Overview

Three models Effect of local (pH, Olsen P and Age) and management practice (Number of species that were sown) on:

The total observed vegetation characteristic
The sown (observed and sown) vegetation characteristic
The sown (observed and sown) vegetation characteristic relative to the total observed vegetation

Overview data

Model	Unit	Variable	Data type
Total number of species	count	n_obs	counted
Number of sown species	count	n_sown	counted
Relative number of sown species	count	n_sown	counted
———	——	———-	——
Total cover	%	cover_obs	continuous ≥ 0
Cover of sown species	%	cover_sown	continuous ≥ 0
Relative cover of sown species	%	cover_sown	continuous ≥ 0
———	——	———-	——
Total nectar	g sugar/m²/year	nectar_obs	continuous ≥ 0
Nectar of sown species	g sugar/m²/year	nectar_sown	continuous ≥ 0
Relative nectar of sown species	g sugar/m²/year	nectar_sown	continuous ≥ 0
———	——	———-	——
Total grass cover	%	grass_obs	continuous ≥ 0
Sown grass cover	%	grass_sown	continuous ≥ 0
Relative grass cover	%	grass_sown	continuous ≥ 0
———	——	———-	——
Total CWM Competitive	%	C_weighted_observed	continuous (0–100)
Sown CWM Competitive	%	C_weighted_sown	continuous (0–100)
Relative CWM Competitive	%	C_weighted_sown	continuous (0–100)
Total CWM Stress-tolerant	%	S_weighted_observed	continuous (0–100)
Sown CWM Stress-tolerant	%	S_weighted_sown	continuous (0–100)
Relative CWM Stress-tolerant	%	S_weighted_sown	continuous (0–100)
Total CWM Ruderal	%	R_weighted_observed	continuous (0–100)
Sown CWM Ruderal	%	R_weighted_sown	continuous (0–100)
Relative CWM Ruderal	%	R_weighted_sown	continuous (0–100)
———	——	———-	——
Total CWM Ellenberg N	–	Ell_N_weighted_obs	continuous ≥ 0
Sown CWM Ellenberg N	–	Ell_N_weighted_sown	continuous ≥ 0
Relative CWM Ellenberg N	–	Ell_N_weighted_sown	continuous ≥ 0
Total CWM Ellenberg R	–	Ell_R_weighted_obs	continuous ≥ 0
Sown CWM Ellenberg R	–	Ell_R_weighted_sown	continuous ≥ 0
Relative CWM Ellenberg R	–	Ell_R_weighted_sown	continuous ≥ 0
Total CWM Ellenberg M	–	Ell_M_weighted_obs	continuous ≥ 0
Sown CWM Ellenberg M	–	Ell_M_weighted_sown	continuous ≥ 0
Relative CWM Ellenberg M	–	Ell_M_weighted_sown	continuous ≥ 0
Total CWM Ellenberg T	–	Ell_T_weighted_obs	continuous ≥ 0
Sown CWM Ellenberg T	–	Ell_T_weighted_sown	continuous ≥ 0
Relative CWM Ellenberg T	–	Ell_T_weighted_sown	continuous ≥ 0

Concrete vragen

Wat vinden jullie van de voorgestelde modellen
Lijkt het gebruik van een offset in het model om de relatieve contributie van de ingezaaide soorten te bekijken jullie ok? Proportionele data is niet evident om in een model te gebruiken
Resiudelen zijn niet altijd normaal verdeeld, ik weet niet zo goed welke transformaties ik hier het best moet uitvoeren.

En uiteraard, zeer open voor alle algemen feedback! :)

Hieronder vinden jullie telkens de modellen en de “check for residuals and outliers”

Dit gaat echt puur over de aanpak/modellen - daarom heb ik de output van modellen er nog niet bijgezet :).

Number of species

Check for overdisperion

Number of observed species observed + Number of sown species

## Counted data >> overdispersion?
m_pois_obs <- glm(n_obs ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022), family = poisson, df)
m_pois_sown <- glm(n_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022), family = poisson, df)

disp_obs <- sum(residuals(m_pois_obs, type = "pearson")^2) /df.residual(m_pois_obs)
disp_sown <- sum(residuals(m_pois_sown, type = "pearson")^2) /df.residual(m_pois_sown)

disp_obs

## [1] 1.305257

disp_sown

## [1] 2.503273

Model proposition

Overdisperion so Negative Binomial models

m_obs <- glmmTMB(n_obs ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022), family = nbinom2, df)
m_sown <- glmmTMB(n_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022), family = nbinom2, df)
m_sown_prop <- glmmTMB(n_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022) 
                       + offset(log(n_obs)), family = nbinom2, df)

Total cover

Model proposition

m_cover_obs <- glm(
  cover_obs ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_cover_sown <- glm(
  cover_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),data = df)

m_cover_sown_prop <- glm(
  cover_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022)
  + offset(log(cover_obs)), family = gaussian(),data = df)

Check for normality of residuals

# Observed
res_obs <- simulateResiduals(fittedModel = m_cover_obs, n = 200)
plot(res_obs)

# Sown
res_sown <- simulateResiduals(fittedModel = m_cover_sown, n = 200)
plot(res_sown)

# Sown relative
res_sown_prop <- simulateResiduals(fittedModel = m_cover_sown_prop, n = 200)
plot(res_sown_prop)

Grass cover

Model proposition

m_grass_obs <- glm(
  grass_cover_obs ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_grass_sown <- glm(
  grass_cover_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_grass_sown_prop <- glm(
  grass_cover_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022)
  + offset(log(grass_cover_obs)), family = gaussian(),data = df)

