation, biodiversity, protection, productivity, and conservation. While drawing upon a foundation of existing classification and mapping efforts, the new systems must be much more dynamic, adaptive, and incremental; be able to infer nested ecosystem-landscape-watershed and regional level processes; be fully integrated, multi-scale, and mappable; and be of known accuracy for reliable interpretations and predictions. Furthermore, resource managers and policy makers require the ability to analyze how ecological classification and mapping systems relate to and affect policy options, land tenure, guidelines and regulations, planning alternatives, and management scenarios in terms of real world issues such as cost, wood supply, forest and conservation practices and fire, insect and disease protection strategies.

The conceptual design

Geomorphological-Hydrological spatial entities

The conceptual design aims to adopt, adapt and integrate a variety of previously described approaches for delineating fundamental spatial entities. We are trying to build on past successes to create a tool kit to support automation of conceptual approaches that are integral to most current manual inventory systems. The design adopts the concept of a multi-level hierarchy of spatial entities that is fundamental to manual ecological land classification procedures (SBLC, 1969; Wiken, 1981) and has been embraced by many existing inventory systems, particularly in the forestry sector (Boyacioglu, 1974; Archibald et al., 1984.) It accepts the importance of explicitly defining hydrological connectivity between adjacent spatial entities. Such connectivity has been assigned manually for entities such as the terrain mapping units (TMUs) of the ITC system of geomorphic mapping (Meijerink, 1988) and computed automatically from digital elevation models following procedures outlined by various authors including Band (1986a,b,c; 1989a,b,c), Heil (1980), Hutchinson (1988) and Miller (1984).

Our design incorporates both hydrological and geomorphic criteria to define and delineate spatial entities at all scales. It recognizes that spatial entities delineated solely on the basis of geomorphic shape are insufficient to meet many current inventory and modeling needs as they lack the information required to establish linkages, interactions and flows between spatial entities. Similarly, landform units defined solely on the basis of hydrological criteria, such as those created by the intersection of a complementary network of stream channels and drainage divides (Band, 1986b), are incomplete in that they do not differentiate areas of different surface morphology or relative landscape context. Our approach uses cell to cell hydrological connectivity to compute a complete description of the drainage topology for any given DEM in order to define a mesh of intersecting stream channels and drainage divides. A separate classification of spatial entities is defined using a combination of measures of local surface morphology and relative landform position as per MacMillan et al., (2000). The drainage topology mesh is then overlaid on the geomorphic classification and the resulting intersection defines spatial entities that have characteristic shapes and landform positions as well as explicitly defined hydrological connectivities.

A multi-level, multi-scale hierarchy of spatial entities

Many current natural resource inventory systems define a hierachy of spatial entities, some explicitly and some implicitly. Typically the lowest one or two layers of the hierarchy are used as the basis for operational planning and management. Higher level spatial entities often form the basis for summarizing data to support generalized planning or policy development and, more recenntly, regional, provincial and national reporting.

Most ecological or soil maps produced by human interpreters delineate repeating patterns of landform entities (e.g. toposequences, associations or catenas). They tend to only describe, but not explicitly map, the spatial distribution of the more uniform landform components that make up this repeating cycle. Interestingly, automated procedures for classifying landforms using digital elevation data tend to be more successful when used to delineate individual landform components (e.g. ridges, valleys, toeslopes) than to recognize complex landform asemblages (e.g. hummocky, rolling, undulating) composed of repeating patterns of landform components. Our system aims to classify and map both individual landform components (lowest level features), and the more complex landform types created by characteristic assemblages of individual landform elements (higher level features). It utilizes both bottom up agglomeration and top down stratification to support recognition of these characteristic landform types.

Figure 1. Illustration of a proposed hierarchy of landform entities

Description of the conceptual hierarchy

We are choosing to focus on defining four levels of hiererchical entities (Figure 1), although almost any number of levels could be envisaged and implemented. The choice of four levels is purely arbitrary. It is based on historical use and personal experience where four different levels of spatial discretization have proven useful for different applications while more than four levels have often been found to lead to un-necessary effort and confusion. Of these four, the most significant are the bottom two, herein termed landform elements and landform types.

