Update data input, transformation and modeling scripts because there is new data

10_compile-ref-data.R

@@ -16,9 +16,11 @@
 pkgs <- c("tidyverse", "simplerspec")
 lapply(pkgs, library, character.only = TRUE)
 
+
 ## Load all plant element data =================================================
 
+elements_yam_leaf <- read_csv("data/training/elements/leaf_elements.csv")
+elements_yam_tuber  <- read_csv("data/training/elements/tuber_elements.csv")
+
 
-elements_yam_leaf <- read_csv("data/elements/leaf_elements.csv")
-elements_yam_tuber  <- read_csv("data/elements/tuber_elements.csv")
 

20_summarize-ref-data.R

+################################################################################
+## Decription: 
+################################################################################
\ No newline at end of file

30_read-process-spc-train.R

+################################################################################
+## Author: Philipp Baumann || philipp.baumann@usys.ethz.ch
+## License: GPL-3.0
+## Project: Biophysical, institutional and economic drivers of sustainable 
+##   soil use in yam systems for improved food security in West Africa (YAMSYS)
+## Description: Tune and evaluate PLS regression models to predict Yam 
+##   root and leaf elemental concentrations
+################################################################################
+
+library("tidyverse")
+library("simplerspec")
+
+## Register parallel backend for using multiple cores ==========================
+
+# Allows to tune the models using parallel processing (e.g. use all available
+# cores of a CPU); caret package automatically detects the registered backend
+library("doParallel")
+# Make a cluster with all possible threads (more than physical cores)
+cl <- makeCluster(detectCores())
+# Register backend
+registerDoParallel(cl)
+# Return number of parallel workers
+getDoParWorkers() # 8 threads on MacBook Pro (Retina, 15-inch, Mid 2015);
+# Quadcore processor
+
+
+################################################################################
+## Part 1: Read and pre-process spectra, Read chemical data, and join
+## spectral and chemical data sets
+################################################################################
+
+## List of OPUS files obtained from Bruker ALPHA spectrometer at ETH Zürich
+lf_leaf <- dir("data/training/spectra/leaf", full.names = TRUE)
+lf_tuber <- dir("data/training/spectra/tuber", full.names = TRUE)
+
+## Read files: ETH Zürich
+spc_lst_leaf <- read_opus_univ(
+  fnames = lf_leaf,
+  extract = c("spc", "ScSm", "ScRf"), 
+  parallel = TRUE)
+
+spc_lst_tuber <- read_opus_univ(
+  fnames = lf_tuber,
+  extract = c("spc", "ScSm", "ScRf"),
+  parallel = TRUE)
+
+
+## Spectral data processing pipe ===============================================
+
+## ETH Alpha
+spc_tbl_leaf <- spc_lst_leaf %>%
+  gather_spc() %>%
+  resample_spc(wn_lower = 500, wn_upper = 3996, wn_interval = 2) %>%
+  average_spc() %>%
+  preprocess_spc(select = "sg_1_w21")
+
+spc_tuber_train <- spc_lst_tuber %>%
+  gather_spc() %>%
+  resample_spc(wn_lower = 500, wn_upper = 3996, wn_interval = 2) %>%
+  average_spc() %>%
+  preprocess_spc(select = "sg_1_w21") %>%
+  group_by(sample_id) %>%
+  slice(1L)

