SATELLITE MONITORING of ESTONIAN LANDSCAPES Kiira Aaviksoo and Andrus Meiner Estonian Environment Information Centre Mustamäe tee 33, Tallinn 10616 ESTONIA,

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