Hue (HSV) based snow flagging

[khufkens]

A Hue based filter performs well in finding images with snow and cloud contamination, even on average daily values across an ROI (no image processing needed).

# HSV based snow flagging
library(phenocamr)
library(tidyverse)

if(!file.exists(file.path(tempdir(),"chibougamau_EN_1000_1day.csv"))){
  download_phenocam(site = "chibougamau",
                    out_dir = tempdir(),
                    frequency = 1,
                    roi_id = 1000)
}

# read data
df <- read.table(file.path(tempdir(),"chibougamau_EN_1000_1day.csv"),
                 sep = ",",
                 header = TRUE,
                 stringsAsFactors = FALSE)

df <- df %>% filter(!is.na(r_mean))

hsv <- rgb2hsv(df$r_mean, df$g_mean, df$b_mean)

df$snow_flag <- ifelse(hsv[1,] > 0.55 & hsv[1,] < 0.97,"snow","no snow")

p <- ggplot(df) +
  geom_point(aes(as.Date(date), gcc_90, col = as.factor(snow_flag))) +
  theme_minimal()

print(p)

[khufkens]

A similar test for bartlettir renders no false positives, although the background has snow the trunks limit blue tints and don't trigger flagging. The reverse is true with evergreens which covered with snow turn blue-ish and trigger the snow flagging thresholds on the hue values.

# HSV based snow flagging
library(phenocamr)
library(tidyverse)

if(!file.exists(file.path(tempdir(),"bartlettir_DB_1000_1day.csv"))){
  download_phenocam(site = "bartlettir",
                    out_dir = tempdir(),
                    frequency = 1,
                    roi_id = 1000)
}

# read data
df <- read.table(file.path(tempdir(),"bartlettir_DB_1000_1day.csv"),
                 sep = ",",
                 header = TRUE,
                 stringsAsFactors = FALSE)

df <- df %>% filter(!is.na(r_mean))

hsv <- rgb2hsv(df$r_mean, df$g_mean, df$b_mean)

df$snow_flag <- ifelse(hsv[1,] > 0.55 & hsv[1,] < 0.97,"snow","no snow")

p <- ggplot(df) +
  geom_point(aes(as.Date(date), gcc_90, col = as.factor(snow_flag))) +
  theme_minimal()

print(p)

[khufkens]

The same goes for detecting snow on ground, if nothing else is in the way. So for agricultural fields it allows you to detect snow on the ground (since the signal isn't contaminated by stems).

# HSV based snow flagging
library(phenocamr)
library(tidyverse)

if(!file.exists(file.path(tempdir(),"arsmorris1_AG_1000_1day.csv"))){
  download_phenocam(site = "arsmorris1",
                    out_dir = tempdir(),
                    frequency = 1,
                    roi_id = 1000)
}

# read data
df <- read.table(file.path(tempdir(),"arsmorris1_AG_1000_1day.csv"),
                 sep = ",",
                 header = TRUE,
                 stringsAsFactors = FALSE)

df <- df %>% filter(!is.na(r_mean))

hsv <- rgb2hsv(df$r_mean, df$g_mean, df$b_mean)

df$snow_flag <- ifelse(hsv[1,] > 0.55 & hsv[1,] < 0.97,"snow","no snow")

p <- ggplot(df) +
  geom_point(aes(as.Date(date), gcc_90, col = as.factor(snow_flag))) +
  theme_minimal()

print(p)