Making Sense of Risk Assessment

The work we are doing to bring low cost sensor tech to smallholder agriculture has the potential to do a number of things. Certainly, we will start with the delivery of useful information to farmers based on more precise data from their fields. We will build a service that supports better decision-making and guides farmers to achieve optimal production outcomes. But there is another value being created through our sensor work and the service that we are designing to support it. We are starting to build more accurate risk profiles of the farmers we work with and we think that is pretty damn exciting.

An insight that consistently surfaces when working with farmers in emerging markets is the need for financing: at the end of a season to buy new inputs, during the year to pay school fees, bridging after a poor harvest, to cover the capital expense of moving towards increased production, and on, and on. There is nothing surprising about it. Commercial capital is a vehicle to growth and stability in every economy. In Myanmar, that is no different. While our partner, Proximity Designs, is leading the charge to fill a massive capital deficit across the country, we wanted to see what was working elsewhere and how others were approaching the design of financial products for small farmers.

In Kenya, we spent time with Juhudi Kilimo, which has built a strong loan book through the provision of asset financing and micro insurance to smallholder, traditionally high-risk farmers and Acre, which has been using heavy data analysis build an insurance business across a similar demographic. What was inspiring to the IDEO.org team generally was the sophistication of thinking around risk profiling and the number of techniques being applied or thought about by these companies. We have listed a few of those techniques below:

We know that there are a series of data for individual crops that describe what an optimal output could look like. Generally speaking, a farmer is looking to navigate those variables that impact their ability to achieve that optimal output. How we react to weather events, how much we irrigate and when, what type of fertilizer we apply, at what time, and how, then, all matter. Too much water applied at the wrong time? Expect suboptimal plant health, increased spend on fertilizer, reduced overall income for the farmer and, crucially for what we are talking about here, reduced ability to repay a loan and increased likelihood that there will have to be an insurance payout.

In short, suboptimal decisions make business more difficult for financial service providers. We are building a sensor-based product that will allow farmers to irrigate more precisely based on real time soil data. We will build that into a decision support tool, which allows for more accurate application of fertilizer and pesticides. This product then, and the data it captures, is really acting as a decision monitor, showing us the farmers who reacted to an event optimally and those who reacted suboptimally. Armed with this data, very many things are possible, one of which is better targeting of loan products and the more accurate pricing of insurance and another of which could be a baseline for real contract farming and finally, a futures market for smallholder production. Big goals, sure, but this is not a small challenge. 

WE ARE PUMPED.

- John Collery

Analog Agricultural Sensors

Something we have been very aware of since the beginning of this project is that a sensor doesn't necessarily have to contain batteries and electronics to be a useful tool. In fact, people have been finding creative ways to track and gather feedback about their environment since the beginning of time.

The following is a collection of simple analog sensors we found from our research in Myanmar, if you click on them a description will pop up. Are there any you would add to our list?

Start with the Soil

“It starts with good soil”, was what I heard from a farmer standing in intense heat, deep in the dry zone of northern Myanmar. He crumbled a fistful of it and smiled, “sheepshit, water, and crop rotation”. I thought back to the time and place where my obsession with food, farming, and development more generally became serious. Darina Allen, owner of the Ballymaloe Cookery School in Ireland where I worked briefly, would say something similar to her students on their first day. Holding a fistful of good earth she would implore them to let the soil fall through their fingers, to feel the condition of it. “It all starts with good soil”, she would say. “The food we eat, the pleasure it gives you, your ability as a cook, all of it starts with this good earth”. A million miles from east Cork, in a place that could not be more different, those words rang true once again. 

Good soil is the start of good agriculture and without good agriculture everything else goes away. For the past month, the Future Sense team has spent time with farmers, traders, experts, and academics in Myanmar, Tanzania, Malawi, and Kenya, and we have, variously, been able to draw upon our own experiences from years of work done in agriculture around the world. From our work, a central question has begun to emerge. What are the decisions that a farmer can take to ensure maximum productivity over the long term? That is, can we optimize for yield, while ensuring the future productivity of soil? 

