BIG Data and META-approaches for analysing research data and improving DECISIONS in plant disease MANAGEMENT

Talk given at an online conference on Big Data in Agriculture organized in Ecuador on 10 February 2021

Transcript

1. BIG Data and META-approaches for analysing research data and improving DECISIONS in plant disease MANAGEMENT Emerson M. Del Ponte Jhonatan P. Barro, Kaique S. Alves and Felipe Dalla Lana
2. Epidemiology Research Control methods Chemicals Biological Genetic Cultural Biotechnology Remote sensing Modeling Machine learning Phytopathometry
3. Big data Decision Soil, crops, pests, diseases RISK Strategic or tactical PDM research Epidemiology Field trials (Regionwide) Information Sensors user-input Storage Processing Impact Knowledge
4. Digital farming Remote sensing → Field scale n > 5k? PDM research requires variation in disease and production situations → several fields (locations x years) n > 50? Context for BIG! Source: grupocultivar.com.br
5. Coordinated efforts (industry and public) One or more target (disease) Common treatments (fungicide, biocontrol, etc.) Chemicals from several industries (control bias?) New treatments added over years, some are kept Disease and yield data are obtained The uniform trials (network)
6. 2004 Soybean Rust Soybean White mold 2009 2011 Soybean Target spot Wheat Blast & FHB Fungicides Biocontrol 2018 2019 Wheat Leaf blotches Wheat powdery mildew Cooperative trials
7. Rapid response Yearly summaries Within trial data/analysis "Combined" analysis Objectives & outcomes
8. Few (< 5) experiments Focus on statistical significance (P-value) Vote-counting approach: how many P < 0.05 When combined, same weight is assigned to trials "Not good" trials are eliminated from analysis By tradition in academic research
9. Three examples of our research Data: soybean rust in fungicide trial network Meta-analysis Yield Loss Meta-analysis Fungicide performance Fungicide profitability Monte Carlo Simulation Cooperative trial datasets 1 3 2
10. 210 trials 57 locations 40 fungicides 9 seasons 2004/05 a 2012/13
11. Variability in intercepts and slopes Intercepts slopes
12. Effect of disease pressure and onset time
13. Conclusion 1 Yield loss can be predicted from severity data and is influenced by onset time and severity level
14. 250 field trials 2004 - 2017 (14 years) > 30 Institutions/researchers Example 2: fungicide performance
15. Exploratory results
16. v v
17. Large within-year (spatial) variation
18. MA Results
19. Dual premixes should be encouraged and single a.i. not recommended due to low efficacy Conclusion 2
20. Example 3: premix performance
21. Fungicide a.i. Study code Commercial name CHECK AZOX + BENZ BIXF + TFLX + PROT PICO + BENZ PICO + CYPR PICO + TEBU PYRA + EPOX + FLUX TFLX + CYPR TFLX + PROT Fungicide treatments
22. Exploratory results
23. Exploratory results
24. 77.6 - 85.2 81.4 - 85.6 SBR control (%) Fungicidea Seasons Trials C CI L CI U BIXF + TFLX + PROT 4 115 76.80 74.39 78.98 PICO + BENZ 4 116 74.02 71.24 76.54 AZOX + BENZ 5 144 72.79 69.74 75.53 PYRA + EPOX + FLUX 4 115 72.23 69.56 74.66 TFLX + PROT 6 166 71.96 69.31 74.39 PICO + TEBU 5 149 66.01 63.11 68.69 TFLX + CYPR 5 143 57.89 54.68 60.88 PICO + CYPR 6 169 56.25 53.25 59.06 MA results Dalla Lana et al (2018)
30. Dual premixes declining after 4 years, but triple premix still good Conclusion 3
31. Example 4: economic analysis
32. Start
33. Probability distributions for Monte carlo simulations Severity on untreated plots Soybean Price (2 years) Interceps Slopes Yield-severity relationship
34. Probability distributions for Monte carlo simulations Damage coefficients Empirical Simulated
35. Yield response
38. Profitability over time
39. Profitability over time
40. Rusty profits shiny app Website
41. Conclusion 4 A decision tool for making profit with fungicides taking epidemiological, control and economic factors into account
42. Chemical breeding epidemiology Kaique Alves Felipe Dalla Lana Jhonatan Barro emersondelponte.netlify.app