-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path5c_method_GAM_short.R
207 lines (154 loc) · 7.56 KB
/
5c_method_GAM_short.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
### TRIAS - Method GAM
# df dataframe with time serie for one species - grouped by year
#
# return list with following elements
# - em dataframe with only one row indicating emerging status for the last year
# - model the whole gam model
# - df_n data frame with prediction data
# - plot ggplot figure with original data and interpretation of emerging status
# - deriv1, deriv2 dataframes with 1st & 2nd derivatives of smoother 'year'
# - result result of try (to capture errors)
#########################################################
# GAM lumped data on time series. Only year as covariate
spGAM_lcount <- function(df, method_em = "GAM_lcount", nbyear = 3,
printplot = FALSE, saveplot = FALSE, savemodel = FALSE) {
require(mgcv)
require(gratia)
fyear <- min(df$year) # First year
lyear <- max(df$year) # Last year
spec <- df[[1,1]] # species name (taxonKey)
spn <- spec_names %>% filter(taxonKey == spec) %>% pull(spn) %>% as.character()
print(paste0(spec, "_", spn))
fm <- formula(obs ~ s(year, k = maxk, m = 3, bs = "tp"))
maxk <- max(round((lyear - fyear) / 10, 0), 4) # 1 knot per decade
mindct <- ifelse(n_distinct(df$obs) >= maxk, TRUE, FALSE) # check distinct values
if (method_em == "GAM_lcount_cobs") {
fm <- update(fm, ~. + s(cobs, k = 4))
mindct <- ifelse(mindct == TRUE && n_distinct(df$cobs) >= 4, TRUE, FALSE)
}
# assign NULL values in case something goes wrong later
g1 <- df_n <- g <- em_level_gam <- deriv1 <- deriv2 <- err_result <- aic <- NULL
df_em <- data.frame(taxonKey = spec, eyear = (lyear - nbyear + 1):lyear,
method_em = method_em, em = NA, lcl = NA,
stringsAsFactors = FALSE)
if (mindct) {
result <- tryCatch(expr = {
g1 <- gam(formula = fm,
family = nb(),
data = df, method = "REML")
# Check at p-value of least 1 smoother < 0.1
p_ok <- ifelse(any(summary.gam(g1)$s.pv < 0.1), TRUE, FALSE)
if (p_ok){
# Predict in real scale
df_n <- predict_real_scale(df, g1)
# Calculate first and second derivative + conf. interval
deriv1 <- derivatives(g1, term = "s(year)", type = "central", order = 1,
level = 0.8, n = nrow(df_n), eps = 1e-4)
deriv2 <- derivatives(g1, term = "s(year)", type = "central", order = 2,
level = 0.8, n = nrow(df_n), eps = 1e-4)
# Emerging status based on first and second derivative
em_level_gam <- em_level(deriv1, deriv2)
df_n <- df_n %>% left_join(em_level_gam, by = "year")
# Mean lower confidence limit from the first derivative
df_lcl <- get_lcl(df_deriv = deriv1, nbyear = nbyear, fam = g1$family)
out <- em_gam2em(em_level_gam, nbyear) # get emerging status
df_em <- tibble(taxonKey = spec, eyear = out$year, method_em = method_em,
em = out$em_out) %>%
left_join(df_lcl %>%
select(year, lcl), by = c("eyear" = "year"))
aic <- g1$aic
# Create plot with conf. interval + colour for status
ptitle <- paste0(method_em,"/", spec, "_", spn)
g <- plot_ribbon_em(df_n = df_n, y_axis = "obs", df = df, ptitle = ptitle,
printplot = printplot, saveplot = saveplot)
}
}, error = function(e) e, warning = function(w) w)
if (class(result)[1] %in% c("simpleWarning", "simpleError", "try-error"))
err_result <- result
}else{
err_result <- "Insufficient data"
}
print(err_result)
if (savemodel == FALSE) g1 <- NULL
return(list(df_em = df_em, model = g1, df_n = df_n, em_level_gam = em_level_gam,
