Presentation on theme: "Accurate Regional to Field Scale Yield forecasting of Australian Sugar Cane and Peanut Crops using Remote Sensing and GIS Dr Andrew ROBSON Kingaroy, Queensland,"— Presentation transcript:
Accurate Regional to Field Scale Yield forecasting of Australian Sugar Cane and Peanut Crops using Remote Sensing and GIS Dr Andrew ROBSON Kingaroy, Queensland, Australia, 4610 b.h mob
Importance of accurate yield forecasting: - At the farm level: provide information to growers that support improved management strategies that optimise productivity (i.e. PA). - At a regional level : provide essential pre-harvest information to support decisions regarding harvesting, transporting, marketing and forward selling.
Most suitable source of imagery: Regional Scale SPOT5 : –10 m spatial resolution –4 band spectral resolution –cost ~ $1/ km 2 –2-3 revisit day improved possibility of capture –3600 km 2 tile encompasses the majority of crops in each region. Bundaberg and IsisBurdekinHerbert Full GIS layers of every crop provided by mills
IKONOS : –Due mainly to cost 3 * 50km2 (~$22/ km2) –3.2 m spatial resolution, 0.8 m PS –4 band spectral resolution (NIR, red, green, blue) –Identify sub metre constraints such as weeds, soldier fly, rat, cane grub, land forming etc. Most suitable source of imagery: Paddock Scale Insect Poor irrigation Weed infestationDiffering Cultivar
Imagery captures between January to early March generally restricted by cloud cover, for all regions. This limits the opportunity to implement alternative management strategy within the same season of imagery capture due to size of cane and growth stage limiting its capacity to respond. However, remote sensing was identified to be highly effective for identifying differences in crop vigour, assisting with coordinated plant and soil sampling to identify the likely driver of reduced production. This information was then used to coordinate remedial action for the following season. Optimal timing of imagery capture
Classified VI image VI- highlights variation in crop vigour False colour image- IR, Red, Green Vegetation Index (VI) (plant structure/ pigment/ water content etc) At the Crop scale: Indentifying variability
Development of Yield maps from sampling points Average GNDVI = 0.61 Predicted Yield = 77.9 TCH GNDVI (NIR-Green)/(NIR+Green) Correlation between GNDVI and TCH (Tonnes of Cane Per Hectare) False colour image of a cane crop with sample points Surrogate yield map derived by correlation algorithm In crop sampling
Entire Cane blocks where the average GNDVI value was extracted. 600ha. Development of generic SPOT5 Yield algorithm Average crop GNDVI Vs Yield from 2008 (n = 39) and 2010 (n= 112).
Validation of generic algorithm at the in- crop level. Classified yield map from in crop samples Classified yield map from generic algorithm Selected point locations within Bundaberg crops. Accuracy of prediction
Regional Prediction of Sugarcane Yield Bundaberg/ Isis/ Herbert growing regions Using 2008/ 2010 algorithm : Yield = * EXP ( * GNDVI) Nth Bundaberg (3544 crops): Pred TCH, act TCH (97%) ISIS (2772 crops): Pred. 84 TCH, act. 84 TCH (100%) Nth Bundaberg (3824 crops): Pred TCH, act TCH (109%) ISIS (4205 crops): Pred TCH, act TCH (118%) Nth Bundaberg (3217 crops): Pred. 88 TCH, act TCH (99%) ISIS (4000 crops): Pred TCH, act. 96 TCH (96%) Herbert (6481 crops:): Pred TCH, act. 55 TCH (103%) Herbert (15463 crops): Pred. 75 TCH, act. 72 TCH (104%) * TCH- Tonnes of Cane Per Hectare
Bundaberg: Comparison of years 2010 ave. GNDVI: ave. GNDVI: ave. GNDVI: 0.584
Generation and Distribution of Yield Maps at the Regional Scale. Can be used to identify sub- regional seasonal and temporal trends
Development of Yield maps from sampling points NDVI (NIR-Red)/(NIR+Red) Correlation between (NDVI) and Yield (Tonnes Per Hectare) False colour image of a peanut crop with sample points Surrogate yield map derived by correlation algorithm In crop sampling
Pod yield vs NDVI (n=352) data from 6 growing seasons and 8 varieties Development of generic algorithm for predicting yield
Assigning blocks as peanut crops Classification of peanut ‘pixels’ South Burnett 5000km2 coverage False colour image Example: Predicted area 6000 ha Ave. NDVI: 0.55 Identifying where the Crops are.