With many systemic issues in mind, there do still exist a set of decisions that when taken at the appropriate moment during a crop cycle should ensure for optimal yield and soil health. While we are not yet at a point of mapping fully the knowledge space within which those decisions exists, we do know that this knowledge space is conditioned by a series of variables, each of which can be measured using low cost sensor technology. Location, rainfall, soil makeup, wind speeds, temperature, historical yields, ambient humidity, crop and input types, time, distance to market, are all things that have a real data set. This data set, when looked at objectively, shows us what optimal yield and soil outcomes could look like. When compared to the real time decisions of a farmer then, we can see the distance between what could happen and what is happening now. 

Though this may sound like a data play, that could not be further from the truth. In order to understand why some decisions are made, time must be spent in the field with farmers, developing a deep understanding of their needs base. The optimal response to a series of data in Myanmar will be vastly different to the optimal response in Tanzania. That is what we are beginning to design for. The grunt work of the Future Sense project has been understanding what we should measure, the hard part, the part that will require us to go deep while remaining culturally appropriate in our design work, is understanding how we should support and facilitate the optimal response to those measurements. How can we use low cost, emerging sensor tech to improve the outcomes of farmers in emerging markets? It is not simple. First we have found what to measure and where, now we must design a response to those measurements which allows farmers to benefit from them.

Darina was right when she said that it all starts with good soil. That is where it starts, but there is more to it. Without good decision making, good soil doesn’t matter and without good soil, good decision making can only get us so far. Where the rubber meets the road for us on the Future Sense team, is thinking about real agricultural ecosystems and the decisions being taken within them and understanding how best to design an intervention that supports farmers in navigating towards their own optimal outcomes.

Guest author: a brief overview of farming and irrigation in Malawi. By William Kamkwamba

The challenge with rain-fed agriculture, which more than 90% of smallholders in Malawi depend upon, is similar to the problem faced by smallholders around the world: rain is not predictable. Some years the rains start and finish early before crops get a chance to mature. In others, infrequent but intense off cycle rains lead to flooding, again reducing the predictability and quality of output.

In order to address these challenges, the Malawi government and NGOs are setting up irrigation schemes. The Ngolowindo Horticultural Cooperative Society is one of the schemes that the government set up in Salima using a canal irrigation system. The problem with this system is that it uses a lot of water, and there is no way for farmers to tell if their plants are receiving too little or too much water. Canal irrigation also tends to water very unevenly, resulting in inconsistent yields and irregular growth patterns.

We're exploring ways that simple moisture sensors might be able to give farmers better feedback on plant health, save irrigation water, and help them design canals that can deliver water more efficiently.

Testing water levels on rice farms

We spent some hours chatting with U Aung Htay, a rice field owner in Pakokku, a dry zone located 30 km northwest from Bagan in Myanmar. About 10 years ago, farmers in the region were mandated to plant rice by the government. Unfortunately, the lack of crop knowledge and resources available forced many farmers to leave their fields, and those who stayed are still learning the skills required.

U Aung's nephew, a successful rice farmer in a neighboring region, taught him all he knows about farming a few years ago. One of the most valuable things he learned, and then taught to his neighbors, was how to make irrigation channels which cut irrigation time in half, and reduce the risk of pest and crop disease. Water management is also a huge concern in the dry zone since it's a scarce and valuable resource. We learned that managing water levels in the rice fields is a challenge that starts from the moment farmers are preparing the land.

When we showed Mr. Htay our water table prototype he was very interested in using a device that could save water and irrigate more efficiently. When he saw the PVC tube he also commented “I can do it myself!”, a good reminder of the potential of creating smart add-ons to current tools and possible DIY solutions. Also, he was immediately concerned about the power consumption, although using solar power seemed the obvious alternative to him. A nice surprise was that simple LED feedback on when to irrigate and when to stop would be helpful since farmers in the area tend to irrigate in the evening and at night. It was a great conversation that helped us confirm some hypotheses, created new questions, and sparked some ideas!