deriv1 = deriv1, deriv2 = deriv2, plot = g, result = err_result,
aic = aic))
}
### GAM lumped count with ncobs (observations native species)
spGAM_lcount_cobs <- function(df, printplot = FALSE, saveplot = FALSE,
savemodel = FALSE, nbyear = 3){
spGAM_lcount(df = df, method_em = "GAM_lcount_cobs", printplot = printplot,
saveplot = saveplot, nbyear = nbyear, savemodel = savemodel)
}
### GAM lumped presence absence (ncell)
spGAM_lpa <- function(df, method_em = "GAM_lpa", printplot = FALSE, nbyear = 3,
saveplot = FALSE, savemodel = FALSE) {
require(mgcv)
require(gratia)
fyear <- min(df$year) # First year
lyear <- max(df$year) # Last year
spec <- df[[1,1]] # species name (taxonKey)
spn <- spec_names %>% filter(taxonKey == spec) %>% pull(spn) %>% as.character()
print(paste0(spec, "_", spn))
maxk <- max(round((lyear - fyear) / 10, 0), 4) # 1 knot per decade
mindct <- ifelse(n_distinct(df$obs) >= maxk, TRUE, FALSE)
fm <- formula(ncell ~ s(year, k = maxk, m = 3, bs = "tp"))
if (method_em == "GAM_lpa_cobs") {
fm <- update(fm, ~. + s(ncobs))
mindct <- ifelse(mindct == TRUE && n_distinct(df$ncobs) >= 4, TRUE, FALSE)
}
g1 <- df_n <- g <- em_level_gam <- deriv1 <- deriv2 <- err_result <- aic <- NULL
df_em <- data.frame(taxonKey = spec, eyear = (lyear - nbyear + 1):lyear,
method_em = method_em, em = NA, lcl = NA,
stringsAsFactors = FALSE)
if (mindct) {
result <- tryCatch(expr = {
g1 <- gam(formula = fm,
family = nb(),
data = df, method = "REML")
# Check at p-value of least 1 smoother < 0.1
s_pv <- summary.gam(g1)$s.pv
p_ok <- ifelse(any(s_pv < 0.1), TRUE, FALSE)
if (p_ok){
# Predict in real scale
df_n <- predict_real_scale(df, g1)
# Calculate first and second derivative + conf. interval
deriv1 <- derivatives(g1, term = "s(year)", type = "central", order = 1,
level = 0.8, n = nrow(df_n), eps = 1e-4)
deriv2 <- derivatives(g1, term = "s(year)", type = "central", order = 2,
level = 0.8, n = nrow(df_n), eps = 1e-4)
# Emerging status based on first and second derivative
em_level_gam <- em_level(deriv1, deriv2)
df_n <- bind_cols(df_n, em_level_gam)
# Mean lower confidence limit from the first derivative
df_lcl <- get_lcl(df_deriv = deriv1, nbyear = nbyear, fam = g1$family)
aic <- g1$aic
# Create plot with conf. interval + colour for status
ptitle <- paste0(method_em,"/", spec, "_", spn, "_", lyear)
g <- plot_ribbon_em(df_n = df_n, y_axis = "ncell", df = df, ptitle = ptitle,
printplot = printplot, saveplot = saveplot)
out <- em_gam2em(em_level_gam, nbyear) # get emerging status
df_em <- tibble(taxonKey = spec, eyear = out$year, method_em = method_em,
em = out$em_out) %>%
left_join(df_lcl %>%
select(year, lcl), by = c("eyear" = "year"))
}
}, error = function(e) e, warning = function(w) w)
if (class(result)[1] %in% c("simpleWarning", "simpleError", "try-error")){
err_result <- result
}
}else{
err_result <- "Insufficient data"
}
print(err_result)
if (savemodel == FALSE) g1 <- NULL
return(list(df_em = df_em, model = g1, df_n = df_n, em_level_gam = em_level_gam,
deriv1 = deriv1, deriv2 = deriv2, plot = g, result = err_result,
aic = aic))
}
## GAM presence / absence with native observations (cobs)
spGAM_lpa_cobs <- function(df, printplot = FALSE,
saveplot = FALSE, savemodel = FALSE, nbyear = 3) {
spGAM_lpa(df = df, method_em = "GAM_lpa_cobs", printplot = printplot,
saveplot = saveplot, savemodel = savemodel, nbyear = nbyear)
}