In HiDef-Aerial-Surveying/stochLAB: Stochastic Collision Risk Model

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

stochLAB

{stochLAB} is a tool to run Collision Risk Models (CRMs) for seabirds on offshore wind farms.

Overview

The {stochLAB} package is an adaptation of the R code developed by Masden (2015) to incorporate variability and uncertainty in the avian collision risk model originally developed by Band (2012). The package is for use by individuals modelling collision risk of seabirds at offshore wind farms. The primary functions take input information on the morphology, behaviour and densities of seabirds as well data pertaining to the proposed wind farm (i.e., turbine dimensions, speed and number).

These collision risk models are useful for marine ornithologists who are working in the offshore wind industry, particularly in UK waters. However, the package itself relies on generic biological and windfarm data and can be applied anywhere (i.e., in any marine environment) as long as the parameters are appropriate for the species and windfarms of interest.

Code developed under {stochLAB} substantially re-factored and re-structured Masden's (heavily script-based) implementation into a user-friendly, streamlined, well documented and easily distributed tool. Furthermore, the package lays down the code infrastructure for easier incorporation of new functionality, e.g. extra parameter sampling features, model expansions, etc.

In addition, previous code underpinning core calculations for the extended model has been replaced by an alternative approach, resulting in significant gains in computational speed over Masden's code. This optimization is particularly beneficial under a stochastic context, when core calculations are called repeatedly during simulations.

For a more detailed overview type ?stochLAB, once installed!

Installation

You can install the released version of stochLAB from CRAN with:

install.packages("stochLAB")

You can install the development version with:

# install.packages("devtools")
devtools::install_github("HiDef-Aerial-Surveying/stochLAB")

This package depends on the following packages, which should be installed automatically:

cli dplyr glue logr magrittr msm pracma purrr rlang stats tibble tidyr

Bug reports

To report any bugs, please log an ISSUE

Input parameters

Many of the input parameters for the stoch_crm() function need to be obtained from developers (e.g., blade pitch, rotor radius, wind speed, etc...). However, there are many parameters around the morphology and biology of birds that are built into the sCRM package for UK seabirds, which can be found HERE. sCRM is an R Shiny application that wraps up the stoch_crm() and band_crm() functions. These biological parameters can be accessed by installing the sCRM package and running sCRM::spp_dflts, which will bring up a tibble object with all the relevant information.

If performing a collision risk assessment in UK waters, default biological data for the following parameters can be obtained from the sCRM package:

flt_speed_pars, body_lt_pars, wing_span_pars, avoid_bsc_pars, avoid_ext_pars, noct_act_pars, flight_type, gen_fhd_boots

Other parameters around the species of interest need to be derived from site-based surveys:

prop_crh_pars, bird_dens_dt, and site_fhd_boots

All wind farm parameters need to be obtained by the wind farm developers:

n_blades, air_gap_pars, rtr_radius_pars, bld_width_pars, bld_pitch_pars, rtn_speed_pars, windspd_pars, trb_wind_avbl, trb_downtime_pars, wf_n_trbs, wf_width, wf_latitude, tidal_offset, season_specs, bld_chord_prf, lrg_arr_corr

The following parameters refer to the outputs:

out_format, out_sampled_pars, out_period, verbose, log_file

More information on input parameter specifics can be found in the vignettes for stoch_crm and band_crm.

Outputs

Once the collision risk model is run, the key outputs are presented as a table which contains the mean, standard deviation and median number of collisions summarised by month, season, or year. Quantiles of the bootstrapped collisions are also presented in the tables. These tables are accessed through calling from the model object. Run the [Examples] to view exemplar outputs.

stochOUT <- stochLAB::stoch_crm(...)

stochOUT$collisions$opt1 #For outputs from option 1 of the stochastic collision risk model 
stochOUT$collisions$opt2 #For outputs from option 2 of the stochastic collision risk model 
stochOUT$collisions$opt3 #For outputs from option 3 of the stochastic collision risk model

Examples

Simple example

This is a basic example of running the stochastic collision model for one seabird species and one turbine/wind-farm scenario, with fictional input parameter data.

library(stochLAB)

# ------------------------------------------------------
# Setting some of the required inputs upfront

b_dens <- data.frame(
  month = month.abb,
  mean = runif(12, 0.8, 1.5),
  sd = runif(12, 0.2, 0.3))

# Generic FHD bootstraps for one species, from Johnson et al (2014)
fhd_boots <- generic_fhd_bootstraps[[1]]

# wind speed vs rotation speed vs pitch
wind_rtn_ptch <- data.frame(
  wind_speed = seq_len(30),
  rtn_speed = 10/(30:1),
  bld_pitch = c(rep(90, 4), rep(0, 8), 5:22))

# wind availability
windavb <- data.frame(
  month = month.abb,
  pctg = runif(12, 85, 98))

# maintenance downtime
dwntm <- data.frame(
  month = month.abb,
  mean = runif(12, 6, 10),
  sd = rep(2, 12))

# seasons specification
seas_dt <- data.frame(
  season_id = c("a", "b", "c"),
  start_month = c("Jan", "May", "Oct"), end_month = c("Apr", "Sep", "Dec"))

