Merge pull request #16 from davidklehr/main

Add the R UDF of spline interpolation

Author: davidfrantz

rstats/ts/spline-reconstruction/Bolten_Spline_FORCE_UDF.R

@@ -0,0 +1,135 @@
+force_rstats_init <- function(dates, sensors, bandnames){
+
+  # Year which should be reconstructed
+  year_to_interpolate <- 2023
+  # Days to predict in this year and the intervall: 60 to 330 (1st March to 26th November)
+  DOYs_to_predict <- seq(60,330,by =10)
+  dates_to_predict <- as.Date(paste(year_to_interpolate, DOYs_to_predict), format = "%Y %j")
+  
+  band_names <- paste(substr(as.character(dates_to_predict),1,4),substr(as.character(dates_to_predict),6,7),substr(as.character(dates_to_predict),9,10), sep ="")
+  return(band_names)
+}
+
+force_rstats_pixel <- function(inarray, dates, sensors, bandnames, nproc){
+  
+  # Year which should be interpolated (same like above)
+  year_to_interpolate <- 2023
+  # Days to predict in this yearand the intervall: 60 to 330 (1st March to 26th November)
+  DOYs_to_predict <- seq(60,330,by =10) 
+  dates_to_predict <- as.Date(paste(year_to_interpolate, DOYs_to_predict), format = "%Y %j")
+
+  # spline variables
+  # smoothing factor for the spline reconstruction
+  smooth_fac <- 0.5
+  # Bolton's variable of maximum weight to assing for the predessesor years
+  # the year of reconstruction has a wheight of 1
+  max_weight <- 0.2
+  
+  # cumulate the DOY to the year of interpolation
+  # start year 2015 (example), because of Sentinel 2 launch date, for e.g. Landsat adjust to your time span
+  start_year <- 2015
+  DOYs_to_predict <- (year_to_interpolate - start_year) * 365 + DOYs_to_predict
+  
+  tryCatch({
+    # grap FORCE no-data case
+    if (all(is.na(inarray[,1]))){
+      return(rep(-9999,length(DOYs_to_predict)))
+    }
+  
+    # calculate cumulative DOYs for the input data 
+    DOYs <- as.numeric(format(dates, "%j"))
+    years <- as.numeric(substr(as.character(dates),1,4))
+    cumulative_DOYs <- (years - start_year) * 365 + DOYs
+    
+    # join the data to a dataframe
+    df <- data.frame(x=cumulative_DOYs,y=inarray[,1])
+    
+    # ------- 1.1 calcualte Mean Function --------------
+    euc.dist <- function(x1, x2) sqrt(sum((x1 - x2) ^ 2))
+    
+    # define Start and endpoints for the three spline reconstuctions
+    DOY_borders_year <- c((year_to_interpolate-start_year)*365 - 180, (year_to_interpolate-start_year+1)*365+180)
+    DOY_borders_b    <- c((year_to_interpolate-start_year-1)*365 - 180, (year_to_interpolate-start_year)*365+180)
+    DOY_borders_bb   <- c((year_to_interpolate-start_year-2)*365 - 180, (year_to_interpolate-start_year-1)*365+180)
+    
+    # create dataframes for reconstuction
+    data_year <- na.exclude(df[df$x %in% seq(DOY_borders_year[1],DOY_borders_year[2]),c(1,2)])
+    data_b   <- na.exclude(df[df$x %in% seq(DOY_borders_b[1] ,DOY_borders_b[2]),c(1,2)])
+    data_bb  <- na.exclude(df[df$x %in% seq(DOY_borders_bb[1],DOY_borders_bb[2]),c(1,2)])
+    
+    # calculate spline model for year of reconstruction and predict
+    DOYs_target_year <- seq(DOY_borders_year[1],DOY_borders_year[2])
+    tryCatch({
+      eval( spline_model_year <<- smooth.spline(data_year$x,data_year$y, spar =smooth_fac) )
+      eval( predict_year <<- predict(spline_model_year, x = DOYs_target_year)$y )
+    }, error = function(err) {
+      return(rep(-9999,length(DOYs_to_predict)))
+    })
+    
+    #calculate d_max
