Open access publications in the SLU publication database

Abstract

It is expected that the ongoing climate change will have a strong influence on the Earth’s
vegetation and cause the advancement of treelines towards the poles and up to higher
elevations. In the Swedish mountains, changes in the positions of alpine treelines have
already been reported, and major changes due to changing climate are predicted for the near
future. Remote sensing techniques have considerable potential to improve the monitoring of
spatially complex treeline ecotones, which are likely to show site dependent responses to
changing climate. Aerial photos provide the longest temporal record of remote sensing data
for studying the historical treeline changes. High spatial resolution and the possibility of
interpreting photos in three-dimensions are the main strengths of aerial photos. The National
Inventory of Landscapes in Sweden (NILS) is a nationwide environmental monitoring
program, which provides sampling infrastructure for monitoring treelines over the Swedish
mountains using high spatial resolution remote sensing data.

The aim of this project was to study the feasibility of visual interpretation of aerial photos for
monitoring changes in treeline ecotones (transition zones between closed forest and treeless
alpine vegetation). More specifically, our aim was to compare possible methods for change
detection using recent and historical aerial photos, and to demonstrate a method that is
applicable to the Swedish mountains using the NILS sampling infrastructure. We also wanted
to evaluate how the technical properties of the aerial photos, particularly the photo scale,
affect the change detection. The results were used for evaluating the potential and costs for a
more comprehensive change detection study including a greater number of NILS squares.

We compared interpretation methods in three 5×5 km NILS squares and studied in total six
5×5 km NILS squares for changes. Colour infrared aerial photos from two points in time
(1975–1980 and 2002–2004) were used. The evaluation of the technical properties of aerial
photos is important as the recent photos have been taken from a height of 4600 m (photo
scale approximately 1:30,000) and the historical photos from 9200 m (1:60,000). The
difference in photo scale could introduce significant errors to the change detection. Therefore
the effect of photo scale was studied in two experiments in a separate study site, which had
been photographed within a few days from both flying heights.

We assessed three visual interpretation methods: complete cover mapping (polygon
interpretation), sample plot method and transect method. To avoid recording false changes
and to minimize the effect of different photo scales, we interpreted the recent and historical
photos at the same time using scanned photos and digital photogrammetric workstations. The
first method, polygon interpretation, was tested by comparing the current NILS
interpretations for the central 1×1 km square within the 5×5 km NILS squares with the
historical photos, and updating polygon borders and attributes if changes were observed. In
the second method the vegetation variables were interpreted for circular sample plots (20 m
radius). We used digital land cover data and GIS techniques for defining a buffer zone
between the forest and the treeless alpine areas in order to focus the interpretation efforts on
the treeline ecotone. The distance between the systematically sampled plots was 250 m within
the treeline buffer. In the third method the variables were interpreted along subjectively
selected transects across the altitudinal treeline transitions. Transects were also used for
interpreting individual trees within the treeline, of which some were measured in the field.

Based on the analyses we can recommend the sample plot method for interpreting tree cover
changes within 5×5 km NILS squares. The main advantage with the method is that it can be applied to any NILS square overlapping the treeline ecotone, and that the sampling can be
done objectively. A treeline buffer zone based on digital land cover data provides one
possible method for pre-stratification. Delineation of polygon borders is not needed and small
sized sample plots are easier to interpret concerning vegetation changes than larger polygons.
This is particularly the case when changes are small. The polygon interpretation using change
control mapping can also be a cost-effective method for change detection within the 1×1 km
squares. The greatest advantage is that in addition to tree cover changes, it can provide
information on the spatial patterns (landscape structure). However, the small changes in the
borders are subjective and difficult to interpret. The transect method can also provide
characterization of treelines, particularly in relation to elevation, but the selection of transects
within NILS squares can be difficult. Furthermore, the result cannot be generalised to larger
areas similar to the method based on probability sampling.

We conclude that only tree cover changes can be interpreted with sufficient accuracy,
because of the relatively small scale historical photos. Larger scale photos and more
extensive field control would be necessary to interpret changes in small trees and shrubs. The
experiments with photo scale show that the greatest source of uncertainty is related to the
different photo scale of the recent and historical photos. The comparison of two independent
interpretations made at different scales would lead to detection of substantial false changes,
particularly if interpretations would be made by different interpreters. However, we think that
concurrent comparison of digital aerial photos from two points in time provides meaningful
change detection for tree cover although some errors due to scale effects should be expected.

The interpretation of six NILS squares by the sample plot method suggests that tree cover has
increased within the treeline ecotone in three of the squares. The results demonstrate the
potential of visual interpretation of aerial photos for treeline change detection. The
information on tree cover changes could supplement field based monitoring of treeline
elevations. Therefore, we consider that there is potential for a more comprehensive study to
examine tree cover changes in a greater number of NILS squares in the Swedish mountains.
Then, however, more emphasis should be given to the panchromatic black and white photos
from the 1960s, which have been taken from approximately the same flying height (4600 m)
as the most recent CIR photos. We estimate that a large scale study, involving the
interpretation of approximately 50–60 NILS squares for three points in time would require a
budget of about 1 million SEK. Airborne laser scanning and larger scale aerial photos are
potential data sources for monitoring future changes in treeline ecotones.