In the previous chapter, it was shown that the preservation of biodiversity and natural areas is of primary importance to some Park stakeholder groups, while the preservation of the Latvian cultural landscape is of primary importance to others. Furthermore, the preservation of the Latvian cultural landscape, while mandated as one of the original Park goals, is not high on the agenda of the current GNP Administration. As stated in the November 1999 Management Plan for Gauja National Park (Petersen 1999), “[t]he traditional rural landscape with its harmony and cosiness is of secondary significance after the primary – the ancient valley of the Gauja River and its tributary valleys.” It was shown in Chapter 2 that this traditional rural landscape, however, is of great importance to some stakeholders.
The focus of this chapter is to understand how the GNP landscape has changed from the late Soviet era (1985) through modern times (2002), and more specifically, in what ways the changes have influenced both the natural and the cultural landscapes in the Park. The changes in landuse and landcover (LU/LC) composition and pattern are analyzed using remotely sensed data in a Geographic Information Systems (GIS) context, and discussed in terms of their effects on GNP’s overall natural and cultural landscapes. The timing of these changes is examined to evaluate the effects of major events and land policy changes on the landscapes in the Park.
Research Goals
1) Characterize LU/LC at multiple time points (i.e., 1985, 1994, 1999, and 2002) in GNP through an assembled satellite image time-series by classifying landuse and landcover, and performing landcover change detections.
2) Examine how changes in composition and spatial structure of LU/LC have affected the protected natural and cultural landscapes within the Park throughout the study period.
Methods
The primary data used to characterize LU/LC in GNP at multiple time points is a time-series of four Landsat Thematic Mapper (TM) satellite images. Remote sensing techniques are used for these analyses due to the suitability of such data. Landsat Thematic Mapper (TM) satellite data were selected to monitor the landcover change, because of Landsat’s sufficient historical archive, areal coverage, and its appropriate spectral, temporal, and radiometric resolutions. Wessels et al. (2004) makes the point that the appropriateness of satellite image resolutions in addressing biodiversity conservation depends both on the patch sizes of important habitats, and on the specific types of land cover changes threatening the biodiversity in that region. Wessels et al. (2004) used MODIS (250 and 500 meter
resolutions) to map fractional landcovers in the Greater Yellowstone Ecosystem of the USA, and reported limited success in mapping smaller habitats (patches smaller than MODIS pixels), often important for biodiversity. Wessels (2004) used Landsat TM as reference data, which he claims was sufficient for the habitat mapping. Evans & Moran (2002) reports that high spatial resolutions, referred to as resolutions between 1 and 30 meters, are appropriate for detecting spatial patterns in spatially complex landscapes. In addition, Pedlowski et al.
(2005) successfully used Landsat TM to monitor deforestation in conservation areas in Rondônia, Brazil.
The image dates in this study were chosen to obtain temporal coverage from the late Soviet era through today, based on the availability of cloud-free images at leaf-on times of year. The specific images used were:
• 1985/06/25, Path 186, Row 20
• 1994/07/11, Path 187, Row 20
• 1999/08/02, Path 187, Row 20
• 2002/05/22, Path 187, Row 20
The 1985 and 1994 images were chosen because they straddle Latvia’s independence (1991). The 1999 image date was chosen because 1999 essentially marks the end of the land
restitution process in Latvia: close to 100 percent of claims had been settled by this date. Finally, the 2002 image gives insight into post-restitution processes taking place in GNP.
Fieldwork was conducted in the summer of 2001 to collect ground-control data for satellite image classification. A 1:100,000 scale topographic map of GNP and a set of orthophotos at 1:10,000 scale, taken in June and August of 1997, were used to navigate throughout the study site. The orthophotos were also used in conjunction with in-situ visualization to identify locations of at least 100 x 100m in area of uniform landcover type. During fieldwork, the exact landcover types at these locations were assessed, along with the collection of Global Positioning Systems (GPS) coordinates to define the center of these plots of uniform landcover. GPS data were also collected from nearby base stations, and Differential GPS was used to post-process the GPS data to increase the accuracy. At least 150 GPS points were collected and averaged for each location to further increase accuracy to
within +/- 3 meters. This accuracy was far more than necessary, because the goal was simply to be able to identify ground cover samples on the satellite images (30 meter resolution) for classification training purposes. In addition to collecting GPS data for all ground-control samples, the orthophotos were annotated using ESRI’s ArcView with a laptop in the field to double check the locations. Ground control data were collected in GNP in the summer of 2001 to support LU/LC classification. One-half of the ground control samples of each landcover type were used for training purposes, and the other one-half were used for accuracy assessment. The classification system was based on the USGS Land Use/Land Cover Classification System for Use with Remote Sensor Data, but tailored to the specific study area and the goals of the research. Samples for 16 landcover classes were recorded:
1. predominantly spruce forest 2. predominantly pine forest
3. predominantly coniferous forest (mixed pine and spruce only) 4. predominantly hardwood forest
5. mixed forest (conifers and hardwood) 6. wet hardwood forest
7. young hardwood forest 8. young pine forest 9. partially cute pine
10.partially cut coniferous forest 11.partially cut hardwood forest 12.shrubs
14.tall grass 15.short grass
16.Hogweed, a nuisance weed that GNP researchers are interested in mapping.
Note that predominance in the above classes is defined by at least 30 percent of the specified forest type. Upon classification of the satellite images (described below), the LU/LC classes were
aggregated into the following 7 basic landcover types through a recoding operation: 1. predominantly hardwood forest
2. predominantly coniferous forest 3. shrubs
4. crops/pasture/grass 5. built-up
6. wetlands 7. water.
For most analyses, these two forest classes were again merged through another recoding operation, as the primary goal was to consider basic landuse change categories relating to changes in the Park’s natural and cultural landscapes. For these analyses, the landcover classes used were:
1. forest 2. shrubs 3. crops/pasture/grass 4. built-up 5. wetlands 6. water.