Landform elements

Landform elements represent the lowest level of the hierarchy and are meant to portray spatial entities that exhibit a relatively restricted range of morphological attributes and, by extension, an equally restricted set of internal characteristics (moisture status, soil type, lithology). The original LandMapR program segmented DEMs into 15 different landform elements that were in turn grouped into four major classes of upper, mid and lower slopes and depressions. We have revised this classification somewhat (Appendix 1) to improve the recognition of depressions and to add 3 new classes to recognize a number of important riverine entities, specifically active channels, channel banks or levees, and flood plains.

Landform types

Landform types represent assemblages of repeating patterns of landform elements with a characteristic pattern and scale of repetition. These are meant to equate to the toposequences, catenas or associations that are the most frequently mapped bio-physical spatial entities used for managing forested land. We are still experimenting with identifying how many landform types it is necessary and feasible to define and with the most appropriate methods for describing and defining them. The initial proposal is for 34 classes in 6 categories (Appendix 2). Each number in Appendix 2 represents a proposed class of landform type. The total number of potential classes is 68 if both dissected and non-dissected versions of all classes are defined. This initial set of proposed landform types is adapted from landform types used to describe soil polygons in the recently completed digital soil map of Alberta (CAESA, 1998). It also attempts to be compatible with the SOTER major landform types defined for use at smaller scales with global and national soils and terrain databases (FAO, 1995).

Physiographic systems

Physiographic systems are the geomorphic equivalent of land systems. We propose to describe and map physiographic systems by agglomerating similar, and adjacent, landform types defined at the immediately lower level. Agglomeration will be guided by reference to generalized spatial entities created by applying a revised version of the LandMapR program (MacMillan et al., 2000) to small scale, coarse resolution DEMs.

Physiographic regions

Physiographic regions are the geomorphic equivalent of eco-regions. They are delineated by applying the LandMapR program to very coarse resolution DEM data sets (1-5 km) and then agglomerating the resulting spatial entities into the equivalent of landform types, at the level of physiographic regions. It is expected that final delineations of Physiographic systems and physiographic regions will utilize manual interpretation to complete and perfect the processes of agglomeration and generalization. Human vision and interpretation is still considered superior to automated processes at this scale of generalization. Additionally, the small number of spatial entities typically defined at these small scales do not justify complicated and time-consuming automated processing, when manual delineation can be achieved faster and with greater accuracy.

Implementation of the conceptual design

The revised LandMapR model

The concepts are being implemented using an expanded and revised version of a heuristic, fuzzy-logic based landform model (LandMapR) previously developed for classifying individual farm fields in support of precision agriculture (MacMillan et al., 2000). This model was itself an extension of a seven unit hillslope model described by Pennock et al., (1987, 1994). The LandMapR model has been used with some success to classify landscapes over a variety of scales (1:5,000 up to 1:1 million) and DEM resolutions (5x5 m up to 5x5 km). This initial success led us to consider adapting it to classify a hierarchy of landform entities operating over a full range of spatial scales (Figure 2).

Figure 2. Illustration of results of applying the original LandMapR model to DEMs of successively finer resolution.

The role of hydrological topology in the classification

Calculation of cell to cell flow topology is a critical component that differs in some important ways from most existing implementations. A conventional D8 alogrithm (Wilson, 1996; Jenson and Dominique, 1988) is used to compute cell to cell downslope flow directions but special routines are included to assign appropriate flow directions to flat cells and to record the full topology of flow into, and out of, depressions. Most programs that use the D8 algorithm consider depressions to be "spurious pits" that represent artifacts of the grid representation of continuous terrain (Hutchinson, 1989). These programs typically include procedures to remove all depressions by filling them to their overspill elevation prior to computing final flow directions for fully integrated flow.

Our approach considers depressions to be real and significant landscape features. The pit removing procedures store information on the location, volume, area and overspill locations of all pits, prior to removing them. Pits are removed by reversing the flow drections of cells along the flow path from the selected pour point to the depression centre. Data on the original flow directions and upslope areas counts are retained in a separate location for all cells that are affected by changes in flow direction. Also stored is an indication of the hydrological conditions (mm of runoff) at the point in time at which the depression became full and overspilled. The final data set can display both initial and final flow networks representing disrupted flow into initial depressions in the DEM and completely integrated flow with no depressions. More importantly, it can also display the flow network at any state between initial, disrupted flow and final, integrated flow by selecting a desired value for mm of runoff required to fill depressions or by specifying a threshold value for area or volume of depressions to remove.

