**4. Processing techniques**

Various data like multi-temporal optical / thermal satellite images of Landsat ETM+, ASTER, ENVISAT systems and image processing / GIS techniques are used for the analysis, Figure 3. To compile the data from the various sources of information, the following problems had to be overcome:


Fig. 2. Climate - rainfall (B) vs hydrology water level(A) measurements of Macro Prespa

http://www.gdem.aster.ersdac.or.jp/ have been also included in the analysis.

The Landsat images that have been used in the current study have been acquired from the USGS / NASA: http://edcsns17.cr.usgs.gov/NewEarthExplorer/ . SRTM DEM data that are available from the USGS server at http://dds.cr.usgs.gov/srtm/ and ASTER-DEM

ENVISAT / MERIS, ASTER satellite systems are relatively new systems and their data are also evaluated as far as monitoring the lake water systems is concerned. Much emphasis is given to extract information concerning a variety of parameters like land cover change that influences (indirectly) lake water quality. The effects of anthropogenic land cover modification on lakes are also analyzed by incorporating detailed information about land surface properties derived from Earth Observation data. The data inventory that is prepared and reported in the current submission includes acquisition of land cover maps, geological maps, compilation of hydrogeological maps based on analysis of relevant data, compilation of digital elevation models, analysis of multi-temporal satellite data and land cover maps. The amount of the above information is reviewed and analyzed and the result of the

Various data like multi-temporal optical / thermal satellite images of Landsat ETM+, ASTER, ENVISAT systems and image processing / GIS techniques are used for the analysis, Figure 3. To compile the data from the various sources of information, the following

(A / B left) and Vegoritis lake (A / B right)

compilation is shown in the form of various maps.

Different scales of maps, charts and imagery.

**4. Processing techniques** 

problems had to be overcome:

Different coordinate systems.

The interpretation process is set up and activated upon receipt of each image. There are several factors which can influence the quality of RS images and can affect whether or not they are even worth acquiring, e.g.: *Weather, Smoke, Time, Sensors and Sensor performance.* Analysts should be familiar with these factors when interpreting the RS data.

Flawed images, or those having too much cloud cover, were rejected and alternatives sought. Only images with less than 10% cloud cover for the watershed areas or lakes of our pilot study area were used for the analysis. For example 5 out of the 7 collected images for the year 2011 shown in Fig. 4 were used while scenes D and H were rejected. This criterion significantly reduced the pool of images suitable for analysis but most years had at least one winter-spring / one summer-autumn image that met the criterion. From the pool of 38 images we selected 10 Landsat TM images, 2 Landsat ETM and 1 MSS images for reference spanning a ~ 35- year period (1974-2011). Another restriction refers to the scan line problem of the LANDSAT ETM scanner which made practically unusable these images for the analysis of Vegoritis lake. However, Prespa lakes are recorded properly and so Landsat ETM images with acquisition dates after the 2003 have been included in the analysis.

Fig. 3. Data and methods of analysis

All Landsat images were registered to the Greek Geodetic Datum of 1987 (EGSA '87) using the Landsat 17 January 2011 scene as a reference. The root mean square error (RMSE) for positional accuracy was generally less than 0.5 pixels (~ 10 m for Landsat TM). A nearestneigbour resampling scheme was used to preserve the original brightness values of the images. The most rigorous method of radiometric calibration involves the use of radiative transfer models to produce an absolute correction. However, some data required to perform such a calibration are unavailable for historic images. We tested simple radiometric correction techniques such as dark pixel subtraction, Sun angle correction and normalization of multi temporal images to a single reference scene but found these insufficient as the area should have: (1) similar elevation to the rest of the scene, (2) minimal vegetation, (3) a relatively flat surface, and (4) constant pattern or general appearance over time.

Monitoring Lake Ecosystems Using Integrated Remote

**5. Information gathering from remote sensing** 

climate will be felt within a lake.

The problems that are faced are related to: The fact that maps are not readily available There is lack of updated information

Digital data are in different scales or coordinate systems

attempted.

**5.1 Lakes** 

**5.1.1 Lake inventory** 

Sensing / Gis Techniques: An Assessment in the Region of West Macedonia, Greece 191

Various ratios of the Landsat bands have been calculated as these are related to SDT measurements. The TM3/TM1 ratio has been tested because previous investigators found it to be a strong predictor of SDT (Cox et al., 1998; Lathrop, 1992), but this was not confirmed by our analysis. All results have been stored to the raster database. Conversion of raster to vector of the lake water surfaces gave the opportunity to identify and store in the database the spatial variability of quantity / quality data of the lakes. GIS techniques have been used

to overlay the results obtained from the multi temporal analysis, Figure 14.

