SATELLITE MONITORING of ESTONIAN LANDSCAPES

1
SATELLITE MONITORING ofESTONIAN LANDSCAPES

• Kiira Aaviksoo and Andrus Meiner
• Estonian Environment Information CentreMustamäe
  tee 33, Tallinn 10616 ESTONIA,
• kiira_at_envinst.ee, andrus.meiner_at_ic.envir.ee

2
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

3
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

4
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

5
LOCATION OF MONITORING SITES
Lahemaa NP
Saarejärve integrated monitoring area
Vilsandi NP
Soomaa NP
Alam-Pedja NR
Karula NP
6
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

7
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

8
CHARACTERIZATION OF METHODOLOGYillustrated by
Vilsandi monitoring area

• Total area is 467 km2, core area 51 and buffer
  49
• Average count of land cover patches was
• 8531 (gt 1 ha 2082) in 1980s, and
• 10516 (gt 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

9
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

10
LAND COVER TYPES (III, IV level) in Vilsandi
(1986 and 1998)
11
LAND COVER CLASSES (II level) in Vilsandi (1986
and 1998)
12
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

13
FRAGMENTATION(arable land)
1986
1998
14
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

15
MAIN TRENDS IN LAND COVER CLASSESVilsandi
monitoring area 1986 - 1998,
16
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

17
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

18
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

19
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