Project title 1 Evaluation of MIR data from SPOT4/VEGETATION for the monitoring of climatic phenomena impact on vegetation Principal investigator: M-Christine IMBERTI Co-investigator: Frédéric BIARD
Stockholm meeting : 18-20/11/96 2 1. CONTEXT AND OBJECTIVES 2. GENERAL APPROACH - Methodology - Test site - Description of data - Simulation techniques - Calculation of Vegetation Index 3. PRELIMINARY RESULTS - Global interest of MIR low resolution - Interest for basic zoning - Interest for diagnosis 4. PRELIMINARY CONCLUSIONS 5. FUTURE PLAN WORK
The operational context of the study 3 A country struck by a climatic phenomena such as a drought needs to measure the gravity of the situation In both space and time Independent diagnosis Climatic Impact Monitoring Information System, using: Space derived, repetitive, low resolution satellite data (NOAA/AVHRR) Other data: land use, agro-climatic data & crop calendars
Aims of the investigation 4 To test the advantages of MIR SPOT4/VEGETATION spectral band for: The improvement of basic zoning (substitution of non space-derived data) A better evaluation of climatic phenomena impact on agricultural areas Advantages of MIR compared to VIS and NIR Advantages of MIR in areas with high concentration of invariable elements
General approach 5 To study the interest of MIR band for improving basic zoning Comparison between the zoning obtained with MIR to the existing one To study the interest of MIR band, in addition or to correct the vegetation index, for temporal and spatial climatic phenomena impact assesment Comparison of the impact results obtained either with or without MIR, to determine the advantages or drawbacks of MIR for impact mapping and early delivery of results
Complement HR MIR 5b To study the interest of high resolution MIR band for local inventory of invariant elements Feasability study of inventory of invariant elements on HR Spot 4 data To study the interest of using combined data from SPOT4 (HR MIR and VGT MIR data) in CIMIS Conclusions about the feasability of using combined data in regions where invariant elements are important
6 Methodology Phase 1a NOAA IMAGES REF 94 & 96 TM 94 IMAGES TM 96 IMAGES STUDY HR MIR SIMULATION VGT 94 SIMULATION VGT 96 HRMIR ZONATION ZONATION TEST TEST OF INVARIANTS INVENTORY EVALUATION ZONATION TEST OF DIAGNOSIS 96 REF DIAGNOSIS 96 EVALUATION DIAGNOSIS STUDY OF VARIATION SIGNIFICANCE SYNTHESIS
Choice of a new Test Site 7 Why ROMANIA replaces SPAIN? SPAIN has not suffered from climatic hurts this year Southern ROMANIA has been affected by rough meteorology in 96 A long and harsh winter badly damaged fields An early heat and long drought affected main crops Operational projects conducted by in Romania Available data over the area Good knowledge of requirements
Location of the new Test Site 8 South-west of ROMANIA (north region of Craiova) Coller une carte de localisation...