Check for normality of residuals

Probleem zit hem hier in heel veel nullen denk ik >> 17 nul waardes voor “sown”

“zero inflation test” - p < 0.05

# Observed
res_obs <- simulateResiduals(fittedModel = m_grass_obs, n = 200)
plot(res_obs)

# Sown
res_sown <- simulateResiduals(fittedModel = m_grass_sown, n = 200)
plot(res_sown)

## Warning in newton(lsp = lsp, X = G$X, y = G$y, Eb = G$Eb, UrS = G$UrS, L = G$L,
## : Fitting terminated with step failure - check results carefully

testZeroInflation(res_sown)

## 
##  DHARMa zero-inflation test via comparison to expected zeros with
##  simulation under H0 = fitted model
## 
## data:  simulationOutput
## ratioObsSim = Inf, p-value < 2.2e-16
## alternative hypothesis: two.sided

# Sown relative
res_sown_prop <- simulateResiduals(fittedModel = m_grass_sown_prop, n = 200)
plot(res_sown_prop)

testZeroInflation(res_sown)

## 
##  DHARMa zero-inflation test via comparison to expected zeros with
##  simulation under H0 = fitted model
## 
## data:  simulationOutput
## ratioObsSim = Inf, p-value < 2.2e-16
## alternative hypothesis: two.sided

Potential Nectar Value

Model proposition

m_nectar_obs <- glm(
  nectar_obs ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_nectar_sown <- glm(
  nectar_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_nectar_sown_prop <- glm(
  nectar_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022)
  + offset(log(nectar_obs)), family = gaussian(),data = df)

Check for normality of residuals

# Observed
res_obs <- simulateResiduals(fittedModel = m_nectar_obs, n = 200)
plot(res_obs)

# Sown
res_sown <- simulateResiduals(fittedModel = m_nectar_sown, n = 200)
plot(res_sown)

# Sown relative
res_sown_prop <- simulateResiduals(fittedModel = m_nectar_sown_prop, n = 200)
plot(res_sown_prop)

CSR

Competitive

Model proposition

m_CSRC_obs <- glm(
  C_weighted_observed ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_CSRC_sown <- glm(
  C_weighted_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_CSRC_sown_prop <- glm(
  C_weighted_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022)
  + offset(log(C_weighted_observed)), family = gaussian(),data = df)

Check for normality of residuals

# Observed
res_obs <- simulateResiduals(fittedModel = m_CSRC_obs, n = 200)
plot(res_obs)

# Sown
res_sown <- simulateResiduals(fittedModel = m_CSRC_sown, n = 200)
plot(res_sown)

## Warning in newton(lsp = lsp, X = G$X, y = G$y, Eb = G$Eb, UrS = G$UrS, L = G$L,
## : Fitting terminated with step failure - check results carefully

# Sown relative
res_sown_prop <- simulateResiduals(fittedModel = m_CSRC_sown_prop, n = 200)
plot(res_sown_prop)

Stress tolerant

Model proposition

m_CSRS_obs <- glm(
  S_weighted_observed ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_CSRS_sown <- glm(
  S_weighted_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_CSRS_sown_prop <- glm(
  S_weighted_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022)
  + offset(log(S_weighted_observed)), family = gaussian(),data = df)

Check for normality of residuals

# Observed
res_obs <- simulateResiduals(fittedModel = m_CSRS_obs, n = 200)
plot(res_obs)

# Sown
res_sown <- simulateResiduals(fittedModel = m_CSRS_sown, n = 200)
plot(res_sown)

## qu = 0.75, log(sigma) = -2.85182 : outer Newton did not converge fully.

## qu = 0.75, log(sigma) = -2.834878 : outer Newton did not converge fully.

## Warning in newton(lsp = lsp, X = G$X, y = G$y, Eb = G$Eb, UrS = G$UrS, L = G$L,
## : Fitting terminated with step failure - check results carefully

# Sown relative
res_sown_prop <- simulateResiduals(fittedModel = m_CSRS_sown_prop, n = 200)
plot(res_sown_prop)

## qu = 0.25, log(sigma) = -4.047707 : outer Newton did not converge fully.

## qu = 0.5, log(sigma) = -3.040892 : outer Newton did not converge fully.

Ruderal

Model proposition

m_CSRR_obs <- glm(
  R_weighted_observed ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_CSRR_sown <- glm(
  R_weighted_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022),
  family = gaussian(),  data = df)

m_CSRR_sown_prop <- glm(
  R_weighted_sown ~ scale(sown_mixt) + scale(age) + scale(olsenp_2022) + scale(ph_2022)
  + offset(log(R_weighted_observed)), family = gaussian(),data = df)

Check for normality of residuals

# Observed
res_obs <- simulateResiduals(fittedModel = m_CSRR_obs, n = 200)
plot(res_obs)

# Sown
res_sown <- simulateResiduals(fittedModel = m_CSRR_sown, n = 200)
plot(res_sown)

# Sown relative
res_sown_prop <- simulateResiduals(fittedModel = m_CSRR_sown_prop, n = 200)
plot(res_sown_prop)

FS_Ghent_Models_V1

Ellen

2026-02-11

Overview

Overview data

Concrete vragen

Number of species

Check for overdisperion

Model proposition

Total cover

Model proposition

Check for normality of residuals

Grass cover

Model proposition

Check for normality of residuals

Potential Nectar Value

Model proposition

Check for normality of residuals

CSR

Competitive

Model proposition

Check for normality of residuals

Stress tolerant

Model proposition

Check for normality of residuals

Ruderal

Model proposition

Check for normality of residuals