Myo Myint - Meeting with the Experts

We spoke to Myo Myint, head of Proximity’s Farm Advisory Services, a couple of times during our time in Yangon. He has been working on agriculture in Myanmar since the 1970s, first as a graduate student and later with the government’s Agriculture Research Institute and as head of the Plant Protection Department where he had responsibility for rolling out initiatives across the entire country. Having retired from that role in 2004 he was compelled to come back to agriculture having witness the devastation brought by Cyclone Nargis in 2008. With Proximity he has been working since then to bring a staggering number of farmers out of poverty, rolling out interventions from the deeply complex - responding to increased salinity in the Delta region - to those that are deceptively simple - adding an extra crop to farmers’ planting schedule, allowing them to increase their income levels significantly. His focus on and dedication to best fit, climate smart, and farmer centric interventions is something that resonated deeply with the IDEO.org team.

It does though always weigh heavy on the mind that when introducing concepts to someone with the experience and knowledge that Myo Myint has, the possibility always exists that he will rubbish them. That he’ll draw from that well of experience to say say no, this isn’t possible. But he didn’t. We introduced the idea of a low cost, smart sensor network, and he immediately jumped on it.  

- What about using sensors to build an early warning system for disease prevention? 
- Could we use computer vision to identify changes in soil type?
- How about packaging the data from sensors up into a risk profile for a financial services product?

Yes. Yes. Yes.

“Other countries are flying in a plane and, until recently, we have been playing catch up in a car”, he said. “For years I thought that sensor technology and precision agriculture was like a fairytale. Now it is possible”. It was an inspiring conversation.

Part of what makes Proximity an exciting partner is the range of verticals they are across. What makes them feel very, very unique is the depth of knowledge they have within each vertical and the success that tapping and directing that knowledge has allowed them to achieve.

We are beyond excited to see where our partnership with Proximity can go and honored to be able to work with and learn from someone like Myo Myint.

The Spoilage Sensor

We want to test the simple value proposition of preventing crop spoilage using affordable off-the-shelf sensors. So, we created the Spoilage Sensor, a sensor that tests for the ideal conditions that Aflatoxin and other agents of spoilage thrive in.

We learned that identifying some conditions for spoilage can be quite simple. Some agents of spoilage really only require high humidity and high temperature to trigger. Inside the Spoilage Sensor, we used a $4 SHT (temperature and humidity) sensor that could notify the farmer or storage owner when the conditions are met.

We have also equipped the Spoilage Sensor with GPS so that we can test other possible value propositions in transport from storage and sale.

 

The Magic Stick

One of the ways we want to test multiple value propositions in agriculture is a prototype dubbed the Magic Stick. The Magic Stick can read temperature and humidity and compare the values over location and time. The experience is quite simple - farmers stick the tip of the Magic Stick into the soil of their field and press the button on the top. The device gives immediate feedback to the farmer using a green (for positive results) or red (for negative) LED. We also want to try this out to help predict crop spoilage by sticking it into piles of stored grain.

Analogous to having the stick send data to the internet, we decided to make the device write to a local SD card which we have easy access to. The data captured is a time-series .CSV file containing latitude, longitude, temperature and humidity that can be visualized. 

The Magic Stick will hopefully give the farmers a better picture into the health of their crops and yields.

 

 

Smart Alternate Wetting and Drying for Rice Fields

During our research we learned about a method recently developed by IRRI, that implies lowering irrigation water consumption in rice fields, called Alternate Wetting and Drying irrigation. In AWD the field is alternately flooded and not flooded, resulting in more rice grain yield per hectare, less harm to the environment, reduction in irrigation cost and potentially a lower incidence of mosquito-borne diseases.