40_build-spc-models-tuber-train.R

+################################################################################
+## Author: Philipp Baumann || philipp.baumann@usys.ethz.ch
+## License: GPL-3.0
+## Project: Biophysical, institutional and economic drivers of sustainable 
+##   soil use in yam systems for improved food security in West Africa (YAMSYS)
+## Description: Tune and evaluate PLS regression models to predict yam 
+##   tuber elemental concentrations
+################################################################################
+
+
+## Register parallel backend for using multiple cores ==========================
+
+# Allows to tune the models using parallel processing (e.g. use all available
+# cores of a CPU); caret package automatically detects the registered backend
+library("doParallel")
+# Make a cluster with all possible threads (more than physical cores)
+cl <- makeCluster(detectCores())
+# Register backend
+registerDoParallel(cl)
+# Return number of parallel workers
+getDoParWorkers() # 8 threads on MacBook Pro (Retina, 15-inch, Mid 2015);
+# Quadcore processor
+
+## Prepare final data for spectral modeling ====================================
+
+# Join spectra tibble and chemical reference analysis tibble
+spc_elements_tuber <- join_spc_chem(
+  spc_tbl = spc_tuber_train, chem_tbl = elements_yam_tuber, by = "sample_id")
+
+spc_elements_leaf <- join_spc_chem(
+  spc_tbl = spc_tbl_leaf, chem_tbl = elements_yam_leaf, by = "sample_id")
+
+
+
+################################################################################
+## Tune and cross-validate training models for yam tubers
+################################################################################
+
+## Run PLS regression models for all measured soil properties
+## Use 5 times repeated 10-fold cross-validation for model tuning and
+## use by resampling repeat averaged (mean) held-out predictions at
+## finally chosen of PLS regression components for model evaluation;
+## argument `resampling_method` = "rep_kfold_cv" creates 5 times repeated
+## cross-validation; always exclude missing values for samples that have missing
+## values in the target soil property variable
+## =============================================================================
+
+## Use 5 times repeated 10-fold cross-validation
+
+# Total N
+pls_tuber_N <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$N), ],
+  response = N,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total C
+# Omit outlier at > 800 g / kg C
+pls_tuber_C <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$C) &
+    spc_elements_tuber$C < 800, ],
+  response = C,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total S
+pls_tuber_S <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$S), ],
+  response = S,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total P
+# Remove outlier at > 4000 mg P / kg
+pls_tuber_P <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$P) & 
+    spc_elements_tuber$P < 4000, ],
+  response = P,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total K
+pls_tuber_K <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$K), ],
+  response = K,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Ca
+# Remove outlier at > 2000 mg Ca / kg
+pls_tuber_Ca <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$Ca) &
+    spc_elements_tuber$Ca < 2000, ],
+  response = Ca,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Mg
+# Remove outlier at > 3000 mg Mg / kg
+pls_tuber_Mg <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$Mg) &
+    spc_elements_tuber$Mg < 3000, ],
+  response = Mg,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Fe
+# Remove outliers at > 150 mg Fe / kg
+pls_tuber_Fe <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$Fe) &
+    spc_elements_tuber$Fe < 150, ],
+  response = Fe,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Zn
+# Remove outliers at > 75mg Zn / kg
+pls_tuber_Zn <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$Zn) &
+    spc_elements_tuber$Zn < 75, ],
+  response = Zn,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Cu
+pls_tuber_Cu <- fit_pls(
+  spec_chem = spc_elements_tuber[!is.na(spc_elements_tuber$Cu), ],
+  response = Cu,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+
+################################################################################
+## Write all pls models (R objects) into separate files
+## using the saveRDS function
+## .Rds files are R data files that contain a single R object, in our case
+## it is a list from the output of the fit_pls() function
+## These files can be directly loaded into a new R session where e.g. calibrated
+## properties are predicted based on new sample spectra
+################################################################################
+
+saveRDS(pls_tuber_N, "models/training/tuber/pls_tuber_N.Rds")
+saveRDS(pls_tuber_C, "models/training/tuber/pls_tuber_C.Rds")
+saveRDS(pls_tuber_S, "models/training/tuber/pls_tuber_S.Rds")
+saveRDS(pls_tuber_P, "models/training/tuber/pls_tuber_P.Rds")
+saveRDS(pls_tuber_K, "models/training/tuber/pls_tuber_K.Rds")
+saveRDS(pls_tuber_Ca, "models/training/tuber/pls_tuber_Ca.Rds")
+saveRDS(pls_tuber_Mg, "models/training/tuber/pls_tuber_Mg.Rds")
+saveRDS(pls_tuber_Fe, "models/training/tuber/pls_tuber_Fe.Rds")
+saveRDS(pls_tuber_Zn, "models/training/tuber/pls_tuber_Zn.Rds")
+saveRDS(pls_tuber_Cu, "models/training/tuber/pls_tuber_Cu.Rds")
+
+

