Components of Precision Agriculture Precision Ag. Technologies and Agronomic Crop Management R.L. (Bob) Nielsen Purdue Univ, Agronomy Dept. West Lafayette, Indiana Equipment control Equipment monitoring Spatial data Mapping & GIS software Email: rnielsen@purdue.edu Twitter: @PurdueCornGuy Web: www.kingcorn.org/cafe GPS Precision Ag Agronomic knowledge Purdue Univ 1 Purdue Univ 2 Many forms of spatial data Spatial data Spatial data layers can be... Soil type, texture, elevation, drainage Grain yield & moisture Aerial imagery Satellite imagery Optical reflectance Soil test nutrients Soil test ph Electrical conductivity Manually mapped spatial features (e.g., wet spots, weeds) As applied logged rates of fertilizer, seeds, and pesticides Planter performance Tractor performance Spatially dense Many data points per unit area of land. e.g., Grain yield data sets often consist of 300 to 600 data points per acre. Spatially sparse Fewer data points per unit area of land. e.g., 2.5 ac grid soil sampling results in an average of 0.4 data point per acre. Non-sampled areas within a field represent spatial holes in the dataset with unknown values. Purdue Univ 3 Purdue Univ 4 GIS software Mapping & GIS software Spatial data density affects the Fills in the spatial holes by mathematically estimating the values in the non-sampled areas based on spatial relationships among the sampled areas. Dense data sets have fewer holes per unit land area than do sparse data sets. Thus, less interpolation is required. Thus, the resulting rasterized (smoothed) map is intuitively more believable. accuracy of the subsequent interpolated (rasterized) spatial maps. accuracy of subsequent management decisions based on the interpolated spatial maps. yield and dollar consequences of the management decisions based on the interpolated spatial maps. Purdue Univ 5 Purdue Univ 6 Purdue Univ 1
Raw yield data = Spatially dense Yield data every second or two Corn yield map of 47 ac field 12 row head; 3 to 4 mph Yield data collected every 2 seconds at 3.5 mph equals 1 data point every 10 ft of linear travel. Equal to 6,824 data points on 47 ac Equal to 145 data points/ac with a 12 row head. Purdue Univ 7 Purdue Univ 8 Interpolated yield data map Interpolated yield data map Purdue Univ 9 Purdue Univ 10 Little practical difference between raw & interpolated maps* Soil sampling = Spatially sparse One common spatial density for grid soil sampling programs is one per 2.5 acres. A single sample might consist of 8 to 12 soil cores pulled from within a ~ 40ft diameter circle and bulked into a single sample for nutrient analysis. Image: RLNielsen * With respect to spatially dense data Purdue Univ 11 Purdue Univ 16 Purdue Univ 2
Soil sampling data = Spatially sparse 25 ac field, one soil sample per 2.5 acres 12 soil samples, each ~ 40ft diameter area Everything else represents spatial holes that must be interpolated with GIS. That s a lot of interpolation! Purdue Univ 17 Let me illustrate w/ an example using bare soil imagery & soil test organic matter estimates Satellite image of a 30-acre field. Infra-red portion identifies dark & light soils Thus, used as a proxy for soil organic matter Both 2.5 acre & half-acre grid soil samples Organic matter values were interpolated to develop smoothed rasterized soil organic matter maps. Purdue Univ 18 Satellite image of bare soil & soil map units Blount SiL Aerial image vs. 2.5-ac SOM map Pewamo SiCL Blount SiL Blount SiL Surface soil reflectance = Proxy for SOM Rasterized SOM map interpolated using 2.5 ac soil sample grid points Satellite IR re classified w/ brown color gradient, Fld M2, DavisPAC Purdue Univ Blount SiL 19 Surface soil reflectance pretty good proxy for soil organic matter Purdue Univ 20 Aerial image vs. half-ac SOM map Closer approximation to reality Let s consider soil test Bray P1 Rasterized map based on 2.5 ac samples Rasterized map based on 0.5 ac samples Surface soil reflectance = Proxy for SOM Rasterized SOM map interpolated with half ac soil sample grid points Red = Deficient (less than 15 ppm) Gray = Adequate (15 to 30 ppm) Blue = High (greater than 30 ppm) Purdue Univ 21 Purdue Univ 22 Purdue Univ 3