Establishing landform context

The program uses hydrological connectivity and hydrological features to establish contextual attributes of landform position that are critical features of the classification. We consider depressional catchments to define the most appropriate local window within which to evaluate landform context to aid in landform classification. Catchments with flow into local closed depressions effectively define the dimensions and shape of one ridge to ridge (or trough to trough) repeat cycle in the landscape. Small, local depressions also represent important locations where surface runoff is temporarily impounded, or at least slows down, leading to increased deposition of sediments and elevated levels of infiltration into the soil. Regional context is established by classifying coarser resolution DEMs into upper, mid, lower and depressional landform elements using the LandMapR program. This helps to establish whether a given local entity occurs on a plateau, plain, basin or hillslope in the regional context.

Establishing hydrologic response units

The final networks of channels and ridge lines defined for completely integrated flow are used to delineate a skeleton for hydrological response units according to procedures suggested by Band (1986b). The intersecting divide and channel network defining the basic skeleton is overlaid on the landform elements defined by application of the LandMapR program. The resulting spatial entities draw attributes of geomorphic shape and landfrom context from the landform elements and hydrological connectivity from the hydrological skeleton. The hydrological skeleton can also be used as a basis for aggregating spatial entities from the sub-catchment level to successively larger and more generalized catchment entities.

Classifying regions into landform types

Defining landform types is an exercise in pattern analysis and contextual classification. A critical component of this activity is defining the window, or region, within which to evaluate context. We are experimenting with both bottom-up agglomeration and top-down subdivision.

In the bottom-up approach, each depressional catchment is regarded as defining an appropriate search window or classification region. The horizontal and vertical dimensions of the landscape encompassed by the catchment are used to provide an indication of the wavelength (horizontal ridge to ridge distance) and amplitude (pit to peak elevation change) of the catchment entity. These are the principal diagnostic criteria for differentiating different landform types (e.g. hummocky versus rolling). A suite of summary statistics describing the percent distribution of morphological indices (e.g. slope length, gradient, aspect, relief, landform elements) within each catchment provides additional information to assist in assigning each catchment to a defined landform type. Additional data on the shape of the catchment entities (e.g. long, straight versus short or arcuate) can be extracted from the ridges and channels defined for each catchment. A fuzzy classification approach is used to apply a heuristic rule base to the morphometric data for each catchment entity. All grid cells within a catchment are assigned a value corresponding to the most likely fuzzy classification. Catchments below a specified minimum size may be dissolved and assigned the classification of the most similar adjacent catchment.

In the top-down approach, the LandMapR model is applied to a DEM of significantly coarser resolution than that used to define the lower level landform elements. The landform elements of the coarser resolution DEM (500 m to 1 km) are reclassified into the four simple classes of upper, mid, lower and depression. Each cell in the finer resolution DEM contains a pointer to the cell in the coarser resolution DEM that directly contains it. This pointer establishes whether the finer resolution cell (and the catchment to which it belongs) occurs within an upper, mid, lower or depressional landscape position in the larger context. This can help differentiate level plains, for example, from level plateau or inclined-undulating from simple undulating or rolling landform types. A procedure we have not yet tested is to go back to the landform element classification data and over-ride the landform type classification (e.g. hummocky) with the original landform element classification for instances of level or depressional landform elements that exceed some minimum defined extent.

Defining higher-level landform entities

Higher level landform entities are defined using coarser resolution DEM data sets in a manner that is otherwise similar to that used to classify landform elements and landform types. Physiographic systems are amalgamations of landform types where the spatial extent of entities to be amalgamated is guided by the extent of landform elements and landform types computed using coarser resolution DEM data (100 m to 500 m). Similarly, physiographic regions are defined by applying the LandMapR classification procedures to very coarse resolution DEM data sets (500 m to 5 km) and then amalgamating the resulting small scale landform elements into the equivalent of landform types at the regional scale.

Discussion

We are still in the process of establishing, applying and evaluating the classification system and the rules used to apply it. Much of the software is still incomplete or only partially functional. The numbers and kinds of spatial entities and their definitions are also not fixed. None the less, a few observations are still possible.

Firstly, it has become increasingly clear that DEM data with horizontal dimensions of 5-10 m and relative vertical accuracies of +/- 0.3 m are required to properly represent, describe and classify landform entities at the level of landform elements and landform types. DEMs with larger horizontal grid spacings (e.g. 25 m, 100 m) and poorer vertical resolution (1-10 m) tend to generalize and abstract the landscape too much, while DEMs with a finer mesh (e.g 1-4 m) provide too much local noise that disrupts the landform signal of interest.