Currently, farmers constantly monitor the depth of ponded water on the field using a field water tube and a meter. Once the water drops under certain level below the soil surface they irrigate again. This “analog sensor “ captured our attention, and we started thinking about the potential of making this practice smart. Besides simplifying the monitoring process for farmers, once we start collecting data on AWD, it's possible to get a better understanding on the water needed for irrigation according to the correlation between different soil types, weather and crop growth stage. This can lead to better strategies for organizing irrigation within a scheme, and the development of automatic AWDI systems. There are also opportunities for AWD impact tracking, crossing data with yields quality and Malaria and Japanese encephalitis outbreaks in the region.

To test this idea we developed a prototype that uses a liquid sensor,  GPS and a SD card to track water levels across several fields over time. We want to experiment with different types of feedback for the farmers, starting by a simple color indicator, red – Time to irrigate!, green – OK, and blue - Over flooded; to potentially integrating the device with feature phones via text messages.

We're excited to go to the field to test our prototypes! Stay tuned, we will be sharing our learnings on the road.

 

Computational Detection for Soil Type

We were lucky to have a chat with Simon King from the IDEO Chicago office. Simon grew up on a farm and is an IDEO agriculture champion who spends his nights researching how technology intersects with farming. During our conversation, he had a great thought that Computer Vision could be used for a range of ideas in agriculture. 

Computer Vision (CV for short) is explained on Wikipedia as - a field that includes methods for acquiring, processing, analyzing, and understanding images. Simply, it helps computers 'see' objects and forms in photos or video. CV is a field that goes very deep and has many uses, notably technology like body tracking, facial recognition and edge detection.

This triggered the idea that edge detection might be able to detect soil type from a simple photograph. If loamy soil is heavier and fuller it would have less edges whereas sandy and loose would have many edges. We conducted a simple experiment using openFrameworks, a C++ based creative programming framework and it's openCV library. 

If we can find a way to create consistent images of multiple soil types this technique might be able to help farmers quickly understand how to amend their soil and prepare it appropriately for different crop types. Analysis of the image could also help farmers understand returns on investment for their soil if we can help them prepare for specific crops.

You can download and fork the code at Github

Initial Prototyping and Exploration

It's been a packed week of testing out sensors and tech we've found, trying to figure out what will work in our field tests.

Building an Ideas Funnel

On day one of this project we decided that one of our outcomes should be a defensible business case. That is, whatever technology we end up using and whatever intervention we design, it must be economically viable. 

But here’s the thing. We don’t yet know what we are designing and we have lots technologies, potential partners, and focus areas that need vetting. Based on the work that Lionel did, we have a world of sensors but no way yet to vet them. So what we need is some way to distinguish between different categories of opportunities and ultimately asking what value creation in any project within a specific category might look like. Yea. Tricky.

Having conducted broad research, a few potential project categories have begun to emerge and with them, likely provocations to build a value creation hypothesis around:

Snapshot_March169.jpg


Right now, this lens feels ok, but it will evolve as our thinking does, no doubt.


Boiling the Ocean

This project is starting broad! Lionel, our design intern here at IDEO.org has been cranking through a volume of research on both the wider world of emerging sensor tech and applications for sensors in developing economies.

Agriculture has begun to emerge as a space where there was a need for automation and the potential for efficiency gains through the application of appropriate sensor technology.

That being said, we see the need to explore the world a little more and understand what kinds of sensors are out there.

Lionel is going deep. His research is looking at the quantified self, remote diagnostics, emergency response triggers, women’s safety, water quality tracking, and various agricultural applications.

He is also asking what applications for remote sensor tech might exist domestically, looking specifically at the community he grew up in.  Some very cool possibilities emerged from this work – using noise sensors to trigger gunshot emergency response, hacking public lighting to illuminate unsafe areas, and on, and on, and on.

Some seriously cool stuff out there. Onwards now.