theory of predicting crop response to non-limiting nitrogen
DESCRIPTION
Theory of Predicting Crop Response to Non-Limiting Nitrogen. What do N Rich Strips Say About N Rate Algorithms? + Quite a Bit of Geostatistics. What do N Rich Strips Say About N Rate Algorithms? – Part II, with a little geostatistics. Nitrogen Rich Strip. - PowerPoint PPT PresentationTRANSCRIPT
Theory of Predicting Crop Response to Non-Limiting Nitrogen
What do N Rich Strips Say About N Rate Algorithms? + Quite a Bit of Geostatistics
What do N Rich Strips Say About N Rate Algorithms? – Part II,
with a little geostatistics
Nitrogen Rich Strip
Apply one Non-Limiting Nitrogen strip across the field between preplant fertilization and shortly after emergence. Use this as a reference strip to determine N rate.
Concept first proposed in 1994by Dr. James Schepers
NRich (N Reference) Strip Enables Paired Comparison of Field Practice N Fertility and
Non-Limiting N FertilityPaired sampling of N Rich and Field Rate NDVI
from either IKONIS or GreenSeeker imagery
Measure Nrich NDVI and calculate expected yield
Measure Field Rate NDVI
Fp
NRichNDVI NDVI
NDVIRI
How Should We Interpret RINDVI
• In 2007, we examined RINDVI indirectly by transforming the data to potential yield
• This year, I will examine RINDVI directly and compare measured RI to RI predicted by the OSU algorithm
• The goal of this is to:– Better understand the relationship between FpNDVI
and RINDVI
– provide a method for evaluating algorithms based on measured paired comparisons of vegetative growth through part of the season
Model of RINDV for three crops and 6,216 data points
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
0.00 0.20 0.40 0.60 0.80 1.00
FpNDVI
RI N
DVI
A0 = 81A1 = 13.25A2 = 1.426R2 = 0.7624
FpNDVI1AcoshFpNDVI0ARI
2A
1Fp
RILim
0NDVI
NDVI
1Fp
RILim
1NDVI
NDVI
RI Model for WheatMarshall - GreenSeeker
0
0.5
1
1.5
2
2.5
0 0.2 0.4 0.6 0.8 1
FpNDVI
RI N
DVI
FpNDVI1AcoshFpNDVI0ARI
2A
A0 = 21.4A1 = 8.5A2 = 1.19R2 = 0.767
Miller IKONIS
0
0.5
1
1.5
2
2.5
0 0.2 0.4 0.6 0.8 1
Fp NDVI
RI N
DVI
FpNDVI1AcoshFpNDVI0ARI
2A
A0 = 33.75A1 = 9.6A2 = 1.41R2 = 0.250
Wheat RI Model from Experimental Data
Lahoma 2006
0
0.5
1
1.5
2
2.5
0 0.2 0.4 0.6 0.8 1
FpNDVI
RI N
DVI
Lahoma 2005
0
0.5
1
1.5
2
2.5
3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
FpNDVIR
I ND
VI
RI DataPowerCosh
Lahoma 2004
0
0.5
1
1.5
2
2.5
3
3.5
0 0.2 0.4 0.6 0.8 1
Fp NDVI
RI N
DVI
Location and Year Effects on RI
RI NDVI Curves Wheat - 3 Years at Two Locations
0
0.5
1
1.5
2
2.5
3
3.5
0 0.2 0.4 0.6 0.8 1Fp NDVI
RI N
DVI
Lahoma2004Lahoma2005Lahoma2006Efaw2003Efaw 2004Efaw 2005
Comparison of RI Curves Constructed from NRich Strips and Field Experiments
0
0.5
1
1.5
2
2.5
3
0 0.2 0.4 0.6 0.8 1Fp NDVI
RI N
DVI
NRich StripsNRich StripField ExptsField Experiments
RI NDVI Corn Model Calculated from Field Averages of FpNDVI and NRich NDVI
Field Average RI - Corn at 20 Locations/Years
0.00
0.50
1.00
1.50
2.00
2.50
0 0.2 0.4 0.6 0.8 1
FpNDVI
RI N
DVI
Model RI
RI Measured R2 = 0.9117
0
1
2
3
4
5
6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
NDVI
Yiel
d, M
g/ha
Yield Potential Model for All Crops Wheat Data Shown in Graph
NDVIbeaYld
00.5
11.5
22.5
33.5
0 0.2 0.4 0.6 0.8 1
Field Rate NDVI
Pote
ntia
l Yie
ld, M
g/ha
YP0YPN = RINDVI YP0YPN = YPmaxSoil/Crop DivideRI/YPmax Divide
Yield with additional N predicted by Response Index
Yield increase with additional N limited to the maximum potential yield
Response Index Theory for Fertilizer N Response
ttanConsRIYPNRIYPN
NDVI
NDVI
maxYPYPN
NDVI
2aNDVI
NDVI Fp1acoshFp0aRI
Comparing RI NDVI Model to OSU Model
0
0.5
1
1.5
2
2.5
0 0.2 0.4 0.6 0.8 1
FpNDVI
RI N
DVI
RIOSU AlgorithmNew RI Model?