50_read-process-spc-test.R

+################################################################################
+## Author: Philipp Baumann || philipp.baumann@usys.ethz.ch
+## License: GPL-3.0
+## Project: Biophysical, institutional and economic drivers of sustainable 
+##   soil use in yam systems for improved food security in West Africa (YAMSYS)
+## Description: Process tuber test data; data set of Patricia Schwitter
+################################################################################
+
+## Register parallel backend for using multiple cores ==========================
+
+# Allows to tune the models using parallel processing (e.g. use all available
+# cores of a CPU); caret package automatically detects the registered backend
+library("doParallel")
+# Make a cluster with all possible threads (more than physical cores)
+cl <- makeCluster(detectCores())
+# Register backend
+registerDoParallel(cl)
+# Return number of parallel workers
+getDoParWorkers() # 8 threads on MacBook Pro (Retina, 15-inch, Mid 2015);
+# Quadcore processor
+
+################################################################################
+## Part 1: Read and pre-process spectra, Read chemical data, and join
+## spectral and chemical data sets
+################################################################################
+
+## List of OPUS files obtained from Bruker ALPHA spectrometer at ETH Zürich
+lf_tuber_test <- dir("data/test/tuber/spectra", full.names = TRUE)
+lf_tuber_beatrice <- dir("data/test_beatrice", full.names = TRUE)
+
+## Read files into list
+spc_lst_tuber_test <- read_opus_univ(
+  fnames = lf_tuber_test,
+  extract = "spc",
+  parallel = TRUE)
+
+spc_lst_tuber_beatrice <- read_opus_univ(
+  fnames = lf_tuber_beatrice,
+  extract = "spc",
+  parallel = TRUE)
+
+
+## Spectral data processing pipeline ===========================================
+
+spc_tuber_test <- spc_lst_tuber_test %>%
+  gather_spc() %>%
+  resample_spc(wn_lower = 500, wn_upper = 3996, wn_interval = 2) %>%
+  average_spc() %>%
+  preprocess_spc(select = "sg_1_w21")
+
+spc_tuber_beatrice <- spc_lst_tuber_beatrice %>%
+  gather_spc() %>%
+  resample_spc(wn_lower = 500, wn_upper = 3996, wn_interval = 2) %>%
+  average_spc() %>%
+  preprocess_spc(select = "sg_1_w21")
\ No newline at end of file

60_explore-spc-training-test.R

+
+
+## Plot training and test spectra ==============================================
+
+pdf(file = "out/figs/spc-train-test-expl.pdf", width = 5, height = 5)
+plot_spc_ext(
+  spc_tbl_l = list(
+    "Training set" = spc_tuber_train %>%
+      mutate(group_id = "Comparison test and training spectra"), 
+    "Test set" = spc_tuber_test %>%
+      mutate(group_id = "Comparison test and training spectra")),
+  lcols_spc = c("spc", "spc_pre"), group_id = "group_id")
+dev.off()
\ No newline at end of file

