the impact of drainage management technology in agriculture: a
TRANSCRIPT
The Impact of Drainage ManagementTechnology in Agriculture:
A Spatial Panel Data Model
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown,Adela Nistor and Jess Lowenberg-DeBoer
Purdue Agricultural Economics
August 27, 2009
III World Conference of Spatial EconometricsBarcelona 2009
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
field water management
A DOUBLE-EDGED SWORD
More is not always better when it comes to water and row-crops.Depending on the stage of growth, lack or excess of watermaybe harmful
Consequences of wet conditions (e.g. corn):field inaccessible for operations such as planting or spraying offertilizer or pesticides
late planting causes yield loss despite fast growing varieties.late application of nitrogen may further delay planting / untimelyspraying of pesticides may increase weed and disease stress onplant
standing water on field may asphyxiate young roots and/or causedevelopment of diseases
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
field water management
DRAINAGE 101
Solutions?dry conditions⇒ irrigation: gravity, sprinkler or dripwet conditions⇒ drainage : conventional or controlled
Drainage:conventional: a network of tiles and ditches which maintain thewater table below a fixed level by evacuating the excess water tothe nearest creekcontrolled: upgrades the conventional with a system of logsallowing for water table to be set at a specified level.
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
field water management
ACTUAL AND EXPECTED BENEFITS
lower level for heavy machinery to enter the field whenoperations are requiredhigher level during off periods (winter and mid-summer)
Expected benefits:technical - yield increase
better control of field operations timingmaximizes water availability at critical stages of growth whilepreventing flooding
environmental - limit chemical runoffpotentially reduces pollution of downstream rivers and water bodies(e.g. Gulf of Mexico)
ä Private and public interestse.g. Environmental Quality Incentives Program for Indiana
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
a multidisciplinary project
THE DRAINAGE PROJECT
USDA-CSRES Grant #2004-51130-03111“Drainage Water Management Impacts on Watershed Nitrate Load,Soil Quality and Farm Profitability”
Work in the Agricultural Engineering and Agricultural Economicsdepartments at Purdue University
four sites: Davis Purdue Agricultural Center (DPAC) and threeprivately-operated farms in Indianadata collected with yield monitor systems as far back as 1996 forDPACcontrolled water management system functional from 2005 on
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
a multidisciplinary project
THE AGECON SIDE
Jason P. Brown (2005) (M.S. thesis): site-specific yield responseto controlled drainage - cross-sectionAdela Nistor (2007) (Ph.D. dissertation): yield response andcost-benefit analysis - panel datacurrent: refine the yield response analysis using more advancedspatial panel techniques and updated data
this paper: Field W - DPAC farm - Randolph county, IN, USA
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
building the grids
COLLECTING YIELD DATA
data collected with AgLeader yieldmonitor linked to a GPS mounted onthe combinecorn was planted in
East: 1996, 1998, (2000), 2002,2005-2008West: 1996, 1998, 2001, 2003,2005-2008
ä Years with controlled drainage in red
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
building the grids
THE ART OF CLEANING DATA
Measurement errors are numerous.Clean-up is done on the basis of“combine dynamics” criteria:
minimum and maximum yieldminimum and maximum combinespeedgrain flow delaystart- and end-pass delay
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
building the grids
“GRIDDING” DATA
Why aggregate?need for data spatially balanced in alldirectionsimproves precision of estimator (butinduces heteroskedasticity)
How to aggregate?
cell size: average combine passwidth over years (e.g. 5m)each cell contains the average yield,the standard deviation and the countof the points falling within its limits
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
the left-hand side
YIELD RESPONSE MODEL REGRESSORS
Left-hand side variables:elevation
interpolated from field measuresusing IDW power 1 method
conventional/controlled drainagedummy variable D
D =
1 if controlled0 if free flow
precipitationtotal rainfall during critical growthperiod (i.e. July 5 to Sept. 5)
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
crop yield response
A SIMPLE LINEAR MODEL
Heady and Dillon (1961) provide a review of algebraic functionalforms for crop response estimation
ä critical factors - rain, elevation and slope
This paper - linear model with year fixed effects, rainfall, elevation anddrainage dummyInteractions:
D × year - differential response of controlled drainage from oneyear to another?D × elev - ...from one location to another?rain × elev
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
crop yield response
CORRECTION FOR HETEROSKEDASTICITY
linear crop response model
yield = α + year βββ1 + D × year βββ2
+γ1 elev + γ2 D × elev + δ rain × elev
Because of aggregation method, the LHS variable is actually averageyield whose variance varies over grid cells.
Var(yield i ) =σ2
ini
i = 1, · · · ,N (1)
ä inherent heteroskedasticity Correction: divide LHS and RHS of equation (??) by σi√ni
, thestandard error.
