TECHNOLOGY FOR CORINE Land Cover data updating

K.Charvát, J.Hodač, S. Holý
Help Service Mapping
Brdičkova 1916, CZ-15500 Praha 5, Czech Republic
e-mail: charvat@topol.cz, http://www.topol.cz

INTRODUCTION

Environment protection belongs to human greatest concerns. These efforts requires accurate information about environment. One of the richest, most objective and accurate information sources are Remote Sensing (RS) data linked together with Geographic Information Systems (GIS). RS and GIS provide many opportunities that have not yet been used in the Czech Republic.

CORINE Land Cover data, assessed in the Czech Republic in the framework of this international project in 1994 - 1995, are one of the sources for GIS foundation at the Ministry of the Environment (ME). This data has not been widely used for purposes of governmental authorities yet, because they are, in our opinion, to rough. Land cover mapping was done with a minimum unit corresponding to 25 hectares, which seems to be insufficient resolution for our country.

For this reason and after detailed consideration of all chances offered by the original CORINE project, we decided in the framework of a tender for the project Methodology for Updating of CORINE Land Cover Database for the Czech Republic and Its Application in 1997 to offer the development and verification of methodology for detailed land cover layer production. Such a layer should suit needs of the ME better.

At this point, we would like to emphasise our understanding of significance of the CORINE Land Cover data for the purposes of EC planning. So that we developed a technology for digital method for the CORINE database updating, which is also described in this paper.

PROJECT RESOURCES AND TOOLS

This paragraph describe in greater detail data sources chosen for the project as well as software tools selected for data production.

SATELLITE DATA

The CORINE updating project was accompanied with two other project at the ME, where remote sensing data were required data sources. So that, demands for all projects were collected and following 7 multi-spectral images were jointly bought to minimise expenses:

Attention was paid to the selection of particular images. To minimise purchase risks resulting from cloudy weather, black-and-white Quick Look overview scenes were obtained for the period from May to September 1997. Unfortunately, no suitable LANDSAT TM image covering the western part of Bohemia was found among them and therefore it was decided to purchase two high quality images from the previous year.

ORIGINAL CORINE LAND COVER DATA IN SCALE 1:100,000

The ME provided for needs of technology design and data updating the original data created in 1994-5. Geographic co-ordinate system (lambda-fi) of Krassovsky ellipsoid was used for this data. The original layer consists of about 34,000 polygons for the whole area of the Czech Republic.

TOPOGRAPHIC MAPS

Topographic maps are additional sources of information during satellite image interpretations. Topographic maps in scale 1:50,000 were sufficient in most of areas for updating CORINE database in scale 1:100,000. Topographic maps in scales 1:25,000 were proven to be excellent sources for detailed interpretation. Rather complicated map sheet system disqualified tourist maps, which provides also required information. Some up-to-date Basic Maps of the Czech Republic in scale 1:25,000 and 1:10,000 (in complicated regions) were also used. We should also mention that Basic Maps in lesser scales were insufficient for satellite data interpretation purposes.

SOFTWARE TOOLS

We used our TopoL GIS system for all data processing procedures. This system was able to accomplish all tasks required for data preparation, interpretation as well as its final check. In final phase of export data for ARC/INFO, data were exported into ArcView ShapeFile format using TopoL and PC ACR/INFO was used to build coverages.

TopoL GIS is a general LIS/GIS software developed in our company, which is running on a standard PC computers in Windows environment. It allows creation, maintenance, analyses and presentation of geographical data. TopoL integrates classical vector-based GIS approach with image processing tools. It is able to handle vector and raster data equally well and its functionality includes vector data overlays, database analyses as well as image processing and raster classification. Extensive care was focused on topology and support for topographic structures of vector data. Digitising capabilities of TopoL are easy to use and the system is able to solve topologic relations in real-time while data are digitised. TopoL can import and export data in many industry-standard formats.

DIGITAL METHOD FOR CORINE DATA UPDATING

UPDATING SCOPE

The whole project also included the assessment of changes in CORINE Land Cover database in scale 1:100,000. The original method of data capture was based on the analogue interpretation of enlarged LANDSAT TM and SPOT XS satellite images. Boundaries and land cover codes were drawn on slides and then digitised. We should notice that such a method is time-consuming and highly dependent on interpreter care in comparison with interactive data capture on a display. So that, we selected full digital method for data processing. TopoL GIS system was able to display satellite data as well as original CORINE vector data in one window allowing direct data updating on screen.