70_evaluate-test.R

+################################################################################
+## Evaluate the the Yam tuber testing set based on the training models
+################################################################################
+
+
+## Read test reference tuber element data (ICP-OES) ============================
+
+elements_yam_tuber_test <- read_csv(
+  "data/test/tuber/elements/tuber_elements_test.csv")
+
+## Read the yam tuber training models ==========================================
+
+pls_tuber_N <- readRDS(file = "models/training/tuber/pls_tuber_N.Rds")
+pls_tuber_C <- readRDS(file = "models/training/tuber/pls_tuber_C.Rds")
+pls_tuber_S <- readRDS(file = "models/training/tuber/pls_tuber_S.Rds")
+pls_tuber_P <- readRDS(file = "models/training/tuber/pls_tuber_P.Rds")
+pls_tuber_K <- readRDS(file = "models/training/tuber/pls_tuber_K.Rds")
+pls_tuber_Ca <- readRDS(file = "models/training/tuber/pls_tuber_Ca.Rds")
+pls_tuber_Mg <- readRDS(file = "models/training/tuber/pls_tuber_Mg.Rds")
+pls_tuber_Fe <- readRDS(file = "models/training/tuber/pls_tuber_Fe.Rds")
+pls_tuber_Zn <- readRDS(file = "models/training/tuber/pls_tuber_Zn.Rds")
+pls_tuber_Cu <- readRDS(file = "models/training/tuber/pls_tuber_Cu.Rds")
+
+# Nest all models in a list for prediction
+pls_tuber_all <- list(
+  "pls_tuber_N" = pls_tuber_N,
+  "pls_tuber_C" = pls_tuber_C,
+  "pls_tuber_S" = pls_tuber_S,
+  "pls_tuber_P" = pls_tuber_P,
+  "pls_tuber_K" = pls_tuber_K,
+  "pls_tuber_Ca" = pls_tuber_Ca,
+  "pls_tuber_Mg" = pls_tuber_Mg,
+  "pls_tuber_Fe" = pls_tuber_Fe,
+  "pls_tuber_Zn" = pls_tuber_Zn,
+  "pls_tuber_Cu" = pls_tuber_Cu
+)
+
+## Predict tuber nutrients based on training model and new spectra =============
+
+pred_tuber_test <- predict_from_spc(
+  model_list = pls_tuber_all, 
+  spc_tbl = spc_tuber_test
+  )
+
+# Join test set predictions and test reference data
+predobs_tuber_test <- inner_join(x = pred_tuber_test,
+  y = elements_yam_tuber_test)
+
+# Gather all predicted elements in one column
+gather(predobs_tuber_test, key = element_pred, value = pred,
+  pls_tuber_N, pls_tuber_C, pls_tuber_S, pls_tuber_P, pls_tuber_K,
+  pls_tuber_Ca, pls_tuber_Mg, pls_tuber_Fe, pls_tuber_Zn, pls_tuber_Cu)
+
+measure_obs <- c("N", "C", "P", "K", "Ca", "Mg", "Fe", "Zn", "Cu")
+measure_pred <- c("pls_tuber_N", "pls_tuber_C", "pls_tuber_P", "pls_tuber_K",
+  "pls_tuber_Ca", "pls_tuber_Mg", "pls_tuber_Fe", "pls_tuber_Zn",
+  "pls_tuber_Cu")
+
+evaluate_multimodels <- function(data, cols_obs, cols_pred) {
+}
+
+library("data.table")
+
+obs_test_long <- data.table::melt(as.data.table(predobs_tuber_test),
+  measure = measure_obs, variable.name = "variable_obs",
+  value.name = "value_obs", variable.factor = FALSE) %>%
+  select(sample_id, variable_obs, value_obs) %>%
+  mutate(variable = variable_obs) %>%
+  as.tibble()
+
+pred_test_long <- data.table::melt(as.data.table(predobs_tuber_test),
+  measure = measure_pred, variable.name = "variable_pred",
+  value.name = "value_pred", variable.factor = FALSE) %>%
+  select(sample_id, variable_pred, value_pred) %>%
+  as.tibble()
+
+predobs_tuber_test_long <- bind_cols(obs_test_long,
+  pred_test_long)
+
+# Perform model evaluation per variable
+model_eval_train <- predobs_tuber_test_long %>%
+  group_by(variable) %>%
+  nest() %>%
+  mutate(model_eval = map(data,
+    ~ summary_df(.x, x = "value_obs", y = "value_pred"))) %>%
+  unnest(model_eval) %>%
+  select(-data)
+
+## Graphical model evaluation
+predobs_tuber_test_long %>%
+  ggplot(aes(x = value_obs, y = value_pred)) +
+  geom_point() +
+  facet_wrap(~ variable_obs, scales = "free")
+
+################################################################################
+
+plot_spc_ext(spc_tbl_l = 
+  list("Training ref." = spc_tbl_tuber %>% mutate(group_id = "Test"), 
+       "Beatrice test" = spc_tuber_beatrice %>% mutate(group_id = "Test")),
+  lcols_spc = c("spc","spc_pre"),
+  group_id = "group_id")
\ No newline at end of file

