Potential of optical satellite multi
Transcription
Potential of optical satellite multi
Potential of optical satellite multi-sensor time series data for landslide mapping covering large areas Robert Behling, Sigrid Roessner, Hermann Kaufmann and Birgit Kleinschmit1 GFZ Potsdam, Section 1.4 – Remote Sensing 1TU Berlin, Department of Geoinformation in Environmental Planning Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Overview • Landslide situation in Southern Kyrgyzstan • Remote sensing based multi-temporal landslide inventory • Automated landslide identification using - RapidEye data (recent activity) - Multi-sensor data (backdated activity) • Conclusions Kyrgyzstan Study Area Region of high landslide activity in Southern Kyrgyzstan Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Area of high landslide activity along Eastern rim of Fergana Basin Landslides represent most severe natural hazard in Southern Kyrgyzstan Several thousand known landslide events since 1950 More than 300 victims since 1990 Frequent destruction of houses and infrastructure Big need for systematic inventory of landslide events in space and time Landslides reported by Ministry of Emergency Situations of Kyrgyzstan between 2005 and 2010 Areas of known landslide activity Study area for landslide analysis at regjonal scale Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Dominant process type – deep seated landslides Rapid displacement of quaternary loess during 15 minutes period in March 1994 (50 victims) Displacement of clay-rich tertiary sediments during period of several days in June 1998 • High number of complex rotational/translational slides – often reactivation and repetitive failure • Regional spatial distribution determined by lithology and neotectonic structures • Landslide initiation by complex interplay between predisposing and triggering factors • Investigation of landslide activity since 1950-ies, mostly in areas close to settlements • Need for systematic inventory of landslide events and quantitative process understanding Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Landslide inventory – main prerequisite for hazard assessment Landslide inventory Predisposing factors - Lithology - Structural and neotectonic setting - Relief - Human interference Triggering factors -Precipitation/ seasonal snowmelt and related infiltration - Seismic activity/ earthquakes Hazard assessment after Guzzetti et al. 2005 Spatial probability Temporal probability Magnitude probability Risk elements Risk assessment - Infrastructure - Census data Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Development of landslide prone slope - Kainama Field picture of Kainama landslide taken on 18th of June 2004 Rapid landslide failure on 26th of April 2004 – landslide runout crossed river and buried parts of settlement (33 casualities) 16. Juni 2004 26. Mai 2002 R‐G‐B: Red‐NIR‐Green 14. Juni 2003 Development of Kainama landslide between 2002 und 2004 – interpreted from multi-temporal ASTER data Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Remote Sensing Data for Landslide Inventory 1980 LANDSAT multi/pan 30m/15m SPOT multi/pan 20-10m/10-2,5m IRS-1C/D multi/pan 23,5m/5,8m ASTER multi 90-15m ALOS multi/pan 10/2,5m RapidEye multi 6,5m 1990 2000 2010 TM ETM+ 1 2 3 4 5 General Availability Limited Quality Available for study area 1986-2013: ~690 datasets, around 80 acquisition dates between April and September nearly annual coverage since 1989 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Challenges: Seasonal Variability of Landslides R-G-B: color infrared RapidEye: 09.05.2009, 5m Foto June ALOS: 02.07.2009, 10m RapidEye: 29.09.2009, 5m Foto Sept Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Challenges: Long-Term Variability of Landslides Landsat TM R-G-B: 7-4-1 TM: Sept 1989, 30m RapidEye: Sept 2011, 5m TM: Juni 1993, 30m TM: Aug 1998, 30m Foto 2012/09/13 R-G-B: true color Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Overall approach for automated landslide detection Challenges High seasonal and long-term surface variability Large area of ~12 000 km² High amount of multi-sensor satellite data Automated, efficient and robust techniques suitable for multi-sensor time-series-analysis Pre-processing, Change Detection, Segmentation, etc. Pre-processing Multi-temporal Optical data Cloud masks Automated co-registration Derivation of landslide objects Spatio-temporal landslide inventoy -> bi- & multi-temporal change detection -> Segmentation -> Quantitative and qualitative Characterization of landslide objects Thematic data Pixel and object-based change detection Multi-temporal NDVI data Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping NDVI – Normalized Difference Vegetation Index ASTER: June 2004 RE: May 2009 RE: Sept 2011 NDVI [-0.8 , 0.5] ASTER: June 2003 2 km NDVI NIR RED NIR RED [-1;1] Analysis of vegetation cover Index derivable in all sensors Reducing radiometric differences Low NDVI -> absence of vegetation -> landslide? Fieldfoto: June 2004 Landslide from April 2004 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Landslide Identification – Temporal NDVI Trajectories NDVI [-0.8 , 0.5] ASTER: June 2003 1 ASTER: June 2004 RE: May 2009 RE: Sept 2011 3 2 4 2 km NDVI [-1,1] 1.0 1 2 3 4 Fieldfoto: June 2004 0.5 