Skip to content

Commit

Permalink
accuracy matrices and plotting results
Browse files Browse the repository at this point in the history
  • Loading branch information
@jonathom
jonathom committed Mar 21, 2021
1 parent 19fe227 commit c2c27c6
Show file tree
Hide file tree
Showing 13 changed files with 289 additions and 186 deletions.
171 changes: 102 additions & 69 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -80,17 +80,10 @@ prod <- read_sf("./yearly_deforestation/PRODES_cropped.shp")
``` r
# whats in it?
# plot(st)
plot(as(st[,,,6], "Raster"))
plot(prod["YEAR"], add = TRUE)
```

![](main_files/figure-markdown_github/plot-AOI-1.png)

``` r
plot(prod["YEAR"])
```

![](main_files/figure-markdown_github/plot-AOI-2.png)
![](main_files/figure-markdown_github/plot-AOI-1.png)

Prepare for Gapfill
-------------------
Expand Down Expand Up @@ -186,94 +179,134 @@ bfast_on_tile <- function(gapfill_matrix, by, ts, order) {
return(result)
}

bfast_monthly <- bfast_on_tile(gf_monthly$fill, by = .08333333, ts = 84, order = 3)
bfast_quarter <- bfast_on_tile(gf_quarterly$fill, by = 0.25, ts = 28, order = 2)
bfast_monthly3 <- bfast_on_tile(gf_monthly$fill, by = .08333333, ts = 84, order = 3)
bfast_monthly2 <- bfast_on_tile(gf_monthly$fill, by = .08333333, ts = 84, order = 2)
bfast_quarter2 <- bfast_on_tile(gf_quarterly$fill, by = 0.25, ts = 28, order = 2)
saveRDS(bfast_monthly3, "bfast_monthly3.rds")
saveRDS(bfast_monthly2, "bfast_monthly2.rds")
saveRDS(bfast_quarter2, "bfast_quarter2.rds")
# warning: too few observations in history period
```

``` r
# plot(prod["YEAR"])
# bfast_monthly3 <- readRDS("bfast_monthly3.rds")
bfast_monthly2 <- readRDS("bfast_monthly2.rds")
bfast_quarter2 <- readRDS("bfast_quarter2.rds")
```

``` r
# rasterize PRODES data
ras <- rasterize(prod, as(st[,,,5], "Raster"), "YEAR")
ras.m <- aperm(matrix(ras[], ncol = 140), c(2,1))
ras.m[ras.m < 2019] <- FALSE
ras.m[ras.m == 2019] <- TRUE
ras.m[is.na(ras.m)] <- FALSE
prodes <- aperm(matrix(ras[], ncol = 140), c(2,1))
prodes[prodes < 2019] <- FALSE
prodes[prodes == 2019] <- TRUE
prodes[is.na(prodes)] <- FALSE

# to mask out previous deforestation
prodes_prev <- aperm(matrix(ras[], ncol = 140), c(2,1))
prodes_prev[prodes_prev < 2019] <- TRUE
prodes_prev[prodes_prev == 2019] <- FALSE
prodes_prev[is.na(prodes_prev)] <- FALSE

bfast_monthly2_prev <- bfast_monthly2
bfast_monthly2_prev[prodes_prev == 1] <- NA

bfast_quarter2_prev <- bfast_quarter2
bfast_quarter2_prev[prodes_prev == 1] <- NA
```

Image(bfast_monthly)
``` r
table1 <- addmargins(table(bfast_monthly2, prodes))
table2 <- addmargins(table(bfast_quarter2, prodes))

accuracies <- function(table1) {
# overall accuracy
P0 <- (table1[1] + table1[5]) / table1[9]
# error of omission, wrong in respect to reference data
# table1[2] / table1[3] # FALSE
# table1[4] / table1[6] # TRUE
# error of commission, wrong in respect to classified data
# table1[4] / table1[7] # FALSE
# table1[2] / table1[8] # TRUE
# producer's accuracy, Probability of classifying a pixel correctly
pa_f <- table1[1] / table1[3] # FALSE
pa_t <- table1[5] / table1[6] # TRUE
# user's accuracy, Probability of a pixel being the classified type
ua_f <- table1[1] / table1[7] # FALSE
ua_t <- table1[5] / table1[8] # TRUE
# kappa
# chance that both TRUE / FALSE randomly
tr <- (table1[8] / table1[9]) * (table1[6] / table1[9])
fr <- (table1[7] / table1[9]) * (table1[3] / table1[9])
Pe <- tr + fr
kappa <- (P0 - Pe) / (1 - Pe)

return(list("Overall Accuracy" = P0*100, "Prod. Acc. FALSE" = pa_f, "Prod. Acc. TRUE" = pa_t, "User's Acc. FALSE" = ua_f, "User's Acc. TRUE" = ua_t, "Kappa" = kappa))
}
```

![](main_files/figure-markdown_github/plot-bfast-aoi-1.png)
Results
=======

