Hi FWI. I have actually worked on both Earth Observation projects and projects looking at other forms of remote sensing for assessing water quality in aquaculture (but don’t have the technical skillset to participate in your challenge). A few (hopefully helpful) points:
any sort of useful model is going to require calibrating against background data on the variables you select (particularly as you appear to be working with small, shallow freshwater pools which have different appearance in visual spectrum imagery that probably represents differences in factors other than those you’re quantifying e.g. mineral content and depth)
since obviously any past water quality shared is going to be highly correlated with the current water quality in a given pool, it would make sense to evaluate models on their ability to pick up change from previous levels in future samples using future images rather than simply on predicting which pools have the most ammonia....
the more observations you can share, the better the chance the model actually works
the variables you’ve highlighted are theoretically detectable using EO, but they’re relatively weak indicators perturbed by other stronger indicators and things like weather (at least it’s normally sunny there!). Depending on how much water you collect at how many points in the farm, presumably there’s some natural variation in the samples you collect too.
free satellite imagery such as Copernicus typically has 1 pixel representing 10-20m on the ground so some of your lakes might be only about 2-4 pixels accross. Large pixel sizes don’t necessarily stop major changes in water quality being picked up in multispectral imagery (and aren’t necessarily an issue if you’re measuring the sea surrounding a coastal fish farm, but it’s going to significantly affect the fidelity of your results. Unfortunately, I suspect commercial imagery (spatial resolutions more of the order of 0.5m pixels) is outside your budget
if you’re currently only occasionally collecting data, satellite revisit rate should be fine
I wouldn’t be hugely optimistic about success in the short term as I suspect the scope of what you’re looking at is a lot more subtle than “spot the effects of leachate on the massive lake” and the data you have so far may not be enough
Hi FWI. I have actually worked on both Earth Observation projects and projects looking at other forms of remote sensing for assessing water quality in aquaculture (but don’t have the technical skillset to participate in your challenge). A few (hopefully helpful) points:
any sort of useful model is going to require calibrating against background data on the variables you select (particularly as you appear to be working with small, shallow freshwater pools which have different appearance in visual spectrum imagery that probably represents differences in factors other than those you’re quantifying e.g. mineral content and depth)
since obviously any past water quality shared is going to be highly correlated with the current water quality in a given pool, it would make sense to evaluate models on their ability to pick up change from previous levels in future samples using future images rather than simply on predicting which pools have the most ammonia....
the more observations you can share, the better the chance the model actually works
the variables you’ve highlighted are theoretically detectable using EO, but they’re relatively weak indicators perturbed by other stronger indicators and things like weather (at least it’s normally sunny there!). Depending on how much water you collect at how many points in the farm, presumably there’s some natural variation in the samples you collect too.
free satellite imagery such as Copernicus typically has 1 pixel representing 10-20m on the ground so some of your lakes might be only about 2-4 pixels accross. Large pixel sizes don’t necessarily stop major changes in water quality being picked up in multispectral imagery (and aren’t necessarily an issue if you’re measuring the sea surrounding a coastal fish farm, but it’s going to significantly affect the fidelity of your results. Unfortunately, I suspect commercial imagery (spatial resolutions more of the order of 0.5m pixels) is outside your budget
if you’re currently only occasionally collecting data, satellite revisit rate should be fine
I wouldn’t be hugely optimistic about success in the short term as I suspect the scope of what you’re looking at is a lot more subtle than “spot the effects of leachate on the massive lake” and the data you have so far may not be enough