How Much Data is Enough Data…? Mike R. Duncan, Ph.D. Sarah Lepp, B.Sc. Gregor Maclean
The 3D Vineyard Engine (2006) • Creation of a 3D data-driven visualization model from 8pt/m2 LiDAR data. • The vineyard engine was connected to a database containing attribute data of over 70,000 vines • Different colours in the left image are varietals.
Vine Tracking Systems (2007) Reynold’s GPS mapped Sentinel Vines. Converted the 3D Vineyard Engine to Google Earth with 300,000 vines in the DB.
Real-Time Temperature Tracking (2007)
PrAgMatic (2007-2012)
Ensemble spectra for 16 sensors 0.5 1.0 1.5 2.0 2.5 3.0 3.5 - 1 Shows that the temperature field is as turbulent as -5/3 Kolmogorov air turbulence. - 2 Characterized by sudden - 3 violent changes. - 4 Suggests a need for high frequency sampling to capture all the variability. - 5 - 6
Sudden Violent Changes in T(t) - Shocks • A sudden rapid drop in temperature. • Shock events occupy 1% of our time series. 28 23 18 13 8 03/05/2012 0:00 04/05/2012 0:00 05/05/2012 0:00 Data from the Niagara College R-T Temperature Sensor Network – May 3 rd , 2012
Deltas/Shocks can be very ‘bad’ • The following Delta – or Shock – Event happened on March 27 th , 2012 and killed much of Ontario’s tender fruit crop. 9 Temperature in deg C 7 5 3 1 26/03/2012 0:00 26/03/2012 12:00 27/03/2012 0:00 27/03/2012 12:00 28/03/2012 0:00 -1 -3 -5 Data from the Niagara College R-T Temperature Sensor Network – March 27 th , 2012
Result of the March 27 th Shock Event Marketed Production of Fruit Crops, Ontario, 1985-2015 (million lbs) 1,200 March 27 1,000 Killer cold event 800 reduced production 600 by >300 million lbs. 400 200 0 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 apples grapes peaches strawberries pears sour cherries sweet cherries plums & prunes raspberries Total 9
Switching to Grains Yield Monitor Vines/grapes are a very managed yield which Good GPS device to map makes it hard to verify the Yield Monitor output terrain/yield relationships. In 2009 we started looking at grains and grain farms. Well tuned and well Opened the door to making maintained harvester. map tools Yellow-Gold Farms in Parkhill Ontario. Keep in mind that there’s a lag between when the harvester takes in yield and when the monitor sees it – this distance can be largish – so you need to correct for it.
Premise: Corn Yield Performance Follows Landforms LandMapR 4 2002 Corn Yield Landform Map for map for Frasier field Frasier Field in overlaid on 3D field Parkhill Ontario surface data
LandMapR: Pits (Blues) and Knolls (Yellow) Tops of Knolls Pits/Gulleys
LandMapR: High (Green) and Low (Red) Yields Lower Yield Higher Yield
Schuyler – 2001 Corn Yield/Landform Distributions 0.035 0.03 0.025 0.02 Landform 1 Landform 2 0.015 Landform 3 Landform 4 0.01 0.005 0 20 40 60 80 100 120 140 160 -0.005
Schuyler – 2010 Corn Yield/Landform Distributions 0.02 0.015 0.01 Landform 1 Landform 2 Landform 3 Landform 4 0.005 0 50 100 150 200 250 300 -0.005
Performance of Yield by Landform: The key to Variable Rate Farming Management Zones 0.35 The yield Gaps (differences) in population between over and under performing cells The correct way 0.3 probability index to compare is a 0.25 landforms and reclassification of crop performance yield values into 0.2 is via a yield performance probability map. 0.15 values relative to LandForm 1 the average LandForm 2 0.1 This graph shows performance of LandForm 3 that landform 1 the crop for a LandForm 4 0.05 shows a year. performance 0 1 2 3 4 deficit throughout The values cluster -0.05 the full range of and the clusters yield values. closely follow the -0.1 landforms -0.15 1 = P<20%, 2 = P<40%, 3 = P<70%, 4 = P=100%
LandMapR Watersheds, Wetness, Streams Local Watersheds Global Watersheds Wetness and Streamflow
PAAO Project – Precision Ag Advancement for Ontario Rx maps with validation Goals: wireless built-in! Co-operator data transfer & data submitted analyze data layers + with transparent collect geospatial mathematics for data to fill gaps teaching farmers Rx maps: implemented with industry direction, support This project was funded in part through Growing Forward 2, a federal-provincial-territorial initiative. The Agricultural Adaptation Council assists in the delivery of Growing Forward 2 in Ontario .