Description of Test Site 9 Physical caracteristics Large range of altitude : centered over piedmont area (Carpathian chain at north, Danubian plain at south) Mixed Land Cover Wooded areas (north), agricultural lands (south), grasslands Invariants elements with agricultural land Variability of climatic impacts Depending on localisation (north/south) and crop types
Description of Input Data 10 Specific images for VGT simulation 7 Landsat5/TM scenes Red, NIR, MIR System corrected Cloud free or few clouds 4 acquisitions in 1994 3 acquisitions in 1996 Projection system : Lambert Azimuthal Equal Area (55 E, 50 N) Suited to Romania VGT simulation bands Resampled resolution : 1 km² / pixel 01/04 04/06 07/08 08/09 22/04 09/06 27/07
Description of Reference Data 11 Low resolution NOAA/AVHRR data 6 NOAA NDVI mosaics NDVI = (NIR-Vis)/(NIR+Vis) Week or decade NDVI synthesis 3 images in 1994 April, June, August 3 images in 1996 April, June, July Projection system : Lambert Azimuthal Equal Area (55 E, 50 N) Suited to Romania Resampled resolution : 1 km² / pixel
Description of Reference Data 12 Reference basic Zoning and 96 Diagnosis Basic Zoning in Agricultural Monitoring Units (A.M.U) Interpreted from NOAA / NDVI synthesis computed over 1993, 1994 and 1995 Complemented with NDVI differences over the 3 years 1996 crop condition Diagnosis Interpreted from NDVI differences between 1996 and a reference year (normal growing conditions)
Description of Reference Data 13 Other data collected on the Test Site Climatic descriptors Such as temperatures and precipitations Agricultural statistics Such as cropped areas, production, yields... Available at administrative unit level
Simulation techniques 14 Simulated Low Resolution Images : Pixel size : +/- 1Km² Bands : Red, NIR, MIR Referenced : Geographic (WGS84) Superimposed : << 1 pixel Superimposition...(+/- 10 GCP) Resampling...scrolling window (average 33 x 33 pixels) Geo-referencement...linear model using the 4 corner points Landsat / TM images : Pixel size : 30 x 30 m Selected bands : 3 (Red), 4 (NIR), 5 (MIR) Preprocessing level : 1B (system corrected)
Calculation of NDVI (mir) 15 NDVI (mir) = (NIR - MIR) / (NIR+MIR) Interest of a normalised difference index using MIR MIR: second most informative spectral band after NIR (Lallemand & al.) Already proposed and used in several scientific studies (Baret & al., etc.)
16 Global interest of MIR low resolution (1) Linear relations between NDVI and NDVI(mir) calculated on representative sample of 50 x 50 pixels (reference year 1994) April June August September 160 200 200 200 NDVI (PIR ; MIR) DN 150 130 110 90 y = 0.7066x - 7.1108 R 2 = 0.547 NDVI (PIR ; MIR) DN 180 160 y = 0.7546x - 9.4407 R 2 = 0.7886 NDVI (PIR ; MIR) DN 180 160 y = 0.9369x - 44.423 R 2 = 0.7998 NDVI (PIR ; MIR) DN 180 160 y = 0.8151x - 24.827 R 2 = 0.8841 80 70 80 80 80 60 60 60 60 160 180 200 160 180 200 220 160 180 200 220 160 180 200 220 NDVI (PIR ; Vis) DN NDVI (PIR ; Vis) DN NDVI (PIR ; Vis) DN NDVI (PIR ; Vis) DN NDVI and NDVI(mir) are highly correlated more correlated during high vegetation activity periods (summer months) : r² = 0.8 higher cluster dispersion in early vegetation period (spring) Hypothesis : possible informative diferential of NDVIs in early crop season
Global interest of MIR low resolution (2) 17 Why such a high level of correlation between NDVI and NDVI(mir)? NDVIs have not a linear evolution Need to study directly the spectral band relations (same sample, june 1994) TM4 (DN) 130 110 90 80 70 60 50 y = 0.0767x + 78.733 R 2 = 0.0056 40 10 30 50 70 90 110 130 TM3 (DN) TM5 (DN) 180 160 80 60 y = 0.1989x + 70.812 R 2 = 0.0158 40 40 60 80 TM4 (DN) TM5 (DN) y = 1.4227x + 34.885 180 R 2 = 0.7694 160 80 60 40 20 0 10 30 50 70 90 110 TM3 (DN) Poor correlation High correlation Complementarity between Visible and NIR NIR and MIR Similar behaviour between Visible and MIR Hypothesis : Global similar behaviour due to bands properties NDVI and NDVI(mir) offer a range of complementarity
Global interest of MIR low resolution (3) 18 However, is there any specific interest of NDVI (mir)? Study of the residuals NDVI (mir) = f(ndvi) Calculation of: Linear Estimated (NDVI(mir)) - Observed (NDVI(mir)) Cartography of the differences Analysis of their geographic coherence
Global interest of MIR low resolution (4) 19 MIR contains its own information GEOGRAPHIC COHERENCE April 1994 June 1994 depending on : crop period relief (north, south) hydrography / moisture land cover... August 1994 September 1994
20 Global interest of MIR low resolution (4) Preliminary conclusions: Low Resolution NDVI (mir) does not carry much more information than NDVI (vis) but some of this information seems different thus complementary Hypothesis : Assessment of NDVI(mir) complementary information Better recognition of cropped areas More pertinent in early crop season (earlier diagnosis?)