DATA PREPARATION

Syntheses in false colours were prepared for LANDSAT TM satellite data in following two spectral band combinations:

band 4 (red) band 5 (green) band 3 (blue)

band 4 (red) band 3 (green) band 2 (blue)

Syntheses in false colours were prepared for SPOT XS satellite data in the following spectral band combination:

band 3 (red) band 2 (green) band 1 (blue)

If both satellite data were available, combined syntheses were prepared in the following spectral band combination:

band 4 TM (red) band 5 TM (green) band 2 XS (blue)

For easier data manipulation and transfer, all raster data were cut into map sheets in scale 1:100,000. If LANDSAT TM images were only available for a particular map sheet, images were not geometrically transformed and were interpreted in their local co-ordinate system (because of decreased quality after transformation). SPOT XS images were supplied transformed into S-JTSK. Because of discovered inaccuracies (about 50 meters), these distortions were eliminated by transformation using similarity method. If both images were available, LANDSAT TM syntheses were geometrically transformed using affinity method (with10 m of pixel size) and SPOT syntheses were transformed using equalisation of their positions with LANDSAT images.

All original vector data were transformed into S-JTSK co-ordinate system. Data were also cut into map sheets in scale 1:100,000. If LANDSAT TM images were only available for a particular map sheet, original data were transformed into local co-ordinate system using 10-15 identical points. Reverse key set were used for transformation of updated data into S-JTSK co-ordinate system. After all transformations, coincidence of images and vector data was checked.

CORINE CLASS DEFINITION

Class definition and class code assignment were used in conformity with CORINE Land Cover - Technical Guide. About 40 basic classes were identified at the whole territory of the Czech Republic. We should note that classes 211 (non-irrigated arable land) and 212 (irrigated arable land) were included into one class. Detailed identification of these classes would have required additional satellite images and spectrozonal aerial photos, which was beyond expenses covered by this project. Hop-fields (plentiful in our country) were included into the class 222 (fruit gardens and orchards).

DATA INTERPRETATION

The interpretation team were made familiar with interpretation key, most frequent errors in the original data as well as principles for data updating. Well-organised work-course and supplying single interpreters with required information (data sources, updated surrounding map sheets if available) were essential for proper project completion. For interpretation process technology defined several rules for screen scale, line updating (removal, adding, minimum area size checking) and area definition points updating (point removal, adding and class code updating).

UPDATING PROCESS CONCLUSIONS

Following experience were identified after data updating. New line capture were done three times more frequently than line removal. Total amount of polygons in updated data increased by 10-15%. Checking of the minimum area size rule during interpretation were very useful, because insensitive removal of some areas in final checking phase was avoided. The final project report included the comparison of original and updated data using coloured printed maps.

METHOD FOR DETAILED LAND COVER DATA CREATION

This part of the project was devoted to the development and verification of methodology for detailed land cover layer production. Such a layer should suit needs of the ME better for following types of land cover: built-up areas, industrial zones, waste areas, water resorts, railways and roads and building sites.

For these reasons, we designed a methodology for the creation of land cover layer with minimum size of area of 4 hectares. This rule is not strictly defined, but it results from the effort to provide land cover categories demanded at the ME. This rule is flexible and depends of identified land cover category. For example, it is very difficult to classify larger forests by this great detail (especially in our country with colourful forest cover), but it is necessary to find all forest areas larger than expected minimum size of 4 ha. Similarly, it is difficult to identify industrial zones in large cities, but it is necessary to find them in countryside. Road and railways should be identified in greater detail than in the CORINE project, where only corridors wider than 100 meters were classified.

For the reason of detailed identification of railways, roads and rivers, we defined separate layer of polylines which define roads (I. and II. class) , railways and important rivers. In our methodology, this layer is created in the first phase of interpretation. Afterwards, it serves as basis for polygon layer creation, because communications forms countryside element borders with different land cover very often.

DATA PREPARATION

To verify detailed land cover identification method, 16 topographic map sheets in scale 1:100,000 were selected for the data creation. Three different regions of 4-6 map sheets were chosen:

    1. Northwest Bohemia - this region was selected for frequent changes (coal mine activities) and its high ecological exposition;
    2. Southwest Bohemia - this region was selected for colourful countryside with many different land cover regions and mountain resort of Šumava
    3. South Moravia - region affected by flooding in last year and completely covered by both LANDSAT TM and SPOT XS

Satellite data were prepared in similar manner like for CORINE database updating, so that syntheses were used for both parts of the project. Topographic maps in scale 1:25,000 and Basic Maps in 1:10,000 scale were also used. Original CORINE data, topographic maps in scale 1:50,000 and tourist maps were used as optional data sources.