``` r
Image(bfast_quarter)
Image(bfast_monthly2) + ggtitle("Monthly Data") + theme(plot.title = element_text(size=22))
Image(bfast_quarter2) + ggtitle("Quarterly Data") + theme(plot.title = element_text(size=22))
Image(prodes) + ggtitle("PRODES Data") + theme(plot.title = element_text(size=22))

table1
```

![](main_files/figure-markdown_github/plot-bfast-aoi-2.png)
## prodes
## bfast_monthly2 0 1 Sum
## FALSE 15599 170 15769
## TRUE 1734 2097 3831
## Sum 17333 2267 19600

``` r
Image(ras.m)
table2
```

![](main_files/figure-markdown_github/plot-bfast-aoi-3.png)
## prodes
## bfast_quarter2 0 1 Sum
## FALSE 15662 374 16036
## TRUE 1671 1893 3564
## Sum 17333 2267 19600

``` r
table(bfast_monthly, ras.m)
array(c(accuracies(table1), accuracies(table2)), dim = c(6,2), dimnames = list(c("Overall Accuracy", "Prod. Acc. FALSE", "Prod. Acc. TRUE", "User's Acc. FALSE", "User's Acc. TRUE", "Kappa"), c("monthly", "quarterly")))
```

## ras.m
## bfast_monthly 0 1
## FALSE 15646 162
## TRUE 1687 2105
## monthly quarterly
## Overall Accuracy 90.28571 89.56633
## Prod. Acc. FALSE 0.8999596 0.9035943
## Prod. Acc. TRUE 0.925011 0.8350243
## User's Acc. FALSE 0.9892194 0.9766775
## User's Acc. TRUE 0.5473767 0.5311448
## Kappa 0.6346743 0.5915353

``` r
table(bfast_quarter, ras.m)
```
# remove already deforested areas
ras[] <- bfast_monthly2
plot(ras)
plot(st_geometry(prod), add = TRUE)

## ras.m
## bfast_quarter 0 1
## FALSE 15662 374
## TRUE 1671 1893
ras[] <- bfast_quarter2
plot(ras)
plot(st_geometry(prod), add = TRUE)

old code
--------

``` r
# single pixel
defo_test <- st[st_geometry(prod[4,])]
defo_test_ts <- as.vector(defo_test[,3,3,][[1]])
test_ts <- as.ts(zoo(defo_test_ts, seq(2013, by = .08333333, length.out = 84)))
test <- bfastmonitor(test_ts, start = c(2018, 1))
plot(test)

# mean of this polygon
mean_test <- st_apply(st[st_geometry(prod[2,])], "attributes", mean, na.rm = TRUE) # aggregate
defo_test_ts <- mean_test[][[1]]
test_ts <- as.ts(zoo(defo_test_ts, seq(2013, by = .08333333, length.out = 84)))
test <- bfastmonitor(test_ts, start = c(2018, 1))
plot(test)

# calculate breaks for every single pixel
vec <- c(0)
count <- 1
for (i in 1:5) {
for (j in 1:12) {
ts <- defo_test[,i,j,][[1]]
test_ts <- as.ts(zoo(ts, seq(2013, by = .08333333, length.out = 84)))