90_model-test-train-fusion.R

+spc_elements_tuber_test <- spc_tuber_test %>%
+  inner_join(x = ., y = elements_yam_tuber_test)
+
+
+spc_elements_tuber_all <- spc_elements_tuber %>%
+  select(sample_id, spc_pre, C, N, P, K, Ca, Mg, Fe, Zn, Cu) %>%
+  bind_rows(spc_elements_tuber_test %>% 
+    select(sample_id, spc_pre, C, N, P, K, Ca, Mg, Fe, Zn, Cu))
+
+
+pls_tuber_N_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$N), ],
+  response = N,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+pls_tuber_C_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$C) &
+    spc_elements_tuber_all$C < 800 & spc_elements_tuber_all$C > 383, ],
+  response = C,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total P
+# Remove outlier at > 4000 mg P / kg
+pls_tuber_P_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$P) & 
+    spc_elements_tuber_all$P < 4000, ],
+  response = P,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total K
+pls_tuber_K_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$K), ],
+  response = K,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Ca
+# Remove outlier at > 2000 mg Ca / kg
+pls_tuber_Ca_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$Ca) &
+    spc_elements_tuber_all$Ca < 2000, ],
+  response = Ca,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Mg
+# Remove outlier at > 3000 mg Mg / kg
+pls_tuber_Mg_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$Mg) &
+    spc_elements_tuber_all$Mg < 3000, ],
+  response = Mg,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Fe
+# Remove outliers at > 150 mg Fe / kg
+pls_tuber_Fe_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$Fe) &
+    spc_elements_tuber_all$Fe < 150, ],
+  response = Fe,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Zn
+# Remove outliers at > 75mg Zn / kg
+pls_tuber_Zn_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$Zn) &
+    spc_elements_tuber_all$Zn < 75, ],
+  response = Zn,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+# Total Cu
+pls_tuber_Cu_all <- fit_pls(
+  spec_chem = spc_elements_tuber_all[!is.na(spc_elements_tuber_all$Cu), ],
+  response = Cu,
+  evaluation_method = "resampling",
+  tuning_method = "resampling",
+  resampling_method = "rep_kfold_cv",
+  pls_ncomp_max = 10
+)
+
+
+pdf(file = "out/figs/model-eval-yam-tuber-all.pdf", width = 10, height = 15)
+cowplot::plot_grid(pls_tuber_C_all$p_model, pls_tuber_N_all$p_model,
+  pls_tuber_P_all$p_model, pls_tuber_K_all$p_model, pls_tuber_Ca_all$p_model,
+  pls_tuber_Mg_all$p_model, pls_tuber_Zn_all$p_model, pls_tuber_Cu_all$p_model,
+  ncol = 2)
+dev.off()
+