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
spatial panel
THE WHYS AND THE PROS
Why panel?control for (unobserved) individual heterogeneity,
hereby reducing omitted variable bias
Why random effects?yield monitor data is a sample rather than a
population (Griffin et al., 2005)
Why spatial?contemporaneous spatial dependence between
observations at each point in time and spatial heterogeneity mayarise when panel data include a location component (Anselin, 1988;Elhorst, 2009)
Why autoregressive error?spatial autocorrelation is due to omitted
variables rather than to the effect of corn yield grid cells on each other(Anselin et al., 2004; Lowenberg-DeBoer et al., 2006)
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
spatial panel
NOT SO SIMPLE ANYMORE...
Following Baltagi et al. (2007), the random effects model withspatially autocorrelated error components can be formulated for eachtime period t as:
yt = Xtβββ + ut , t = 1, . . . ,Tut = u1 + u2t
u1 = ρ1WNu1 +µµµ, µµµiid∼ N (0, σ2
µ)
u2t = ρ2WNu2t + ννν t , ννν tiid∼ N (0, σ2
ν) (2)
The vector of errors ut is composed of two spatially autocorrelatedcomponents, one time-invariant and unit-specific u1, and onetime-varying u2t , with two separate autocorrelation parameters.
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
spatial panel
Stacking the model for each time period so that the slower index istime and rearranging equation (??) into a single equation yield thefollowing reduced form:
y = Xβββ + (INT − ρ1WNT )µµµ+ (INT − ρ2WNT )ννν (3) Note:Because of the time invariance, µµµ = ιιιT ⊗µµµ where ιιιT is a vectorof ones of dimension T .WNT = IT ⊗WN is block diagonal
calculations of (INT − ρr WNT )−1 and |INT − ρr WNT | involve N × Nmatrix instead of NT × NT (r = 1, 2)renders estimation more computationally manageable
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
spatial panel
MAXIMUM LIKELIHOOD ESTIMATION
Concentrated likelihood function:
LLc(ρ1, ρ2, φ) = −C − NT2
ln σ2ν(ρ1, ρ2, φ)− 1
2ln detΣΣΣu(ρ1, ρ2, φ)
where
φ =σ2
µ
σ2ν
σ2ν =
u(βββ)′ΣΣΣ−1u u(βββ)
NTβββ = (X′ΣΣΣ−1
u X)−1X′ΣΣΣ−1u y
ΣΣΣ−1u = σ2
νΩΩΩ−1u
ln detΣΣΣu = ln det[Tφ(A′A)−1 + (B′B)−1] + (T − 1) ln det(B′B)−1
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
exploratory spatial analysis
SUMMARY STATISTICS
N/S 1996 1998 2002 2005 2006 2007 2008Minimum 47/59 76/113 11/10 79/77 106/96 50/33 110/106Maximum 131/131 200/189 98/105 237/221 218/219 163/154 239/238Mean 97/99 145/144 51/45 154/174 175/172 107/107 192/192SD 12/13 17/15 19/19 30/22 20/20 20/22 23/22(Whole field)Minimum 47 76 10 77 96 33 106Maximum 131 200 105 237 219 163 239Mean 98 145 48 164 174 107 192SD 13 16 19 28 20 21 23rain (in) 3.60 4.03 2.53 5.67 3.78 8.24 5.38
Table: Corn yield (bu.a−1) and precipitation - Davis, Field W, EAST
N/S 1996 1998 2001 2003 2005 2006 2007 2008Minimum 32/35 106/56 104/112 56/52 79/89 86/81 51/52 107/124Maximum 119/121 200/208 227/232 184/188 206/209 220/216 151/141 239/236Mean 81/89 151/138 177/175 137/123 150/156 167/155 110/104 196/189SD 18/13 20/22 20/19 22/28 22/19 22/24 16/18 21/21Whole fieldMinimum 32 56 104 52 79 81 51 107Maximum 121 208 232 188 209 220 151 239Mean 85 145 176 130 153 162 107 193SD 16 22 19 26 21 24 18 21rain (in) 3.60 4.03 4.96 7.33 5.67 3.78 8.24 5.38
Table: Corn yield (bu.a−1) and precipitation - Davis, Field W, WEST
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
exploratory spatial analysis
SPATIAL AUTOCORRELATION?