if (length(unique(test_ts)) > 1) {
test_bf <- bfastmonitor(test_ts, start = c(2018, 1))
vec[count] <- test_bf$breakpoint
} else {
# vec[count] <- 99
}
count <- count + 1
}
}
vec
array(c(accuracies(addmargins(table(bfast_monthly2_prev, prodes))), accuracies(addmargins(table(bfast_quarter2_prev, prodes)))), dim = c(6,2), dimnames = list(c("Overall Accuracy", "Prod. Acc. FALSE", "Prod. Acc. TRUE", "User's Acc. FALSE", "User's Acc. TRUE", "Kappa"), c("monthly", "quarterly")))
```

Results
=======
## monthly quarterly
## Overall Accuracy 91.26871 89.14326
## Prod. Acc. FALSE 0.9109448 0.8994075
## Prod. Acc. TRUE 0.925011 0.8350243
## User's Acc. FALSE 0.9884956 0.9747229
## User's Acc. TRUE 0.5948936 0.5399315
## Kappa 0.6751202 0.5948577

<img src="main_files/figure-markdown_github/results-1.png" width="33%" /><img src="main_files/figure-markdown_github/results-2.png" width="33%" /><img src="main_files/figure-markdown_github/results-3.png" width="33%" /><img src="main_files/figure-markdown_github/results-4.png" width="33%" /><img src="main_files/figure-markdown_github/results-5.png" width="33%" />

Discussion
==========
Expand Down
Binary file added bfast_monthly2.rds
View file
Binary file not shown.
Binary file added bfast_monthly3.rds
View file
Binary file not shown.
Binary file added bfast_quarter2.rds
View file
Binary file not shown.
136 changes: 83 additions & 53 deletions main.Rmd
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,9 +2,9 @@
title: "Deforestation Detection"
author: "Brian Pondi, Jonathan Bahlmann"
date: "2/12/2021"
output:
md_document:
variant: markdown_github
output:
md_document:
variant: markdown_github
---

```{r setup, include=FALSE}
Expand Down Expand Up @@ -63,8 +63,6 @@ prod <- read_sf("./yearly_deforestation/PRODES_cropped.shp")
```{r plot-AOI, warning=FALSE}
# whats in it?
# plot(st)
plot(as(st[,,,6], "Raster"))
plot(prod["YEAR"], add = TRUE)
plot(prod["YEAR"])
```

Expand Down Expand Up @@ -121,7 +119,7 @@ plot(bf1)

## Calculate BFAST Over AOI

```{r bfast-aoi, warning=FALSE}
```{r bfast-aoi, warning=FALSE, eval=FALSE}
bfast_on_tile <- function(gapfill_matrix, by, ts, order) {
dims <- dim(gapfill_matrix)
result <- matrix(rep(FALSE, dims[1]*dims[2]), ncol = dims[1])
Expand All @@ -141,65 +139,97 @@ bfast_on_tile <- function(gapfill_matrix, by, ts, order) {
return(result)
}
bfast_monthly <- bfast_on_tile(gf_monthly$fill, by = .08333333, ts = 84, order = 3)
bfast_quarter <- bfast_on_tile(gf_quarterly$fill, by = 0.25, ts = 28, order = 2)
bfast_monthly3 <- bfast_on_tile(gf_monthly$fill, by = .08333333, ts = 84, order = 3)
bfast_monthly2 <- bfast_on_tile(gf_monthly$fill, by = .08333333, ts = 84, order = 2)
bfast_quarter2 <- bfast_on_tile(gf_quarterly$fill, by = 0.25, ts = 28, order = 2)
saveRDS(bfast_monthly3, "bfast_monthly3.rds")
saveRDS(bfast_monthly2, "bfast_monthly2.rds")
saveRDS(bfast_quarter2, "bfast_quarter2.rds")
# warning: too few observations in history period
```

```{r load-bfast-tiles}
# bfast_monthly3 <- readRDS("bfast_monthly3.rds")
bfast_monthly2 <- readRDS("bfast_monthly2.rds")
bfast_quarter2 <- readRDS("bfast_quarter2.rds")
```

