mwwg meeting notes.912

advertisement
1
Notes from meeting of Midwest Weather Working Group (MWWG)
August 3-4, 2012 – Blackstone Room, Westin Providence Hotel – Providence, RI
Purposes:


Discuss creation of a uniform set of standards for design of pest-warning systems
(diseases, insects, weeds) through participation in the developing AgGateway online
system.
Consider how to advance these ideas via collaborative grant proposals.
Leaders:
Roger Magarey, USDA/APHIS, Raleigh, NC
Joe Russo, ZedX Inc., Bellefonte, PA
Mark Gleason, Iowa State University
Other attendees:
Kiersten Bekoscke, Cornell University/Geneva
John Brightwell, North Carolina State University
Len Coop, Oregon State University
Erick DeWolf, Kansas State University
Frank Forcella, USDA/ARS, Morris, MN
Karen Garrett, Kansas State University
Glen Koehler, University of Maine
Sara Legler, Universitá Cattolica del Sacro Cuore, Piacenza, Italy
Curt Petzoldt, Cornell University/Geneva
Damon Smith, University of Wisconsin-Madison
Carla Thomas, UC-Davis
Art Tuttle, University of Massachusetts-Amherst
2
August 3 notes:
Roger Magarey (RM):





A key question in implementing pest advisory systems is how to compare models with
substantially different inputs and outputs.
Many different computer languages are used for running these models.
A related question is how to preserve knowledge of warning systems when knowledge
experts retire or change jobs.
There are at least 2 main sources of diversity in models: 1) difference in model design
and 2) inherent differences in the biology of the pest systems.
Creation of uniform standard for these systems would be helpful as follows:
o To help researchers evaluate and compare models
o For implementation by extension
o For use by consultants and other private service providers.
Organization and activity of the AgGateway online system:










Goal is to propose and develop uniform (harmonized) standards for use of various
agricultural technologies utilizing automation, including warning systems.
Create an online glossary of agricultural terms to standardize usage.
AgGateway terms:
API = Application Program Interface
XML = Executable Markup Language – for communication between different web
programs
NCEAS = Ecological metadata standards
Main envisioned applications are for control of irrigation, fertilization applications, and
pesticide sprays.
Main impetus for AgGateway has come from private sector (industry and service
providers) and government.
Ag Gateway intends to harmonize its standards with those of ISO.
2.5 years ago, a consortium of ag technology companies met to discuss formation of Ag
Gateway.
Joe Russo:




Data Exchange Project Information Flow
MISs: IT companies
Consultants, growers, service providers
Extension could secure a place in the system.
3




NPDN uses a set of standard terms, as AgGateway aspires to do.
ISO standards would be for equipment only (e.g., precision ag field equipment).
FMIs: Intermediate language/standards – can be applied for pest-warning systems
There are many types of sensors per unit of electronic/autonomous/semi-autonomous
agricultural equipment.
Len Coop:

Why not crop models as well as pest models for the AgGateway system?
Joe Russo:



Ag clients demand 1) reliability, 2) consistency and 3) accuracy
NRCS has crop management zones; can we similarly create pest management zones to
provide assurance that pest models behave reliably within designated geographic zones?
NEON: National Ecological Observatory Network
Erick DeWolf:


How could updates be made to the AgGateway pest models?
JR: Must be vetted in order to be accepted. Quality control would be done by university
experts, but exact mechanism is not yet clear.
Erick DeWolf:


Risk maps are needed.
Could the existing Fusarium head blight online disease advisory be converted to
AgGateway? It has already been implemented on other websites overseas.
Carla Thomas:


Works with many systems.
Can all the strengths of some existing systems be re-created on AgGateway?
Erick DeWolf:


There are problems with adapting a warning system to a new region without getting
detailed local expert advice on implementation.
Response variables may differ profoundly depending on goals of a particular pest
modeling effort.
Len Coop:

Automated tractor example is OK, but this is quite different from a pest model.
4