21 I Evaluation of the interest of MIR for basic zoning Methodology (1) Identification of discrepancies Zones appearing with NDVI (mir) and not with NDVI Zones appearing with NDVI and not with NDVI (mir) Analysis of the discrepancies Synthesis Identification of factors contributing to these discrepancies (land cover, soil, relief, climate, crop calendar, date,...) Identification of favourable factors to the use of MIR data Identification of constraints to the use of MIR data
22 I Evaluation of the interest of MIR for basic zoning Methodology (2) : zoning provided by a multitemporal unsupervised classification on 1994 vegetation indices images. Unsupervised classification : Clustering : ISODATA 5 classes convergency = 0.99 NDVI 5 zones NDVI(mir) (5) wooded areas (4)pasture / grasslands (3) mixed crops (1) & (2) intensive crops NDVI NDVI(mir)
23 I Evaluation of the interest of MIR for basic zoning Results (3) : identification of the main discrepancies between the two zonings Map of basic zoning differences (per class) NDVI zoning Main differences legend NDVI NDVI(mir) class 2 class 1 Difference class 3 class 1 class 3 class 4 NDVI(mir) zoning The main discrepancies are localised in cropped areas differences between mixed crops and intensive crops differences between intensive crops themselves...
24 I Evaluation of the interest of MIR for basic zoning Preliminary Conclusions (4) Existant but poor interest of Low Resolution NDVI(mir) for basic zoning Nevertheless : Assessment of NDVI(mir) complementary information for basic zoning Easier zonation on NDVI(mir) data (clear definition of zones) Complementary information on cropped areas
II VEGETATION Preparatory Programme 25 Evaluation of the interest of MIR for diagnosis Methodology (1) Identification of discrepancies Sharper diagnosis with NDVI (mir) (more correct or detailed or evident or sooner, according to surface or intensity) Weaker diagnosis with NDVI (mir) Analysis of the discrepancies Synthesis Identification of factors contributing to these discrepancies (land cover, soil, relief, climate, crop calendar, date,...) Identification of favourable factors to the use of MIR data Identification of constraints to the use of MIR data
26 II Evaluation of the interest of MIR for diagnosis Methodology (1) : temporal analysis of 1994 and 1996 NDVIs relations per zone 94 & 96 NDVI relation 94/96 (june) 210 200 190 y = 0.1804x + 136.35 R 2 = 0.0277 NDVI 1996 180 170 160 150 Zone 1 Zone 2 Zone 3 Zone 4 Zone 5 130 130 150 170 190 210 NDVI 1994 NDVI images Extracting NDVI values with classification results masks relation 94/96 per zone 94 & 96 NDVI(mir) relation 94/96 (june) 150 y = 0.2777x + 82.025 R 2 = 0.0519 NDVI(mir) 1996 130 110 Zone 1 Zone 2 Zone 3 Zone 4 Zone 5 90 80 80 160 NDVI(mir) 1994 NDVI(mir) images Extracting NDVI(mir) values with classification results masks relation 94/96 per zone
II VEGETATION Preparatory Programme 27 Evaluation of the interest of MIR for diagnosis (2) Study of the interest of NDVIs for 1996 diagnosis in an agricultural area (zone 1) Appraisal of temporal behaviour April June August NDVI NDVI 1996 180 y = 0.3364x + 90.84 R 2 = 0.1558 160 80 60 60 80 160 180 NDVI 1994 NDVI 1996 220 200 y = 0.2089x + 129.45 R 2 = 0.0354 180 160 160 180 200 NDVI 1994 NDVI 1996 220 200 180 160 80 y = 0.446x + 89.833 R 2 = 0.1415 60 160 180 200 220 NDVI 1994 NDVI(mir) NDVI(mir) 1996 y = 0.3229x + 