Next, it is also clear that coarser resolution DEM data (100 m to 5 km) already available for most portions of the world can prove useful for establishing global to regional context for classifications developed using finer resolution data. For most jurisdictions, high resolution DEM data, such as that obtained using airborne lasers will likely be available for only a limited area of specific interest. Coarser resolution ETOPO5, GTOPO30, DTED, or (soon) SRTM data sets (Figure 1) will be needed to provide data on the regional context within which the finer resolution data occur. It is conceivable that existing global data sets could be processed in their entirety to establish a single global to regional scale context within which to nest detailed classifications produced using high resolution data from LIDAR or other sources.

In our experience, it is not the absolute accuracy of the DEM that most influences the results of landform classification. Rather, it is the degree to which point to point relationships in the DEM describe a smoothed abstraction of the landscape at the scale of interest to the viewer or classifier. We have found it necessary to filter most DEM data sets two or three times with mean filters ranging in size from 7x7 to 3x3 in order to reduce the effects of local noise and bring out the longer range signal. An increasing number of authors are recommending the use of wavelet functions or fast Fourier transforms and their inverses as more effective tools for reducing noise and bringing out signal in DEM data sets (Herrington and Pellegrini, 2000). We agree with the need to smooth and abstract DEM data sets prior to classification and intend to investigate more effective tools for doing this.

The landform-based spatial entities defined by our procedures lack information on climate, vegetation and the geology or lithology underlying them. Our entities are defined purely on physiographic and hydrological criteria. It will be necessary to add information from other sources to create complete ecological units with comprehensive information on climate, vegetation, parent material, drainage and soils. It is our hope and belief, however, that the landform-based units will offer a reasonable first approximation of a useful, stable framework of spatial entities for many different types of ecologically oriented mapping. Intersection with other themes of information may require subdividing some of the initially defined landform entities, but the basic spatial pattern is expected to remain valid and useful. After all, geomorphic shape and context are the main determinants in defining the spatial extent of observable physical entities that humans seem to want to recognize, describe and manage.

Conclusions

We are in the process of developing a conceptual design for an integrated natural resource inventory and information tool kit. One of the tools calculates a large number of terrain derivatives from digital elevation data. These derivatives are used to apply an automated classification of a standardized set of physiographic mapping units for a range of scales. The toolkit adapts and builds upon many existing concepts and classification procedures. It also incorporate several new concepts and approaches presently being researched.

We believe that a tool kit for processing digital elevation data to delineate a multi-level hierarchy of physiographic mapping units is possible, practical and commercially desirable. Restrictions related to the quality and availability of digital elevation data are rapidly disappearing. Recent technological advances make acquisition of very high-resolution digital elevation data (< 1 m horizontal and < 0.3 m vertical) for very large areas both feasible and affordable. Progress in developing techniques and algorithms for analyzing DEM data to automatically compute terrain mapping units complements the progress made toward increasing the availability and lowering the costs of DEM data.

Most previous efforts in automated landform classification from digital elevation data have focussed on classifying specific landform facets at a single scale of interest (Irwin et al., 1997; Fels and Matson, 1996; Franklin, 1987; Graff and Usery, 1993; Burrough et al., 2000; MacMillan et al., 2000; Pennock et al., 1987, 1994; Skidmore, 1990). Multi-scale classification is considerably less common but one ingenious approach has been described by Wood (1996). Our approach utilizes the concepts of expanding kernels described by Wood (1996) and Fels and Matson (1986) but implements them in a different way. Rather than simply expanding the dimensions of the search radius within a single DEM, we advocate creating a hierachical pyrimid of DEM data sets of decreasing spatial resolution and processing each DEM separately to classify landscape components at a variety of scales. We are also experiencing some success with recognizing higher level landform types composed of a repeating pattern and scale of landform elements. The work is still in its early stages, but we hope to produce a functional toolkit that will function as an add-on, or extension, to commercial GIS software to facilitate automated landform classification in support of a variety of ecological classification and mapping efforts.

Acknowledgements

Funding for portions of this work was provided by the Alberta Research Council, Forest Resources. The initial concepts and programs upon which the approach is based were developed under previous projects funded by Agriculture and Agri-Food Canada with the paticipation of Alberta Agriculture, Food and Rural Development and industry partners Agrium Inc, Westco Ltd. and Norwest Soil Research. The original classification scheme expanded upon by the current LandMapR program was largely conceived of by Dr. W.W. Pettapiece.

