Comment by prometheus76 | Hacker News Reader

Comment by prometheus76 | Hacker News Reader

It would have been nice to link to the actual paper instead of a third-party commentary piece.

https://www.nature.com/articles/s41467-019-12808-z

Just the title of the paper is dubious. "New elevation data" should be "the output of a neural network trained on data from the US coastline, extrapolated into wildly different climates and regions".

The first paragraph says "Even with sharp, immediate cuts to carbon emissions, it could rise another 0.5 m this century."

Then later in the paper share this wildly broad range of predictions: "Central estimates in the recent literature broadly agree that global mean sea level is likely to rise 20–30 cm by 20503,4,5,6,7,8,9,10. End-of-century projections diverge more, with typical central estimates ranging from 50–70 cm under representative concentration pathway (RCP) 4.5 and 70–100 cm under RCP 8.53,9,10,12, though more recent projections incorporating Antarctic ice sheet dynamics indicate that sea levels may rise 70–100 cm under RCP 4.5 and 100–180 cm under RCP 8.5, and could even exceed 2 m or more in far-tail scenarios"

Summary: neural network using the highest levels predicted in the most recent models, trained on data from areas with climates and landscapes wildly different than where the model is applied produces maps that freak people out and fuel political agendas.

Nice analysis! I'm interested in learning how to read these papers better.

Which section suggests "trained on data from areas with climates and landscapes wildly different than where the model is applied"?

EDIT: I see it now in "Methods". Here's the LIDAR dataset they used, Digital Coast coastal Lidar by NOAA: https://coast.noaa.gov/digitalcoast/data/coastallidar.html

In this article, we present ECWL exposure assessments that address this problem by employing CoastalDEM, a new DEM developed using a neural network to perform nonlinear, nonparametric regression analysis of SRTM error. This model incorporates 23 variables, including population and vegetation indices, and was trained using lidar-derived elevation data in the US as ground truth.

I don't share your summary of this paper. I would summarize it as follows: for USA and Australia, high-resolution/precision lidar-based maps are available of coastal areas. For other regions of the world it is not the case (or severely limited). However, gaining an insight into the elevation of land is crucial to determine a region's vulnerability to sea-level rise. NASA’s SRTM has almost global coverage of elevation levels, but is known to be too low resolution to be meaningful for this application (esp. in urban areas). A neural network was trained on the USA lidar data to augment the SRTM data (i.e. make the resolution higher). The network was verified on the Australian lidar dataset (and they got a good match; the model was already published elsewhere before [1]). edit: I should add that the point of this paper was to then have the newly-derived elevation maps be exposed to sea-level change. This is where the maps with flooded cities come from.

Now, you can discuss whether these datasets are representative of other areas, whether the methods make sense, whether the assumed sea-level rise is realistic or if the outcomes of this analysis have value with arguments. But you give none.

To address your other points:

> "Even with sharp, immediate cuts to carbon emissions, it could rise another 0.5 m this century." Then later in the paper share this wildly broad range of predictions

The authors don't contradict themselves here. You even quote them: "End-of-century projections diverge more, with typical central estimates ranging from 50–70 cm under representative concentration pathway (RCP) 4.5". The RCP 4.5 pathway involves major cuts to GHG emissions.

> neural network using the highest levels predicted in the most recent models

Their neural network has nothing to do with the sea-level rise prediction. It simply tries to (better) estimate the current land elevation.

> trained on data from areas with climates and landscapes wildly different

Also verified on coastline that is "wildly different" from the USA coastal line.

> produces maps that freak people out

The truth is hard, sometimes

> and fuel political agendas

Any findings to the contrary would equally "fuel political agendas", namely those of climate change deniers. Damned if you do, damned if you don't.

[1] https://www.sciencedirect.com/science/article/abs/pii/S00344...

Reply on news.ycombinator.com