quantChange/quantChangeALL.R at main · JuanchoData/quantChange · GitHub

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
## install the package
#install.packages('bfast')

## load the required packages
library(bfast)

## read the data
data = read.csv('yankee.csv')

## BFAST can't handle missing values
## Insert a row for July 2019
data[nrow(data) +1,] = c("July 1, 2019", NA)

## sort the df
data = data[order(as.Date(data$system.time_start, format = "%b %e, %Y")),]
rownames(data) = 1:nrow(data)

data$mesic = as.numeric(data$mesic)

## define a function to interpolate
replace_na_with_mean <- function(df) {
  for (i in 1:ncol(df)) {
    na_indices <- which(is.na(df[, i]))
    for (j in na_indices) {
      if (j == 1) {
        df[j, i] <- df[j + 1, i]
      } else if (j == nrow(df)) {
        df[j, i] <- df[j - 1, i]
      } else {
        df[j, i] <- mean(c(df[j - 1, i], df[j + 1, i]))
      }
    }
  }
  return(df)
}

## apply the interpolation function to fill NAs
data.fill = replace_na_with_mean(data)

## Embrace the NAs! insert some more for 2016
data.2016 = cbind(c("Jun 1, 2016", "July 1, 2016", "Aug 1, 2016", "Sep 1, 2016"),
                  c(rep(NA,4)))

colnames(data.2016) = c("system.time_start", "mesic")
data.NAs = rbind(data, data.2016)
data.NAs$date = as.Date(data.NAs$system.time_start, format = "%b %e, %Y")
data.NAs = data.NAs[order(data.NAs$date),]

## create a time series object
data.ts = ts(data.fill$mesic, frequency = 4)
data.ts.NA = ts(data.NAs$mesic, frequency = 4)

## assigning time units to observations
#tsp(data.ts.NA) = c(as.Date("2004-06-01"),as.Date("2020-09-01"), 4 )


plot(data.ts)
plot(data.ts.NA)

## apply the BFAST function
fit = bfast(data.ts, h = 0.15, season = "dummy")
#fit.na = bfast(data.ts.NA, h = 0.15, season = "dummy") ## can't handle missing data

## test the sensitivity to the prop of observations used to detect breakpoint; < 0.5
fit.05 = bfast(data.ts, h = 0.05, season = "dummy")
fit.25 = bfast(data.ts, h = 0.25, season = "dummy")
fit.35 = bfast(data.ts, h = 0.35, season = "dummy")
fit.45 = bfast(data.ts, h = 0.45, season = "dummy")

## plot the output
plot(fit, ANOVA = T)

#plot(fit.05, ANOVA = T) ## no diffs
#plot(fit.25, ANOVA = T) ## no diffs
#plot(fit.35, ANOVA = T) ## no diffs
#plot(fit.45, ANOVA = T) ## no diffs
#plot(fit.NA, ANOVA = T) ## doesn't work

######### Bayesian Change Point (for comparison) #############

##
#install.packages('bcp')
library(bcp)

## include a 'year' variable
data.fill$year = as.numeric(format(as.Date(data.fill$system.time_start, format = "%b %e, %Y"), "%Y"))

## univariate
results.bcp = bcp(data.fill$mesic)
plot(results.bcp,main = "WRP Mesic Probabilities of Change - Yankee Fork",xlab = "Date", xaxlab = data.fill$year )

summary(results.bcp)

## Multivariate - use PDSI as a predictor with time

## data with drought index predictor
pdsi = read.csv('pdsi_Yankee.csv')

## remove 2016 to have corresponding values with the time series
pdsi$year = as.numeric(format(as.Date(pdsi$system.time_start, format = "%b %e, %Y"), "%Y"))
pdsi.no2016 = subset(pdsi, year != 2016)

data.multiv = cbind(data.fill, pdsi.no2016$pdsi)
results.bcp.multiv = bcp(data.multiv$mesic, data.multiv$`pdsi.no2016$pdsi`)
plot(results.bcp.multiv,main = "WRP Mesic Probabilities of Change - Yankee Fork Multivariate",xlab = "Date", xaxlab = data.fill$year )

summary(results.bcp.multiv)

## with NAs  - DOESNT WORK
#results.bcp.na = bcp(data.NAs$mesic)
#plot(results.bcp,main = "WRP Mesic Probabilities of Change - Yankee Fork",xlab = "Date", xaxlab = data.NAs$date )

## Test sensitivity of the w0 and p0 params - quant diffs minimal, no qualitative diffs
#results.bcp.w1 = bcp(data.fill$mesic, w0 = 0.1)
#results.bcp.w3 = bcp(data.fill$mesic, w0 = 0.3)
#results.bcp.w4 = bcp(data.fill$mesic, w0 = 0.4)
#results.bcp.w5 = bcp(data.fill$mesic, w0 = 0.5)
#results.bcp.w6 = bcp(data.fill$mesic, w0 = 0.6)
#results.bcp.w7 = bcp(data.fill$mesic, w0 = 0.7)
#results.bcp.w8 = bcp(data.fill$mesic, w0 = 0.8)
#results.bcp.w9 = bcp(data.fill$mesic, w0 = 0.9)
#
#results.bcp.p1 = bcp(data.fill$mesic, p0 = 0.1)
#results.bcp.p3 = bcp(data.fill$mesic, p0 = 0.3)
#results.bcp.p4 = bcp(data.fill$mesic, p0 = 0.4)
#results.bcp.p5 = bcp(data.fill$mesic, p0 = 0.5)
#results.bcp.p6 = bcp(data.fill$mesic, p0 = 0.6)
#results.bcp.p7 = bcp(data.fill$mesic, p0 = 0.7)
#results.bcp.p8 = bcp(data.fill$mesic, p0 = 0.8)
#results.bcp.p9 = bcp(data.fill$mesic, p0 = 0.9)
#
#plot(results.bcp.w1, main = "WRP Mesic Probabilities of Change - w0 = 0.1",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.w3, main = "WRP Mesic Probabilities of Change - w0 = 0.3",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.w4, main = "WRP Mesic Probabilities of Change - w0 = 0.4",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.w5, main = "WRP Mesic Probabilities of Change - w0 = 0.5",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.w6, main = "WRP Mesic Probabilities of Change - w0 = 0.6",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.w7, main = "WRP Mesic Probabilities of Change - w0 = 0.7",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.w8, main = "WRP Mesic Probabilities of Change - w0 = 0.8",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.w9, main = "WRP Mesic Probabilities of Change - w0 = 0.9",xlab = "Date", xaxlab = data.fill$year )
#
#plot(results.bcp.p1, main = "WRP Mesic Probabilities of Change - p0 = 0.1",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.p3, main = "WRP Mesic Probabilities of Change - p0 = 0.3",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.p4, main = "WRP Mesic Probabilities of Change - p0 = 0.4",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.p5, main = "WRP Mesic Probabilities of Change - p0 = 0.5",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.p6, main = "WRP Mesic Probabilities of Change - p0 = 0.6",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.p7, main = "WRP Mesic Probabilities of Change - p0 = 0.7",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.p8, main = "WRP Mesic Probabilities of Change - p0 = 0.8",xlab = "Date", xaxlab = data.fill$year )
#plot(results.bcp.p9, main = "WRP Mesic Probabilities of Change - p0 = 0.9",xlab = "Date", xaxlab = data.fill$year )

#install.packages('Rbeast')
library(Rbeast)

results.beast = beast(data.fill$mesic, start = as.Date('2004-6-1'), deltat = "3 months", dump.ci = T) ##not quite right - missing 2016 data....
plot(results.beast, interactive = F)

## missing data
results.beast.na = beast(as.numeric(data.ts.NA), start = as.Date('2004-1-1'), deltat = "3 months")
plot(results.beast.na, interactive = F) ## missing data leads to higher likelihood of negative slopes, and an extra breakpoint estimated in the trend

## Parameters
results.beast.na.unifPrec = beast(as.numeric(data.ts.NA), start = as.Date('2004-1-1'), deltat = "3 months", precPriorType = 'uniform')
plot(results.beast.na.unifPrec, interactive = F)

results.beast.na.compPrec = beast(as.numeric(data.ts.NA), start = as.Date('2004-1-1'), deltat = "3 months", precPriorType = 'componentwise')
plot(results.beast.na.compPrec, interactive = F)

## BSTS
library(CausalImpact)

## Include PDSI as a predictor
data.bsts=  cbind(data.fill$mesic, pdsi.no2016$pdsi)
data.bsts.time =  zoo(cbind(data.fill$mesic, pdsi$pdsi), as.Date(data.fill$system.time_start,format = "%b %e, %Y")) ## includes missing data period

## define pre and post restoration periods
pre.period = c(1,32)
post.period = c(33, 64)

## with date
pre.period.date = as.Date(c("Jun 1, 2004", "Sep 1,2011"), format = "%b %e, %Y")
post.period.date = as.Date(c("Jun 1, 2012", "Sep 1,2020"), format = "%b %e, %Y")

impact = CausalImpact(data.bsts, pre.period, post.period, model.args = list(nseasons = 4))
impact.time = CausalImpact(data.bsts.time, pre.period.date, post.period.date, model.args = list(nseasons = 4))
plot(impact)
plot(impact.time)

summary(impact)
summary(impact.time) ## nearly identical results AND includes period of missing data...


library(terra)
ras = rast("Yankee.tif")
plot(ras)