It seems scary to have an automated tractor driving a sprayer off local-spatial-scale
interpolations.
We need an example of adaption of a pest-warning system in AgGateway.
Carla Thomas:

There are no algorithms of models on AgGateway.
Erick DeWolf:

Could we test “orphan” warning system models (i.e., validated but not currently
implemented) on AgGateway?
Carla Thomas:


Could we define some common lanisguage terms for pest models? For example, from a
downy mildew “orphan” warning system?
Could AgGateway adopt the “orphan” systems?
Breakout groups. Attendees subdivided into 3 groups for focused discussions on “end
users” (Group 1), _______ (Group 2), and __________ (Group 3).
Notes from Group 1 discussion (End Users):
Joe Russo:

System could highlight the key things for a user to pay attention to; further details could
be accessible on request.
Curt Petzoldt:





System should make it clear not only what a user should do, but also why; this is an
extension responsibility.
It should be adaptable to a mobile environment, including hand-held devices, so
information would need to be provided in small, selective segments. A warning could be
presented as a readily noticed color or sound.
This adaptation could redefine extension function, including typing explanatory depth to
the data reports and advisories.
In the future, a dashboard system could be used for delivering disease advisories, whith
extension (explanatory) links tacked to the advisories.
A challenge would come in dealing with the “gray area” between obvious no-spray and
do-spray zones – spray or not?
5
Glen Koehler:


A 2-way flow of information is needed because end users must have advice based on
local conditions (for example, how much scab is in leaves on an orchard floor?).
Could growers keep their spray records in The Cloud?
Joe Russo:





A logic matrix such as a Markov chain is a pathway for making local, grower-specific
data.
The concept of a “virtual consultant” could also include a grower’s past actions.
The user could have a dashboard of risk tools, including: date sprayed; fungicide decay
rate; etc. There could be built-in (automated) “gears” to speed up decision-making. For
example, automated algorithms relating fungicide decay rate to rainfall.
Social networking could also be built in for end users, including tweets.
There would be a 3-way look at data: 1) Pest observations; 2) Pesticide product
information; and 3) timeline (click-accessible).
Questions arose concerning the AgGateway format.
Glen Koehler:

Potentially, grants could be awarded to growers who reduced pesticide spray frequency
or amount – a form of “enriched advice.”
Group 1 general discussion:









Number of sprays saved could be documented.
The system could also track how much of a maximum threshold remains (e.g., for a given
pesticide class).
As an example, three alternative models for the same pathosystem could be displayed to
extension educator or researcher. Then it would be helpful to track what each model
advised vs. what the grower did.
An anonymous system could be used for tracking multiple growers. Results could then be
compared for one growers vs. unidentified other growers.
Another benefit is that it could speed up the grower/educator learning curves thanks to
more rapid and useful feedback.
Pest information could be embedded (via links) into action-oriented recommendations for
growers, since they must monitor all issues (not only a single pest) simultaneously.
Will “audience fatigue” be a risk with such a system?
Use forecast as well as hindcast weather data in AgGateway?
Candidate disease-warning systems for validation of system could include: scab, fire
blight, sooty blotch and flyspeck, gray leaf spot, wheat rust, and/or onion downy mildew.
6



Should insect pest-warning systems be included in a validation effort?
Extension needs a mechanism for getting key information from the field.
A “scorecard” could be developed to chart financial savings and/or IPM compliance. This
could permit real-time total quality improvement.
Glen Koehler:



Snippets of information from Extension state or regional production guides in disease
advisories – “enriched” advice.
Grower confidentiality and anonymity must be assured in a system that contains
information from numerous growers.
Adoption of the system would need to be incentivized with dollar savings.
Group 3 report (Roger Magarey summarized):