Fig. 6. Extracted surfaces of Macro Prespa lake, using the available map coastline.

High resolution of about 0.5 m ortho-photos available through the WMS service of Greek Cadastral Agency of Greece have been also used to acquire information and verify the results obtained from the analysis of Landsat data. The GIS system gives the opportunity of using the ortho-photos as a background while overlaying any type of GIS data and updating the information. All processing techniques have been applied using the TNTmips Image Processing / GIS S/W system (www.microimages.com ). Our case study is intended to give a recent example of the practical applications of RS and GIS to lake monitoring. The RS study is placed first, followed by the GIS study, and finally an integrated interpretation is

Lake physics plays a fundamental role in limnology as temperature structures, circulation patterns and turbulent mixing, all set the environment in which the biology and chemistry within a lake operate. It is also through physics that the initial impact of any changes in

Delineation of water bodies is essential for the estimation of the water balance of the area. Water authorities need to know date, location, extent and variations of these water bodies. The test area covers a broad region while the transnational Prespa lakes basin is included.

 Accurate measurements of surface areas of Macro – Micro Prespa lakes are lacking. The 17th of January 2011 Landsat image has been used to make an inventory of all the lakes of the region at a scale of ~ 1:50000, Figure 3. The lake water surfaces have been extracted using classification of infrared bands & conversion of raster to vector techniques. There is up to date information which is readily available in a digital format for the whole of the translational

Fig. 4. Available acquisitions of Landsat images for the year 2011: L5:1/January B. L5 17/January C. L7 10/February D. 30/March E. L5 7/April F. 23/ April G.2/June

Data fusion techniques have been used in creating enhanced images at ~ 15 meters resolution for the Landsat ETM images for the watershed areas of the lakes. Digital image processing techniques are applied, plus some necessary image enhancement. The next step in extracting image data for lakes used an unsupervised classification method based on a clustering K Means algorithm with 10 classes and 20 maximum iterations that were then aggregated to land and water classes. Raster to vector conversion techniques were used to outline the polygons of the water surfaces. Due to intense topographic relief of the area shadows were also classified as water surfaces especially in winter scenes. These were eliminated as they have small area extent using vector editing techniques. Auxiliary information was also used to guide AOI selection and correction and this included the use of vector (GIS) layers of map coastlines, bathymetric maps and sampling point locations, Figure 5. Each Lake\_AOI polygon was assigned a unique identification number and database fields so as to join the satellite data to the observation database.

The Lake\_AOI polygons were used to create the water-only images of the lakes. Multitemporal water-only images of Macro Prespa, Vegoritis and Ohrid lakes have been created and these have been stored as a raster database. Metadata information describing the image acquisition information was also included. Lake surfaces have then been further analyzed using unsupervised classification techniques. Self-organizing Map Classifier – unsupervised classification using neural network techniques proved quite effective in analyzing the lake water surfaces. Available SDT and Cl data are not readily available for the lakes of our region and so some ground measurements are used just for general verification purposes.

Fig. 5. Auxiliary information: Lake bathymetry of Macro Prespa lake (left), Vegoritis lake (right)

Various ratios of the Landsat bands have been calculated as these are related to SDT measurements. The TM3/TM1 ratio has been tested because previous investigators found it to be a strong predictor of SDT (Cox et al., 1998; Lathrop, 1992), but this was not confirmed by our analysis. All results have been stored to the raster database. Conversion of raster to vector of the lake water surfaces gave the opportunity to identify and store in the database the spatial variability of quantity / quality data of the lakes. GIS techniques have been used to overlay the results obtained from the multi temporal analysis, Figure 14.

Fig. 6. Extracted surfaces of Macro Prespa lake, using the available map coastline.

High resolution of about 0.5 m ortho-photos available through the WMS service of Greek Cadastral Agency of Greece have been also used to acquire information and verify the results obtained from the analysis of Landsat data. The GIS system gives the opportunity of using the ortho-photos as a background while overlaying any type of GIS data and updating the information. All processing techniques have been applied using the TNTmips Image Processing / GIS S/W system (www.microimages.com ). Our case study is intended to give a recent example of the practical applications of RS and GIS to lake monitoring. The RS study is placed first, followed by the GIS study, and finally an integrated interpretation is attempted.
