Mercurial > repos > ecology > srs_diversity_maps
diff prosail-master/R/Lib_SpectralIndices.R @ 0:9adccd3da70c draft default tip
planemo upload for repository https://github.com/Marie59/Sentinel_2A/srs_tools commit b32737c1642aa02cc672534e42c5cb4abe0cd3e7
author | ecology |
---|---|
date | Mon, 09 Jan 2023 13:37:37 +0000 |
parents | |
children |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/prosail-master/R/Lib_SpectralIndices.R Mon Jan 09 13:37:37 2023 +0000 @@ -0,0 +1,679 @@ +# ============================================================================== = +# prosail +# Lib_spectralindices.R +# ============================================================================== = +# PROGRAMMERS: +# Jean-Baptiste FERET <jb.feret@teledetection.fr> +# Florian de BOISSIEU <fdeboiss@gmail.com> +# Copyright 2019/11 Jean-Baptiste FERET +# ============================================================================== = +# This Library includes aims at computing spectral indices from reflectance data +# ============================================================================== = + +#" This function computes Area under curve for continuum removed reflectances +#" See Malenovský et al. (2013) for details +#" http://dx.doi.org/10.1016/j.rse.2012.12.015 +#" +#" @param refl RasterBrick, RasterStack or list. Raster bands in the order of sensorbands. +#" @param sensorbands numeric. vector containing central wavelength for each spectral band in the image +#" @param aucminmax list. wavelengths of lower and upper boundaries ("CRmin" and "CRmax") +#" @param reflfactor numeric. multiplying factor used to write reflectance in image (==10000 for S2) +#" +#" @return aucval raster +#" @export +auc <- function(refl, sensorbands, aucminmax, reflfactor = 1) { + + aucbands <- list() + aucbands[["CRmin"]] <- sensorbands[get_closest_bands(sensorbands, aucminmax[["CRmin"]])] + aucbands[["CRmax"]] <- sensorbands[get_closest_bands(sensorbands, aucminmax[["CRmax"]])] + bands <- get_closest_bands(sensorbands, aucbands) + for (i in bands[["CRmin"]]:bands[["CRmax"]]) { + if (is.na(match(i, bands))) { + aucbands[[paste("B", i, sep = "")]] <- sensorbands[i] + } + } + # compute continuum removal for all spectral bands + cr <- cr_wl(refl = refl, sensorbands = sensorbands, + crbands = aucbands, reflfactor = reflfactor) + + wl <- sort(unlist(aucbands), decreasing = FALSE) + aucval <- 0.5 * (1 - cr[[1]]) * (wl[2] - wl[1]) + for (i in 2:length(cr)) { + aucval <- aucval + 0.5 * (2 - cr[[i - 1]] - cr[[i]]) * (wl[i + 1] - wl[i]) + } + aucval <- aucval + 0.5 * (1 - cr[[length(cr)]]) * (wl[i + 2] - wl[i + 1]) + return(aucval) +} + +#" This function extracts boundaries to be used to compute continuum from reflectance data +#" +#" @param refl RasterBrick, RasterStack or list. Raster bands in the order of sensorbands. +#" @param sensorbands numeric. vector containing central wavelength for each spectral band in the image +#" @param crbands list. list of spectral bands (central wavelength) including CRmin and CRmax +#" @param reflfactor numeric. multiplying factor used to write reflectance in image ( == 10000 for S2) +#" +#" @return crminmax list. list of rasters corresponding to minimum and maximum wavelengths +#" @export +crbound <- function(refl, sensorbands, crbands, reflfactor = 1) { + + # get closest spectral bands from CR1 and CR2 + bands <- get_closest_bands(sensorbands, list(crbands[["CRmin"]], crbands[["CRmax"]])) + wl <- sensorbands[bands] + # get equation for line going from CR1 to CR2 + crminmax <- readrasterbands(refl = refl, bands = bands, reflfactor = reflfactor) + names(crminmax) <- paste("wl_", as.character(wl), sep = "") + return(crminmax) +} + +#" This function extracts boundaries to be used to compute continuum from