For input data, hourly format was preferred (e.g., T, RH, leaf wetness duration, solafr
radiation (latter because weed germination is impacted by soil temperature).
LWD limitations were discussed.
National Weather Service standards for data collection: plant pathologists and
entomologists often do not follow these standards.
Better idea: use NWS data from regional and local weather stations with corrections to
crop canopy environment.
Information displayed for growers could be simplified to days rather than hours, because
hourly data is often too much information for growers.
Output options: grid vs. point. Len Coop: they use 800 x 800 m grid in Pacific NW.
Risk can be categorized into a few useful distinctions: e.g., high/medium/low.
Action can be represented as “Get ready to do something” to suppress pest/disease risk.
An analogy was made to sequential sampling for decision-making.
Metadata:
o Proportion of missing data
o What information is needed about the model?
o When?
o Where?
o What crops and cultivars?
o Time step?
o Where output is published?
o Where is output received for use (range of devices)?
o Sensitivity analysis of input parameters
o Usage statistics; what is use pattern?
Where to go next?
7







Publicly available models
Sources cited and acknowledged
IPM/Pipe pests (Pipe-fitters)
Climate Change proposals
Online repository for models – with code available to users
Related groups: NC-179 (Ag Weather and Climate); APS Epidemiology Committee;
NC1191 (Weeds and Microclimate)
Planning Grant proposal
Erick DeWolf:

Standardization of models is needed, via cross-talking.
Roger Magarey:

Discuss with USDA as to whether they fund modeling proposals? Could propose to
preserve and archive date online and establish standards.
Joe Russo:

Build a website to demonstrate what is needed for sending in your model according to
common standards.
Carla Thomas:




Who owns the AgGateway website? Are there precedents?
MWWG could host a “white paper” on the NCIPM website; then, anyone could write
proposals around the general subject.
A vision is needed from this group; where we are and where we want to go. We must
emphasize providing proof of concept, and also our track records.
A precedent for “white papers” came from the WWWG in Pacific NW:
o What was working and what was not.
o But had a poor handle on model uncertainties.
o Most of time was spent running, but not using output from, networks.
Karen Garrett:


NSF Research Coordination Networks: Her team submitted a proposal that was
unsuccessful, but we could borrow from the stated goals of that proposal.
NSF rejected proposal because it did not advance a new conceptual framework for
infectious diseases.
Possible “hook” for our proposed project: It could integrate across the pest spectrum (disease,
weeds, and insect pests).
8
SCRI CAP or SREP – Could team up with Pipe-fitters and WWWG.
SCRI RPIs: Bob Seem tried this category. They wanted A LOT of private-sector involvement. It
was tough to raise the 100% match.
Carla Thomas:


Start with planning grant? About $50,000 (SCRI)
Collaboration between WWWG and MWWG?
NIMBS - National Institute for Mathematical and Biological Synthesis: Are our proposed ideas
mathematically interesting or innovative?
NIFA Foundational Program: Would we need to change the project’s emphasis?
Use MWWG as a springboard?
Global Sustainability – Opens opportunities for international collaborations.
Len Coop:





Gave a short presentation on Western Weather Working Group (WWWG)
Website: uspest.org
He gave examples of new features and outputs from the site.
RTMA: real-time mesoscale analysis: runs once per hour for one hour; Penn State uses
these data in their PestWatch online site.
IPPC V2 – A virtual weather data network.
August 4 (8:30 am to noon) – PROPOSAL BRAINSTORMING SESSION
Attending: Mark Gleason, Roger Magarey, Joe Russo, John Brightwell, Frank Forcella,
Karen Garrett, Glen Koehler, Erick DeWolf, Damon Smith
NSF program: Advances in Biological Informatics – September 2013 deadline?
Joe Russo:

A goal of the group could be to build a glossary as a step toward ISO harmonization.
Erick DeWolf:


Could culminate in a Letter to the Editor in Phytopathology.
Barriers to using “orphan models” include a lack of standards, so we can’t pick up and
modularize models.
9
Joe Russo:

Start with a “legal” definition of terms (AgGateway approach)
Roger Magarey:




Uncertainty as a unifying theme for a proposal. Can we quantify uncertainty?
NSF: Decision-making under uncertainty.
1) Inputs; 2) Model choice (choose case studies); 3) Glossaries; 4) Output units. What is
the confidence interval?
Partition errors into false positives and false negatives where appropriate.
Joe Russo:



Growers like Low-Medium-High rating of uncertainty.
Risk map: Relate to a crop’s market value.
Spatial scale vs. error magnitude: how does this relationship change?
Erick DeWolf:



Bayesian theory to help answer question. When do you switch tactics/
Models are valuable in gray areas of decision-making matrices.
Break models out of academic “jail.”
Outputs:






Glossary
Include all 3 major pest groups
Build a shell
Proof of concept
Create a model clearinghouse
Interactivity. Growers should be able to plug in their own costs.
Erick DeWolf:

What is the cost of doing nothing (no action to remediate pest risk)? Can we enable
growers to quantify this?
Mine existing data to develop decision rules.
Karen Garrett:


Formalize how expert opinion is elicited: Another proposal goal. “The Elicitator”
Opinion replacing field data.
10
Roger Magarey:


Make models that are easy to use, easy to share, and show application across multiple
decision support systems (diseases, insects, weeds).
Growers’ decision-making and interface must be simplified, otherwise the system will
not be used. Example: Penn State Apple Orchard Consultant (expert system)
More growers today are computer-savvy than in the PSAOC era (20 years ago).
Erick DeWolf:

His Bayesian decision theory work is designed to help simplify growers’ choices.
Carla Thomas:

Reality check: The group is this room is a modeling group. Is decision theory too far
away from the goal of common modeling standards?
Roger Magarey:

Planning grant can be used to help develop modeling standards to fit the AgGateway
framework. A “white paper” could result from that effort.
Planning grants can evolve out of Working Groups. Regional IPM Centers will have “visioning”
sessions to determine future priorites.
We can have an email round-robin as Farm Bill RFAs and Regional IPM Centers RFAs are
released.
Erick DeWolf:

Do developing-world growers want decision-support advice via cell phone? Or will we
just repeat mistakes made in the US?
SCRI Planning Grant: Grower input is a key.
Carla Thomas:
2 basic audiences:


Corporate farms – They have their own advisors
Smaller farms – Need external advisors
PIPEs have evolved to focus tightly on their relevant constituents/clientele.
Proposal should target both large- and small-scale growers
11
Glen Koehler:

We want to “re-invent” expert systems.
Len Coop:

Think broader than just pests and diseases.
Mark Gleason:

Include one or more Communities of Practice.
Len Coop:

New tools to re-invent old ideas.
Roger Magarey:
Decision tree from Penn State people: How to capture knowledge from programmers in a webbased decision-support system? Lower cost and retains information.

This is a form of mind map.
Planning grant: Get right players at Planning Grant-funded meeting; will they write support
letters later on?
Damon Smith:



Previous planning grant meetings resulted in large changes to proposal priorities after
hearing from growers.
Surveys can be done of both growers and crop consultants.
Planning meeting should be 2 to 3 days to allow enough time for all constituencies to be
heard.
Roger Magarey:

How to frame a proposal: Use Joe Russo’s pest model box. Include a box for extension.
o How to use models: inputs, outputs, etc.
o Organize extension information into AgGateway standards.
o Redefine extension role in the modern world (retains role as one among several
authoritative sources).
o Extension is shrinking but can still be a primary source of management
information.
o Extension = those who will maintain the knowledge systems for crops and/or
regions (for example, what equipment or pesticide to use, and when). Could also
act as a liaison with large-scale markets and growers.
12
o eXtension: peer-reviewed extension products (electronic only).
o Could weather-based models be incorporated into NRCS standards/ If so, this
could be a bridge between NRCS and IPM.
Frank Forcella:

Model system candidates: apple diseases; IPIPE (corn rust, Goss’ wilt); weeds (giant
ragweed and mare’s tail in corn and soybeans)
Who will lead proposal efforts?
Silence reigned…
Frank Forcella: Happy to contribute.
Roger Magarey: Off his job mission (invasive pests)?
Download