PlantVillage Nuru is a publically supported, and publically developed application that uses a digital assistant to help farmers diagnose crop disease in the field, without an internet connection. Developed at Penn State University the app uses Google’s Tensorflow machine learning tool and a database of images collected by crop disease experts across the world. The app is based on extensive research comparing the accuracy of machine learning models to human experts and extension work. This is continual research and the app will be constantly updated. The app also allows for a blended model where images are examined by AI and human intelligence through a cloud system. This app was developed with International Institute of Tropical Agriculture and the United Nations Food and Agricultural Organization. We welcome further collaboration with public institutions. This app is a public good and is not commercial or backed by venture capitalists. We do not have advertisements or collect farmer data to sell to third parties. You can donate if you like https://plantvillage.psu.edu/. In addition to the diagnostic tool the app contains the library of knowledge that is on PlantVillage, the largest open access library of crop health knowledge in the world.
This is not just because Western Kenya is featured 🙂 , but as the programme summary puts it:
“The world’s population is set to grow from 7.7 to 11 billion by the end of this century. The challenge is to produce enough food to feed this number of people. In the 1960s the Green Revolution provided answers to similar problems – but the projected population growth of the future is on a much greater scale than before, and so new measures are required. In east Africa they’re working to reduce the amount of food that’s lost before it even gets to market – globally this stands at around 30 per cent. In the United States scientists are working to improve the natural process of photosynthesis – to make plants themselves function more efficiently. And in India they’re working to preserve genetic diversity – conserving rice varieties that can flourish in salt water or in conditions of drought.”
According to USAID research standards, quality data
must exhibit five key attributes, V-TRIP; i.e. Validity, Timeliness,
Reliability, Integrity and Precision.
Valid data is data that shows a true representation
of the measure of interest (indicator), and its changes can credibly be
associated with the interventions in question. It should be free of sampling
and non-sampling errors. The validity of data is achieved by developing proper
data collection tools and their subsequent effective use during the data
collection exercise. Soon after the exercise, the integrity of the data in
question must be protected, and this is often linked to the capabilities of the
management system in place to reduce the possibility of introducing bias either
by transcription error or deliberate manipulation during data entry and
cleaning.
In reality though, isn’t this just idealism? During
a data collection exercise, there are usually many factors in play that may hinder
the collection of valid data. Therefore, what follows data collection is often
putting in hard work on preserving the integrity of invalid data.
The challenge does not start at the field but from
the planning of a research exercise, and most of it has to do with time. There
are key areas that are associated with the issue of time that must be dealt
with precisely to avoid these “time-bottlenecks”. I conducted a mini-research
for purposes of getting opinions and experiences (basically qualitative in nature)
from fellow researchers and here are some of the concerns raised:-
1. Training
period. Proper planning ultimately determines the level of success of any given
project. Therefore, it is advisable to spend as much time as needed to prepare,
so that on execution every possible angle of challenges and risks will have
been mitigated or prepared for. Training of enumerators is part of planning.
Spending only a day on training enumerators who are going to carry out a seven day
survey only to end up getting 50% of the responses wrong doesn’t make sense.
Wouldn’t it be wiser to spend two days on training and increase the precision
and validity of collected data to 90%? Release enumerators to the field only
when you are sure they will bring you, not good but excellent data.
2. Data
collection period. A questionnaire that takes an hour during a mock survey in
the training venue will not take the same amount of time in the field. It will
take probably a half an hour more. Therefore it is not logical to expect an
enumerator to bring back eight questionnaires at the end of the day. The plan
must consider sampling method used and time of travel to access target
participants. All these are about TIME. If you give unreasonable targets the
enumerator will use unethical means (compromised integrity) to reach the target
and the result will most likely affect the validity of the data.
3. Sample
size versus daily target. Often, the aspect of bias and assumption among
enumerators comes in when they have had to ask the same questions over and over
and keep getting the same answers. By the time they are administering the
one-hour-long, fifth questionnaire of the day they have basically switched to
‘auto-drive-mode’. What they do is to assume responses to some questions will
be similar to what they got before and therefore they do not pose these
questions to the respondent but fill in the assumed response. This is also
reported to happen whenever respondents look unsettled, seem to be in a rush or
when the enumerator is tired, feels like they are far from reaching set daily
target or are running out of time as the day concludes.
Is it possible to deal with these “time bottlenecks”
to beat the issue of validity and integrity at the level of data collection?
Several suggestions were put across but there was no
single standing solution. The suggested approaches must be combined to move
from 90% that can be achieved with proper planning, to 98%. First of all
sufficient time must be allocated and used in the planning phase. Train, carry
out mock-survey, re-train, pre-test with a sample of targeted respondents then
re-train. Ask questions and engage trainees.
It helps in gauging their level of understanding of the tool, their
confidence on the tool and their level of preparation to undertake the
exercise. Do not depend on getting phone calls to clarify issues for
enumerators after deployment to field. Network reception may be terrible or
something else may render it impossible to communicate, then enumerators will
make-do with guesswork. Attain an
excellent mark before deployment.
Secondly, allocate sufficient time for the survey.
Don’t give unreasonable targets because enumerators will hit the target but
will deliver invalid data. It was also suggested that sound recording would be
a great tool for confirming validity.
Have you felt the urge to know a bit more about sustainable development?
Do you need a little more on your CV?
Can’t quite manage all the fees for some university courses?
We have all the answers here:
The third choice here is ‘The Age of Sustainable Development‘ delivered by Professor Jeffery Sachs. Jeffrey D. Sachs is a world-renowned economics professor, leader in sustainable development, senior UN advisor, bestselling author, and syndicated columnist. Professor Sachs serves as the Director of The Earth Institute, Quetelet Professor of Sustainable Development, and Professor of Health Policy and Management at Columbia University.
