Earth Observation for Agriculture – State of the - PowerPoint Presentation

Slide1

Earth Observation for Agriculture – State of the Art –

F. BaretINRA-EMMAHAvignon, France

1Slide2

Outlook

The several needs for agricultureObservational RequirementsVariables targeted / accessibleSpatialTemporalRetrieval of key variables from S2 observationsGeneric algorithm

Specific algorithmAssimilationConclusion/recommandations2Slide3

The several needs for agriculture

Regional/International

Local

Statistics

Control

Precision agriculture

Farmers

Tools

Seeds

Fertilizer

Pesticide

Dealers

Insurance

Governments

Food Industry

Cooperatives

Consultants

Traders

Governments

Food Industry

3Slide4

From

observations to applications

Structure

Biochemical

content

Soil

Atmosphere

Canopy

Functioning

Models

Assimilation of radiances

Biophysical

variables

estimates

(

Products

)

Assimilation of

Products

Need

for biophysical products (LAI, fAPAR, fCover, Albedo) and their dynamicsUsed as indicators for decision making Input to crop process modelsSmooth expected temporal course (allows smoothing / real time estimates)Allows validation

Provide uncertainties

Need for crop classification4Slide5

Observational requirements: Variables targetted (and accessible!)

Biophysical variables of interest:LAI (actually GAI)Green fraction (FAPAR, FCOVER)

Chlorophyll contentWater contentSoil related characteristics Crop residue estimates

5Slide6

Spectral requirements

Correction for the atmosphereSampling the absorption of main leaf constituants

6Slide7

Observational requirements:

Spatial resolutionPrecision agriculture: intra-field variability

Other applications:FieldsSpecies (regional assessment of production)

Number

of patches/pixel

Purity

of pixel

Variability

within

pixel

Large differences between 10-20-60 m with 100-250-1000m

7Slide8

Observational requirements:

Revisit frequencyProviding information on crop state at specific stages (± 1 week)Monitoring crops for resources management

Green Fraction

Green Fraction

Getting information every 100°C.day:

One month in winter

5 days in summer

Accounting for clouds (≈50%

occurence

)

8Slide9

Retrieval of key variables from S2: Generic

algorithmsApplicable everywhere with variable accuracy but good consistency

Allows continuity with hectometric/kilometric observationsBased on simple assumptions on canopy structure

9Slide10

Retrieval of key variables from S2: Generic

algorithms applied to several sensors

Capacity to build a consistent time series from multiple sensors Virtual constellation Possible spectral sensitivity residual effects

Time

SPOT4

Rapideye

IRS

SPOT4

Landsat

Landsat

SPOT4

SPOT4

DMC

Grassland_1

Shrubland

Forest (

oak

)

Grassland_2

10Slide11

Retrieval of key variables from S2: Specific

algorithms

Need knowledge of land-use (species / cultivars)

On the fly land-use (continuously updated)

Allows using prior distribution of canopy characteristics

Canopy Structure

Leaf properties (structure, chlorophyll, SLA, water, surface effects …)

Need calibration over

detailed

radiative

transfer model

Comprehensive experiments

11Slide12

Calibration over radiative transfer models

Generic

(

Turbid

)

Specific

(3D)

Measured

LAI

Measured

LAI

Measured

LAI

Measured

LAIEstimated LAI

Estimated

LAIEstimated LAIEstimated LAI

MaizeVineyardFrom Lopez-Lozano, 2007

Better use more realistic 3D model than turbid medium (generic) model

12Slide13

Calibration over experiments

Green Fraction

Use of (HT) phenotyping / agronomical

Experiments

13

Characterize specific structural traits Slide14

Combination with crop

models

?

Variables of interest

Radiance observations

Process model

(dynamic)

Model

Parameters

Diagnostic variables

Radiative Transfer Model

Ancillary Information/data

Assimilation allows to:

input additional information in the system:

Knowledge on some processes

Exploitation of ancillary data (climate, soil, …)

exploit the temporal dimension: process model as a link between dates

access specific processes / outputs (biomass, yield, nitrogen balance)

Run process models in prognostic mode : simulations for other conditions

14Slide15

Combination with crop

modelsExample of assimilation

Question: How to optimize the nitrogen amount for a field crop ?

Inputs: Climate (past) Soil (Prior knowledge of characteristics, but no spatial variability)

Technical practices (sowing date, …)

Crop model (STICS) and some crop parameters

3 flights with CASI instrument

Outputs:

Map of nitrogen content (QN)Slide16

Assimilation of (RS) observations

Prior distribution of

inputsClimate past'

Soil

Cultural

Pract

.

Crop

model

Prior distribution of

outputs

LAI, Cab

200 000

cas

Cost function

Remote sensing

Estimates

LAI, Cab

Posterior distribution of

inputs

1 000

cases

16

Actual QN (kg/ha)

Posterior QN (kg/ha)

Flight 1

Flight 2

Flight 3

Actual QN (kg/ha)

Prior QN (kg/ha)

Flight 1

Flight 2

Flight 3Slide17

Conclusion & Recommandations

Organize the validation / calibration to capitalize on the work doneBuild an archive (anomalies)Fusion with other missions for improved revisit frequency at the level of biophysical variables (or higher) productsdecametric missions (Rapid-eye, DMC, Venµs

, , SPOT6/7, LDCM…)hectometric resolution observations (PROBA-V, S3 …)Development of algorithms for:Top of canopy fused products at 10 m resolution and original resolutionson the fly classification (continuously updated)specific products per crop/cultivarPatch (object) oriented algorithm to take into accountthe continuity within patchesThe variability within patches (texture)

Development of combination of S2 data with crop models (Assimilation)Improved description of canopy structure by models in relation to functionSimplification of crop models (meta-model)17

S2 very well adapted to requirements for agriculture

Following issues to be solved: