bb1999

IDENTIFYING POTENTIAL WITHIN-FIELD MANAGEMENT ZONES FROM COTTON YIELD ESTIMATES: 

IDENTIFYING POTENTIAL WITHIN-FIELD MANAGEMENT ZONES FROM COTTON YIELD ESTIMATES Broughton Boydell & A.B. McBratney T h e U n i v e r s i t y o f S y d n e y Australian Centre for Precision Agriculture

Format: 

Format Background Objectives Analytical methodology and results Accuracy and precision of estimates

Cotton in Australia: 

Cotton in Australia High input per Hectare crop Centralised and organised growing region Under environmental pressure

Yield Data F28: 

Yield Data F28 Area = 40 ha Statistics: Average = 9.9 CV = 33 Spatial statistics: Range = 120 m Structure ~2.19 1 sigma

Bare soil aerial photograph: 

Bare soil aerial photograph Evidence of temporal stability and management opportunity.

Precision Agriculture for Cotton: 

Precision Agriculture for Cotton Still no commercial yield mapping Soil sampling very expensive therefore not viable to base extensive investigations on soil sampling.

Satellite based yield estimations: 

Satellite based yield estimations 1998 launch of “Farsite” Remote sensing yield estimation service. Landsat based TM index (proprietary). NDVI Scene acquired at “cut-out” (top flower). 2 months prior to picking. Maximum possible yield.

Satellite based yield estimations: 

Satellite based yield estimations Yield estimates for large areas. Landsat archives mean yield estimates for many years past. May be used to “Grow the information pool”.

The Study: 

The Study Collect 11 consecutive years data Analyze data set to investigate presence of regions of similarity. Group multiples years of data into regions of similarity Evaluate how many consecutive years may be required to give good stable zone estimation.

Analysis: How many groups in the data set?: 

Analysis: How many groups in the data set? Used the FuZ-ME program to calculate: Fuzziness partition index FPI. 0 No membership sharing “hard classification” 1 More membership sharing between classes “Fuzzy”

Analysis: How many groups in the data set?: 

Analysis: How many groups in the data set? Used the FuZ-ME program to calculate: Mean partition entropy MPE 0 More order in grouping 1 More disorganised grouping

Results: How many groups in the data set?: 

Results: How many groups in the data set?

Results: Clustering.: 

Results: Clustering. low High

Mean cluster cotton yield bale(227kg)/Ha. For each year in the study. Field 107: 

Mean cluster cotton yield bale(227kg)/Ha. For each year in the study. Field 107 Dryland

Slide16: 

Mean cluster cotton yield bale(227kg)/Ha. For each year in the study. Field 110 Dryland

Analysis: How similar are the two maps?: 

Analysis: How similar are the two maps? Simple correlation and the Kappa index of agreement. KIA which measures the degree of agreement on a scale from zero to one. If two responses tend to agree, then most of the counts are on the diagonal of a contingency table

Results: If 11 years is stable?: 

Results: If 11 years is stable? How many years to give a close estimate to the stable map? KIA:

Slide19: 

KIA Field 110 Dryland years

Mean cluster cotton yield bale(227kg)/Ha. Dryland vs Irrigated management. Field 110: 

Mean cluster cotton yield bale(227kg)/Ha. Dryland vs Irrigated management. Field 110

Slide21: 

Yield estimations for dryland vs. irrigated type years. 4.5 Field 110

Slide22: 

KIA Field 110 Irrigated years only

Discussion: KIA: 

Discussion: KIA Irregular management years may be regarded as outliers for the purpose of farm planning Dryland vs Irrigated plans

Conclusions: 

Conclusions Stable yield zone patterns may emerge from multi-year yield estimates. Irrigated and 'dry' years need a different kind of management 5-years data (±2 years) seem to give reasonably stable estimates of yield zones. This approach essentially short circuits the 3-5 year wait for yield maps and gives producers an alternative and more immediate management option

A closer look at Farsite accuracy: 

A closer look at Farsite accuracy Comparison of FS and monitored values Correlation's average 0.60 on a 25m block. Outliers appear on field boundaries Where yield data is of questionable integrity and in regions of catastrophic yield loss (late season insects such as mites).

A closer look at Farsite accuracy: 

A closer look at Farsite accuracy

A closer look at Farsite accuracy: 

A closer look at Farsite accuracy

Slide29: 

“Farmscan” Yield Monitor Farsite 14/2/99

Conclusions: 

Conclusions Management zones for: directed sampling of soil “ “ insects “ “ weeds “ management of water Intelligent use of information which is immediately available for implementation.

Acknowledgements: 

AUSTRALIAN CENTRE FOR PRECISION AGRICUTLURE Acknowledgements CRDC and ACCRC for funding this research Grower cooperators; Peter Glennie, Nick Barton, National Mutual Farmscan, IAMA, John Deere Precision Farming Group, NESPAL & The University of Georgia, Zycom, Fugro. Brett Whelan and the ACPA