References used

Archibald JH, Ferguson NB, Haag RW, Hay WK, and O'Leary DJ. 1984. An integrated resource inventory of Deep Basin Area (NTS 83L). Physical land classification, forage inventory and ecological land classification. Vol. 1. Resource Inventory and Appraisal. Resource Evaluation and Planning. Alberta Energy and Natural Resources. ENR Report No. T78. Edmonton, AB. 339 p.

Boyacioglu E. 1974. Manual of biophysical land classification and evalaution of capability. Alberta Energy and natural resources, Edmonton, AB.

Baldwin J, Fisher P, Wood J, and Langford M. 1996. Modelling environmental cognition of the view with GIS. (in) 3^rd International Conference/Workshop on Integrating GIS and Environmental Modeling. Santa Fe, New Mexico, Jan 21-25, 1996. National Centre for Geographic Information and Analysis, Santa Barbara, CA, USA. CD-ROM.

Band LE. 1986a. Topographic partitioning of watersheds with digital elevation data. Water Resources Research 22:15-24

Band LE. 1986b. Analysis and representation of drainage basin structure with digital elevation data (in): Proceedings of the second international symposium on spatial data handling. International Geographical Union, Commission on Geographical Data Sensing and Processing and the International Cartographic Association. Seattle Wa. 437-450.

Band LE. 1989a. A terrain based watershed information system. Hydrological Processes 3:151-162.

Band LE. 1989b. Automated topographic and ecounit extraction from mountainous and forested watersheds. AI Applications 3:1-11.

Band LE. 1989c. Spatial aggregation of complex terrain. Geographical Analysis 21:279-293.

Band LE, and Wood EF. 1988. Strategies for large scale distributed hydrologic simulation. Applied Mathematics and Computation 27:23-37.

Burrough PA, van Gaans PFM, and MacMillan RA. 2000. High resolution landform classification using fuzzy-k means, Fuzzy Sets and Systems 113:37-52.

CAESA-Soil Inventory Working Group. 1998. AGRASID: Agricultural Region of Alberta Soil Inventory Database (Version 1.0). Edited by JA Brierley, BD Walker, PE Smith and WL Nikiforuk. Alberta Agriculture Food and Rural Development, publications. CD-ROM.

FAO. 1995. Global and national soils and terrain digital databases (SOTER): Procedures manual. van Engelen VWP, and Ting-tiang W. (eds.). World Soil Rsources report 74 Rev 1. Land and Water Division. Food and Agriculture Organization of the United Nations. ISSN 0532-0488. ISRIC, Wageningen, 128 p.

Fels JE, and Matson KC. 1996. A cognitively based approach for hydro-geomorphic land classification using digital terrain models, (in) 3^rd International Conference/Workshop on Integrating GIS and Environmental Modeling, Santa Fe, New Mexico, Jan 21-25, 1996, National Centre for Geographic Information and Analysis, Santa Barbara, CA, USA. CD-ROM (1996).

Franklin SE. 1987. Geomorphic processing of digital elevation models, Computers and Geosciences 13:603-609.

Graff LH, and Usery EL. 1993. Automated classification of generic terrain features in digital elevation models, Photogrammetric Engineering and Remote Sensing 59:1409-1407.

Heil RJ. 1980. The digital terrain model as a data base for hydrological and geomorphological analysis. (in): Auto-Carto IV. Vol. II. Proceedings of the international symposium on cartography and computing: Applications in health and environment. Reston. Va. American Congress on Surveying and Mapping. American Society of Photogrammetry. Falls Church. Va. 132-139.

Herrington L, and Pellegrini G. 2000. An advanced shape of country classifier: Extraction of surface features from DEMs. 4^th international conference on GIS and environmental modeling (GIS/EM4). Problems, Prospects and Research Needs. Banff, Alberta, Canada. September 2-8, 2000. http://www.colorado.edu/cires/banff/upload/126/> Accessed July 5, 2000.

Hutchinson MF. 1988. Calculation of hydrologically sound digital elevation models. (in): Proceedings of the third international symposium on spatial data handling. August 17-19, 1988. Sydney, Australia. International Geographical Union, Commission on Geographical Data Sensing and Processing. Department of Geography, the Ohio State University. Columbus, Ohio. 117-133.

