This document discusses using data and climate modeling to provide agricultural climate services. It notes that climate explains 30-60% of yield variation for various crops. It aims to provide farmers with information on critical climate-related decisions like when to sow and what to plant. The services analyze historical climate and crop data to understand limiting factors, predict seasonal climate, and predict crop yields. Formats and participatory forecasting are used to communicate forecasts to farmers. Initial results found the services helped farmers in Colombia save $3.6 million in 2014 and many farmers in Latin America now receive monthly agro-climatic advice.
3. Córdoba: 56% (irrigated)
Temperature in grain filling
phase
Meta: 29% (rainfed)
Rainfall distribution
during vegetative phase
Casanare: 32% (irrigated)
Radiation during reproductive
phase
Tolima: 41% (irrigated)
Radiation during grain
filling
Huila: 28% (irrigated)
Temperature during
flowering
Climate explains 30-60 % yield variation
Meta: 61% (irrigated)
Temperature during
reproductive phase
4. In this context, farmers
plan their crop based on
what occurred last year
7. 7
Understanding user needs
What are the most critical climate-related decisions?
• Whether to sow
• When to sow?
• What to sow?
8. Understanding user capacities
37% participants reported that
“format 1” was clear yet DID
NOT produce a good
interpretation
14% participants reported
“format 1” was confusing and
consequently gave an
erroneous or incomplete
interpretation
95 participants from 6 sites:
21 farmers
70 profesionals
4 others
Muñoz and Howland (2016)
23% participants reported that
“format 3” was clear yet DID
NOT produce a good
interpretation
23% participants reported
“format 3” was confusing and
consequently gave an
erroneous or incomplete
interpretation
9. Data-driven agronomy for rice in Colombia
Data
collection
Data
analysis
Insight
validation
Delerce et al. (2016)
10. Learning from historical observations to
determine limiting factors
Machine learning analysis with n=1,240
plot-level observations for irrigated rice
Delerce et al. (2016)
26.7 % explained variance
Note importance of cultivar, and 3 main
climate variables
11. Learning from historical observations to
determine limiting factors
Yet more interesting. We can determine
why some farms are underperforming
“Cluster 10”
Delerce et al. (2016)
And recommend ’best suited’ cultivars
for given climate types
12. Perform seasonal predictions
1. Understand predictability
2. Develop local climate
predictions for 3-6 months
3. Automate predictions for
operational purposes
13. Understanding seasonal climate
predictability
• Understanding what
predictive skill we have
in different regions and
periods
• Using that knowledge
to systematically inform
on conditions for the
next 4-6 months
18. • From not knowing how to manage climate variability,
to having a team of 6 people tasked with delivering
agro-climatic information
• Many of their 24,000 farmers receive monthly agro-
climatic advice
The impacts
• A team of 2 agro-climatologists running models,
leading discussion committees, and producing
bulletins
• Many of their ~7,000 farmers systematically receive
agro-climatic advice
• We estimate 300,000 farmers in Latin America are currently receiving
agro-climatic information resulting from our work
This is related to the below. We found that the most important for them is:
Whether to sow
When to sow?
What varieties to sow?
From here on the ppt goes step by step: Step 1 is to understand users needs. We carried out workshops with farmers and technicians to understand their capacities and information needs.
This slide tries to give a sense on how these insights are developed. Here, the analysis shows that cultivar is the most important variable, followed by a bunch of climatic variables.
Developing these insights is a continuous process starting from data collection, then data analysis and then validation of these insights with experts
This slide tries to give a sense on how these insights are developed. Here, the analysis shows that cultivar is the most important variable, followed by a bunch of climatic variables.
Developing these insights is a continuous process starting from data collection, then data analysis and then validation of these insights with experts
Here the idea is to give more detail about the “variety” being most important variable. Not only we are able to say that “variety” is an important factor determining productivity (this may be obvious), but also we are able to determine those farms which are underperforming and why (i.e. because having climates which are similar to other farms, they are growing varieties that do not perform well under those conditions).
We try to understand the extent to which we can predict rainfall. The idea is that the graph shows that there is variation in Kendall correlation (which tells you how good a climate forecast is).
This tries to give an idea of yield variation across planting dates. If you focus on explaining the black line (mean yield across a large number of simulations) you could say that the model is allowing us to identify dates in which yield is likely low, and people shouldn’t plant their crop on those dates.
A final slide showing one major project impact in 2014. We saved many rice farmers from crop failure.
This is self explanatory, hopefully. It shows project-level impacts in Colombia