Object-based terrain classification as tool for improving the quality of the digital geomorphological maps: a case study in Retezat-Godeanu Range (Southern Carpathians).

Torok-Oance, Marcel Francisc ; Ardelean, Florina-Minodora ; Voiculescu, Mircea 等

Abstract: The geomorphological mapping is a difficult and time-consuming task, but the geomorphological maps are necessary tools in many practical activities. The conversion of the already done analogue geomorphological maps into digital form could be a useful alternative. The result of this conversion process presents some errors which are observed during the integration of geomorphological data with other spatial data, like Digital Elevation Models (DEMs) or high resolution satellite images. The most frequent errors are the inaccurate location and the imprecise dimensions of the landforms. These errors are related both with the degree of generalization of mapping, due to scale of the analogue maps, and with the traditional mapping methods. This study presents a method for the correction of the errors which consists in the semiautomatically extraction of the main specific landforms from DEMs in an Object-based image analysis (OBIA) environment and the use of them as check-points for a visual cross-check against the digital geomorphological map. This method allowed a more precise and faster way to remove the errors and to obtain a corrected digital geomorphological map.

Key words: geomorphological map, DEM, landform classification, OBIA

1. INTRODUCTION

The work is part of a project that aims to achieve medium-scale digital geomorphological map coverage for the entire Romanian territory (Badea et al., 2008). Conversion of the analogue geomorphological maps in digital format is not a simple process of digitization. Geomorphological mapping and editing maps with classical methods lead in many times to errors that are observed when these data are integrated in a geodatabase with other spatial data. The most common and severe errors are related to the precise position of the landforms, as well as to the area occupied by them and to their level of detail that depends mainly on the scale of analogue map. At present the correction and updating of the geomorphological maps are done manually, in a difficult and time consuming process, by visual analysis using digital topographic maps, DEMs and airphotos.

Because of the laborious error correction process, in most of the cases the conversion of analogue geomorphological maps in digital format is made without validating them. Most often the maps are just scanned and georeferenced. There are many such geodatabases consisting of geomorphologic raster maps, available to users through web-GIS services.

In order to simplify the validation of the digital geomorphological maps we proposed a new method: the use of terrain classification method in an Object-based image analysis (OBIA) environment for the extraction of the main specific landforms from the DEM and their use as benchmarks to make faster corrections of the geomorphological maps.

The method could be improved by using more accurate DEMs and by developing new algorithms for the automatic extraction of other specific landform from DEMs. Another important step for the further research is to provide strict definition of landform segments (Minar and Evans, 2008).

2. THE STUDY AREA AND DATA

2.1 The study area

The study area is located in the western part of the Southern Carpathians, in the central sector of the Retezat-Godeanu Range (Fig. 1). The elevation of this mountainous area ranges from 650 m a.s.l, to 2509 m a.s.l. The major elements of the study area's relief are represented by planation surfaces, glacial and periglacial landforms.

2.2 Data

We used a 30 m horizontal resolution SPOT-HRS DEM which represents the most suitable DEM, Romanian--wide cover, for the geomorphological studies regarding the accuracy of the altitude data and the quality of the surface morphology representation. (Torok-Oance et al., 2010).

The geomorphological maps at scale 1:200000 with national--wide cover were made in analogue form by the Institute of Geography, Romanian Academy, during the period 1977-1990. For our approach we have used the map sheet L34-XXIX (Orsova), scanned and georeferenced in the national reference system Stereo 1970.

3. METHODS AND RESULTS

In order to develop a more precise and faster way for removing the errors and to obtain a corrected digital geomorphological map we followed more steps (Fig.2). First we used the surface analysis tools implemented in ArcGIS 9.3 software to derive several land-surface models from the initial DEM: slopes model, mean curvature model, maximum curvature model, minimum curvature model and runoff model. The land-surface models represented the input data in the object-based terrain classification in order to detect specific landforms which are characteristic for this study area.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

The OBIA requires first the transformation of the raster models from the pixel level to the object or spatial primitives through the image segmentation process. Both segmentation and classification were made with Definiens v.8 software. The segmentation was made at different levels (scales) in order to obtain different size land surface segments. The detected specific landforms were: ridges, planation surfaces, glacial cirques and steep slopes (scarps). Based on the acquired knowledge about these landforms morphology we made the multi-resolution segmentation using the curvature model in the case of ridges and glacial cirques and the slope model in the case of planation surfaces and scarps.

The classification process of the objects obtained through segmentation took into account more factors: the mean slope and the runoff for the planation surfaces (Torok-Oance et al., 2009), the mean slope for the scarps or headwalls, the mean curvature for the ridges and glacial cirques (Eisank et al., 2010) (Fig. 3). All the classification results were validated with accurate field data.

The visual cross-check of the classification results against geomorphological map allowed to detect easily the errors and to make the correction faster (Fig. 4).

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

4. CONCLUSION

The method presented in this study demonstrates that the object-based terrain classification could be a useful tool for the assessment and improvement of the accuracy of the geomorphological maps. The results of the classification must be first validated with field data because some untested algorithms could lead to misclassifications compromising the final result. The limitation of the method is that the transferability of the classification system to another area is steel difficult and requires the modification of the classification's parameters.

5. ACKNOWLEDGEMENTS

This work was supported by CNMP, project number PNII--GEOMORF 32-140/2008 and by CNCSIS--UEFISCSU, project number PNII--IDEI 1075/2009.

6. REFERENCES

Badea, A.; Moise, C.; Dana, I.F.; Olteanu, V.G.; Mamulea, A.A. & Badea, R. (2008). Complex Method for Developing the Digital Geomorphologic Map of Romania at 1:200.000 Scale. 0049-0050, Annals of DAAAM for 2008 & Proceedings of the 19th International DAAAM Symposium, ISBN 978-3-901509-68-1, ISSN 1726-9679, pp 025, Editor B. Katalinic, Published by DAAAM International, Vienna

Eisank, C.; Dragut, L.; Gotz, J. & Blasche, T. (2010) Developing a semantic model of glacial landforms for object-based terrain classification--the example of glacial cirques, ISPRS Archives, Vol. XXXVIII-d/C7

Minar, J.; Evans, I. (2008) Elementary forms for land surface segmentation: The theoretical basis of terrain analysis and geomorphological mapping, Geomorphology, Vol. 95, Issues 3-4, 15 (March 2008), (236-259), ISSN 0169-555X

Torok-Oance, M.; Ardelean, F.; Onaca, A.; Voiculescu, M. & Urdea, P. (2010). The Evaluation of Different Types of Digital Elevation Models for Geomorphological Applications in Mountain Areas. 1413-1415, Annals of DAAAM for 2010 & Proceedings of the list International DAAAM Symposium, ISBN 978-3-901509-73-5, ISSN 17269679, pp 0707, Editor B. Katalinic, Published by DAAAM International, Vienna

Torok-Oance, M.; Ardelean, F.; & Onaca, A. (2009). The Semi automated Identification of the Planation Surfaces on the Basis of the Digital Terrain Model. Case Study: The Mehedinti Mountains (Southern Carpathians). Geographical phorum, Vol. VIII (8), (5-13), ISSN 1583-1523