a new operational tool derived from remotely- sensed vegetation metrics for drought monitoring and...

A New Operational Tool Derived from Remotely-sensed Vegetation Metrics for Drought Monitoring

and Yield Estimation

Ping Zhang

Department of Geography and Environment

Dissertation CommitteeBruce T. Anderson

Ranga Myneni

Mathew Barlow

Nathan Phillips

Curtis Woodcock

Jan, 2007 Ph.D. Defense Ping Zhang 2

Outline

Introduction Derivation of the Climate-Variability Impact Index (CVII)

and its application in drought diagnosis and monitoring Modeling crop yield from CVII and its potential

application in near real-time crop monitoring Application of the CVII for early crop yield forecasting in

drought-stricken regions – Case studies in US Concluding remarks and future directions


Backgrounds

Climate control on vegetation Agriculture drought indices based on meteorological data

Rainfall deficiency index Palmer’s Crop Moisture Index (CMI) (1968): U.S. Department of

Agriculture Standardized Precipitation Index (SPI) (1993): National Drought

Mitigation Center US drought monitor map (1999): US Drought Monitor

Agriculture drought indices based on satellite data Vegetation indices from Landsat (1980s) Vegetation Condition Index (VCI) from AVHRR NDVI (1995) Standardized Vegetation Index from AVHRR NDVI (2002)

Research questions Can we identify and quantify the climatic impact on vegetation at regional

to global scales using remote sensing data? And can we construct an operational tool for real-time drought monitoring

and early yield estimation using these data?


Datasets and compilations

Domains: Global vegetation pixels and crops in U.S. Scales: county, crop reporting district, state, country, and global Datasets:

Satellite-based vegetation indices including AVHRR, MODIS LAI and NDVI Survey-based crop yield estimations from USDA NASS and FAO STAT Climate data including CMAP precipitation, NCEP temperature, and ISCCP

radiation Drought indices based on meteorological records including SPI and US drought

monitor maps GIS boundary maps from the National Atlas of the United States


Outline






0 2 4 6 8 10 120

1

2

3

4

5

0.0

0.2

0.4

0.6

0.8

1.0Deciduous Broadleaf Forest

0 2 4 6 8 10 120

1

2

3

4

5

0.0

0.2

0.4

0.6

0.8

1.0Cropland

Part One: Characteristics of MODIS vegetation indices

).,('10/)),('),('()(minminmax

mlLmlLmlLlT

L' (l,m) 1

N p

1

Ny

L(l, p,m,y)y

p


Part One: Characteristics of MODIS vegetation indices


Part One: Construction of a Climate Impact Index

Climate Impact SensitivityGiven estimates of the growing season length, as well as the annual growth, we can construct a climatological-based Climate Impact Index (CII). This index is designed to identify the monthly contribution of growing-season activity to annual production (normalized for a given land cover type), which in turn provides an estimate of the vulnerability of a particular region to climate variability during the growing season.

)(),()(),()(12

1

'12

1

' lGmlLpGmpLpCIImm



0 2 4 6 8 10 120.00

0.05

0.10

0.15

0.20

0.25Normalized Climatological LAI Sequence

North Africa North-east Asia Europe North America

0 2 4 6 8 10 120.00

0.05

0.10

0.15

0.20

0.25

Normalized Climatological NDVI Sequence

North Africa North-east Asia Europe North America


Part One: Construction of a Climate-Variability Impact Index

Monitoring of Drought To see how the CII can be used for agricultural monitoring, we

used AVHRR LAI to generate a Climate-Variability Impact Index (CVII). Here the CVII was designed to quantify the percentage of the climatological annual grid-point production either gained or lost due to climatic variability in a particular month.

CVII(p,m,y) 100 L(p,m,y) L'(p,m)

L'(p)


Part One: Application of CVII in drought monitoring

1988 June 1988 June-July1984 August 1984 August-September


Part One: Summary

MODIS LAI can quantify spatial differences between productivity of ecosystems as well as seasonal variations within ecosystems.

A Climatic Impact Index (CII) is derived to provide additional information regarding the potential sensitivity of vegetation to changes in climatic variables by accounting for the length of growing season.

Major drought events are well-captured by the similarly derived Climate-Variability Impact Index (CVII).

Zhang, P. et al., Climate related vegetation characteristics derived from MODIS LAI and NDVI, J. Geophys. Res., VOL. 109, 2004.