```{r plot-bfast-aoi, warning=FALSE}
# plot(prod["YEAR"])
# rasterize PRODES data
ras <- rasterize(prod, as(st[,,,5], "Raster"), "YEAR")
ras.m <- aperm(matrix(ras[], ncol = 140), c(2,1))
ras.m[ras.m < 2019] <- FALSE
ras.m[ras.m == 2019] <- TRUE
ras.m[is.na(ras.m)] <- FALSE
Image(bfast_monthly)
Image(bfast_quarter)
Image(ras.m)
table(bfast_monthly, ras.m)
table(bfast_quarter, ras.m)
prodes <- aperm(matrix(ras[], ncol = 140), c(2,1))
prodes[prodes < 2019] <- FALSE
prodes[prodes == 2019] <- TRUE
prodes[is.na(prodes)] <- FALSE
# to mask out previous deforestation
prodes_prev <- aperm(matrix(ras[], ncol = 140), c(2,1))
prodes_prev[prodes_prev < 2019] <- TRUE
prodes_prev[prodes_prev == 2019] <- FALSE
prodes_prev[is.na(prodes_prev)] <- FALSE
bfast_monthly2_prev <- bfast_monthly2
bfast_monthly2_prev[prodes_prev == 1] <- NA
bfast_quarter2_prev <- bfast_quarter2
bfast_quarter2_prev[prodes_prev == 1] <- NA
```

## old code
```{r, eval=FALSE}
# single pixel
defo_test <- st[st_geometry(prod[4,])]
defo_test_ts <- as.vector(defo_test[,3,3,][[1]])
test_ts <- as.ts(zoo(defo_test_ts, seq(2013, by = .08333333, length.out = 84)))
test <- bfastmonitor(test_ts, start = c(2018, 1))
plot(test)
# mean of this polygon
mean_test <- st_apply(st[st_geometry(prod[2,])], "attributes", mean, na.rm = TRUE) # aggregate
defo_test_ts <- mean_test[][[1]]
test_ts <- as.ts(zoo(defo_test_ts, seq(2013, by = .08333333, length.out = 84)))
test <- bfastmonitor(test_ts, start = c(2018, 1))
plot(test)
# calculate breaks for every single pixel
vec <- c(0)
count <- 1
for (i in 1:5) {
for (j in 1:12) {
ts <- defo_test[,i,j,][[1]]
test_ts <- as.ts(zoo(ts, seq(2013, by = .08333333, length.out = 84)))
if (length(unique(test_ts)) > 1) {
test_bf <- bfastmonitor(test_ts, start = c(2018, 1))
vec[count] <- test_bf$breakpoint
} else {
# vec[count] <- 99
}
count <- count + 1
}
```{r aoi-tables}
table1 <- addmargins(table(bfast_monthly2, prodes))
table2 <- addmargins(table(bfast_quarter2, prodes))
accuracies <- function(table1) {
# overall accuracy
P0 <- (table1[1] + table1[5]) / table1[9]
# error of omission, wrong in respect to reference data
# table1[2] / table1[3] # FALSE
# table1[4] / table1[6] # TRUE
# error of commission, wrong in respect to classified data
# table1[4] / table1[7] # FALSE
# table1[2] / table1[8] # TRUE
# producer's accuracy, Probability of classifying a pixel correctly
pa_f <- table1[1] / table1[3] # FALSE
pa_t <- table1[5] / table1[6] # TRUE
# user's accuracy, Probability of a pixel being the classified type
ua_f <- table1[1] / table1[7] # FALSE
ua_t <- table1[5] / table1[8] # TRUE
# kappa
# chance that both TRUE / FALSE randomly
tr <- (table1[8] / table1[9]) * (table1[6] / table1[9])
fr <- (table1[7] / table1[9]) * (table1[3] / table1[9])
Pe <- tr + fr
kappa <- (P0 - Pe) / (1 - Pe)
return(list("Overall Accuracy" = P0*100, "Prod. Acc. FALSE" = pa_f, "Prod. Acc. TRUE" = pa_t, "User's Acc. FALSE" = ua_f, "User's Acc. TRUE" = ua_t, "Kappa" = kappa))
}
vec
```

# Results

```{r results, fig.show="hold", out.width="33%"}
Image(bfast_monthly2) + ggtitle("Monthly Data") + theme(plot.title = element_text(size=22))
Image(bfast_quarter2) + ggtitle("Quarterly Data") + theme(plot.title = element_text(size=22))
Image(prodes) + ggtitle("PRODES Data") + theme(plot.title = element_text(size=22))
table1
table2
array(c(accuracies(table1), accuracies(table2)), dim = c(6,2), dimnames = list(c("Overall Accuracy", "Prod. Acc. FALSE", "Prod. Acc. TRUE", "User's Acc. FALSE", "User's Acc. TRUE", "Kappa"), c("monthly", "quarterly")))
# remove already deforested areas
ras[] <- bfast_monthly2
plot(ras)
plot(st_geometry(prod), add = TRUE)
ras[] <- bfast_quarter2
plot(ras)
plot(st_geometry(prod), add = TRUE)
array(c(accuracies(addmargins(table(bfast_monthly2_prev, prodes))), accuracies(addmargins(table(bfast_quarter2_prev, prodes)))), dim = c(6,2), dimnames = list(c("Overall Accuracy", "Prod. Acc. FALSE", "Prod. Acc. TRUE", "User's Acc. FALSE", "User's Acc. TRUE", "Kappa"), c("monthly", "quarterly")))
```

# Discussion

# Conclusion
Expand Down
Loading

0 comments on commit c2c27c6

Please sign in to comment.