“Approximating” the derivative of date points in R

问题

So I have a time series of MODIS NDVI values (vegetation values from 0-1 for the non-geographic geeks), and I'm trying to approximate the derivative by using a for loop.

This is a sample of my data:

> m2001
   date  value    valnorm
1     1 0.4834 0.03460912
2    17 0.4844 0.03664495
3    33 0.5006 0.06962541
4    49 0.4796 0.02687296
5    65 0.5128 0.09446254
6    81 0.4915 0.05109935
7    97 0.4664 0.00000000
8   113 0.5345 0.13864007
9   129 0.8771 0.83611564
10  145 0.9529 0.99043160
11  161 0.9250 0.93363192
12  177 0.9450 0.97434853
13  193 0.9491 0.98269544
14  209 0.9434 0.97109121
15  225 0.9576 1.00000000
16  241 0.8992 0.88110749
17  257 0.9115 0.90614821
18  273 0.8361 0.75264658
19  289 0.5725 0.21600163
20  305 0.5188 0.10667752
21  321 0.5467 0.16347720
22  337 0.5484 0.16693811
23  353 0.5427 0.15533388

Column 1 is the julian day of the pixel value
Column 2 is the raw NDVI value
Column 3 is the NDVI stretched from 0-1 (it's a normalization technique, since NDVI rarely actually gets to 1 or 0).

I'm still very new to programming and R, but I think I've managed to piece together a tenuous grasp on it. What I'm trying to do is create a new column with values that would give me some idea of the local slope of data points.

The function I've come up with is this:

deriv <- function(x1=1:23, x2=1){
    for (i in x1){
    i1 <- c(x1[i-1], x1[i], x1[i+1])
    i2 <- c(x2[i-1], x2[i], x2[i+1])
        deriv.func <- lm(i2~i1, na.action=NULL)
    } return(deriv.func$coef[[2]])
}

What happens when I run it is this:

> deriv <- function(x1=1:23, x2=1){
+ for (i in x1){
+     i1 <- c(x1[i-1], x1[i], x1[i+1])
+     i2 <- c(x2[i-1], x2[i], x2[i+1])
+ deriv.func <- lm(i2~i1, na.action=NULL)
+ } return(deriv.func$coef[[2]])
Error: unexpected symbol in:
"deriv.func <- lm(i2~i1, na.action=NULL)
} return"
> }
Error: unexpected '}' in "}"
>

I'm not sure what I'm doing wrong, as I can get it to parse when I fill in a value for i

> i=6
> x1=m2001$date
> x2=m2001$valnorm
>     i1 <- c(x1[i-1], x1[i], x1[i+1])
>     i2 <- c(x2[i-1], x2[i], x2[i+1])
> i1
[1] 33 49 65
> i2
[1] 0.06962541 0.02687296 0.09446254
> lm(i2 ~ i1)

Call:
lm(formula = i2 ~ i1)

Coefficients:
(Intercept)           i1  
  0.0256218    0.0007762  

> func <- lm(i2 ~ i1)
> func$coef[[2]]
[1] 0.0007761604

Any ideas? Thanks a ton.

回答1:

Try putting 'return' on a new line.

    } 
    return(deriv.func$coef[[2]])
}

回答2:

Well, after looking (a lot) more into the for loop, I got it to do what I want.

deriv <- function(x1=1:23, x2=1){
  n=length(x1)
  deriv.func <- character(length = n)
    for (i in 1:n) {
    i1 <- c(x1[i-1], x1[i], x1[i+1])
    i2 <- c(x2[i-1], x2[i], x2[i+1])
        derivate <- lm(i2~i1)
        deriv.func[i] <- derivate$coef[[2]]*
    }
  return(deriv.func)
}

Thanks for the help, and the tip in the right direction, @dbaseman!
Ideas that made a difference:

making sure I had space allocated for the iterator deriv.func <- character(length = n).
making sure the intermediate variables didn't overwrite the output.

来源：https://stackoverflow.com/questions/10266308/approximating-the-derivative-of-date-points-in-r

标签

function

for-loop

time-series

derivative