Outline






Part Two: the Relation between CVII and Yield Estimation

Correlation between CVII and Yield at Different Scales Local scales (county- and CRD-level)

Representative regions: Illinois and North Dakota Representative crops: corn and spring wheat Study period: 2000-2004

Regional scales (state-level) Representative regions: 3 regions in US Representative crops: corn, spring wheat, and winter wheat Study period: 1985-1989

National scales Representative regions: 4 European countries Representative crops: winter wheat Study period: 1982-1999



m

m

mpL

mpLympLypCVII

),('

),('),,(100),(

Y

YY 100



Prediction of Yield with CVII Homogeneous prediction

Train the model based on 2000-2003 data in Illinois Predict the corn yield of 2004 in Illinois

Heterogeneous prediction Train the model based on 1985-1989 data in winter wheat regions in

US Predict the winter wheat yield of 1985-1989 in 4 European countries Predict the winter wheat yield of 1982-2000 in 4 European countries



22 )()(

)()(

yyxx

yyxxr

ii

ii

EN

ECHSS

0.50

0.75

1.00

1.25

1.50

1.75

0.50 0.75 1.00 1.25 1.50 1.75

2000-2003 2004

Model prediction

US

DA

Est

ima

tes

R=0.97HSS=1

0.6 0.8 1.0 1.2 1.4

0.6

0.8

1.0

1.2

1.4

R=0.69 HSS=0.61

FA

O E

stim

ates

Model Prediction



Operational Tool for Agriculture Monitoring To study the relation between monthly CVII and growing-season

yield, we examine the timing of the relationship between CVII and crop yields. Here, we construct a statistical model for yield:

This linear equation is examined four times with different regression selection procedures. Ideal model Cumulative model Chronological model Fixed-coefficient model

iiCVIIY


Part Two: Operational Tool for Agriculture Monitoring

Ideal Model Provides information about which months within the

phenological cycle are most strongly related to the overall yield Ideal model contains no redundant predictor and provides the

best correlation between CVII and yield. Important predictors are different for various crop types and

study areas

Cumulative Model CVII is progressively accumulated month by month over the growing

season. At any point in the growing season, the cumulative CVII represents the variations in the total growth up to that point.

Coefficient will change over the growing season.



Chronological Model CVII series for each month of the growing season are added in the

model chronologically.

Coefficients will change with each subsequent CVII value added during the course of the growing season

Fixed-coefficients Model Additional predictor variable is regressed with the residual of Y, instead

of Y itself. Calculates partial F statistic to test whether the addition of one

particular predictor variable adds significantly to the prediction of Y achieved using the pre-existing predictor variables.

nnCVIICVIICVIIY ...2211



Evolution of the R-square of the models as a function of the number of predictors.



1982 1984 1986 1988 1990 1992 1994 1996 1998

0.4

0.6

0.8

1.0

1.2

1.4

USDA Estimate vs Chronologial (R=.82) and Fix-coefficient (R=.83) Prediction (LAI)

Chronological Fix-coefficent USDA estiamte

1982 1984 1986 1988 1990 1992 1994 1996 1998

0.4

0.6

0.8

1.0

1.2

1.4

1.6

USDA Estimate vs Chronologial (R=.61) and Fix-coefficient (R=.54) Prediction (NPP)

Chronological Fix-coefficent USDA estiamte


A quantitative index, Climate-Variability Impact Index, is introduced to study the relationship between the remotely-sensed data and crop yields.

Integrated CVII is highly correlated with crop yields at different scales.

Based on this correlation, models can be applied to produce homogeneous and heterogeneous yield forecasts. single-crop CVII-production relationship may be quasi-independent of

location CVII-production relationship appears to be crop-independent for certain

crop types CVII can be exploited for crop growth monitoring by

investigating the timing of the relationship between the CVII and crop yields.

Zhang, P. et al., Potential monitoring of crop yield using a satellite-based climate-variability impact index , Agricultural and Forest Meteorology,132,344-358, 2005.



Outline






Part Three: Application of CVII at US

Extreme drought occurred in 2005 over Illinois the April-September rainfall ranked 10th lowest in the past 113 years [NCDC]

US drought in 2006 centered in North and South Dakota the persistence of anomalous warmth made the summer the second warmest

June-August period in the continental US in the past 110 years combined with the below-average precipitation, large parts of US were under

drought conditions [NOAA]


Model

Unstandardized Coefficients

95% Confidence Interval

B Std. Error Lower Upper

1 Constant 1.00 0.007 0.985 1.014

CVII 0.024 0.001 0.021 0.026

2 Constant 1.009 0.012 0.986 1.032

CVII 0.021 0.001 0.019 0.023

3 Constant 1.003 0.006 0.991 1.015

CVII 0.022 0.001 0.020 0.023

At CRD scale: Yield=0.023*CVII+1.011


Part Three: Case study in Illinois


Part Three: Case study in Illinois

Model predictions vs. USDA estimates of normalized yield in 102 Illinois counties. Yields of ‘1’ represent average conditions. Bold symbols represent the state-wide average; light symbols represent individual counties. The bold triangle represents the state-wide estimate made by USDA released in August (dark color) and September (red color) 2005.

