FOREST MONITORING AND BIOMASS ESTIMATION FOR REDD+ WITH INSAR Svein Solberg, Johannes May, Belachew Gizachew, Wiley Bogren, Johannes Breidenbach Norwegian Institute of Bioeconomy Research GFOI R&D and GOFC-GOLD Land Cover Science Meeting THE HAGUE, 31. Oct. 4. Nov. 2016
InSAR data 2000: SRTM C and X ~2012: TanDEM-X
Step 1: ΔH 1 = DEM TDX DEM SRTM_C 15.11.2016 3
STEP 2: REMOVING ARTEFACTS IN SRTM C-BAND using ANOVA model ΔH 1 = DEM TDX DEM SRTM_C = C artifact + ΔH time + ΔH penetration + e 1 = b 0 + L1 + L2 + B1 + B2 + e 2 15.11.2016 4
ΔH 1 ANOVA RESULTS: R 2 = 0.18 Source DF SS MS F Value Pr > F Belts B1 3 168366515 56122172 10700000 <.0001 Belts B2 3 15617771 5205924 992341 <.0001 Lines L1 814 28974668 35595 6785 <.0001 Lines L2 535 32178017 60146 11465 <.0001 Error 2.07*10 8 1088151112 5 Corrected Total 2.07*10 8 1333288083 ΔH corr1 = ΔH 1 - C artifact C artifact 15.11.2016 5
PENETRATION DIFFERENCE DEPENDING ON FOREST COVER AND LAND COVER TYPE X-band C-band
STEP 3: REMOVING PENETRATION DIFFERENCES AND ARTIFACTS OF SRTM-X using GLM model ΔH 2 = DEM SRTM_X - DEM SRTM_C_corr1 + e 3 = X artifact + ΔH penetration + e 3 = b 0 + XL i + FC*b j + e 3 15.11.2016 7
GLM RESULTS: ΔH 2 R 2 = 0.53 Source DF SS MS F Pr > F XL 2400 956806646 398669 40921 <.0001 forest_cover * Land_cover 8 1347970 168496 17295 <.0001 Error 8.84E+07 861031631 10 Corrected Total 8.84E+07 1819186246 Residual e 3 GLM model: X artifact + ΔH penetration ΔH 2 = b0 + XL i + FC* j + e 3, L1 = 1 km lines 330.2 XLi = X-band errors as lines FC* j = penetration difference per forest 15.11.2016 8
Penetration difference X-C, m, C TO X PENETRATION CORRECTION MODEL 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0 20 40 60 80 100 Forest cover % Evergreen Broadleaf Forest Woody Savanna Savannas Grasslands Permanent Wetlands Croplands Cropland/Natural Vegetation Mosaic others / mixed 15.11.2016 9
MAKING A SIMULATED SRTM X-BAND DEM DEM SRTM_X_sim = DEM SRTM_C - C artifact X artifact Global Forest Cover 2000 (Hansen et al.) MODIS land cover lifting of c-band dem 15.11.2016 10
STEP 4: CHECKING FOR REMAINING BIAS OR RAMP ERRORS 18 728 cells systematically distributed Zero forest cover No forest cover changeno-forest points systematically distributed over Uganda: Average North-South slope East-West slope ΔH 0.9 mm 8 mm 16 mm 15.11.2016 11
STEP 5: VOID FILLING IN STEP TERRAIN (> 30 DEGREES), 1% OF AREA 15.11.2016 12
MEAN HEIGHT CHANGE FOR LANDSAT CHANGE CATEGORIES Land cover type loss no change gain Evergreen Broadleaf Forest -8.8-1.0 0.6 Woody Savanna -3.7-0.1 0.7 Savannas -1.2-0.3 1.8 Grasslands -2.2-0.1-0.4 Permanent Wetlands -3.6 0.1-0.1 Croplands -1.4-0.1 0.8 Cropland/Natural Vegetation Mosaic -3.5-0.3 1.2 others / mixed -3.6 0.4 1.4 For example: Evergreen Broadleaf Forest loss category with Landsat corresponds to 8.8 * 18.4 t/ha/m = 162 t/ha in AGB loss = ca 162 t/ha CO 2 emission Table x. Height change estimates from the ANOVA used for filling of void areas and pixels having unreliable height change estimates 15.11.2016 13
COMPARISON AND SYNERGY WITH LANDSAT: FOREST GROWTH IN PROTECTED AREAS InSAR Landsat 15.11.2016 14
FOREST GROWTH IN PROTECTED AREAS (2) 15.11.2016 15
STEP 6: FROM ΔH TO ΔAGB TO ΔB TO ΔC BIOMASS AGAINST INSAR HEIGHT
Savannahs: Noisy relationships due to differences in stem taper
STEP 7: UNCERTAINTY ESTIMATION WITH MONTE CARLO 45 random samples; each containing approximately 4 million pixels (1% of the data) 5 times processing of each sample = 225 processing batches In each processing we varied the correction factors randomly according to their uncertainty Sequential processing aggregating errors through the 5 steps: 1. error removal of C-band SRTM, 2. correction from C to X-band SRTM, 3. replacing voids and extreme, illegal values with values specific the given land cover and forest change category, 4. recalculating ΔH to ΔAGB, 5. expansion of ΔAGB to ΔB, 15.11.2016 18
UGANDA: CHANGE 2000-2012 Forest height decrease 2000 2013: ΔH = 33 cm Corresponding CO 2 emission ΔCO 2 = 27 mill t/year 95% confidence interval = ± 10.5 mill t/year
A NOVEL METHOD FOR DIRECT ESTIMATION OF FOREST CARBON CHANGES: Conventional method E = A EF InSAR method E = A H EF H
CONCLUSIONS SRTM and Tandem-X can be used for estimating 12 year changes as a Reference Emission Level in REDD+, and for forest C stocks at large scale