SATELLITE MONITORING of ESTONIAN LANDSCAPES Kiira Aaviksoo and Andrus Meiner Estonian Environment Information Centre Mustamäe tee 33, Tallinn 10616 ESTONIA, [email protected], [email protected]
Apr 01, 2015
SATELLITE MONITORING of
ESTONIAN LANDSCAPES
Kiira Aaviksoo and Andrus Meiner
Estonian Environment Information CentreMustamäe tee 33, Tallinn 10616 ESTONIA, [email protected], [email protected]
BACKGROUND
1994: Estonian Environmental Monitoring Program
Landscape monitoring was not present in the program. The proposal for development of methodology for landscape monitoring was submitted
1996: Subprogram: Monitoring of Landscapes Landscape monitoring was organized by three monitoring
projects, incl. Remote Sensing of Landscapes
2000: Subprogram: Monitoring of Nature biodiversity
Project: Satellite Monitoring of Landscapes
INITIAL TASKS to elaborate hierarchical land cover
classification scheme, which supports on local level pecularities of Estonian landscapes and corresponds on regional level with internationally applied analogues
to produce satellite maps of recent (90/2000s) and historical (80s) environmental conditions
to determine the change of land cover and landscape diversity
to bring forth ongoing trends on class level and give the prognosis
PRESENT STATE OF SATELLITE MONITORING
6 monitoring sites Soomaa, Saarejärve, Alam-Pedja, Lahemaa, Vilsandi, Karula
Sites consist of the core area and the buffer zone
the core area is one of the permanent national monitoring sites with mostly natural and semi-natural land cover types, usually a protected area
the buffer is the 3 km wide zone around the core area, containing different land cover types
Resources 2 fulltime employees Landsat TM imagery, aerial photos, topographic maps,
training areas Pentium workstations 128Mb RAM, Windows 98 PCI EASI/PACE, ARC/INFO, Idrisi, ArcView, Fragstats
LOCATION OF MONITORING SITES
Lahemaa NP
Vilsandi NP
Soomaa NP Alam-Pedja NR
Karula NP
Saarejärve integrated monitoring area
RESULTSI After 5 years of monitoring work, 4481 km2 (10%)
of Estonia has been monitored Classification system developed so far has:
I level - 8 landscape types II level- 21 land cover classes III level - 58 land cover types, with additional IV level
subtypes
Mapping accuracy was enhanced by integrating GIS in spectral-based image processing (masking)
Estimation of landscape diversity used parameters show increase in landscape fragmentation,
especially in the buffer zones the main reason is increase of patch number and decrease of
their area
RESULTS II
Main trends in monitoring areas (and in Estonian nature as a whole) can be brought forth:
afforestation the increasing of coniferous stands in forests
(hypothesis) the decreasing of clear-cut areas in core areas and
increasing in buffer zones the increasing of grassland at the expense of arable land the increasing of fallow land at the expense of
abandoned fields and cultivated grasslands the overgrowing of natural grasslands and fallow land
with shrubs and young trees the decreasing of arable lands
CHARACTERIZATION OF METHODOLOGY
illustrated by Vilsandi monitoring area:
Total area is 467 km2, core area 51% and buffer 49%
Average count of land cover patches was: 8531 (> 1 ha: 2082) in 1980s, and 10516 (> 1 ha: 2272) in 1990s
Mean patch size (without water in 1986 and 1998): core area - 5.4 / 5.3 ha buffer zone - 12.7 / 9.96 ha
In total were mapped 36 land cover (sub)types Accuracy of change map: overall = 84%, KIA = 73% Field work on 84 sites (LC description, GPS, photo) Problematic land cover types:
alvar grasslands, fallow lands, wooded meadows, shrublands
METHODOLOGY IProcessing the satellite imagery
Elaboration of classification scheme III and IV level - mapping (map) II level - for monitoring land cover and diversity (map)
Classification masks forest and natural grasslands mires (fens, swamps, bogs) agricultural areas (crops, cultivated grasslands) water surfaces
Image processing pre-processing (geometric correction) histogram normalisation of two dates pre-classification (hybrid classification with ancillary data) ground truth (filed visit of training areas, GPS, photography) final classification under masks and accuracy assessment
LAND COVER TYPES (III, IV level)