1996 1998 2002 2005 2006 2007 2008EAST 0.48*** 0.68*** 0.62*** 0.63*** 0.56*** 0.73*** 0.62***
1996 1998 2001 2003 2005 2006 2007 2008WEST 0.71*** 0.83*** 0.37*** 0.62*** 0.58*** 0.62*** 0.67*** 0.61***
Table: Moran’s I (yields), Davis, Field W
ä Evidence of strong within year positive spatial autocorrelation
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
reports
yield RE SEM-RE KKP BEP
intercept 104.30 *** 104.21 *** 104.23 *** 104.33 ***( 0.407 ) ( 0.434 ) ( 0.434 ) ( 0.438 )
years *** *** *** ***
elev 13.12 *** 14.10 *** 14.05 *** 13.76 ***( 0.963 ) ( 0.961 ) ( 0.959 ) ( 0.958 )
year05× D 15.62 *** 12.23 *** 12.22 *** 12.17 ***( 1.744 ) ( 1.842 ) ( 1.837 ) ( 1.828 )
year06× D 0.89 2.19 1.96 1.62( 2.066 ) ( 2.420 ) ( 2.409 ) ( 2.386 )
year07× D 1.35 0.90 0.89 0.84( 1.096 ) ( 1.123 ) ( 1.12 ) ( 1.114 )
year08× D 33.70 *** 32.57 *** 32.60 *** 32.72 ***( 1.694 ) ( 1.856 ) ( 1.850 ) ( 1.834 )
elev× D 1.69 *** 1.72 *** 1.72 *** 1.74 ***( 0.250 ) ( 0.233 ) ( 0.232 ) ( 0.234 )
elev× rain -5.71 *** -5.97 *** -5.96 *** -5.93 ***( 0.209 ) ( 0.202 ) ( 0.202 ) ( 0.201 )
ρµ — — 0.527 0.949ρν — 0.538 0.527 0.477φ — — — 0.0146Log Likelihood -63747.6 *** -62644.0 -62631.6 -62565.0LR vs BEP 2364.1 *** 157.4 *** 132.0 *** —
Table: Estimation results for Davis field W - East
ä BEP model stands implying both time-variant and time-invariantspillovers
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
reports
yield RE SEM-RE KKP BEP
intercept 76.80 *** 76.45 *** — 76.45 ***( 0.417 ) ( 0.462 ) ( 0.462 )
years *** *** ***
elev 30.36 *** 28.97 *** — 29.00 ***( 1.890 ) ( 1.973 ) ( 1.974 )
year05× D 1.89 1.41 — 1.37( 3.056 ) ( 3.344 ) ( 3.346 )
year06× D 35.94 *** 37.46 *** — 37.46 ***( 2.864 ) ( 3.190 ) ( 3.191 )
year07× D 2.31 1.77 — 1.75( 2.417 ) ( 2.446 ) ( 2.446 )
year08× D 13.766 *** 14.52 *** — 14.51 ***( 0.964 ) ( 0.963 ) ( 0.963 )
elev× D -33.76 *** -33.90 *** — -33.87 ***( 2.288 ) ( 2.278 ) ( 2.278 )
elev× rain -7.02 *** -6.69 *** — -6.69 ***( 0.469 ) ( 0.486 ) ( 0.486 )
ρµ — — — -0.562ρν — 0.568 — 0.57Log Likelihood -70651.1 -69101.7 — -69100.7LR vs BEP 3100.8 *** 1.91 — —
Table: Estimation results for Davis field W - West
ä SEM-RE model is the final specification implying only transitoryspillovers
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
reports
MARGINAL EFFECTS
The marginal effect of the watermanagement system in year t(t = 2005, . . . , 2008) depends directlyon elevation:
(∂yield∂D
)t= β2t · yeart + γ2 · elev
∆yield 2005 2006 2007 2008EASTbu.ac−1 14.1 2.0 2.0 34.7% 8.1 1.2 1.9 18.1WESTbu.ac−1 -17.7 19.8 -17.7 -3.2% -11.8 11.9 -16.1 -1.6
Table: Average effect of controlleddrainage on yields
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
IN A FEW WORDS
impact of controlled drainage varies:from year to year: positive or negative, with large variancefrom field to field:
predominantly negative for the West sidepositive across all years for East side
within field (direct function of elevation)
year to year difference explained by significance of year × Dinteraction
when significant, positive effect on yieldsinsignificant in 2006 and 2007 for East and 2005 and 2007 for West
presence of spatial autocorrelation in the error termhowever, autocorrelation of the ramdom effects relevant for East
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model
FUTURE DIRECTIONS
1 fine tune and improve the agronomic model specification inparticular on the inclusion of precipitations
take into account not only quantity but variation of rainfall overcritical growth period or yearuse precipitation data after logs are installed in the spring afterplanting. hypothesis: if rainfall is low, there is nothing to hold back and
controlled drainage would have no impact
2 estimate similar models for the three other sites in the project3 discuss results with agricultural engineers and agronomists for
interpretation
Benoıt A. Delbecq, Raymond J.G.M Florax, Jason P. Brown, Adela Nistor and Jess Lowenberg-DeBoerThe Impact of Drainage Management Technology in Agriculture: A Spatial Panel Data Model