7 ms·

I’m a bit surprised to see a single geodata theme packaged as a whole “thing.” And goodness the marketing fluff is quite repetitive given they have so little to

I’m a bit surprised to see a single geodata theme packaged as a whole “thing.” And goodness the marketing fluff is quite repetitive given they have so little to talk about. Call me jaded but it feels like someone’s really announcing, “I shipped something! Give me my promotion!” > Our Pollen API is based on a model that calculates the seasonality and daily amount of pollen grains on a 1x1 km2 grid in over 65 countries worldwide, supporting an up to 5-day forecast, 3 plant types, and 15 different plant species. The model uses various inputs like land cover, climatological data, annual pollen production per plant, and more to deliver reliable output predictions of the local pollen level and exposure risk. The maps look like what you get when you throw Ordinary Kriging at a dataset of very limited coverage. I’d be skeptical of this model without thorough analysis.

Raed667 3y ago
I'm disappointed as i thought they'd use actual sensors instead of guessing using old data.
- INTPenis 3y ago
  Yeah same here, a much more challenging project would be to aggregate data from the thousands of individual pollen services in various countries. I know of at least two here in Sweden. Isn't this what we should use AI for?
  - Cthulhu_ 3y ago
    > Isn't this what we should use AI for? Nah, this is just data collection and statistics.
    - INTPenis 3y ago
      No I mean we should use the potential flexibility of AI to gather data from many varying APIs that might change at any moment without prior warning. Some of these national pollen sites may not even have an API, may require scraping.
- Waterluvian 3y ago
  There’s nothing wrong with building a model using related data if the thing you care about is hard/expensive to measure at that scale (which I don’t know if that’s true or not). But you’d want to validate your model using actual pollen samples. The data here is pretty much worthless without a paper explaining the model and the validation.
  - zo1 3y ago
    Imagine if they said that about street view? At this point, with so much capital at google, it probably makes sense to just have measuring equipment shipped, installed and have hosted/maintained at these locations. At the very least, snap some kalmann filters together with these "models" and adjust your model predictions with real-time actual values.
- flir 3y ago
  I feel like once upon a time, "blanket the world with a mesh of pollen sensors" would absolutely have been Google's approach.
  - Raed667 3y ago
    Or use an open-data approach from the millions of hobby/scientific weather stations. That would make it harder to sell the data though so why bother ?
    - flir 3y ago
      I think crowdsourcing data's more of an open source thing. I don't think Google's ever been into that? Their approach is more "we're gonna scan all the books in the library/photograph every street in the world". (Interested to hear counterexamples).
      - gbear605 3y ago
        They’ve crowdsourced a lot of their Maps data for information on specific businesses through the Local Guides program.
        goosinmouse 3y ago
        Also we local guides provide streetview maps where there previously were none.
      - dylan604 3y ago
        crowdsourcing data is exactly what the Googs does. CAPTCHA to train it's models of data of dubious origins is the SEP (someone else's problem) nobody wants to talk about
  - vineyardmike 3y ago
    Once upon a time, google was obsessed with aggregating the worlds data. Today, google is all about AI generated or sourced data. And building a complex model to estimate pollen levels is definitely what they’d do. (Plus the cost of those sensors would not be justified for such a small product).
- xnx 3y ago
  If Google deployed a bunch of sensors of any type the headlines would certainly be "Google is now spying on the air!". Google has put a air quality sensors in some of its Streetview cars (https://www.google.com/earth/outreach/special-projects/air-quality/ https://www.google.com/earth/outreach/special-projects/air-q...). Hopefully, they will do the same for Waymo vehicles.
gorbachev 3y ago
I have doubts about the accuracy of the data as well. Pollen can be very localized. I'd love to see an independent study on it.
taco_emoji 3y ago
Yeah given that they don't describe any sort of innovation on the sensing side of thing, I'm guessing they just trained a neural net with nebulous inputs and expect everybody to believe in the output.

Raed667 3y ago

I'm disappointed as i thought they'd use actual sensors instead of guessing using old data.

INTPenis 3y ago
Yeah same here, a much more challenging project would be to aggregate data from the thousands of individual pollen services in various countries. I know of at least two here in Sweden. Isn't this what we should use AI for?
- Cthulhu_ 3y ago
  > Isn't this what we should use AI for? Nah, this is just data collection and statistics.
  - INTPenis 3y ago
    No I mean we should use the potential flexibility of AI to gather data from many varying APIs that might change at any moment without prior warning. Some of these national pollen sites may not even have an API, may require scraping.
Waterluvian 3y ago
There’s nothing wrong with building a model using related data if the thing you care about is hard/expensive to measure at that scale (which I don’t know if that’s true or not). But you’d want to validate your model using actual pollen samples. The data here is pretty much worthless without a paper explaining the model and the validation.
- zo1 3y ago
  Imagine if they said that about street view? At this point, with so much capital at google, it probably makes sense to just have measuring equipment shipped, installed and have hosted/maintained at these locations. At the very least, snap some kalmann filters together with these "models" and adjust your model predictions with real-time actual values.
flir 3y ago
I feel like once upon a time, "blanket the world with a mesh of pollen sensors" would absolutely have been Google's approach.
- Raed667 3y ago
  Or use an open-data approach from the millions of hobby/scientific weather stations. That would make it harder to sell the data though so why bother ?
  - flir 3y ago
    I think crowdsourcing data's more of an open source thing. I don't think Google's ever been into that? Their approach is more "we're gonna scan all the books in the library/photograph every street in the world". (Interested to hear counterexamples).
    - gbear605 3y ago
      They’ve crowdsourced a lot of their Maps data for information on specific businesses through the Local Guides program.
      - goosinmouse 3y ago
        Also we local guides provide streetview maps where there previously were none.
    - dylan604 3y ago
      crowdsourcing data is exactly what the Googs does. CAPTCHA to train it's models of data of dubious origins is the SEP (someone else's problem) nobody wants to talk about
- vineyardmike 3y ago
  Once upon a time, google was obsessed with aggregating the worlds data. Today, google is all about AI generated or sourced data. And building a complex model to estimate pollen levels is definitely what they’d do. (Plus the cost of those sensors would not be justified for such a small product).
xnx 3y ago
If Google deployed a bunch of sensors of any type the headlines would certainly be "Google is now spying on the air!". Google has put a air quality sensors in some of its Streetview cars (https://www.google.com/earth/outreach/special-projects/air-quality/ https://www.google.com/earth/outreach/special-projects/air-q...). Hopefully, they will do the same for Waymo vehicles.

gorbachev 3y ago

I have doubts about the accuracy of the data as well. Pollen can be very localized. I'd love to see an independent study on it.

taco_emoji 3y ago

Yeah given that they don't describe any sort of innovation on the sensing side of thing, I'm guessing they just trained a neural net with nebulous inputs and expect everybody to believe in the output.