Skip to content

Latest commit

 

History

History
81 lines (59 loc) · 3.66 KB

Notes.md

File metadata and controls

81 lines (59 loc) · 3.66 KB

Rain Details

  • QPEs: Quantitative Precipitation Estimates

  • CRPS: Continuous Ranked Probability Score

      - H(x):		Heaviside step function

  				=0 if x < 0

  				=1 if x >= 0
  			It is the cumulative distribution function of a random variable which is almost surely 0.
  - z:			The actual recorded gauge value (in mm)
  - N:			Testing dataset size

  • HCA: Hydrometeor Classifaction Algorithm

      	HA: hail
  	HR: heavy rain, etc.

  • KDP: Specific Differential Phase

      	- Good explanation at http://www.erh.noaa.gov/rah/downloads/Dual_Pol/KDP_v1.pdf
  	- The dual polarity has two radar being sent from the observation place to the storm cloud. One radar is horizontal, one is vertical. When they go through a certain medium, like rain or hail, they get slow. They slow differently, though, so there is a difference in where they end up. KDP is the horizontal pulse minus the vertical pulse.
  	- KDP will be positive if the medium droplets are oval elongated horizontally and negative if the medium droplets are oval elongated vertically and near 0 if perfectly round.
  	- The more dense the medium (heavy rain), the more shift. In other words, as KDP increases absolutely, so should the expected rain amount.
  	- ranges from -2 to 7

  • dbZ: Decibals relative to Z

      	5: Hardly noticeable
  	10: Light mist
  	...
  	35: Moderate rain
  	...
  	65: Extreme/large hail

  • QC: Quality-controlled (reflectivity)

  • RhoHV: Rho (correlation coefficient), H (horizontal), V (vertical)

  • ZDR: Differential Reflectivity

      	- Good explanation at http://www.erh.noaa.gov/rah/downloads/Dual_Pol/ZDR_v1.pdf
  	- Measurement in decibals of the log of the ratio of horiz power to vertical power
  	- Ranges -7.9 to 7.9

  • NEXRAD

    • Polarimetric radar data
    • US National Weather Service's weather radar network
    • Err by biological echoes (birds, bats, etc.), and drops may evaporate or blow off by the time they reach the ground

  • MADIS

    • Rain gauge data
    • Err by siting, wind, or splashing

Variable Notes

  • Expected

    • Peaks of common millimeters: 0,1,2,3,14,28,43,57,72,86,100, and so on for this 14/15 mm difference pattern
    • DistanceToRadar is the only explanatory variable that stays static from one radar measurement to the next. Interesting that the goal is to predict static from varying. What would happen if I collapsed the varying variables to be one measurement per Id. Mean, variance, or both. Create variable like RR1.mean, RR1.sd. I think I should try this

  • TimeToEnd

    • na: no missing values
    • dist: pretty much uniformly distributed between 0 and 60 = .012 to .016. The exceptions are 0 (.0018) and 61 (.003762)
    • cor: near 0 correlation with all other integer/numeric explanatory variables
    • as TimeToEnd goes up (approaches 60), the percentage of non-0 Expected (i.e., there is some rain) goes up.
      • 0:2mm and 61mm are the exceptions.
      • goes up from about .245 to .268 so for small but noticeable difference
      • nonlinear increase
    • Given there was some rain, as TimeToEnd increases, the average amount rained decreases ever so slightly (24.79 to 20.16); outliers are 0 and 61 mm; the median is practically always 1.3 exactly.
    • There is a strong correlation between the first (per Id) recorded TimeToEnd with the first (per Id) recorded RadarQualityIndex. Why?

  • DistanceToRadar

    • na: no missing values
    • dist: pretty much uniformly distributed between 0 and 100
    • cor: near 0 correlations with other integer/numeric explanatory variables (though all other explanatory variables vary from one reading to the next for a given Id whereas DistanceToRadar is the same)
    • I thought that the closer the Distance, the better the RadarQualityIndex, but there is no evidence of that