+    mean_year <- mean(na.exclude(data_year$y))
+    mean_prediction <- rep(mean_year,length(DOYs_target_year))
+    d_max = euc.dist(mean_prediction, predict_year) / 10000
+    
+    
+    # --------- 1.2 spline for precessor years ------------
+    # one year before
+    # predict with DOYs of year of reconstruction, for difference calculation 
+    # between the two spline reconstructions
+    tryCatch({
+      eval( spline_model_b <<- smooth.spline(data_b$x+365,data_b$y, spar =smooth_fac) )
+      eval( predict_b <<- predict(spline_model_b, x = DOYs_target_year)$y )
+    }, error = function(err) {
+      return(rep(-9999,length(DOYs_to_predict)))
+    })
+    d_b = euc.dist(predict_year, predict_b) / 10000
+    
+    # two years before
+    # predict with DOYs of year of reconstruction, for difference calculation 
+    # between the two spline reconstructions
+    tryCatch({
+      eval( spline_model_bb <<- smooth.spline(data_bb$x+(365*2),data_bb$y, spar =smooth_fac) )
+      eval( predict_bb <<- predict(spline_model_bb, x = DOYs_target_year)$y )
+    }, error = function(err) {
+      return(rep(-9999,length(DOYs_to_predict)))
+    })
+    d_bb = euc.dist(predict_year, predict_bb) / 10000
+    
+    # ---------- 1.3 Calculate weights -------------------
+    # one year before
+    if (d_max != 0) {
+      weight_b = max_weight*(1-(d_b/d_max))
+      if (weight_b < 0){
+        weight_b = 0
+      }
+    } else {weight_b = 0}
+    
+    # two years before
+    if (d_max != 0) {
+      weight_bb = max_weight*(1-(d_bb/d_max))
+      if (weight_bb < 0){
+        weight_bb = 0
+      }
+    } else {weight_bb = 0}
+    
+    #----------- 1.4 apply weights and calculate weighted spline --------------
+    # combine the time series to one year and assign weights to the new data points
+    combined_x <- c(data_year$x , (data_b$x+365)[weight_b>0] , (data_bb$x + (365*2))[weight_bb>0])
+    combined_y <- c(data_year$y , data_b$y[weight_b>0]       , data_bb$y[weight_bb>0])
+    vec_weights <- c(rep(1,length(data_year$x)),
+                     rep(weight_b,length(data_b$x))[weight_b>0],
+                     rep(weight_bb,length(data_bb$x))[weight_bb>0])
+    
+    # calculate weighted spline
+    tryCatch({
+      eval( spline_model_combined <<- smooth.spline(x=combined_x, y=combined_y, w = vec_weights , spar =smooth_fac) )
+      eval( predict_combined <<- predict(spline_model_combined, x = DOYs_to_predict)$y )
+    }, error = function(err) {
+      return(rep(-9999,length(DOYs_to_predict)))
+    })
+    return(predict_combined)
+  
+  }, error = function(err) {
+    return(rep(-9999,length(DOYs_to_predict)))
+  })
+}

rstats/ts/spline-reconstruction/readme.md

@@ -0,0 +1,20 @@
+# Time series reconstruction for forest using splines
+
+©
+Copyright 2024, David Klehr
+
+## Run with
+
+- program: ``force-higher-level``
+- submodule: ``TSA``
+- DATE_RANGE: ``xxxx-07-01 yyyy-06-31``
+    * xxxx = three years before your target year
+    * yyyy = one year after your target year
+    * e.g. for target year 2022: ``2019-07-01 2023-06-31``
+- UDF type: ``RSTATS_TYPE = PIXEL``
+- required parameters:``none``
+- required R libraries: ``none``
+
+## References
+
+- Bolton, D.K., Gray, J.M., Melaas, E.K., Moon, M., Eklundh, L., Friedl, M.A., 2020. **Continental-scale land surface phenology from harmonized Landsat 8 and Sentinel-2 imagery**. *Remote Sensing of Environment 240*, 111685. https://doi.org/10.1016/j.rse.2020.111685.