2000 2002 2004 20060.4

0.6

0.8

1.0

1.2

1.4

1.6

Model Prediction

Yie

ld a

nom

aly

USDA Estimates


2000 2001 2002 2003 2004 2005 20060.0

0.5

1.0

1.5

2.0

0.0

0.5

1.0

1.5

2.0

Corn Wheat

Pro

duct

ion

Ano

mal

y

Model Prediction vs. USDA Production

Part Three: Case study in North and South Dakota

Model predictions vs. USDA estimates of normalized yield in North and South Dakota. Yields of ‘1’ represent average conditions. Open symbols represent the USDA estimates and filled symbols represent our model predictions. Bold symbols represent the state-wide average; light symbols represent individual counties.



USDA monitors the crop conditions and yields via monthly-conducted Objective Yield Surveys from August to November.

Measurement of stalks, ears, kernel row length, and ear diameter are obtained monthly according to the development stage

NASS then make yield forecasts at state-level by applying corn models (blue bars). The CVII model predictions (red bars) are based upon the CVII values at the end of July,

August, and September.

2000 2001 2002 2003 2004 2005 20060

90

100

110

120

130

Co

rn Y

ield

(b

ush

el/a

cre

) CVII predict (Apr-Jul) CVII predict (Apr-Aug) CVII predict (Apr-Sep) Actual yield NASS estimate (Aug) NASS estimate (Sep) NASS estimate (Oct)



Aug Sep Oct0

5

10

15

20

Ave

rag

e o

f ab

solu

te d

iffe

ren

ce (

%)

Availabe time

CVII prediction NASS estimate

The average (absolute) percentage difference of CVII model predictions and NASS estimates to the actual corn yield in South Dakota over the 2000-2006 period.


Part Three: Operational Product Development and Dissemination

http://ecrop.bu.edu/home.html


Part Three: Operational Product Development and Dissemination


Part Three: Application of CVII in US

Drought-monitoring indices based upon meteorological data alone can both overestimate (2005) and underestimate (2002) vegetation variations in drought-stricken regions.

The need for monitoring vegetation growth directly when estimating yield.

CVII model can provide significant predictability (less than 10% error) at the state-average level at least 1-2 months prior to the start of the harvest.

Satellites can provide a secondary, independent estimate that can pinpoint regions where agricultural failure is greatest.

The cost effectiveness and repetitive, near-global view of earth’s surface suggest this LAI-based CVII may significantly improve crop monitoring and yield estimation at regional scales.

Zhang et al., 2006. Monitoring 2005 corn belt yields from space. EOS, Transactions, AGU, 87(15), p150. Zhang et al., 2007. Application of a satellite-based climate-variability impact index for crop yield forecasting in drought-stricken regions, Agricultural and Forest Meteorology, submitted.


Outline






Concluding remarks and future directions

ConclusionThe findings of this dissertation indicate that the new remotely-sensed vegetation indices can be used to identify sensitive regions which are particularly susceptible to

agricultural failure arising from month-to-month climate variations; provide both fine-scale and aggregated information on yield for

various crops; determine when the in-season predictive value plateaus and which

months provide the greatest predictive capacity; perform near real-time drought monitoring and famine prediction at

regional and global scale; provide earlier yield forecasts for crop production before harvest

and pinpoint regions where agricultural failure is greatest.


Future directions

Validation of the CVII-production relationship Utilization of fine temporal scale data with longer coverage Validation of the CVII model using in situ data Validation of the CVII model with more crops and locations

Interactions between growing-season CVII and climatic variations in semi-arid regions The impacts of three local driving factors including precipitation,

temperature, and solar radiation The impacts of remote climate variables including sea surface

temperature, surface pressure, stream function, and vertical velocity


Contributions of the Complete Research

Development of Climate Impact Index (CII) to provide information regarding the potential sensitivity of vegetation to changes in climatic variables by accounting for the length of growing season.

Development of Climate-Variability Impact Index (CVII) to quantify the percentage of the climatological production either gained or lost due to climatic variability during the growing season. Explain more than 50% of the variance in crop yields at different scales. Train linear model using historical crop production data and satellite data Apply CVII-models to produce homogeneous and heterogeneous yield forecasts. Apply CVII-models to provide early yield estimation at drought-stricken regions

before the end of the growing season. Examination of the effect of local- and large-scale climate

variability on vegetation productivity and hence yield in specific regions.


Thank you!

a new operational tool derived from remotely- sensed vegetation metrics for drought monitoring and...

Documents

realtime drought monitoring

drought diagnosis

climate variability

climate impact indexpart

monitoring modeling

standardized vegetation

climatic impact

early crop yield forecasting