In dulvrq/ddl8s1ts: Map forest disturbance using Landsat 8 and Sentinel-1

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

ddl8s1ts

ddl8s1ts is an R package for detecting and mapping forest disturbance using Landsat 8 and Sentinel-1 time seires. Harmonic regression models are applied to both Landsat 8 and Sentinel-1 time series to predict disturbance probability. Disturbance is detected based on the time series of disturbance probabilities predicted by Random Forests (RF).

• The functions are tested only on Windows OS.
• The process takes very long time to map large area and uses much memory and CPUs in parallel computing.

Installation

You can install the latest version from GitHub using devtools:

devtools::install_github("dulvrq/ddl8s1ts")
library(ddl8s1ts)

Example

Currently, mapDisturbanceL8S1() is an only usable function in this package. All other functions are called by this function. Several data are required for running mapDisturbanceL8S1() to detect and map disturbance. The followings are key variables that should be prepared beforehand.

suppressWarnings(library(raster, quietly = T))
library(ddl8s1ts)
dir_main <- dirname(dirname(getwd()))
# ls_l8 ls_s1
dir_l8 <- paste0(dir_main, "/GIS_data/_for_share/Landsat_SR")
list_l8  <- list.files(dir_l8, "tif$", full.names = T)
list_l8  <- gsub("OutPut_", "", list_l8)
dir_s1 <- paste0(dir_main, "/GIS_data/_for_share/Sentinel-1_VV_VH")
list_s1  <- list.files(dir_s1, "tif$", full.names = T)
list_s1  <- gsub("OutPut_", "", list_s1)

Landsat 8 data

ls_l8: requires a file list of 6 bands images of Landsat 8 (B2, B3, B4, B5, B6, and B7). All images must have the same extent (origin, resolution, and crs). These should be surface reflectance products, may be from Google Earth Engine (GEE) or other preprocessing chains. NoData values should be NA, not 0 (sometimes occurred in GEE output). An exmaple of this product is the following.

# an example data in this package
head(ls_l8, 1)

When you use your own data, it is also acceptable to use a filename provided by list.files as follows.

# use list.files to list files in your directory
head(basename(list_l8))

Sentinel-1 data

ls_s1: requires a file list of 2 bands images of Sentinel-1 (VV and VH). All images must have the same extent (origin, resolution, and crs). These might be from GEE or other preprocessing chains. An exmaple of this product is the following.

# an example data in this package
head(ls_s1, 1)

Of course you can use a filename provided by list.files in the same manner.

# use list.files to list files in your directory
head(basename(list_s1))

Julian day of Landsat 8 and Sentinel-1 data

l8_doys: is a list of julian day of Landsat 8. This should be the same order and length as ls_l8.
s1_doys: is a list of julian day of Sentinel-1. This should be the same order and length as ls_s1.

# calculate julian day from filenames
fname_l8 <- sapply(ls_l8, function(x) names(x)[1])
fname_s1 <- sapply(ls_s1, function(x) names(x)[1])
n <- 18
l8_dates <- paste(substring(fname_l8, n,   n+3),
                  substring(fname_l8, n+4, n+5),
                  substring(fname_l8, n+6, n+7), sep = "-")
l8_doys <- as.numeric(substring(fname_l8, n, n+3)) + as.numeric(strftime(l8_dates, format = "%j")) / 365
s1_dates <- paste(substring(fname_s1, n,   n+3),
                  substring(fname_s1, n+4, n+5),
                  substring(fname_s1, n+6, n+7), sep = "-")
s1_doys <- as.numeric(substring(fname_s1, n, n+3)) + as.numeric(strftime(s1_dates, format = "%j")) / 365

head(l8_doys)
head(s1_doys)

Reference data for disturbance detection

dt_ref: is a dataframe of reference data for RF modeling. This should contain column x, y, and date. x and y should indicate locations of reference samples in the same crs as Landsat 8 or Sentinel-1. date should indicate the timing of disturbance. If there is no disturbance at the sample location, use NA.

# an example data in this package
head(dt_ref)

Others

ls_dem: a filename of DEM with 2 bands (elevation and slope). Optional. If DEM will be not used for model, use NULL.
dir_save: a directory for save the results. NULL is acceptable.
VI : a list of spectral index used for Landsat 8.
rf_model: use NULL to build RF model. To skip RF modeling, provide a list of two RF models.
startDOY: a start date of disturbance detection.
endDOY: a end date of disturbance detection.
mmu: a minimum mapping unit (pixel) for final mapping. Use NULL to avoid mmu.
only_rf: If TRUE, only build RF models. If FALSE, build RF models and map disturbance detection.
max_cores: a maximum cores used for parallel processing.
threshold: a threshold to detect disturbance from time series disturbance probabilities.

ls_dem <- NULL # do not use DEM in RF models
dir_save <- getwd()
startDOY <- 2016 # detect disturbance from 2016/01/01
endDOY   <- 2018 # detect disturbance until 2017/12/31
mmu      <- 4
VI <- c("NBR", "TCA8", "TCB8", "TCG8","TCW8") # NBR, TCA, TCB, TCG, and TCW
max_cores <- 20
threshold <- 0.5

example 1. Full implementation of building RF models and mapping disturbance

mapDisturbanceL8S1() has 2 processing parts: RF model building and disturbance mapping. A simple implementation is to process both as follows. Please note that this function uses parallel processing and consume much memory and CPUs. Furthermore, this process takes very long time.

# use only_rf = F & rf_model = NULL to run entire process.
mapDisturbanceL8S1(ls_l8, ls_s1, l8_doys, s1_doys, dt_ref, ls_dem, dir_save, VI,
                   rf_model = NULL, startDOY, endDOY, mmu, only_rf = F, max_cores, threshold)

example 2. Full implementation only with Landsat 8

If you would like to use either of Landsat 8 or Sentinel-1, use NULL to either of file list.

# use ls_s1 = NULL.
mapDisturbanceL8S1(ls_l8, NULL, l8_doys, NULL, dt_ref, ls_dem, dir_save, VI,
                   rf_model = NULL, startDOY, endDOY, mmu, only_rf = F, max_cores, threshold)

example 3. Implementation of RF modeling

If you would like to run only RF model building, such as in case that samples are from large extent but would like to map only small extent, then use only_rf = TRUE. This only generate RF models in a subfolder.

# use only_rf = T.
mapDisturbanceL8S1(ls_l8, ls_s1, l8_doys, s1_doys, dt_ref, ls_dem, dir_save, VI,
                   rf_model = NULL, startDOY, endDOY, mmu, only_rf = T, max_cores, threshold)

example 4. Implementation of disturbance mapping

If you would like to run only disturbance mapping, such as in case of mapping only small extent, then provide a list of two RF models (Landsat and Sentinel-1).

# provide rf_model, a list of two RF models.
mapDisturbanceL8S1(ls_l8, ls_s1, l8_doys, s1_doys, dt_ref, ls_dem, dir_save, VI,
                   rf_model = rf_model, startDOY, endDOY, mmu, only_rf = F, max_cores, threshold)

Other note:

• It is a good idea to mask forest area before running this function.
• A subfolder is automatically generated for saving RF results and temporal tile maps.
• 50% of reference samples are used for RF modeling, and the rest are used for evaluation.

Reference

Shimizu, K. Ota, T. Mizoue, N. (2019) Detecting Forest Changes Using Dense Landsat 8 and Sentinel-1 Time Series Data in Tropical Seasonal Forests. Remote Sensing 11: 1899. doi.org/10.3390/rs11161899

dulvrq/ddl8s1ts documentation built on May 2, 2021, 5:18 a.m.