reflectance data +#" +#" @param refl RasterBrick, RasterStack or list. Raster bands in the order of sensorbands. +#" @param sensorbands numeric. vector containing central wavelength for each spectral band in the image +#" @param crbands list. list of spectral bands (central wavelength) including CRmin and CRmax +#" @param reflfactor numeric. multiplying factor used to write reflectance in image ( == 10000 for S2) +#" +#" @return outlier_iqr numeric. band numbers of original sensor corresponding to S2 +#" @importFrom progress progress_bar +#" @export +cr_wl <- function(refl, sensorbands, crbands, reflfactor = 1) { + + # Make sure CRmin and CRmax are correctly defined + if (is.na(match("CRmin", names(crbands))) || is.na(match("CRmax", names(crbands)))) { + stop("Please define CRmin and CRmax (CRmin<CRmax) as spectral bands in crbands") + } + if (crbands[["CRmax"]] < crbands[["CRmin"]]) { + stop("Please define CRmin < CRmax in crbands") + } + # extract CRmin and CRmax + crminmax <- crbound(refl, sensorbands, crbands, reflfactor = reflfactor) + # extract other bands and compute CR + crmin <- sensorbands[get_closest_bands(sensorbands, crbands[["CRmin"]])] + crmax <- sensorbands[get_closest_bands(sensorbands, crbands[["CRmax"]])] + crbands[["CRmin"]] <- NULL + crbands[["CRmax"]] <- NULL + cr <- list() + # initiate progress bar + pgbarlength <- length(crbands) + pb <- progress_bar$new( + format = "Computing continuum removal [:bar] :percent in :elapsedfull, estimated time remaining :eta", + total = pgbarlength, clear = FALSE, width = 100) + # computation for each band + for (band in crbands) { + pb$tick() + bandrank <- get_closest_bands(sensorbands, band) + raster2cr <- readrasterbands(refl = refl, bands = bandrank, reflfactor = reflfactor) + cr[[as.character(band)]] <- computecr(wlmin = crmin, wlmax = crmax, + wltarget = band, boundaries = crminmax, + target = raster2cr) + } + return(cr) +} + +#" This function computes continuum removal value for a spectral band of interest, +#" based on lower and upper wavelengths corresponding to boundaries of the continuum +#" +#" @param wlmin numeric. wavelength of the spectral band corresponding to minimum boundary +#" @param wlmax numeric. wavelength of the spectral band corresponding to maximum boundary +#" @param wltarget numeric. wavelength of the spectral band for which cr is computed +#" @param boundaries list. raster objects corresponding to minimum and maximum wavelengths +#" @param target list. raster object corresponding target wavelength +#" +#" @return cr list. raster object corresponding to continuum removed value +#" @export +computecr <- function(wlmin, wlmax, wltarget, boundaries, target) { + + cr <- target / (boundaries[[1]] + (wltarget - wlmin) * (boundaries[[2]] - boundaries[[1]]) / (wlmax - wlmin)) + return(cr) +} + +#" this function produces a spectral index from an expression defining a spectral index +#" +#" @param refl RasterBrick, RasterStack or list. Raster bands in the order of sensorbands. +#" @param sensorbands numeric. wavelength in nanometers of the spectral bands of refl. +#" @param expressindex character. expression corresponding to the spectral index to compute +#" @param listbands list. list of spectral bands defined in the "expressindex" variable +#" @param reflfactor numeric. multiplying factor used to write reflectance in image ( == 10000 for S2) +#" @param nameindex character. name for the index to be computed, provided in the raster layer +#" +#" @return numeric. band numbers of original sensor corresponding to S2 +#" @export +spectralindices_fromexpression <- function(refl, sensorbands, expressindex, listbands, reflfactor = 1, nameindex = NULL) { + + # define which bands to be used in the spectral index + bands <- get_closest_bands(sensorbands, listbands) + + classraster <- class(refl) + if (classraster == "RasterBrick" || classraster == "RasterStack" || classraster == "stars") { + # if !reflfactor == 1 then apply a reflectance factor + if (classraster == "stars") { + refl <- refl[bands] + } else { + refl <- raster::subset(refl, bands) + } + if (!reflfactor == 1) { + refl <- refl / reflfactor + } + } else if (is.list(refl)) { + refl <- raster::stack(refl[bands]) # checks that all rasters have same crs/extent + if (!reflfactor == 1) { + refl <- refl / reflfactor + } + } else { + stop("refl is expected to be a RasterStack, RasterBrick, Stars object or a list of rasters") + } + names(refl) <- gsub(pattern = "B", replacement = "Band", x = names(bands)) + + nbbands <- unique(as.numeric(gsub(pattern = "B", + replacement = "", + x = unlist(regmatches(expressindex, + gregexpr("B[[:digit:]] + ", + expressindex)))))) + sortband <- sort(nbbands, index.return = TRUE, decreasing = TRUE) + matches <- unique(unlist(regmatches(expressindex, gregexpr("B[[:digit:]] + ", expressindex))))[sortband$ix] + replaces <- paste("refl[['Band", gsub(pattern = "B", replacement = "", x = matches), "']]", sep = "") + expressindex_final <- expressindex + for (bb in 1:seq_along(matches)) { + expressindex_final <- gsub(pattern = matches[bb], replacement = replaces[bb], x = expressindex_final) + } + si <- eval(parse(text = expressindex_final)) + if (!is.null(nameindex)) { + names(si) <- nameindex + } + return(si) +} + +#" this function aims at computing spectral indices from Sensor reflectance data in raster object +#" it computes the spectral indices based on their computation with Sentinel-2 +#" and assumes that the bands of the S2 data follow this order +#" wavelength = {496.6, 560.0, 664.5, 703.9, 740.2, 782.5, 835.1, 864.8, 1613.7, 2202.4} +#" Full description of the indices can be found here: +#" https://www.sentinel-hub.com/eotaxonomy/indices +#" +#" @param refl RasterBrick, RasterStack or list. Raster bands in the order of sensorbands. +#" @param sensorbands numeric. wavelength in nanometers of the spectral bands of refl. +#" @param sel_indices list. list of spectral indices to be computed +#" @param stackout logical. If TRUE returns a stack, otherwise a list of rasters. +#" @param reflfactor numeric. multiplying factor used to write reflectance in image ( == 10000 for S2) +#" +#" @return list. includes +#" - spectralindices: List of spectral indices computed from the reflectance initially provided +#" - listindices: list of spectral indices computable with the function +#" @importFrom methods is +#" @importFrom raster stack brick +#" @export + +computespectralindices_raster <- function(refl, sensorbands, sel_indices = "ALL", stackout = TRUE, reflfactor = 1) { + + s2bands <- c("B2" = 496.6, "B3" = 560.0, "B4" = 664.5, "B5" = 703.9, "B6" = 740.2, + "B7" = 782.5, "B8" = 835.1, "B8A" = 864.8, "B11" = 1613.7, "B12" = 2202.4) + + spectralindices <- list() + sen2s2 <- get_closest_bands(sensorbands, s2bands) + classraster <- class(refl) + if (classraster == "RasterBrick" || classraster == "RasterStack" || classraster == "stars") { + # if !reflfactor == 1 then apply a reflectance factor + if (classraster == "stars") { + refl <- refl[sen2s2] + } else { + refl <- raster::subset(refl, sen2s2) + } + if (!reflfactor == 1) { + refl <- refl / reflfactor + } + } else if (is.list(refl)) { + refl <- raster::stack(refl[sen2s2]) # checks that all rasters have same crs/extent + if (!reflfactor == 1) { + refl <- refl / reflfactor + } + } else { + stop("refl is expected to be a RasterStack, RasterBrick, Stars object or a list of rasters") + } + names(refl) <- names(sen2s2) + + indexall <- list() + + # inelegant but meeeeh + listindices <- list("ARI1", "ARI2", "ARVI", "BAI", "BAIS2", "CCCI", "CHL_RE", "CRI1", "CRI2", "EVI", "EVI2", + "GRVI1", "GNDVI", "IRECI", "LAI_SAVI", "MCARI", "mNDVI705", "MSAVI2", + "MSI", "mSR705", "MTCI", "nBR_RAW", "NDI_45", "NDII", "NDSI", "NDVI", "NDVI_G", + "NDVI705", "NDWI1", "NDWI2", "PSRI", "PSRI_NIR", "RE_NDVI", "RE_NDWI", "S2REP", + "SAVI", "SIPI", "SR", "CR_SWIR") + if (sel_indices[1] == "ALL") { + sel_indices <- listindices + } + if ("ARI1" %in% sel_indices) { + ari1 <- (1 / refl[["B3"]]) - (1 / refl[["B5"]]) + spectralindices$ARI1 <- ari1 + } + if ("ARI2" %in% sel_indices) { + ari2 <- (refl[["B8"]] / refl[["B2"]]) - (refl[["B8"]] / refl[["B3"]]) + spectralindices$ARI2 <- ari2 + } + if ("ARVI" %in% sel_indices) { + arvi <- (refl[["B8"]] - (2 * refl[["B4"]] - refl[["B2"]])) / (refl[["B8"]] + (2 * refl[["B4"]] - refl[["B2"]])) + spectralindices$ARVI <- arvi + } + if ("BAI" %in% sel_indices) { + bai <- (1 / ((0.1 - refl[["B4"]])**2 + (0.06 - refl[["B8"]])**2)) + spectralindices$BAI <- bai + } + if ("BAIS2" %in% sel_indices) { + bais2 <- (1 - ((refl[["B6"]] * refl[["B7"]] * refl[["B8A"]]) / refl[["B4"]])**0.5) * ((refl[["B12"]] - refl[["B8A"]]) / ((refl[["B12"]] + refl[["B8A"]])**0.5) + 1) + spectralindices$BAIS2 <- bais2 + } + if ("CCCI" %in% sel_indices) { + ccci <- ((refl[["B8"]] - refl[["B5"]]) / (refl[["B8"]] + refl[["B5"]])) / ((refl[["B8"]] - refl[["B4"]]) / (refl[["B8"]] + refl[["B4"]])) + spectralindices$CCCI <- ccci + } + if ("CHL_RE" %in% sel_indices) { + chl_re <- refl[["B5"]] / refl[["B8"]] + spectralindices$CHL_RE <- chl_re + } + if ("CRI1" %in% sel_indices) { + cri1 <- (1 / refl[["B2"]]) - (1 / refl[["B3"]]) + spectralindices$CRI1 <- cri1 + } + if ("CRI2" %in% sel_indices) { + cri2 <- (1 / refl[["B2"]]) - (1 / refl[["B5"]]) + spectralindices$CRI2 <- cri2 + } + if ("EVI" %in% sel_indices) { + evi <- 2.5 * (refl[["B8"]] - refl[["B4"]]) / ((refl[["B8"]] + 6 * refl[["B4"]] - 7.5 * refl[["B2"]] + 1)) + spectralindices$EVI <- evi + } + if ("EVI2" %in% sel_indices) { + evi2 <- 2.5 * (refl[["B8"]] - refl[["B4"]]) / (refl[["B8"]] + 2.4 * refl[["B4"]] + 1) + spectralindices$EVI2 <- evi2 + } + if ("GRVI1" %in% sel_indices) { + grvi1 <- (refl[["B4"]] - refl[["B3"]]) / (refl[["B4"]] + refl[["B3"]]) + spectralindices$GRVI1 <- grvi1 + } + if ("GNDVI" %in% sel_indices) { + gndvi <- (refl[["B8"]] - refl[["B3"]]) / (refl[["B8"]] + refl[["B3"]]) + spectralindices$GNDVI <- gndvi + } + if ("IRECI" %in% sel_indices) { + ireci <- (refl[["B7"]] - refl[["B4"]]) * (refl[["B6"]] / (refl[["B5"]])) + spectralindices$IRECI <- ireci + } + if ("LAI_SAVI" %in% sel_indices) { + lai_savi <- - log(0.371 + 1.5 * (refl[["B8"]] - refl[["B4"]]) / (refl[["B8"]] + refl[["B4"]] + 0.5)) / 2.4 + spectralindices$LAI_SAVI <- lai_savi + } + if ("MCARI" %in% sel_indices) { + mcari <- (1 - 0.2 * (refl[["B5"]] - refl[["B3"]]) / (refl[["B5"]] - refl[["B4"]])) + spectralindices$MCARI <- mcari + } + if ("mNDVI705" %in% sel_indices) { + mndvi705 <- (refl[["B6"]] - refl[["B5"]]) / (refl[["B6"]] + refl[["B5"]] - 2 * refl[["B2"]]) + spectralindices$mNDVI705 <- mndvi705 + } + if ("MSAVI2" %in% sel_indices) { + msavi2 <- ((refl[["B8"]] + 1) - 0.5 * sqrt(((2 * refl[["B8"]]) - 1)**2 + 8 * refl[["B4"]])) + spectralindices$MSAVI2 <- msavi2 + } + if ("MSI" %in% sel_indices) { + msi <- refl[["B11"]] / refl[["B8A"]] + spectralindices$MSI <- msi + } + if ("mSR705" %in% sel_indices) { + msr705 <- (refl[["B6"]] - refl[["B2"]]) / (refl[["B5"]] - refl[["B2"]]) + spectralindices$mSR705 <- msr705 + } + if ("MTCI" %in% sel_indices) { + mtci <- (refl[["B6"]] - refl[["B5"]]) / (refl[["B5"]] + refl[["B4"]]) + spectralindices$MTCI <- mtci + } + if ("nBR_RAW" %in% sel_indices) { + nbr_raw <- (refl[["B8"]] - refl[["B12"]]) / (refl[["B8"]] + refl[["B12"]]) + spectralindices$nBR_RAW <- nbr_raw + } + if ("NDI_45" %in% sel_indices) { + ndi_45 <- (refl[["B5"]] - refl[["B4"]]) / (refl[["B5"]] + refl[["B4"]]) + spectralindices$NDI_45 <- ndi_45 + } + if ("NDII" %in% sel_indices) { + ndii <- (refl[["B8A"]] - refl[["B11"]]) / (refl[["B8A"]] + refl[["B11"]]) + spectralindices$NDII <- ndii + } + if ("NDSI" %in% sel_indices) { + ndsi <- (refl[["B3"]] - refl[["B11"]]) / (refl[["B3"]] + refl[["B11"]]) + spectralindices$NDSI <- ndsi + } + if ("NDVI" %in% sel_indices) { + ndvi <- (refl[["B8"]] - refl[["B4"]]) / (refl[["B8"]] + refl[["B4"]]) + spectralindices$NDVI <- ndvi + } + if ("NDVI_G" %in% sel_indices) { + ndvi_g <- refl[["B3"]] * (refl[["B8"]] - refl[["B4"]]) / (refl[["B8"]] + refl[["B4"]]) + spectralindices$NDVI_G <- ndvi_g + } + if ("NDVI705" %in% sel_indices) { + ndvi705 <- (refl[["B6"]] - refl[["B5"]]) / (refl[["B6"]] + refl[["B5"]]) + spectralindices$NDVI705 <- ndvi705 + } + if ("NDWI" %in% sel_indices) { + ndwi <- (refl[["B3"]] - refl[["B8"]]) / (refl[["B3"]] + refl[["B8"]]) + spectralindices$NDWI <- ndwi + } + if ("NDWI1" %in% sel_indices) { + ndwi1 <- (refl[["B8A"]] - refl[["B11"]]) / (refl[["B8A"]] + refl[["B11"]]) + spectralindices$NDWI1 <- ndwi1 + } + if ("NDWI2" %in% sel_indices) { + ndwi2 <- (refl[["B8A"]] - refl[["B12"]]) / (refl[["B8A"]] + refl[["B12"]]) + spectralindices$NDWI2 <- ndwi2 + } + if ("PSRI" %in% sel_indices) { + psri <- (refl[["B4"]] - refl[["B2"]]) / (refl[["B5"]]) + spectralindices$PSRI <- psri + } + if ("PSRI_NIR" %in% sel_indices) { + psri_nir <- (refl[["B4"]] - refl[["B2"]]) / (refl[["B8"]]) + spectralindices$PSRI_NIR <- psri_nir + } + if ("RE_NDVI" %in% sel_indices) { + re_ndvi <- (refl[["B8"]] - refl[["B6"]]) / (refl[["B8"]] + refl[["B6"]]) + spectralindices$RE_NDVI <- re_ndvi + } + if ("RE_NDWI" %in% sel_indices) { + re_ndwi <- (refl[["B4"]] - refl[["B6"]]) / (refl[["B4"]] + refl[["B6"]]) + spectralindices$RE_NDWI <- re_ndwi + } + if ("S2REP" %in% sel_indices) { + s2rep <- 705 + 35 * (0.5 * (refl[["B8"]] + refl[["B5"]]) - refl[["B6"]]) / (refl[["B7"]] - refl[["B6"]]) + spectralindices$S2REP <- s2rep + } + if ("SAVI" %in% sel_indices) { + savi <- 1.5 * (refl[["B8"]] - refl[["B5"]]) / (refl[["B8"]] + refl[["B5"]] + 0.5) + spectralindices$SAVI <- savi + } + if ("SIPI" %in% sel_indices) { + sipi <- (refl[["B8"]] - refl[["B2"]]) / (refl[["B8"]] - refl[["B4"]]) + spectralindices$SIPI <- sipi + } + if ("SR" %in% sel_indices) { + sr <- refl[["B8"]] / refl[["B4"]] + spectralindices$SR <- sr + } + if ("TCARI" %in% sel_indices) { + sr <- refl[["B8"]] / refl[["B4"]] + spectralindices$SR <- sr + } + if ("CR_SWIR" %in% sel_indices) { + cr_swir <- refl[["B11"]] / (refl[["B8A"]] + (s2bands["B11"] - s2bands["B8A"]) * (refl[["B12"]] - refl[["B8A"]]) / (s2bands["B12"] - s2bands["B8A"])) + spectralindices$CR_SWIR <- cr_swir + } + + if (stackout) + spectralindices <- raster::stack(spectralindices) + + res <- list("spectralindices" = spectralindices, "listindices" = listindices) + return(res) +} + +#" this function aims at computing spectral indices from Sensor reflectance data. +#" it computes the spectral indices based on their computation with Sentinel-2 +#" and assumes that the bands of the S2 data follow this order +#" wavelength = {496.6, 560.0, 664.5, 703.9, 740.2, 782.5, 835.1, 864.8, 1613.7, 2202.4} +#" Full description of the indices can be found here: +#" https://www.sentinel-hub.com/eotaxonomy/indices +#" +#" @param refl numeric. reflectance dataset defined in matrix +#" @param sel_indices list. list of spectral indices to be computed +#" @param sensorbands numeric. wavelength of the spectral bands corresponding to the spectral index +#" +#" @return list. includes +#" - spectralindices: List of spectral indices computed from the reflectance initially provided +#" - listindices: list of spectral indices computable with the function +#" @export + +computespectralindices_hs <- function(refl, sensorbands, sel_indices = "ALL") { + + spectralindices <- list() + s2bands <- data.frame("B2" = 496.6, "B3" = 560.0, "B4" = 664.5, "B5" = 703.9, "B6" = 740.2, + "B7" = 782.5, "B8" = 835.1, "B8A" = 864.8, "B11" = 1613.7, "B12" = 2202.4) + + sen2s2 <- get_closest_bands(sensorbands, s2bands) + indexall <- list() + # set zero vaues to >0 in order to avoid problems + selzero <- which(refl == 0) + refl[selzero] <- 0.005 + if (dim(refl)[1] == length(sensorbands)) { + refl <- t(refl) + } + + # inelegant but meeeeh + listindices <- list("ARI1", "ARI2", "ARVI", "BAI", "BAIS2", "CHL_RE", "CRI1", "CRI2", "EVI", "EVI2", + "GRVI1", "GNDVI", "IRECI", "LAI_SAVI", "MCARI", "mNDVI705", "MSAVI2", + "MSI", "mSR705", "MTCI", "nBR_RAW", "NDI_45", "NDII", "NDVI", "NDVI_G", + "NDVI705", "NDWI1", "NDWI2", "PSRI", "PSRI_NIR", "RE_NDVI", "RE_NDWI", "S2REP", + "SAVI", "SIPI", "SR", "CR_SWIR") + if (sel_indices == "ALL") { + sel_indices <- listindices + } + if ("ARI1" %in% sel_indices) { + ari1 <- (1 / refl[, sen2s2[["B3"]]]) - (1 / refl[, sen2s2[["B5"]]]) + spectralindices$ARI1 <- ari1 + } + if ("ARI2" %in% sel_indices) { + ari2 <- (refl[, sen2s2[["B8"]]] / refl[, sen2s2[["B2"]]]) - (refl[, sen2s2[["B8"]]] / refl[, sen2s2[["B3"]]]) + spectralindices$ARI2 <- ari2 + } + if ("ARVI" %in% sel_indices) { + arvi <- (refl[, sen2s2[["B8"]]] - (2 * refl[, sen2s2[["B4"]]] - refl[, sen2s2[["B2"]]])) / (refl[, sen2s2[["B8"]]] + (2 * refl[, sen2s2[["B4"]]] - refl[, sen2s2[["B2"]]])) + spectralindices$ARVI <- arvi + } + if ("BAI" %in% sel_indices) { + bai <- (1 / ((0.1 - refl[, sen2s2[["B4"]]])**2 + (0.06 - refl[, sen2s2[["B8"]]])**2)) + spectralindices$BAI <- bai + } + if ("BAIS2" %in% sel_indices) { + bais2 <- (1 - ((refl[, sen2s2[["B6"]]] * refl[, sen2s2[["B7"]]] * refl[, sen2s2[["B8A"]]]) / refl[, sen2s2[["B4"]]])**0.5) * ((refl[, sen2s2[["B12"]]] - refl[, sen2s2[["B8A"]]]) / ((refl[, sen2s2[["B12"]]] + refl[, sen2s2[["B8A"]]])**0.5) + 1) + spectralindices$BAIS2 <- bais2 + } + if ("CCCI" %in% sel_indices) { + ccci <- ((refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B5"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B5"]]])) / ((refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B4"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B4"]]])) + spectralindices$CCCI <- ccci + } + if ("CHL_RE" %in% sel_indices) { + chl_re <- refl[, sen2s2[["B5"]]] / refl[, sen2s2[["B8"]]] + spectralindices$CHL_RE <- chl_re + } + if ("CRI1" %in% sel_indices) { + cri1 <- (1 / refl[, sen2s2[["B2"]]]) - (1 / refl[, sen2s2[["B3"]]]) + spectralindices$CRI1 <- cri1 + } + if ("CRI2" %in% sel_indices) { + cri2 <- (1 / refl[, sen2s2[["B2"]]]) - (1 / refl[, sen2s2[["B5"]]]) + spectralindices$CRI2 <- cri2 + } + if ("EVI" %in% sel_indices) { + evi <- 2.5 * (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B4"]]]) / ((refl[, sen2s2[["B8"]]] + 6 * refl[, sen2s2[["B4"]]] - 7.5 * refl[, sen2s2[["B2"]]] + 1)) + spectralindices$EVI <- evi + } + if ("EVI2" %in% sel_indices) { + evi2 <- 2.5 * (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B4"]]]) / (refl[, sen2s2[["B8"]]] + 2.4 * refl[, sen2s2[["B4"]]] + 1) + spectralindices$EVI2 <- evi2 + } + if ("GRVI1" %in% sel_indices) { + grvi1 <- (refl[, sen2s2[["B4"]]] - refl[, sen2s2[["B3"]]]) / (refl[, sen2s2[["B4"]]] + refl[, sen2s2[["B3"]]]) + spectralindices$GRVI1 <- grvi1 + } + if ("GNDVI" %in% sel_indices) { + gndvi <- (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B3"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B3"]]]) + spectralindices$GNDVI <- gndvi + } + if ("IRECI" %in% sel_indices) { + ireci <- (refl[, sen2s2[["B7"]]] - refl[, sen2s2[["B4"]]]) * (refl[, sen2s2[["B6"]]] / (refl[, sen2s2[["B5"]]])) + spectralindices$IRECI <- ireci + } + if ("LAI_SAVI" %in% sel_indices) { + lai_savi <- - log(0.371 + 1.5 * (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B4"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B4"]]] + 0.5)) / 2.4 + spectralindices$LAI_SAVI <- lai_savi + } + if ("MCARI" %in% sel_indices) { + mcari <- (1 - 0.2 * (refl[, sen2s2[["B5"]]] - refl[, sen2s2[["B3"]]]) / (refl[, sen2s2[["B5"]]] - refl[, sen2s2[["B4"]]])) + spectralindices$MCARI <- mcari + } + if ("mNDVI705" %in% sel_indices) { + mndvi705 <- (refl[, sen2s2[["B6"]]] - refl[, sen2s2[["B5"]]]) / (refl[, sen2s2[["B6"]]] + refl[, sen2s2[["B5"]]] - 2 * refl[, sen2s2[["B2"]]]) + spectralindices$mNDVI705 <- mndvi705 + } + if ("MSAVI2" %in% sel_indices) { + msavi2 <- ((refl[, sen2s2[["B8"]]] + 1) - 0.5 * sqrt(((2 * refl[, sen2s2[["B8"]]]) - 1)**2 + 8 * refl[, sen2s2[["B4"]]])) + spectralindices$MSAVI2 <- msavi2 + } + if ("MSI" %in% sel_indices) { + msi <- refl[, sen2s2[["B11"]]] / refl[, sen2s2[["B8"]]] + spectralindices$MSI <- msi + } + if ("mSR705" %in% sel_indices) { + msr705 <- (refl[, sen2s2[["B6"]]] - refl[, sen2s2[["B2"]]]) / (refl[, sen2s2[["B5"]]] - refl[, sen2s2[["B2"]]]) + spectralindices$mSR705 <- msr705 + } + if ("MTCI" %in% sel_indices) { + mtci <- (refl[, sen2s2[["B6"]]] - refl[, sen2s2[["B5"]]]) / (refl[, sen2s2[["B5"]]] + refl[, sen2s2[["B4"]]]) + spectralindices$MTCI <- mtci + } + if ("nBR_RAW" %in% sel_indices) { + nbr_raw <- (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B12"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B12"]]]) + spectralindices$nBR_RAW <- nbr_raw + } + if ("NDI_45" %in% sel_indices) { + ndi_45 <- (refl[, sen2s2[["B5"]]] - refl[, sen2s2[["B4"]]]) / (refl[, sen2s2[["B5"]]] + refl[, sen2s2[["B4"]]]) + spectralindices$NDI_45 <- ndi_45 + } + if ("NDII" %in% sel_indices) { + ndii <- (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B11"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B11"]]]) + spectralindices$NDII <- ndii + } + if ("NDSI" %in% sel_indices) { + ndisi <- (refl[, sen2s2[["B3"]]] - refl[, sen2s2[["B11"]]]) / (refl[, sen2s2[["B3"]]] + refl[, sen2s2[["B11"]]]) + spectralindices$NDSI <- ndsi + } + if ("NDVI" %in% sel_indices) { + ndvi <- (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B4"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B4"]]]) + spectralindices$NDVI <- ndvi + } + if ("NDVI_G" %in% sel_indices) { + ndvi_g <- refl[, sen2s2[["B3"]]] * (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B4"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B4"]]]) + spectralindices$NDVI_G <- ndvi_g + } + if ("NDVI705" %in% sel_indices) { + ndvi705 <- (refl[, sen2s2[["B6"]]] - refl[, sen2s2[["B5"]]]) / (refl[, sen2s2[["B6"]]] + refl[, sen2s2[["B5"]]]) + spectralindices$NDVI705 <- ndvi705 + } + if ("NDWI1" %in% sel_indices) { + ndwi1 <- (refl[, sen2s2[["B8A"]]] - refl[, sen2s2[["B11"]]]) / (refl[, sen2s2[["B8A"]]] + refl[, sen2s2[["B11"]]]) + spectralindices$NDWI1 <- ndwi1 + } + if ("NDWI2" %in% sel_indices) { + ndwi2 <- (refl[, sen2s2[["B8A"]]] - refl[, sen2s2[["B12"]]]) / (refl[, sen2s2[["B8A"]]] + refl[, sen2s2[["B12"]]]) + spectralindices$NDWI2 <- ndwi2 + } + if ("PSRI" %in% sel_indices) { + psri <- (refl[, sen2s2[["B4"]]] - refl[, sen2s2[["B2"]]]) / (refl[, sen2s2[["B5"]]]) + spectralindices$PSRI <- psri + } + if ("PSRI_NIR" %in% sel_indices) { + psri_nir <- (refl[, sen2s2[["B4"]]] - refl[, sen2s2[["B2"]]]) / (refl[, sen2s2[["B8"]]]) + spectralindices$PSRI_NIR <- psri_nir + } + if ("RE_NDVI" %in% sel_indices) { + re_ndvi <- (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B6"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B6"]]]) + spectralindices$RE_NDVI <- re_ndvi + } + if ("RE_NDWI" %in% sel_indices) { + re_ndwi <- (refl[, sen2s2[["B4"]]] - refl[, sen2s2[["B6"]]]) / (refl[, sen2s2[["B4"]]] + refl[, sen2s2[["B6"]]]) + spectralindices$RE_NDWI <- re_ndwi + } + if ("S2REP" %in% sel_indices) { + s2rep <- 705 + 35 * (0.5 * (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B5"]]]) - refl[, sen2s2[["B6"]]]) / (refl[, sen2s2[["B7"]]] - refl[, sen2s2[["B6"]]]) + spectralindices$S2REP <- s2rep + } + if ("SAVI" %in% sel_indices) { + savi <- 1.5 * (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B5"]]]) / (refl[, sen2s2[["B8"]]] + refl[, sen2s2[["B5"]]] + 0.5) + spectralindices$SAVI <- savi + } + if ("SIPI" %in% sel_indices) { + sipi <- (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B2"]]]) / (refl[, sen2s2[["B8"]]] - refl[, sen2s2[["B4"]]]) + spectralindices$SIPI <- sipi + } + if ("SR" %in% sel_indices) { + sr <- refl[, sen2s2[["B8"]]] / refl[, sen2s2[["B4"]]] + spectralindices$SR <- sr + } + if ("CR_SWIR" %in% sel_indices) { + cr_swir <- refl[, sen2s2[["B11"]]] / (refl[, sen2s2[["B8A"]]] + (s2bands$B11 - s2bands$B8A) * (refl[, sen2s2[["B12"]]] - refl[, sen2s2[["B8A"]]]) / (s2bands$B12 - s2bands$B8A)) + spectralindices$CR_SWIR <- cr_swir + } + res <- list("spectralindices" = spectralindices, "listindices" = listindices) + return(res) +} + +#" this function identifies the bands of a given sensor with closest match to its spectral characteristics +#" +#" @param sensorbands numeric. wavelength in nanometer of the sensor of interest +#" @param listbands numeric or list. Named vector or list of spectral bands corresponding to sensor +#" +#" @return numeric. band numbers of original sensor +#" @export +get_closest_bands <- function(sensorbands, listbands) { + sapply(listbands, function(x) { + b <- which.min(abs(sensorbands - x)) + names(b) <- "" + b + }) +} + +#" This function computes interquartile range (IQR) criterion, which can be used +#" as a criterion for outlier detection +#" +#" @param distval numeric. vector of distribution of values +#" @param weightirq numeric. weighting factor appplied to IRQ to define lower and upper boudaries for outliers +#" +#" @return outlier_iqr numeric. band numbers of original sensor corresponding to S2 +#" @importFrom stats IQR quantile +#" @export +iqr_outliers <- function(distval, weightirq = 1.5) { + iqr <- IQR(distval, na.rm = TRUE) + range_iqr <- c(quantile(distval, 0.25, na.rm = TRUE), quantile(distval, 0.75, na.rm = TRUE)) + outlier_iqr <- c(range_iqr[1] - weightirq * iqr, range_iqr[2] + weightirq * iqr) + return(outlier_iqr) +} + +#" This function selects bands from a raster or stars object +#" +#" @param refl RasterBrick, RasterStack or list. Raster bands in the order of sensorbands. +#" @param bands numeric. rank of bands to be read in refl +#" @param reflfactor numeric. multiplying factor used to write reflectance in image ( == 10000 for S2) +#" +#" @return robj list. R object (default is raster, stars if refl is stars object) +#" @importFrom raster subset stack +#" @export +readrasterbands <- function(refl, bands, reflfactor = 1) { + + # get equation for line going from CR1 to CR2 + classraster <- class(refl) + if (classraster == "RasterBrick" || classraster == "RasterStack" || classraster == "stars") { + # if !reflfactor == 1 then apply a reflectance factor + if (classraster == "stars") { + robj <- refl[bands] + } else { + robj <- raster::subset(refl, bands) + } + if (!reflfactor == 1) { + robj <- robj / reflfactor + } + } else if (is.list(refl)) { + robj <- raster::stack(refl[bands]) # checks that all rasters have same crs/extent + if (!reflfactor == 1) { + robj <- robj / reflfactor + } + } else { + stop("refl is expected to be a RasterStack, RasterBrick, Stars object or a list of rasters") + } + return(robj) +}