The spatial lesson here is that GIS software will do what you tell it to do, but does not mean that the resulting rasterized maps are spatially accurate. Questionable spatial accuracy of rasterized sparse data naturally leads to questionable accuracy of spatial agronomic decisions made on the basis of the rasterized spatial data. Sparse spatial data GIS Uncertain rasterized spatial accuracy Uncertain rasterized spatial accuracy Good agronomy Uncertain agronomic results Purdue Univ 23 Purdue Univ 24 Difference in P 2 O 5 Spread Maps Spread map based on 2.5 ac samples Spread map based on 0.5 ac samples Soil test P Based on 2.5ac samples P2O5 spread map Soil test P Based on half ac samples P2O5 spread map Red = No P2O5 recommended Yellow = Drawdown P2O5 application rates Green= Replacement of crop removal (60 lbs P2O5) Blue = Crop removal + buildup application rates Purdue Univ 25 Purdue Univ 26 So, the question is Can you afford to soil sample as intensely as a half-acre spatial density? A 2.5 ac grid sample is admittedly more accurate than traditional whole field sampling. Can you afford NOT to soil sample as intensely as a half-acre spatial density? What is the value of an accurate baseline? Maybe it is a one-time investment, followed by less intensive subsequent soil sampling. Alternatively Supplement 2.5 ac grid sampling by using yield data or imagery to identify spatially odd field areas that deserve additional, targeted, soil sampling. Purdue Univ 28 Wintex Purdue 1000 Univ automated soil sampler w/ 29 GPS Purdue Univ 4
Agronomic Factors Supplement 2.5 ac grid samples with additional samples targeting low-yielding areas (red) Labor efficiency Operator fatigue? Document spatial variability Crop input efficiency Identify spatial YLFs e.g., Seed savings w/ automatic row shutoffs Equipment efficiency Yield or profit Manage spatial YLFs Purdue Univ 30 e.g., On the go adjustments to planter down force Precision Ag Potential Purdue Univ YLFs = Yield limiting factors 31 The key to consistently producing high-yielding corn... is the ability to accurately identify AND successfully mitigate the YLFs specific to your farming operation. Take advantage of handheld GPS technologies to map, GPS-tag & document problem areas in your fields. Crop scouting & mapping apps Simple note-taking apps Smartphone cameras Use with other GIS information to help diagnose possible causes of problems Purdue Univ 32 Image: RLNielsen Purdue Univ 33 Take advantage of your dense spatial data sets to help you visualize problem areas and then literally navigate to those areas in the field with your handheld app to diagnose or verify the causes. Remotely sensed imagery supplements yield maps in identifying and locating problem areas within your fields. can identify problem areas prior to harvest. May enable earlier & more accurate crop problem diagnostics and, possibly, in-season mitigation of crop problems (foliar fungicide, late N applic s). does not, however, diagnose the causes of crop problems by itself. can vary in quality among vendors. Digital resolution, spatial accuracy, flight conditions, post-processing of imagery. Purdue Univ 34 Purdue Univ 38 Purdue Univ 5
Remotely sensed imagery Equipment-mounted crop sensors e.g., GreenSeeker, OptRx Satellite imagery Aerial imagery Handheld cameras Professional cameras Drones, UAV? Purdue Univ 39 Image: http://aerialfarmer.blogspot.com/ Aerial NDVI image 31 July 2014 flight, 3+ ft image resolution Red = poor vegetation; green = better vegetation Purdue Univ Image 40 provider: GeoVantage Inc. 2014 Yield Data Aerial NDVI image Reasonably good correlation between late July NDVI and grain yield Yield map Purdue Univ 41 Purdue Univ 42 Useful diagnostic aids, but remember, neither yield maps or imagery will diagnose what caused the low yields without more background information or ground truthing. So, what about variable rate crop inputs? Image: RLNielsen Purdue Univ 43 Purdue Univ 44 Purdue Univ 6
What is the reason we intuitively believe that VR technology will help us? Because we believe that spatially different areas of a field might require different crop input rates to maximize yield or dollar return to that crop input. Variable rate P, K, and lime have tended to be cost-effective because their application rates are strongly correlated with spatial variability in soil test P, K, and ph. In other words, recommended rates for these soil nutrients are based primarily on a single factor. Purdue Univ 45 Purdue Univ 46 Variable rate N or seeding rates tend to be more challenging decisions because yield responses to these inputs are influenced by multiple factors, not simply single soil test variables. Both inputs: Soil characteristics, rainfall (timing, amount), and possibly genetics. Nitrogen: Available soil N supply, source of N, timing of N applic., placement of N. Consequently, it is more difficult to define stable or predictable management zones for these variable crop input decisions. Image: RLNielsen Purdue Univ 47 Purdue Univ 48 My general opinions Be cautious building VR nitrogen prescriptions based on single factors like soil organic matter. Predicting soil N supply is not that simple. Be cautious with VR nitrogen recommendations from computer models. Predicting soil N supply is notoriously difficult. Soil characteristics, biology, moisture, and temperature plus weather forecasting Let s focus on fundamentals Look for link at the Chat n Chew Café; www.kingcorn.org/cafe Purdue Univ 49 Purdue Univ 50 Purdue Univ 7
