Beck doublelogistic parametrization typo

[SveBuch]

Hello,

I'm currently trying to fit a double logistic curve to Sentinel derived NDVI values. When predicting values, based on the fitted curve, the curve was always biased low. It seems as if there is a typo in the formula returned by curvefit. I tracked the error down to the function doubleLog.Beck.

In there, the double-log curve is parametrized as mn + (mx - mn) * (1/(1 + exp(-rsp * (t - sos))) + 1/(1 + exp(rau * (t - eos)))) - 1 while in Beck et al. 2006 (https://doi.org/10.1016/j.rse.2005.10.021) it is parametrized as mn + (mx - mn) * (1/(1 + exp(-rsp * (t - sos))) + 1/(1 + exp(rau * (t - eos))) - 1)

What I don't understand, is, that if I predict with the corrected formula but using the parameters fitted with the one with the typo, the resulting prediction looks quite okay. Maybe you can help me there?

All the best, Sven

Here is a reproducible example:

## doublelog parametrization as in output from curvefit
dlogfun <- function(par, t) {
  mn <- par$mn; mx <- par$mx; sos <- par$sos; rsp <- par$rsp; eos <- par$eos; rau <- par$rau
  mn + (mx - mn) * (1/(1 + exp(-rsp * (t - sos))) + 1/(1 + exp(rau * (t - eos)))) - 1
}

## doublelog parametrization as in Beck et al.
dlogfun2 <- function(par, t) {
  mn <- par$mn; mx <- par$mx; sos <- par$sos; rsp <- par$rsp; eos <- par$eos; rau <- par$rau
  mn + (mx - mn) * (1/(1 + exp(-rsp * (t - sos))) + 1/(1 + exp(rau * (t - eos))) - 1)
}

## modified example from phenofit help
par  = c( mn  = 0.1, mx  = 0.7, sos = 50, rsp = 0.1, eos = 250, rau = 0.1)
t    <- seq(1, 365, 8)
tout <- seq(1, 365, 1)
y <- doubleLog.Beck(par, t)  + rnorm(length(t), 0, 0.1)
fFITs <- curvefit(y, t, tout, "Beck")

## double log-parametrisation from curvefit
fFITs$model

(l_param   <- get_param(list(fFITs))$Beck)

ggplot() +
  geom_point(aes(x = t, y = y)) +
  geom_line(aes(x = tout, y = dlogfun(l_param, tout)), col="red") +
  geom_line(aes(x = tout, y = dlogfun2(l_param, tout))) +
  theme_bw()

[kongdd]

The equation is right in https://github.com/eco-hydro/phenofit/blob/master/R/doubleLogistics_R.R#L148. Could you tell which line do you find the error equation?

[kongdd]

The output is normal:

library(phenofit)
library(ggplot2)

## modified example from phenofit help
par  = c( mn  = 0.1, mx  = 0.7, sos = 50, rsp = 0.1, eos = 250, rau = 0.1)
t    <- seq(1, 365, 8)
tout <- seq(1, 365, 1)
y <- doubleLog.Beck(par, t)  + rnorm(length(t), 0, 0.1)

fit <- curvefit(y, t, tout, "Beck")

## double log-parametrisation from curvefit
par_opt = fit$model$Beck$par
# l_par <- get_param(fit)$Beck # v0.3.5 install from github

ypred = doubleLog.Beck(l_par %>% as.numeric(), tout)

ggplot() +
    geom_point(aes(x = t, y = y)) +
    geom_line(aes(x = tout, y = ypred), col="red") +
    # geom_line(aes(x = tout, y = dlogfun2(l_param, tout))) +
    theme_bw()

[SveBuch]

Ah, I see. The typo is in line 152 https://github.com/eco-hydro/phenofit/blob/master/R/doubleLogistics_R.R#L152, meaning only the attribute of the function is affected. So the fitting is done correctly, but since I used the attribute in a formula, the curve was shifted. Thank you

[kongdd]

Thanks for your report. This bug will influence phenology extraction. I will fix it.