Average NDVI extracted = 0.55 Predicted average Yield: *EXP (3.9659*0.55) = 2.4 T/ha Predicted total yield: 2.4 T/ha * ha = 14,400 t peanut Correlation between NDVI and crop yield t/ha Predicting Average and Total Yield.
Yield Predictions: Crop scale Dryland crop (cv. Walter) Area=317 ha Predicted 849 t : Delivered 874 t (97%) NDVI images of dry land peanut crops overlayed on a false colour image. Dryland crop (cv. Walter) Area=18.9 ha Predicted 76.6 t: Delivered 77.4 t (99%) 1500 m Crop locations within Australia
Yield Prediction at the Block Scale Pred. yld 2.1 t/ha (act. 1.8 t/ha) Pred. yld 3.6 t/ha (act. 2.9 t/ha) Pred. yld 1.2 t/ha (act. 1.8 t/ha) Pred. yld 4.3 t/ha (act. 4.4 t/ha) Pred. yld 4.0 t/ha (3.6 t/ha) Pred. yld 6.87 t/ha (Act. 5.3 t/ha) Pred. yld 4.3 t/ha (Act. 4.1 t/ha) Pred. yld 4.3 t/ha (Act. 3.9 t/ha ) Total area of Peanut (143.6 ha) average predicted yield = 3.43 t/ha: Actual average yield = 3.39 t/ha total predicted yield = tonnes; Actual total yield = 487 tonnes
Production of Peanut Yield Maps at the regional Level.
Surrogate maps provide can assist with harvest segregation for quality based on pod maturity and aflatoxin risk.
Correlation between NDVI and pod maturity (% Black kernel) Total area 17.61ha Total Yield t Ave t/ha 5.75t/ha Maturity: Case Study
Soil temperature variability across zones Tiny tag sensor Portable weather station Soil temperatures measured at strategic locations
the predicted crop maturity date was 158 days using ambient temperature. Substitution of soil temperature data measured over the three week period a predicted maturity range of 2 days between the black and red colour zones ( days from sowing) extrapolated for the entire pod-filling period, a maturity range of 7 days between the black to red colour zones ( days from sowing) 1/t = (T-Tb)/θ where, 1/t is the development rate T is daily mean temperature (°C) Tb is base temperature (°C) below which the rate is zero θ is a constant identifying the thermally modified time for each development stage. APSIM thermal time model
X3: , X2: , R2: , Opportunity for Harvest Segregation Based on Aflatoxin Risk. The optimum conditions for aflatoxin production are low soil moisture, as well as high soil temperature (25 C to 32 C). Stressed plants/ less shading are likely to be exposed to a longer period of high aflatoxin risk.
Avocado- Field sampling based on NDVI maps False colour image of Avocado block sampling locations overlayed Classified NDVI image of Avocado block sampling locations overlayed
Development of commercially relevant maps Map of %commercial yield generated from the correlation between N1RENDVI. Correlation between N1/RENDVI and % commercial yield
Avocado: Regional Forecasting 1,383 ha * Predicted average yield of 13.1 TPH = 18,117 tonnes Predicted Avocado pixels (SAM Classification) Predicted Avocado blocks (Polygons) Derived yield maps (Avocado blocks)
Mapping orchard constraints at the tree level ‘Geo tagging’ individual trees so in field assessments and measurements can be spatially linked (i.e. with a PDA). Can be performed at low cost using GoogleEarth, with points then exported into ArcGIS/ Excel for further analysis.
Mapping Disease Derived tree health map following field survey Ciba Geigy Avocado tree health rating scale False colour image of Avocado orchard with Geo-tagged trees
Conclusions: - Imagery is an effective and efficient TOOL for: - For identifying within crop/ Orchard variability. Assist with the adoption of PA, fertigation, disease and pest monitoring. - Within season yield forecasting at the Block/ Farm and Regional level. - Useful for multiple applications but only as good as the data collected. - Requires involvement from all facets of farming (i.e. grower, agronomist, seed and fertilizer reps, research bodies etc) to ensure maximum use, feasibility and adoption. - GoogleEarth provides a very effective method for image distribution.