0

I am using to verify obsereved climate against model. I saw on verification package three options

  1. when obs is cont and forecast is cont

  2. when obs is binary and forecast binary

  3. when obs is binary and forecast is cont.

I use and understand the first and the second but I want to use the third but I face challenge to develop the script for the following probabilstic data and logical behind go together binary and cont data 

po = c(0.26,0.09,0.48,0.36,0.08,0.95,0.83,0.62,0.16,0.21,0.82,0.61,0.22,0.16,0.27,0.92,0.90,0.88,0.81,0.37,0.86,0.51)
py = c(0.00,0.76,0.27,0.31,0.54,0.76,0.52,0.70,0.31,0.18,0.23,0.81,0.40,0.91,0.01,0.40,0.75,0.79,0.36,0.59,0.71,0.87)

I need to use to plot roc using like the following script

A<- verify(py, po, frcst.type = "prob", obs.type = "binary") 
roc.plot(A, main = "Test 1", binormal = TRUE, plot = "both")

how could develop a script to use both binary and cont? Any help on this scripts

plannapus
  • 18,529
  • 4
  • 72
  • 94
tokicha
  • 13
  • 5
  • Please read about how to [format your data and code in markdown](http://stackoverflow.com/help/formatting) and provide a [reproducible example](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example). – Thomas Dec 16 '13 at 08:52
  • Which packages are the `verify` and `plot.roc` functions contained in? – Richie Cotton Dec 16 '13 at 12:54
  • It exists in verification packages – tokicha Dec 16 '13 at 13:05

1 Answers1

0

What is the data you provide (po, py)? Probabilities?

For cont/cont and binary/binary, you don't need a probabilistic forecast model, for prob/binary you do. If I understand this correctly, cont/cont compares forecast and observation of for example rainfall. Binary/binary compares if the forecast and observation exceed a certain amount. The forecast can be continuous and then be converted to binary, or it can be binary to begin with. All of this can be done with a deterministic forecast model. For prob/binary you have to forecast the probability of something happening, so that is why you need a probabilistic forecast model here.

user1901736
  • 21
  • 3