Hutchinson M F. 1989. A new procedure for gridding elevation and stream line data with automatic removal of spurious pits. Journal of Hydrology. 106: 211-232.

Irwin BJ, Ventura SJ, and Slater BK. 1997. Fuzzy and isodata classification of landform elements from digital terrain data in Pleasant Valley, Wisconsin, Geoderma 77:137-154.

Jenson SK, and Dominique JO. 1988. Extracting topographic structure from digital elevation data for geographic information system analysis. Photogrammetric Engineering and Remote Sensing. 54:1593-1600.

MacMillan RA, Pettapiece WW, Nolan SC, Goddard TW. 2000. A generic procedure for automatically segmenting landforms into landform elements using DEMs, heuristic rules and fuzzy logic. Fuzzy Sets and Systems 113 (1):81-109.

Meijerink AMJ. 1988. Data acquisition and data capture through terrain mapping units. ITC Journal 1:23-44.

Miller SW. 1984. A spatial data structure for hydrologic applications. (in): Proceedings of the international symposium on spatial data handling. Zurich. Vol 1:267-287.

Pennock DJ, Zebarth BJ, and DeJong E. 1987. Landform classification and soil distribution in hummocky terrain, Saskatchewan, Canada. Geoderma 40:297-315.

Pennock DJ, Anderson DW, and DeJong E. 1994. Landscape changes in indicators of soil quality due to cultivation in Saskatchewan, Canada, Geoderma 64:1-19.

Pike RJ. 1988. The geometric signature: quantifying landslide terrain types from digital elevation models. Mathematical Geology 20(5):491-511.

Rowe JS. 1979. Revised working paper on methodology/philosophy of ecological land classification in Canada. (in): Applications of ecological (biophysical) land classification in Canada. Proceedings of the second meeting Canada Committee on Ecological (Biophysical) Land Classification. April 4-7, 1978, Victoria, BC. CDA. Rubec (editor). Ecological Land Classification Series No. 7. Lands Directorate, Environment Canada. Ottawa. p 23-30.

Sub-committee on bio-physical land classification (SBLC). 1969. Guidelines for bio-physical land classification for classification of forest lands and associated wildlands. D.S. Lacate (editor). Ministry of Fisheries and Forestry. Queens Printer for Canada. Catalogue No. Fo. 47-1264. Ottawa. 61 p.

Skidmore AK. 1990. Terrain position as mapped from gridded digital elevation data, Int. J. Geographical Information Systems 4:33-49.

Wood J. 1996. The geomorphological characterisation of digital elevation models, Unpublished Ph.D. Thesis, Department of Geography, University of Leicester, Leicester, UK. <http://www.geog.le.ac.uk/jwo/research/dem_char/thesis/index.html > Accessed 1999 May 1.

Wiken EB, Welch DM, Ironside GR, and Taylor DG. 1981. The northern Yukon: An ecological land survey, Lands Directorate, Environment Canada. Ecological Land Classification Series, No. 6. Minister of Supply and Services Canada. Ottawa. 197 p.

Wilson JP. 1996. GIS-based land surface/subsurface modeling: New potential for new models. In: 3^rd International Conference/Workshop on Integrating GIS and Environmental Modeling. Santa Fe, New Mexico, Jan 21-25, 1996. National Centre for Geographic Information and Analysis, Santa Barbara, CA, USA. CD-ROM

Appendix 1: Definition and description of the proposed landform elements

Appendix 2: Definition and description of the proposed landform types

Authors

Robert A. MacMillan, Ph.D., P.Ag., Principal
LandMapper Environmental Solutions, 7415 118 A Street, Edmonton, AB, Canada T6G 1V4.
Email: bobmacm@telusplanet.net, Tel: +1-780-435-4531, Fax: +1-780-436-1788.

David H. McNabb, Ph.D., Manager, Forest Resources
Alberta Research Council, P.O. Bag 4000, Vegreville, AB, Canada T9C 1T4.
Email: dhmacnabb@arc.ab.ca, Tel: +1-780-632-8264, Fax: +1-780-632-8379.

R. Keith Jones, M.Sc., Principal
R. Keith Jones & Associates, 2554 Bowker Avenue, Victoria, BC V8R 2G1.
Email: kjones@tnet.net, Tel: +1-250-598-2635, Fax: +1-250-598-2630.