Estimating crop biomass in smallholder fields with very high resolution imagery
-
Upload
cimmyt-int -
Category
Education
-
view
646 -
download
0
description
Transcript of Estimating crop biomass in smallholder fields with very high resolution imagery
Estimating crop biomass in smallholder fields with very high resolution imagery
Remote Sensing – Beyond Images WorkshopMexico City – 15 Dec. 2013
P.S. Traore, S.S. Traore, K. Goita, W.M. Bostick, J. Koo
Dryland Systems of West Africa
Possible intensification pathways
Large cities andhigh rural densities
‘Bhoo Chetanaintensification
pathway’?
Large cities andlow rural densities‘Fazendaintensificationpathway’?
• Human and animal population growth• Changes in dietary preferences• Crop-livestock integration• C sequestration• Bio-fuels
Opportunities in biomass productionNoFertNoResidue PK + Residue
M9D3 STAM 59A CSM388 ObatanpaMillet Cotton Sorghum Maize
Yield Biomass Yield Biomass Yield Biomass Yield Biomass2009 1450 5000 1300 2000 1276 4880 2100 39002010 1130 7900 1500 2700 1144 7680 1800 31502011 1150 9900 1300 2550 2112 8100 2600 3450
• Canopy height and optical signal saturation• Tropical cloud cover• Heterogeneous field size and geometries• Mixed crops and trees in fields• Spread of planting dates & phenologies• Heterogeneous soil properties at sub-field
scale & heterogenous stand conditions• Lack of historical calibration data• Lack of commercial seed systems• Dynamic inter-annual land tenure / use & field boundaries
Challenges of biomass estimation
WBSs2DimabiTolonNRGhana
25NOV12
9,081proto-plotsextracted(~91/km2)
Smallholder systems metrics
© DigitalGlobeWorldView28-band50cm PAN200cm MUL
WBSt2NanposelaKoutialaSikassoMali
26OCT12
7,399proto-plots extracted(~38/km2)
Smallholder systems metrics
© DigitalGlobeWorldView28-band50cm PAN200cm MUL
WBSt1SukumbaKoutialaSikassoMali
26OCT12
5,580proto-plots extracted(~38/km2)
Smallholder systems metrics
© DigitalGlobeWorldView28-band50cm PAN200cm MUL
Smallholder systems metrics
Locally dominant crops – cotton belt, Mali
Land use survey, Aboveground biomass measurements
Class Number of samples
Bare Soil 10Cotton 154Grass + pasture + fallow 32Groundnut / legumes 32Maize 51Millet 104Rock Outcrops 2Sorghum 51Wetland + ponds 15Wild vegetation 21
total 472
Number of fields
Crop Age (Day) Crop Biomass (d[DW] m-2)
Avg. Stdev CV (%) Avg Stdev CV (%)Cotton 8 96 3 3 110 69 63Maize 9 78 7 9 143 71 50Millet 8 98 4 4 181 118 65
Sorghum 9 77 6 8 114 71 62Total 34 86 11 13 136 85 62
Biomass-NDVI relationship, crop & sensor-wiseCoton: biomasse=f(NDVI), n=12
R2QB = 0.653
R2SP = 0.716
R2AL = 0.763
R2MD = 0.538
0
100
200
300
400
0.3 0.4 0.5 0.6 0.7
Maïs: biomasse=f(NDVI), n=9
r2QB = 0.366
r2SP = 0.316
r2AL = 0.303
r2MD = 0.148
0
100
200
300
400
0.2 0.3 0.4 0.5 0.6
Sorgho: biomasse=f(NDVI), n=9
R2QB = 0.544
R2SP = 0.389
R2AL = 0.204
R2MD = 0.191
0
100
200
300
0.2 0.3 0.4 0.5 0.6
Mil: biomasse=f(NDVI), n=11
R2QB = 0.702
R2SP = 0.697
R2AL = 0.421
R2MD = 0.440
0
100
200
300
400
0.2 0.3 0.4 0.5 0.6
1000
500
0
QuickBird SPOT ASTER MODISAggr
egat
ed b
iom
ass
estim
ate
(met
ric to
ns)
u=1, aucune connaissance de l’utilisation des terres a prioriu=2, coton et céréales séparéesu=4, coton, maïs, mil, sorgho séparés
u=1
u=2
u=4
Aggregate biomass estimate(co 187, ml 132, mz 63, sg 88)
u=1, no a priori knowledge of land useu=2, cotton and cereals separatedu=4, cotton, maize, millet, sorghum separated
Measured and predicted crop biomass
Contour ridgetillage effectson yield, biomass
Contour ridge tillage effects on NDVI• 38 field pairsmonitored (samecatena class, samefarmer, contiguous,trees removed)• Stdev(NDVI) differsin 82% of pairs (50% inCRT fields)• Mean NDVI differs in87% of pairs (55% in CRTfields)
• Intra-specific variability in reflectance is larger than inter-specific variability (time-specific, with exceptions)
• Spatial uncertainty inherent to biomass predictions does not change significantly from 2 to 30m resolution (time-specific)
• RMSEP (DM) modestly decreases with model complexity• Cloud cover remains a major constraint to peak biomass acquisitions• Discriminating between cotton and cereals important for unbiased
landscape-scale biomass estimates• Tree management is independent of underlying crop type – tree mask
required for crop recognition• Stereoscopic (or lidar) monitoring of canopy height next quick & dirty
improvement for biomass estimates
Learnings
ICRISAT is a member of the CGIAR Consortium
Thank you!