55.208 R 2 = 0.0978 110 90 80 70 60 60 80 NDVI(mir) 1996 150 130 110 90 80 70 y = 0.376x + 65.166 R 2 = 0.1263 60 60 80 160 NDVI(mir) 1996 150 130 110 90 80 70 y = 0.1779x + 92.247 R 2 = 0.0329 60 60 80 160 NDVI(mir) 1994 NDVI(mir) 1994 NDVI(mir) 1994 96 and 94 NDVIs are poorly correlated for both situations (with or without MIR) On cropped areas, both Indices are good indicators of discrepancies between two years NDVI seems to be more sensitive in vegetation periods (June) NDVI(mir) appears to be more convenient for low vegetation coverage periods (April, August)
II VEGETATION Preparatory Programme 28 Evaluation of the interest of MIR for diagnosis (3) study of the interest of NDVIs for 1996 diagnosis at high vegetation level period (june) Assessment of land cover influence Zone 1 Zone 3 Zone 5 NDVI NDVI 1996 220 200 y = 0.2089x + 129.45 R 2 = 0.0354 180 160 160 180 200 NDVI 1996 220 200 y = 0.2948x + 119.4 R 2 = 0.0607 180 160 160 180 200 220 NDVI 1996 220 210 200 190 180 170 160 150 y = 0.305x + 136.64 R 2 = 0.0595 160 180 200 220 NDVI 1994 NDVI 1994 NDVI 1994 NDVI(mir) NDVI(mir) 1996 150 130 110 90 80 70 y = 0.376x + 65.166 R 2 = 0.1263 60 60 80 160 NDVI(mir) 1996 150 y = 0.5632x + 50.776 R 2 = 0.2473 130 110 90 80 80 160 NDVI(mir) 1996 160 y = 1.0399x - 3.1142 150 R 2 = 0.6599 130 110 90 80 80 160 NDVI(mir) 1994 NDVI(mir) 1994 NDVI(mir) 1994 minimum % of «invariant» elements maximum Higly different behaviour of the two NDVIs, depending on land cover type NDVI(mir) seems to be more sensitive to land cover type NDVI(mir) appears to be «more stable» in areas with high proportion of «invariants»
II VEGETATION Preparatory Programme 29 Evaluation of the interest of MIR for diagnosis Results (4) : cartography of significant discrepancies between 1994 and 1996, with the two indices April June August NDVI NDVI(mir)
II VEGETATION Preparatory Programme 30 Evaluation of the interest of MIR for diagnosis Preliminary Conclusions (5) Few differences observed between diagnosis results coming from NDVI and NDVI(mir) Same struck cropped areas Similar level of climatic impact However, NDVI(mir) could offer some advantages Better delineation of different vegetation status zones Focus on cropped areas for a better diagnosis More information on valleys (pertinent information?)
Preliminary conclusions - Phase 1a 31 Existant but poor interest of Low Resolution MIR Nevertheless : Identified NDVI(mir) complementary information Easier zonation on NDVI(mir) data Better discriminating potential on cropped areas (interesting mainly for zoning) More pertinent information in early crop season Caution : Pertinent information (disturbing water content level?) Lack of information or less sensitive to spectral noise?
Future test and analysis 32 III Evaluation of the interest of HR MIR data for detecting invariants Test of inventory of invariant elements with HR MIR Analysis of proportion of invariants per zone on basic zoning Analysis of discrepancies of diagnosis towards these proportions of invariants
Future work plan 33 Phases Period Title Data Prelaunching March 96 - February 97 Preliminary methodological study - 2 series of NOAA data - 7 full Landsat TM images March 97 - December 97 Extension and validation of the method -1 serie of NOAA data - +/- 20 ATSR images Postlaunching March 99 - December 99 Realisation in real context - 1 serie of NOAA data - +/- 20 VEGETATION images