Response to plant population is also influenced by a lot of factors, mostly those related to stress on the crop. One of the most influential factors is likely available soil moisture. Purdue Univ 51 Grain yield (bu/ac) 200 180 160 140 120 100 80 60 40 20 Yield Response to Population 12 trials, AO pop = 24,400 Primarily severe drought stress 55 trials, AO pop = 32,000 Normal range of growing conditions 0 18000 23000 28000 33000 38000 43000 Plant population (plants/ac) Purdue Seeding Rate Trials, 2008 2014 Purdue Univ 52 Other than soil moisture my data has shown very few instances of variability for yield response to population for other kinds of management zones within fields (soils, elevation, etc.). I also cannot document a consistent relationship between general yield level and optimum plant population. Purdue Univ 53 Optimum population (plants/ac) 50000 40000 30000 20000 10000 Optimum Populations vs. Yield Level What one might expect if higher yields required higher populations 0 75 125 175 225 275 Grain yield at optimum population Purdue Univ 54 Optimum population (plants/ac) 50000 40000 30000 Optimum Populations vs. Yield Level 20000 No clear relationship between 10000 optimum population & yield from about 140 bu/ac to about 250 bu/ac 0 120 170 220 270 Grain yield at optimum population Purdue Seeding Rate Trials, 2008 2014 55 trials (excluding severe stress sites) Purdue Univ 55 Here s my opinion on VR seeding Probably no more than two seeding rates necessary within any given field. A lower rate for highly stressed fields / soils. Yields frequently less than ~ 130 bu/ac Seeding rates ~ 26,000 spa A higher rate for everything else. Seeding rates ~ 34,000 spa Purdue Univ 56 Purdue Univ 8
Let s focus on fundamentals Challenges ahead for PA Look for link at the Chat n Chew Café; www.kingcorn.org/cafe Purdue Univ 57 Cost/benefit of PA technologies Compatibility/reliability of PA technologies Rapid cycling of PA technologies Customer support for PA technologies Data overload, data analysis, data quality Data ownership, data privacy Agronomic research to support PA decision-making models. Purdue Univ 58 Adopt PA technology wisely Precision ag technologies can certainly augment agronomic decision-making, but not without the prerequisite fundamental agronomic knowledge behind whatever it is you are wanting to improve. Email: rnielsen@purdue.edu Twitter: @PurdueCornGuy Web: www.kingcorn.org/cafe Knowledge Technology Purdue Univ 59 Purdue Univ 60 Variable Hybrid Planting? Goal: Match hybrid strengths and weaknesses with spatial variability for growing conditions. Hybrid B Hybrid A Success requires that You make the effort to identify Critical YLFs for spatial zones in every field. Water availability, drainage, rooting depth, diseases, plus a gazillion other YLFs. Challenge: Important YLFs tend to vary one year to another because many are influenced by weather patterns. Purdue Univ 63 Purdue Univ 64 Purdue Univ 9
Success requires that You make the effort to identify Critical YLFs for spatial zones in every field. Spatial zones for critical YLFs have been accurately identified and are spatially stable year to year. Low and high yielding zones tend to occur in different places in a field one year to another. Soil map boundaries can be inaccurate. Multiple year data analysis can be misleading. Success requires that You make the effort to identify Critical YLFs for spatial zones in every field. Spatial patterns for critical YLFs have been accurately identified & are stable year to year. Hybrid characteristics relevant to critical YLFs are clearly identified. Simple descriptions like defensive and offensive hybrids are not accurate enough. Purdue Univ 65 Purdue Univ 66 1999 May, dry soil surface 2000 June, wet soil surface Deep Veris EC Purdue Univ 2001 Nov, moist soil surface 69 2003 Oct, moist soil surface Purdue Univ 10