in Vilsandi (1986 and 1998)
LAND COVER CLASSES (II level) in Vilsandi (1986 and 1998)
METHODOLOGY IIEstimation of landscape diversity
Landscape diversity parameters: Measured parameters
• general: count, average, maximum and total size, perimeters
Computed parameters• representing shape: edge index, shape index• representing neighbourhood: mean distance between
patches of the same class• diversity metrics: Shannon diversity index (only
landscape level)
Minimum size of patch for diversity analysis - 1 ha
FRAGMENTATION(arable land) 1986 1998
METHODOLOGY IIIChange detection
Change (or stability) of each class within the monitoring area - comparison of classification results for 2 dates
change database computed: 2 attributes per pixel (T1 and T2)
tally matrix: class changes (off-diagonal elements) and no-changes (diagonal) pixels
percent changes per class
Change in landscape diversity - comparison of diversity metrics for 2 dates
core area buffer zone change statistics
Change prognosis
MAIN TRENDS IN LAND COVER CLASSES
Vilsandi monitoring area 1986 - 1998, %
Land cover class 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 % V19861 - water 99 1 100 282.12 - coastal reedbed 14 86 100 3.43 - barren coast 24 39 20 6 1 10 1 100 4.94 - till coast with sparse vegetation 9 30 20 16 12 10 3 100 10.55 - natural grassland 3 8 62 6 4 7 7 5 100 44.06 - open mire 94 5 1 100 0.37 - treed mire, mire forest 31 67 1 100 0.68 - alvar grassland 1 13 5 58 20 3 100 7.99 - coniferous (juniper) shrubland 1 6 6 12 53 18 1 2 100 6.710 - coniferous forest 1 3 3 1 2 81 1 7 100 37.611 - deciduous forest 11 2 56 30 100 17.012 - mixed forest 1 2 5 32 2 57 100 25.413 - arable land 40 51 9 100 16.114 - cultivated grassland 30 61 9 100 7.115 - fallow land 20 40 40 100 0.316 - settlement, artificial areas 100 100 2.5V1998 (km2) 278.5 7.0 3.7 11.6 35.4 0.5 0.4 10.0 9.5 45.2 13.5 24.9 8.7 12.8 2.2 2.5 466.51986 - 1998 (%) 99 206 76 111 80 154 71 126 142 120 80 98 54 178 723 100
AREAS OF LAND COVER CLASSES IN VILSANDI
1986, 1998, 2010*
Land cover class 2 - coastal reedbed 3 - barren coast 4 - till coast with sparse vegetation 5 - natural grassland 6 - open mire 7 - treed mire, mire forest 8 - alvar grassland 9 - coniferous (juniper) shrubland 10 - coniferous forest 11 - deciduous forest 12 - mixed forest 13 - arable land 14 - cultivated grassland 15 - fallow land 16 - settlement, artificial areas
V2010 = M8698 * V98
0
1000
2000
3000
4000
5000
6000
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Land cover class
Area
(ha)
1986 1998 2010
ADVANTAGES AND DISADVANTAGES of SATELLITE REMOTE SENSING
Satellite remote sensing is a good tool for regular searching and updating of landscape state information
Digital satellite remote sensing data have direct input to GIS
congruous with raster and vector coverages
Landsat TM and ETM satellite data have the best quality/cost ratio for environmental monitoring
good spectral, temporal, spatial and radiometric resolution
0.30 EEK/km2) Satellite maps in context of
GIS help to resolve the problems of everyday tasks in management
• qualitative maps (land cover)• quantitative (statistical) data
Landsat satellite data is greatly dependent from:
clouds water content in soil and vegetation
Satellite mapping does not replace geobotanic mapping
Spectral and spatial resolution is too rough for detail habitat mapping
WHY MASKS?
Spectral signatures of land cover types are too similar
Number of classes and accuracy of map is too small
Spectral similarity was avoided by using GIS coverages as binary masks
Similarity of spectral signatures
0
10
20
30
40
50
60
70
Landsat TM channels
Re
fle
cta
nce
s
crop 11.76 13.19 54.89 57.62
coastal meadow 10.16 11.80 58.44 61.92
open bog 11.00 14.99 60.11 53.57
TM2 TM3 TM4 TM5
NORMALIZATION OF TWO SATELLITE IMAGES OF THE SAME
FENOLOGICAL STATE
Landsat TM 08.06.1988 12.06.1995
Normalisation of histograms around mean using value of standard deviation
normalisation by channel pairs
TM2 1988 and TM2 1995 a.s.o
Spectral signatures of land cover types in 1988 and 1995
0
10
20
30
40
50
60
70
80
90
100
110
120
130
TM2 TM3 TM4 TM5
Landsat TM channels
Re
fle
cta
nc
es
grassland 88
grassland 95
juniper 88
juniper 95
open bog 88
open bog 95