Overview on Land Cover and Land Use Monitoring in Russia

Transcription

1 Russian Academy of Sciences Space Research Institute Overview on Land Cover and Land Use Monitoring in Russia Sergey Bartalev Joint NASA LCLUC Science Team Meeting and GOFC-GOLD/NERIN, NEESPI Workshop Monitoring land cover and land use in boreal and temperate Europe August 25-28, 2010, Tartu, Estonia

2 Russian Academy of Sciences activity related to LCLU satellite monitoring land cover mapping using MODIS data and LAGMA method; agricultural monitoring with focus on arable land and crops mapping; burnt area mapping and severity assessment using MODIS and high-resolution optical data; TerraNorte Information System

3 Some features of R&D at IKI Focus is on national (entire Russia) and sub-continental (Northern Eurasia) monitoring Primary sources of EO data are moderate resolution satellite instruments (mainly MODIS and SPOT-VGT), while the role of high-res. (e.g. Landsat-TM, SPOT-HRV/HRVIR, RapidEye) data for national monitoring is rapidly increasing Focus on long-term time-series data analysis for land cover mapping and monitoring Development of automatic satellite data processing chains to perform monitoring in the routine and repeatable manner

4

5 GLC2000 legend for Northern Eurasia FORESTS WETLANDS NON-VEGETATED LAND COVER TYPES TUNDRA SHRUBLANDS OTHER VEGETATION TYPES AND COMPLEXES GRASSLANDS

6 Main features of GLC2000 Northern Eurasia land cover map 1-km resolution SPOT-Vegetation data for year 2000 Mapping method involves: i. set of advanced spectral-temporal and spectral-angular indexes to distinguish various land cover types ii. clustering and significant human input for labelling and decomposing of ambiguous semantic clusters Advantages: large number of mapped land cover types high level mapping accuracy Disadvantages: limited repeatability

7 Towards better land cover mapping: main directions of consideration - spatial resolution of mapping according to satellite sensors ability (1 km => 250 m) - mapping accuracy - mapping repeatability (annual as the target) - possibility to modify mapping legend (e.g. to increase number of thematic classes)

8 Cloud-free summer MODIS composite

9 Cloud-free winter MODIS composite

10 Classification based on LAGMA method

11 TerraNorte RLC mapping method Thematic source data Satellite data GLC 2000 Forest map Peatlands map Training data preparation Histogram filtration GIS analysis Manual selection Expert evaluation and correction Training samples Spectral mixture modeling Training samples spatial regularization (gridding) Classes signatures for cell-grid nodes Contextual Maximum Likelihood classification Auxiliary thematic products New land cover map GIS analysis Burnt area Croplands Water mask Urban mask

12 Contextual Maximum Likelihood Classification Local spectral-temporal signatures of classes Spectral-temporal MODIS data composites Covariation of metrics Average of metrics Number of samples Metrics for the pixel Maximum likelihood classifier Probabilities for classes

13 TerraNorte RLC Map for 2005 The land cover map for Russia based on MODIS 250 m

14 The Legend of TerraNorte RLC Map

15 The Pareto Boundary method to estimate accuracy of the land cover map Boschetti et al. Analysis of the conflict between omission and commission in low spatial resolution dichotomic thematic products: The Pareto Boundary // Remote Sensing of Environment 91 (2004)

16 TerraNorte RLC accuracy assessment for two test sites 1 2 Site 1: Karelia Republic Pareto optimum: for 250 m resolution for 1000 m resolution Site 2: Komi Republic

17 PVI time-series analysis NIR A PVI Soil line PVI=Distance (A, Soil line) PVI= *RED+0.56*NIR RED Inter-annual PVI dynamic similarity analysis and multi-annual phenological features retrieval 0,4 PVI arable lands natural vegetation 0,3 0,2 0,

18 The features for arable lands mapping with MODIS multi-annual data time-series Features Description Formula Feature Image Histograms Index of shortest vegetation period j j L1/ 2 min ( tl t F) j 1.. N, PVImax PVI( tl) PVI( t F), 2 tl tmax, tf t max Index of vegetation spring development j 1.. MSI min PVI N i spw ij Index of seasonal biomass decrease NSMI const N j 1 N PVI j 1 i sw min sw j PVI i

19 Arable lands map based on MODIS

20 MODIS derived arable lands map vs. HR imagery based fields limits

21 Crop types classification using MODIS 0,35 PVI 0,3 0,25 0,2 Peas Melilot Potato Alfalfa Perennials Spring crops Fallow Rape Winter rye Barley+Peas mixture 0,15 0,1 0, Apr 7-May 27-May 16-Jun 6-Jul 26-Jul 15-Aug 4-Sep 24-Sep Ground-truth Omission (%) , , ,1 Classification , , , , , , ,7 Commission (%) 3,3 3,1 9,8 3,7 0,9 6,5 4,5 9,8 11,1 9,8 93

22 SWVI (x1000) Burnt area mapping using MODIS Multi-annual MODIS data fire day Contaminated pixels detection mean mean-2*sdev mean+2*sdev current year Detection of SWVI statistical anomalies SWVI time-series restoration Combined SWVI and thermal anomalies SWVI time-series Burnt area maps Thermal anomalies

23 Burnt area for year 2009

24 Fires in Central European Russia in 2010 The fires have been mapped using MODIS data and MOD14 thermal anomalies detection algorithm implemented within the Satellite Monitoring Information System of Russian Forest Service The considered period: July 1 August 21, 2010

25 Forest burns severity assessment Ground-truth collection using HR data samples Burns severity assessment using MODIS data Field measurements HR Satellite images Burnt area Burns severity mapping assessment Land cover 12 Усыхающие насаждения Data analysis Число набл Степень повреждения по данным MODIS 180 Burns severity by forest types Число набл Statistical relationships Burn severity assessment 0-0,5 0,0 0,5 1,0 1,5 2,0 2,5 3,0 3,5 4,0 4,5 5,0 Statistics on forest mortality Lost forest area assessment

26 Post-fire assessment of trees mortality Landsat-TM ; RGB:NIR-SWIR-Red Trees mortality < 10% 11-40% 41-80% >80% Forestry districts Burns limits from MODIS

27 Response of forests to drought Vyksunskyy administrative district Semenovskyy administrative district

28 TerraNorte: Data Products on-line