Two x-axis scales on one R plot

Question

I am trying to create a plot that shows two different time scales on the x-axis. The problem is that the two time scales have a complicated relationship.

I would like to show weather data by the day of year and by the thermal units. Thermal units are the accumulation of the mean temperatures of each day. Some days we get a lot of thermal units, some days not so many. I fit a spline to the relationship between day of year and thermal units and used that to predict thermal unit values for each day. So I do have a nice dataset with the following headers: day of year (day), thermal units (gdd), temperature (temp), precipitation (precip).

I created the following figure (may have to open in new window):

with this code:

pdf(file="Climate 2010.pdf", family="Times")
par(mar = c(5,4,4,4) + 0.3)

plot(cobs10$day, cobs10$precip, col="white", type="h", yaxt="n", xaxt="n", ylab="",      
xlab="")
axis(side=3, col="black", labels=FALSE) 
at = axTicks(3)
mtext(side = 3, text = at, at = at, col="black", line = 1, las=0)
mtext("Day of Year", side=3, las=0, line = 3)

par(new=TRUE)
plot(cobs10$gdd, cobs10$temp, type="l", col="red", yaxt="n", ylab="", xlab="Thermal    
Units")
axis(side=2, col='red', labels=FALSE)
at= axTicks(2)
mtext(side=2, text= at, at = at, col = "red", line = 1, las=0)
mtext("Temperature (C)", side=2, las=0, line=3)

par(new=TRUE)
plot(cobs10$gdd, cobs10$precip, type="h", col="blue", yaxt="n", xaxt="n", ylab="",   
xlab="")
axis(side=4, col='blue', labels=FALSE)
at = axTicks(4)
mtext(side = 4, text = at, at = at, col = "blue", line = 1,las=0)
mtext("Precipitation (cm)", side=4, las=0, line = 3)

dev.off()

This is exactly what I want, but I realized the x-axis scales are linear here, and they should not be. I put the top x-axis in by making my precipitation data white and writing over it. See what happens when I make it green:

It's obvious things don't match up. So how can I make the two axes in scale with eachother?

Here's the little dataframe I have been using where the time units are matched up by predicting: cobs10.txt. "gdd" is thermal units

EDIT: Here is some new code that doesn't use par(new=TRUE):

par(mar = c(5,4,4,4) + 0.3)
plot(cobs10$gdd, cobs10$temp, type="l", col="red", yaxt="n", xlab="", ylab="",     
ylim=c(-25, 30))
lines(cobs10$gdd, cobs10$precip, type="h", col="blue", yaxt="n", xlab="", ylab="")

axis(side=3, col="black", at=cobs10$gdd, labels=cobs10$day) 
want<-(c(1, 130, 150, 170, 190, 210, 230, 250, 270, 360))
mtext(side = 3, text = want, at = want, col="black", line = 1, las=0)
mtext("Day of Year", side=3, las=0, line = 3)

axis(side=2, col='red', labels=FALSE)
at= axTicks(2)
mtext(side=2, text= at, at = at, col = "red", line = 1, las=0)
mtext("Temperature (C)", side=2, las=0, line=3)

axis(side=4, col='blue', labels=FALSE)
at = axTicks(4)
mtext(side = 4, text = at, at = at, col = "blue", line = 1,las=0)
mtext("Precipitation (cm)", side=4, las=0, line = 3)

Answer 1

It is almost never a good idea to use par(new=TRUE), it causes more problems than it solves.

You should decide which x axis units you want to use, days or thermal units and create the initial graph using that unit, then use functions like points or lines to add any additional points or lines to the existing graph using the existing units. You can then use the axis function to add another axis, use the original units for the at argument, but then the conversion to the other units for the labels (you need to decide if you want the locations of the ticks to match the original units, or be in locations that give pretty values for the converted units).