Variable Rate The Portal Data ‘Pipeline’ VRx Analytics Inputs RRx Yield YI Elevation Elevation Gridding/Mapping YPI/YPZones Soil Type Upload Cleaning Field (Kriging) Work Chemistry YCI Nutrients Sampling Velocities, EC Partitioning direction, boundaries, etc … Measurement Test Plots
Block Elevation RAW Kriged Cleaned
Data Cleaning Clipping the distribution to remove high and low outliers. Function removes << 1% of the data most of the time. Not the best way to do business.
Delta Clean – Point to Point Differencing Velocity, Azimuth and Yield Y A dV V -V dY -A -Y
Yield Distributions 250 200 150 Corn 1996 Corn 1998 Corn 2000 Corn 2002 100 Corn 2005 Corn 2007 50 0 20.000 40.000 60.000 80.000 100.000 120.000 140.000 160.000 180.000 200.000 220.000 Yield in bu/ac
Normalized Yield Distributions 350 300 250 Single trait seed Histogram Count Corn 1996 200 Corn 1998 Corn 2000 150 Corn 2002 Corn 2005 Corn 2007 100 Multi trait seed 50 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 Normalized Yield
Yield Stability Always Converges onto Areas of the Farm Field 0.7 Percent of Field Area 0.6 Always Over-Performing Area 0.5 0.4 0.3 0.2 0.1 Always Under-Performing Area 0 0 1 2 3 4 5 6 7
Schuyler Yield Distributions Non-normalized distributions show an average yield that varies greatly from year to year 16000 14000 12000 10000 Corn 2001 Corn 2002 8000 Corn 2004 Corn 2005 Corn 2008 6000 Corn 2010 Corn 2011 4000 2000 0 30 80 130 180 230 280 330 Yield in bu/ac
YPI Level Avg YPI YPI YPI Historical Yield Maps Yield Membership Clusters Membership Distributions Cells (6m2) Calculations
Partitioned Yield Histograms for Cul-Owned (Schuyler Farms) 7000 6000 5000 P=7 P=6 4000 P=5 P=4 3000 P=3 P=2 P=1 2000 P=0 1000 0 110 130 150 170 190 210 Yield in bu/ac
Partitioned Yield – Spatial Distribution Green is High White is Low
Summed Yield Histograms for Cul-Owned (Schuyler Farms) 12000 10000 8000 6000 P=6&7 P=3&4&5 P=0&1&2 4000 2000 0 110 130 150 170 190 210 Yield in bu/ac
Partitioned Yield – Spatial Distribution Green is High Pink is Low Yellow is Mid
VRx – Variable Rate Prescription Generator Push Button Operation. Uses a yield map, or yield index map as a pattern and takes a target yield to generate removals. Fertilizers are added to match the removals. There are seven resulting maps.
But this area can’t be any good… it’s full of sand! Yield Performance Index Map Green is High-Performing areas Red is Under-Performing areas.
Variable Rate Over Watersheds: Keeping Track Another complex calculation performed A complex calculation by LandMapR using field topography, or involving historical yield and elevation data. The figure shows local field topography performed watersheds in the field, or where water by the NC Research Crop will first pool under rain or irrigation Portal. inputs.
Return of Sensor Networks! 16 May, 2016 Temperature Event.
Adding in Real-Time Radar Rainfall Estimates
Sudden 4 deg C drop and 35% rise in RH 80 16 May 16, 2016 - 20:58(ish) 75 15 Temperature deg C 70 Humidity RH(%) 14 65 No Prior warning of the RH(Low) sudden drop in either RH% RH(High) 60 13 or in temperature – all is T(Low) T(High) moving along and then a 55 12 drop…. 50 11 45 40 10 16/05/2016 20:09:36 16/05/2016 20:38:24 16/05/2016 21:07:12 16/05/2016 21:36:00 Time
25 30 35 40 45 50 1 (Top sensors are in a canopy – T(T), RH(T), Therm(T)) Evapotranspiration Node Raw Data Output 299 597 895 1193 1491 1789 2087 2385 2683 2981 3279 3577 3875 4173 4471 4769 5067 5365 5663 T(T) 5961 6259 6557 6855 T(M) 7153 7451 7749 8047 8345 T(B) 8643 8941 9239 9537 Therm(T) 9835 10133 10431 10729 11027 11325 Therm(B) 11623 11921 12219 12517 12815 13113 RH(T) 13411 temperatures, as well as inversions. Nightfall – characterized rising humidity and falling 13709 14007 14305 14603 RH(M) 14901 15199 15497 15795 16093 RH(B) 16391 16689 16987 17285 17583 17881 18179 18477 18775 19073 19371 19669 19967 20265 20563 20861 21159 21457 0 10 20 30 40 50 60 70 80
Rover Tracks April 14 (Orange), April 20 (Green) Temperature data points at 3cm and 1.2m mapped onto Google Earth. Data rate is 1Hz. Vehiccle velocity is ~3 mph
Temp Map (High) – Temp Map (Low) – April 20 Kriging at 1m Resolution
Questions?
Recommend
More recommend