0.0 -0.5 -1.0 landslide characteristic variations in vegetation cover over time Landslide from April 2004 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Landslide Identification – Temporal NDVI Trajectories Multitemporal: Variation of vegetation cover in time 6 3 1.0 NDVI [-1,1] Landslide objects between 2003-04 0.5 4 0.0 -0.5 1.0 0.5 1 landslide 2 vegetation 3 field 2 4 water 5 outcrop 6 urban 1 0.0 -0.5 5 2 km Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Automated Landslide Identification approach • Analysis of landslide occurrence for all subsequent images (time period) • Combined pixel- and object-based approach for each time period: 3 main steps: • Bi-temporal vegetation change analysis with multiple thresholds (Castilla et al. 2009) followed by a segmentation • Multi-temporal analysis of revegetation rates • Relief oriented analysis (slope, slope orientation) Behling, R.; Roessner, S.; Kaufmann, H.; Kleinschmit, B. Automated Spatiotemporal Landslide Mapping over Large Areas Using RapidEye Time Series Data. Remote Sensing, 2014, 9, 8026-8055. Behling, R.; Roessner, S.; Segl, K.; Kleinschmit, B.; Kaufmann, H. Robust Automated Image Co-Registration of Optical Multi-Sensor Time Series Data: Database Generation for Multi-Temporal Landslide Detection. Remote Sensing, 2014, 6, 2572-2600. Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping RapidEye - High Resolution Satellite Database Data acquisition within RESA RapidEyeScience Archive Program since 2010 • 6,5m spatial resolution • 5 spectral bands (440 – 850nm) • Acquisition of data in pre-defined time periods of high process activity • Orthorectified Data in 21 25x25km tiles • Best temporal and spatial resolution for regional scale • ~ 600 datasets until 2013 • Status of RESA priority area since 2014 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Identified Landslides – Exemplary Results 2012/06/18 Time period of occurrence: 2009/05/26 – 2011/05/02 2012/05/24 – 2012/06/18 Approach identifies landslides of varying: • Size • Shape • Lithology • Stage of development: fresh failures reactivation relocation Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Identified Landslides – Exemplary Results 2012/06/18 Time period of occurrence: 2009/05/26 – 2011/05/02 2012/05/24 – 2012/06/18 A A Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping field foto taken: 2012/09/10 Identified Landslides – Exemplary Results 2012/06/18 Time period of occurrence: 2009/05/26 – 2011/05/02 2012/05/24 – 2012/06/18 C B B C 2012/09/10 2012/09/10 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Results of Automated Identification at Regional Scale Study area: 12,000 km² Analyzed time period: 2009-2013 600 RapidEye datasets (5 m resolution) SRTM-DEM (30 m resolution) 612 identified landslides Size 125 – 775,000 m² A total of 7.3 million m² affected 2009 - 2013 no major triggering event Shows the need for regular multi-temporal landslide inventories Identification of landslides with different sizes, shapes, types and stages of development in objects form Multi-temporal dynamic landslide inventory as Input for Hazard and Risk analysis Continuation of existing inventories -> Potential for continuous monitoring of landslide activity Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Optical Remote Sensing Database • • • Study area: 12,000 km² 27 years of data coverage: 1986 – 2013; since 1998 nearly annual coverage ~ 670 orthorectified datasets of 4 different multispectral sensors Landsat (E)TM 30 m 49 datasets 1990-1999 & 2009-2013 ASTER 15 m 30 datasets 2000-2008 SPOT 10-20 m 12 datasets 1986 & 2006-2010 RapidEye 5m 592 datasets 2009-2013 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Example Uzgen – Landslide Situation 2012 B A C Field Photo Sept. 2012 A B C Perspective View: RapidEye True color Sept. 2012 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Multi-Temporal Landslide Identification 2000 2010 Data Identified landslides C C B B A A 2 km 32 datasets starting in 1986 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Multi-Temporal Landslide Identification 2000 2010 Data Identified landslides 2 km 32 datasets starting in 1986 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping Multi-Temporal Landslide Identification 2000 2010 Data Identified landslides 32 datasets starting in 1986 Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping 2 km Conclusions and outlook Developed automated approach for landslide identification enables: • Identification of landslides of different sizes, shapes, types and in different stages of development in an object-based form • Multi-temporal dynamic landslide inventory, not only event-based inventory • Spatial and temporal explicit input for hazard and risk analysis • Update of existing inventories -> Monitoring of landslide activity • Analysis at regional scale • Identification of landslide events for variable time span covering the whole time period of multispectral remote sensing data availability • Transfer of methodology to other global hot-spots of landslide activity Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping THANK YOU FOR YOUR ATTENTION Contact: [email protected] [email protected] Behling et al. Potential of optical satellite multi-sensor time series data for landslide mapping