Comparison of OSU Topdress Rate to RI NDVI Model Topdress Rate
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.80
10
20
30
40
50
60OSU Alg Topdress RateNew TD Rate
FpNDVI
N A
ppl.
Rat
e, k
g/ha
?
Measure of undetermined small scale variability and sampling error.
Semivariogram
Distance “range” where data is spatially related.
Overall sample variance.
Indication of spatial strength.
Region where samples remain correlated (i.e. integral scale) or region of high relatednessIntegral
Scale
Results Intermediate Scale Sensing or SamplingData Type Date Nugget Nugget:Sill Range
(m)Lag Size
(m)Lag No.
IntegralScale (m)
Model Error (RMSS)
Model Error(SME)
Yield (bu/acre) 2005 4.636 0.06 24 3.5 7 5.4 1.02 0.00381
Yield (bu/acre) 2006 0 0 24 3.5 7 5.6 0.77 0.00243
Yield (bu/acre) 2007 0 0 22 3.4 7 5.3 1.08 0.00543
NDVIGreenSeeker™ 18-Dec-04 0 0 12 0.7 17 3.8 1.91 0.00028
NDVIGreenSeeker™ 17-Mar-06 0.001 0.62 43 5.4 8 4.5 0.97 -0.00540
NDVIGreenSeeker™ 6-Apr-06 0.005 0.58 72 8.1 9 6.2 0.92 -0.00381
NDVIGreenSeeker™ 4-Mar-07 0.003 0.26 9 1.2 8 3.0 0.95 -0.00021
NDVIGreenSeeker™ 1-Mar-08 0.006 0.58 30 4.7 8 4.0 0.87 -0.00121
NDVIGreenSeeker™ 16-Mar-08 0.010 0.57 30 4.7 7 4.0 0.91 -0.00250
Soil Test NO3-N0-15cm 15-Aug-06 1.200 0.85 181 26.0 7 5.9 1.00 0.00069
Soil Test NO3-N15-30cm 15-Aug-06 0.191 0.82 174 22.0 8 6.3 1.00 -0.00975
Soil Test P 15-Aug-06 0.041 0.48 168 19.0 9 10.5 1.01 -0.01894
Soil Test K 15-Aug-06 0.011 0.46 150 19.0 8 10.1 1.01 0.00416
Soil Test pH 15-Aug-06 0.147 0.36 168 19.0 9 11.7 0.98 0.01491
Soil Test TSS 15-Aug-06 0.068 0.84 158 20.0 8 5.7 1.00 0.00475
Soil Test OM 15-Aug-06 0.012 0.36 158 20.0 8 11.3 0.92 0.00010
Soil ECVeris 0-30cm 2005 0.002 0.04 35 5.0 7 6.5 0.79 -0.00004
Soil ECVeris 0-91cm 2005 0 0 42 6.0 7 7.3 0.80 -0.00005
Recommendations for Measuring RI NDVI• Pair your farmer practice treatment and your NRich
treatments in your experiment design or (in the case of statistical purists) insert an extra farmer practice treatment which is paired with your NRich treatments.
• To maximize spatial relatedness, your sensor measurements from the two treatments should be spaced no more than three to four meters apart.
• At greater distances, relatedness declines and variability (error) in the value of RI increases.
• Remember that beyond the range measurements can be highly related by chance. Between the integral scale and the range, the odds of the measurements being highly related declines rapidly.
Conclusions• Paired comparisions between field N rate and non-
limiting N rate along an NRich strip define the relationship between the existing and optimum N application rate.
• All algorithms purporting to determine N application rate must account for the relationships between FpNDVI and NRich NDVI defined by the NRich strip.
• These relationships vary from year to year and location to location.
• The Power/Cosh model appears to accurately predict the NDVI Response Index as a function of FpNDVI.