Re:The impact of water vapor on afternoon rainfall
by mikhailfranco on Thursday May 07, 2020 @06:10AM
(#23101392)
This rambling account may not do justice to the original research, but I found it just another example of how naive ML correlations find it impossible/difficult/laborious to re-discover basic platitudes of a specialist field.
If you discretize features from a continuous domain, like space-time chunking of "afternoon rain", then it will degenerate into a hybrid CNN-RNN ML system. Wet soil in the morning will depend on "overnight rain" and "morning dew". Then "overnight rain" might suppress afternoon rain, if it drains moisture from the atmosphere etc. etc. There will be various spatial and temporal influences expressed as Bayesian priors.
One fact that I found surprising when I first heard it, but within 10 seconds realized was obvious, is that water is lighter than air (N2 O2 v. H2O). I imagined that clouds need convection currents from hot land causing temperature gradients, for moisture to rise into cooler layers, then condense and rain. But water (MW 18 g/mol) is lighter than air (MW 29 g/mol), so it just rises and rains (unless there is a temperature inversion anti-convection barrier). I imagine many such 'obvious' facts of fluid dynamics that elude CS ML hammer-nail practitioners.
Re:The impact of water vapor on afternoon rainfall
by stirbot on Wednesday May 06, 2020 @01:31PM
(#23093856)
Sadly the active radar sensor on the pictured SMAP satellite failed shortly after launch [0]. I've been looking for news on NASA's NIAC Phase 2 R-MXAS studies which, while it looks like it would require some serious material science breakthroughs, could provide near-continuous soil moisture data over an entire continent which suddenly seems even more useful [1].
Re:The impact of water vapor on afternoon rainfall
by lolc on Wednesday May 06, 2020 @01:21PM
(#23093719)
I can rationally appreciate that creating a weather model is an iterative process but still it does weird me out that something as basic as this is being actively researched. Even after reading this I still have trouble saying "we're just figuring out how moist air leads to precipitation".
Re:The impact of water vapor on afternoon rainfall
by wcarss on Wednesday May 06, 2020 @01:53PM
(#23094165)
"Over the Southern Great Plains, we found that when the wind brings less moisture, dry soils are associated with increases in rainfall amount; and when the wind brings greater moisture, wet soils are associated with increases in rainfall amount. In the current study, we find that, actually, in many regions, the opposite is true for the likelihood of afternoon rainfall," Welty said.
I'm certainly surprised that it's _so_ nuanced as to be completely the opposite in different areas.
If you discretize features from a continuous domain, like space-time chunking of "afternoon rain", then it will degenerate into a hybrid CNN-RNN ML system. Wet soil in the morning will depend on "overnight rain" and "morning dew". Then "overnight rain" might suppress afternoon rain, if it drains moisture from the atmosphere etc. etc. There will be various spatial and temporal influences expressed as Bayesian priors.
One fact that I found surprising when I first heard it, but within 10 seconds realized was obvious, is that water is lighter than air (N2 O2 v. H2O). I imagined that clouds need convection currents from hot land causing temperature gradients, for moisture to rise into cooler layers, then condense and rain. But water (MW 18 g/mol) is lighter than air (MW 29 g/mol), so it just rises and rains (unless there is a temperature inversion anti-convection barrier). I imagine many such 'obvious' facts of fluid dynamics that elude CS ML hammer-nail practitioners.
[0] https://www.nasa.gov/press-release/nasa-soil-moisture-radar-...
[1] https://www.nasa.gov/directorates/spacetech/niac/2019_Phase_...
I'm certainly surprised that it's _so_ nuanced as to be completely the opposite in different areas.