data/elements/raw/tuber_cns.csv

-sample_ID,country,site,category,material,N_perc,C_perc,S_perc
-CI_tb_01_tuber_x1,CI,tb,ref,tuber,1.0095,40.3243,0.1079
-CI_tb_02_tuber,CI,tb,ref,tuber,1.0837,39.1344,0.0942
-CI_tb_03_tuber,CI,tb,ref,tuber,1.074,39.4732,0.0716
-CI_tb_04a_tuber,CI,tb,ref,tuber,1.2283,39.6649,0.0673
-CI_tb_04b_tuber,CI,tb,ref,tuber,1.0218,39.3409,0.0595
-CI_tb_05_tuber_x2,CI,tb,ref,tuber,1.1051,40.3163,0.0702
-CI_tb_06_tuber_x3,CI,tb,ref,tuber,1.2135,40.1294,0.0887
-CI_tb_07_tuber_x4,CI,tb,ref,tuber,0.9133,40.2266,0.0775
-CI_tb_08_tuber_x5,CI,tb,ref,tuber,1.1747,40.2292,0.0749
-CI_tb_01_tuber_x1,CI,tb,ref,tuber,0.9932,40.2671,0.07
-CI_tb_09_tuber,CI,tb,ref,tuber,0.502,39.8867,0.0396
-CI_tb_10_tuber,CI,tb,ref,tuber,0.9931,39.7161,0.0683
-CI_tb_11_tuber_x6,CI,tb,ref,tuber,0.7243,39.8494,0.0673
-CI_tb_12_tuber_x7,CI,tb,ref,tuber,0.9876,40.1559,0.0849
-CI_tb_13_tuber,CI,tb,ref,tuber,1.1646,39.4043,0.0854
-CI_tb_14_tuber_x8,CI,tb,ref,tuber,0.9455,39.799,0.0758
-CI_tb_15_tuber_x9,CI,tb,ref,tuber,0.7852,40.2089,0.0625
-CI_tb_16_tuber,CI,tb,ref,tuber,1.6781,40.0941,0.0907
-CI_tb_01_tuber_x1,CI,tb,ref,tuber,1.021,40.275,0.0851
-CI_tb_18_tuber,CI,tb,ref,tuber,1.5816,39.557,0.0961
-CI_tb_19_tuber,CI,tb,ref,tuber,0.5807,39.5731,0.0461
-CI_tb_20_tuber,CI,tb,ref,tuber,1.0144,39.2474,0.0654
-CI_tb_20_tuber,CI,tb,ref,tuber,1.0335,39.2417,0.0802
-CI_tb_20_tuber,CI,tb,ref,tuber,1.0181,39.2038,0.0769
-CI_sb_01_tuber,CI,sb,ref,tuber,0.6544,39.6101,0.0767
-CI_sb_02_tuber_x1,CI,sb,ref,tuber,1.3493,39.2415,0.1115
-CI_sb_03b_tuber,CI,sb,ref,tuber,1.5752,40.3228,0.1213
-CI_sb_04_tuber,CI,sb,ref,tuber,1.8656,39.3012,0.1655
-CI_sb_05_tuber,CI,sb,ref,tuber,0.9773,39.0753,0.0877
-CI_sb_06_tuber,CI,sb,ref,tuber,1.0049,39.7355,0.088
-CI_sb_07_tuber,CI,sb,ref,tuber,0.8524,39.4246,0.0869
-CI_sb_08_tuber,CI,sb,ref,tuber,0.6186,39.407,0.0781
-CI_sb_01_tuber,CI,sb,ref,tuber,0.6899,39.501,0.083
-CI_sb_09_tuber,CI,sb,ref,tuber,0.9002,39.0967,0.0885
-CI_sb_10_tuber,CI,sb,ref,tuber,0.6767,39.2129,0.1028
-CI_sb_11_tuber,CI,sb,ref,tuber,0.8167,38.8496,0.0836
-CI_sb_12_tuber,CI,sb,ref,tuber,0.6306,39.4329,0.0666
-CI_sb_13_tuber,CI,sb,ref,tuber,0.6248,39.3116,0.0682
-CI_sb_14_tuber,CI,sb,ref,tuber,0.5432,37.3434,0.0542