MONITORING LAND USE AND LAND USE CHANGES IN FRENCH GUIANA BY OPTICAL REMOTE SENSING Photo : Valéry Gond Photo : Valéry Gond Photo Photo : Gaëlle : : Valéry VERGER Gond Gaëlle VERGER ONF, French National Forest Office gaelle.verger@onf.fr
Summary Context and objectives: Objectives and method of the project Implementation : Data Stratification Photo-interpretation Statistical estimates Results and discussion:
Summary Context and objectives: Objectives and method of the project Implementation : Data Stratification Photo-interpretation Statistical estimates Results and discussion:
Ongoing project in French Guiana Objectives of the project In the Kyoto Protocol framework To complement existing inventories made for 1990, 2006 and 2008 time points Provide surface data for 2012 and feed the last French GHG inventory Methods Based on the existing method (developed and approved by IPCC in 2006) Statistical inventory wall to wall mapping On the basis of stratified sample points By visual photo-interpretation of satellite imagery GLCF Landsat Images for year 1990 SPOT images for years 2006, 2008 & 2012 To be combined with emission factors
Summary Context and objectives: Objectives and method of the project Implementation : Data Stratification Photo-interpretation Statistical estimates Results and discussion:
Satellite Data 1990: 8 Landsat Imageries 2008: SPOT 2, SPOT 4 & SPOT 5 2012: SPOT 5 (about 270 imageries, thanks to SEAS) Selection of imageries per frame depending on the cloud coverage SPOT 5 SPOT 4 SPOT 2 Cloud cover < 75 10 25 %
Definition of the strata Based on several existing studies and on experts knowledge (ONF and National Park of Guyane) Strata R (Reinforced sampling) ONF GIS analysis on existing information Roads Agriculture Cities 1km Human settlements Buffer Urban areas Gold mining Mangroves ecological area Strata N (Normal sampling) : all area not in Strata R Strata P : 52 km circle surrounding Petit Saut reservoir
Sampling design Systematic/random grid Square grid - 932 m between points Random origin and inclination Sample definition Strata R : All Grid points Strata N : One point over 5 Strata Total Area (ha) Strata Nbr of points Distance between 2 points (m) Surface of sample points (ha) N 7 030 800 Normal 1 953 4 194 3 600 R 1 526 241 Reinforced 17 543 932 87 TOTAL 8 557 041 19 496 Permanent sampling : same samples in 1990, 2008 and 2012 Good estimation of Land Use areas AND Land Use Changes
Photo interpretation Shifting Cultivation Gold mining Landsat SPOT 5 TM 2006 1990 - CNES GLCF Landsat SPOT 5 TM 2006 1990 - CNES GLCF Petit Saut Reservoir Agriculture extension Landsat SPOT 54 TM 2006 1990 - CNES GLCF Landsat SPOT 54 TM 2006 1990 - CNES GLCF
Photo interpretation 1990 2008 Land Use categories Forest Mangroves Other forests Settlements Gold mining other Settlements Cropland Grassland Wetlands Sea Other lands
Statistical estimates Land use area estimates from Land use proportion in each strata Area estimate Variance Stand. dev. 2 Pij (1 Pij) ij Pij S j Vij S j ij Vi j n S Land use conversion areas estimated from Land use conversion proportions in each strata : Same formulas Estimates + Precision of estimates Sample used = All points without clouds in 1990 And 2006 And 2008 = 16469 points out of 17667. j j V Standard error of the estimate ij
Summary Context and objectives: Objectives and method of the project Implementation : Data Strata Photo-interpretation Statistical estimates Results and discussion:
Results: Land uses 1990 2006 2008 Land use 1990 / 2006 (methodology development) / 2008 Ongoing interpretation for 2012
Results: Forest cover changes Land use changes between forest and non forest (ha) 1990-2008 Strata N Estimate Strata P Estimate Strata R Estimate TOTAL Estimate Stand. error Stand. error Stand. error Stand. error Forest > Forest 6 824 884 176 030 1 028 196 8 029 110 30 971 1 622 4 046 31 276 Forest > Non forest 8 067 34 790 65 662 108 518 8 063 1 589 2 419 8 566 Including Forest > Settlements 8 067 954 21 166 30 188 8 063 287 1 399 8 188 Including Forest > Gold mining 0 868 14 863 15 731 0 274 1 175 1 207 Including Forest > Other settlemen 8 067 87 6 303 14 457 8 063 87 768 8 100 Including Forest > Cropland 0 0 27 281 27 281 0 0 1 584 1 584 Including Forest > Grassland 0 0 2 258 2 258 0 0 460 460 Including Forest > Wetlands 0 33 835 4 892 38 727 0 1 571 677 1 711 Including Mangroves > Wetlands 0 0 2 634 2 634 0 0 497 497 Including Forest > Other lands 0 0 10 066 10 066 0 0 969 969 Including Mangove > Sea 0 0 9 595 9 595 0 0 946 946 Including other Forest > Other land 0 0 470 470 0 0 210 210 Non Forest > Non forest 112 941 1 822 131 605 246 369 29 938 396 3 326 30 125 Non Forest > forest 0 0 12 229 12 229 0 0 1 067 1 067 Including Cropland > Forest 0 0 3 292 3 292 0 0 556 556 Including other non forest > Mangrove 0 0 6 867 6 867 0 0 802 802 A large majority of the forest is unchanged > 98 % A changed observed on 1 point of strata N Large Influence on the estimates and the standard error of the estimates 3 main causes of human deforestation in French Guiana Petit Saut Dam Natural evolution deforestation (Kyoto) Little Non forest to forest changes Mainly in croplands (Shifting cultivation)
Results : Annual deforestation Analysis annual deforestation during 1990/2006 and 2006/2008 Without Petit Saut Dam (temporal event) and without Mangrove to sea conversions (= Natural Deforestation) Forest to other lands Forest to settlements Forest to wetlands Forest to Grassland Forest to cropland 1990-2006 period 2006-2008 period 0 500 1 000 1 500 2 000 hectare / year 2 500 3 000 3 500 Increase of total annual deforestation from 3500 ha/year to 5900 ha/year
Discussion Why this methodology has been chosen It was the methodology used for the NFI and this methodology has been validated by the IPCC for France => Coherence with the national territory Does not require lot of expertise Saving digitization time and bias Existing ArcGIS extension for the analysis Quality control integrated within the statistical process Application of this methodology in the REDD+ framework? Efficiency of the method to monitor Deforestation has been demonstrated Possibility to monitor forest degradation? Further investigations needed Use of SPOT 5 data (2.5 to 10 m resolution) or Very High Resolution sensors (< 1m) Combination with radar data (sensitivity to biomass) / Combination with field data Key issues: Availability and cost of high resolution satellite imageries No localization of the deforestation