Hacker News Reader

Comment by mninm | Hacker News Reader

From the article:

"satellite observations and ground-based ecological work. The former gives an overall view of where forests do and don’t exist on the Earth’s surface. But the latter goes underneath the canopy to determine how many trees exist in a given area in a given type of forest."

It seems that satellites give a good measure of the acreage covered by forests but are not suitable for differentiating individual trees. To do that requires on the ground sampling and as more sampling is done it is being discovered that the actual number of trees differs substantially from previously established estimates.

View on HN

Doesn't it seem that acreage is more relevant?

(I didn't read the article)

Acreage doesn't tell you how much biomass there is. Imagine a 1m tall shrub with big leaves and not much underneath vs. a dense rainforest where the top canopy extends tens of metres up and there's massive undergrowth.

So I guess the answer it depends what you're trying to measure.

So basically we have no way of counting trees from pictures? I mean it shouldn't be that hard, right? You have green pixels and brown ones. We have software that can discern car license plates in urban environments where you have a gazillion of different colors in the picture.

I guess the need never arise for an automated tree counting algorithm.

You'll have green pixels in areas covered by shrubs as well as in areas covered by 40m+ trees intertwined with massive undergrowth.

I'm sure we can count trees from pictures reasonably well, but we can't count tree mass that's entirely obscured by other objects.

To count trees accurately from pictures would require a vast quantity of footage taken from under the canopy.

General forest type can probably be determined quite well and I imagine that ecologists have been estimating tree mass of various types of forest for quite some time?

Recent developments with laser scanning using drones is likely to improve accuracy too.

The whole point of the article is that they've (hopefully) improved the accuracy by using recent developments and combining different types of counts, and that method dramatically increases the count.

From above, the green pixels will obscure the brown ones, so the best you could say is "there are 3 forests in this area". If you are under the canopy itself, you have much larger problems. The colours are very similar, so distinguishing between one tree trunk and another is not the same as distinguishing between a white licence plate and a blue car. The shapes are very similar, so distinguishing between one tree and another is not the same as distinguishing between a car and a house.

Some trees will be obscured by others so you will need a number of different cameras to create a 3D map. You would also need to measure the distance between the cameras to determine what area they cover.

A 10 minute analysis revealing a list of complexities as long as my arm is not a good start. It will probably be easier (let alone more accurate) to just count the trees.

Great answer, thanks dude.

Did you even ever set foot in a natural, dense, wild forest? This isn't like in a timber plantation or the park near your home, trees are fighting for light and overlap vertically, and satellite images happen to be two dimensional. Shouldn't be that hard to understand, right?

You can't read a license plate from a satellite image. Those high-res Google Maps images are from planes. The best satellite imagery is about 1 meter per pixel, and those probably don't cover all the forests in the world.

>So basically we have no way of counting trees from pictures? I mean it shouldn't be that hard, right?

Are you a programmer? Have you never had to deal with someone who thinks they're smarter than they are and thinks they know everything get pushy about the simplicity of doing things that are not as simple as they assume? I have, and I can't say I'm a fan of those conversations or the people who initiate them. Your comment feels a lot like that.

His comment reminded my of xkcd: https://xkcd.com/793/

Not http://xkcd.com/1425/ ?

I guess the "I mean it shouldn't be that hard, right?" struck me the most, but yeah, both comics would apply :)

BTW, funny how the second requirement has become apparently almost as trivial as the first one.

I didn't attack or criticized anyone specifically so no, I don't think I'm smarter than everyone else or knowing everything. I'm just expressing a genuine question as to how we ended up with such a miscalculation and whether there is software that could help in that direction.

Perhaps you need to chill down a bit. This isn't Reddit. We're all here to learn something.

I admit it's rare, but sometimes, with a fresh perspective, and not falling into believing something is best just because it's the way it's always done, what programmers do actually is needlessly complicated and difficult. Examples are the entire J2EE ecosystem from 10 years ago, forcing concepts into baroque relational DB models that don't fit, or XML.

Reply on news.ycombinator.com