DATA INTERPRETATION

Layer of communications and rivers were created in the first phase of interpretation. Satellite image data were accompanied with vector layer of communications created in 1993-4. We could not expect to create this layer without such an additional data in complicated regions (city surroundings). Following objects were identified: highways, first class roads, second class roads, other important roads, railways, main rivers for each water resort, other important rivers.

CORINE class definition were used in this part of the project as well. But we defined three new classes, which are found often in our country: cottage settlements, agriculture buildings, hop-fields. The interpretation team were made familiar with interpretation key and principles for data creation. Technology defined for interpretation process several rules for screen scale, display settings, line creation (snap size, join to polyline layer, minimum area size checking, joins on map frame), area definition points capture (capture method, class code supply rules), final check rules.

UPDATING PROCESS CONCLUSIONS

Following experience were identified after detailed data creation for 16 map sheet of scale 1:100,000:

  1. Total amount of polygons is three times higher than in CORINE Land Cover database. For the extent of created data, this increased size does not constitute any problem. We can expect, that data size for the whole territory of the Czech Republic will be larger than 150 MB. For real data use should be such data divided into several parts.
  2. Topographic maps in scales 1:25,000 were proven to be excellent sources for detailed interpretation. Detailed additional data sources (e.g. Basic Maps in scale 1:10,000) will increase time consumption of interpretation process - amount of necessary maps for one interpreted map sheet will be higher than 100 maps.
  3. Combined syntheses of LANDSAT and SPOT are the most useful for the interpretation process. SPOT images increase data resolution significantly. The most import advantage of combined syntheses is its multi-temporal characteristic (LANDSAT image from August combined with SPOT image from May), which makes the interpretation of agriculture areas easier. LANDSAT syntheses are very useful for forest and agriculture land classification and SPOT images are useful for urban zones.

CONSLUSION

This paper describes technology, which was used in last year for updating CORINE Land Cover data for the Czech Republic. This project was accomplished for the Ministry of Environment of the Czech Republic. Described technology is based on satellite data interpretation and CORINE Land Cover vector data updating using TopoL GIS system.

The project included also design of technology for the creation of detailed data based on CORINE data structure. Such data should help the Ministry and other governmental offices to identify land objects of smaller size than expected in CORINE Land Cover instructions. The project also included verification of this technology for areas with minimum extent of about 4 ha. Data layers were prepared in detailed accuracy (1:50,000) for selected regions - Northwest Bohemia, Southwest Bohemia and flooded regions of South Moravia.

There are two main results of the project: document describing all updating methods and processes as well as updated CORINE Land Cover data for the Czech Republic and detailed data for selected regions of interest. All data (including detailed layers) conform to data structure defined in the CORINE Land Cover project, which consist of 84 classes, of which about 40 could be found in the territory of the Czech Republic. All data provided to the Ministry were in the S-JTSK coordinate system and data formats for ARC/INFO and ArcView.

REFERENCES

  1. Červenka, V., Charvát, K., 1990: Land-use Classes Discrimination with Satellite Images Based on Spectral Knowledge. ISPRS Symposium (Commission VII) -- Global and Environmental Monitoring Techniques and Impacts, Victoria.
  1. Charvát, K. et al., 1995.: TopoL for Windows - New Tool for Remote-Sensing. Proceedings of the 17th International Cartographic Conference, ICC Barcelona, Spain.
  1. Charvát, K. et al., 1996: Technology for Digital Vector Land-Use Map Preparation Based on Remote-Sensing Data. International Archives of Photogrammetry and Remote Sensing. Volume XXXI, Part B2, Commission II, Working Group III (Proceedings of the 18th ISPRS Congress), Vienna, Austria.
  1. Charvát, K., Červenka, V., 1990: Pseudocolour photomaps production using neural networks. International Symposium on Thematic Mapping from Satellite Imagery, Bulletin of French Committee for Cartography, No.127-128.
  1. Heynmann Y. 1990: CORINE Land Cover - Technical Guide.