Everything we eat takes water to make. About 90% of water usage comes to agriculture, and hence food production. Under the crude assumption that food is locally sourced, one can determine how much strain is put on water sources in a specific region. Unnecessary strain can drain rivers and cause a lot of harm.
A historic example are the river basins of Kazakh rivers Amudarja and Sirdarja drained by extensive watering of crops in the Aral region in the 80s. A much recent example is Cape Town, where is predicted tto run out of fresh water next year - predicting to hit the so called Day Zero.
As it is quite possible to drain yourself to death by having an unsustainable water footprint, we developed a platform both to monitor and raise awareness about the it, so that informed actions can be taken - both at the individual or governmental level.
_ We build our work on papers and datasets published at [waterfootprint.org]. (We are not affiliated to the organization) _
What it does
Our proof of concept is a client-server application.
The client takes a shopping receipt as an input by taking a photo using Android device or scanning the receipt. The app extracts information about the amount of food in the shopping list.
This is sent to the server, along with geo-spatial information of the client. Based on this, the server calculates the freshwater footprint of the shopping within the relevant water basin - the geo-spatial info is related to river basin data for which there is known information about relationship of drought to freshwater footprint. _ Here we use the datasets from the Water Scarcity study: [waterfootprint.org/en/resources/waterstat/water-scarcity-statistics/] containing data for about 500 river basins globally _
Assuming that the client is shopping for a daily meal, we then estimate how sustainable his diet is in hers geographic location. We do this by extrapolating the impact on the water footprint, assuming that the whole population would be on the same diet for a period of time. This gives a proxy contribution of such diet to the freshwater footprint in the region and determine if the diet is sustainable
How we built it
We used image to text transcription to read the shopping receipts using both Xerox Tesseract (for testing) and the Google Vision API. We also use Google Geospatial API to retrieve the location which is mapped to the river basin (_ the final version of the app has this disabled as we need to be the same local network which does not allow the client to update the location _). This is used from a client-side Android app, which sends the receipt data scanned from a photograph to a server-side application on a second computer.
The server side app processes the data by making extensive use of the datasets from [waterfootprint.org] and the one provided by Buhler and USDA (US dept. of agriculture). We estimate the water footprint of the shopped for diet based on these. The results are sent back to the client where they are displayed.
Additional notes: We also used USDA API to map common retail products to USDA food categories + content. This allows us to estimate the water footprint just based on the name of the product (e.g. instant calculation of wf for 'Snicker's bar'). (In progress)
Challenges we ran into
We had noted that the combination of [waterfootprint.org] and USDA data gives order of magnitude inconsistent estimate of freshwater footprint for livestock compared to the data provided by Evocco/Buhler. Our position here is that USDA estimates along with [waterfootprint.org] actually undershoot the freshwater footprint estimate and we Evocco. Some of the estimates oscillate wildly.
We would love to have more time to focus on UI and data presentation.
We originally aimed for Day Zero predictions (days when cities run out of water, as is Cape Town suppose to next year). We however did not find a rigorous and transparent calculation of the day zero!
Accomplishments that we're proud of
Implementing both custom client and server-side application. Using OCR to transcribe the receipts. Scrambling independent datasets for data patterns we build our idea around.
What we learned
Implementing client - server solution. OCR. Pandas and data scrambling. We also learned it is really difficult to ping a specific IP on the HackZurich network.
What's next for Food Drain
--- Business side --- As the costs for receipt processing are minuscule, we hope to implement the water footprint calculation on the side of food retailers at minimal overhead. We would love to see the water footprint information printed on receipts. This will raise awareness about the problem and will also help the shopper to monitor the water footprint and potentially make choices.
This is important in critical situations, such as the day zero in Cape town, where the water footprint needs to be put down in a very short span of time. Monitoring our water footprint data by governments can also influence regulatory decisions taken to prevent the day zero case.
This may help with Water footprint assesment by [waterfootprint.org], which may start issuing certificates. Thos can be monetized similarly to carbon emission allowances issued per country but despite this being hackathon (We are not fans of that :)).
The water footprint calculation may be literary saving necks of people in some regions and incentivizing water footprint aware food production. While this does not lead to immediate profit, it can cut significant losses. One can hence the water footprint allowance coupons to short sellers and keep on raising awareness. This is how insurances make money, so in case this moves forward, we will reach out to them.
--- Technical side --- Comparing performance and accuracy of Microsoft Cognitive Services and Google Vision Services Extend to various groups of food products. Improve UI and presentation. Localization.