# ----------------------------------------------------------
# Run stochastic CRM, treating rotor radius, air gap and
# blade width as fixed parameters (i.e. not stochastic)

stoch_crm(
  model_options = c(1, 2, 3),
  n_iter = 1000,
  flt_speed_pars = data.frame(mean = 7.26, sd = 1.5),
  body_lt_pars = data.frame(mean = 0.39, sd = 0.005),
  wing_span_pars = data.frame(mean = 1.08, sd = 0.04),
  avoid_bsc_pars = data.frame(mean = 0.99, sd = 0.001),
  avoid_ext_pars = data.frame(mean = 0.96, sd = 0.002),
  noct_act_pars = data.frame(mean = 0.033, sd = 0.005),
  prop_crh_pars = data.frame(mean = 0.06, sd = 0.009),
  bird_dens_opt = "tnorm",
  bird_dens_dt = b_dens,
  flight_type = "flapping",
  prop_upwind = 0.5,
  gen_fhd_boots = fhd_boots,
  n_blades = 3,
  rtr_radius_pars = data.frame(mean = 80, sd = 0), # sd = 0, rotor radius is fixed
  air_gap_pars = data.frame(mean = 36, sd = 0),    # sd = 0, air gap is fixed
  bld_width_pars = data.frame(mean = 8, sd = 0),   # sd = 0, blade width is fixed
  rtn_pitch_opt = "windSpeedReltn",
  windspd_pars = data.frame(mean = 7.74, sd = 3),
  rtn_pitch_windspd_dt = wind_rtn_ptch,
  trb_wind_avbl = windavb,
  trb_downtime_pars = dwntm,
  wf_n_trbs = 200,
  wf_width = 15,
  wf_latitude = 56.9,
  tidal_offset = 2.5,
  lrg_arr_corr = TRUE,
  verbose = TRUE,
  seed = 1234,
  out_format = "summaries",
  out_sampled_pars = TRUE,
  out_period = "seasons",
  season_specs = seas_dt,
  log_file = paste0(getwd(), "scrm_example.log")
)

Multiscenario example

This is an example usage of stoch_crm() for multiple scenarios. The aim is two-fold:

Suggest how input parameter datasets used in the previous implementation can be reshaped to fit stoch_crm()'s interface. Suggested code is also relevant in the context of multiple scenarios applications, since the wide tabular structure of these datasets is likely the favoured format for users to compile input parameters under different scenarios.
Propose a functional programming framework to run stoch_crm() for multiple species and wind-farm/turbines features.

Please note the example runs on fictional data.

library(stochLAB)

# --------------------------------------------------------- #
# ----      Reshaping into list-column data frames       ----
# --------------------------------------------------------- #
#
# --- bird features
bird_pars <- bird_pars_wide_example %>%
  dplyr::relocate(Flight, .after = dplyr::last_col()) %>%
  tidyr::pivot_longer(AvoidanceBasic:Prop_CRH_ObsSD) %>%
  dplyr::mutate(
    par = dplyr::if_else(grepl("SD|sd|Sd", name), "sd", "mean"),
    feature = gsub("SD|sd|Sd","", name)) %>%
  dplyr::select(-name) %>%
  tidyr::pivot_wider(names_from = par, values_from = value) %>%
  tidyr::nest(pars = c(mean, sd)) %>%
  tidyr::pivot_wider(names_from = feature, values_from = pars) %>%
  tibble::add_column(prop_upwind = 0.5)

# --- bird densities: provided as mean and sd Parameters for Truncated Normal lower
# bounded at 0
dens_pars <- dens_tnorm_wide_example %>%
  tibble::add_column(
    dens_opt = rep("tnorm", nrow(.)),
    .after = 1) %>%
  tidyr::pivot_longer(Jan:DecSD) %>%
  dplyr::mutate(
    par = dplyr::if_else(grepl("SD|sd|Sd", name), "sd", "mean"),
    month = gsub("SD|sd|Sd","", name)) %>%
  dplyr::select(-name) %>%
  tidyr::pivot_wider(names_from = par, values_from = value) %>%
  tidyr::nest(mth_dens = c(month, mean, sd))

# --- FHD data from Johnson et al (2014) for the species under analysis
gen_fhd_boots <- generic_fhd_bootstraps[bird_pars$Species]

# --- seasons definitions (made up)
season_dt <- list(
  Arctic_Tern = data.frame(
    season_id = c("breeding", "feeding", "migrating"),
    start_month = c("May", "Sep", "Jan"),
    end_month = c("Aug", "Dec", "Apr")),
  Black_headed_Gull = data.frame(
    season_id = c("breeding", "feeding", "migrating"),
    start_month = c("Jan", "May", "Oct"),
    end_month = c("Apr", "Sep", "Dec")),
  Black_legged_Kittiwake = data.frame(
    season_id = c("breeding", "feeding", "migrating"),
    start_month = c("Dec", "Mar", "Sep"),
    end_month = c("Feb", "Aug", "Nov")))

# --- turbine parameters
## address operation parameters first
trb_opr_pars <- turb_pars_wide_example %>%
  dplyr::select(TurbineModel, JanOp:DecOpSD) %>%
  tidyr::pivot_longer(JanOp:DecOpSD) %>%
  dplyr::mutate(
    month = substr(name, 1, 3),
    par = dplyr::case_when(
      grepl("SD|sd|Sd", name) ~ "sd",
      grepl("Mean|MEAN|mean", name) ~ "mean",
      TRUE ~ "pctg"
    )) %>%
  dplyr::select(-name) %>%
  tidyr::pivot_wider(names_from = par, values_from = value) %>%
  tidyr::nest(
    wind_avbl = c(month, pctg),
    trb_dwntm = c(month, mean, sd))

## address turbine features and subsequently merge operation parameters
trb_pars <- turb_pars_wide_example %>%
  dplyr::select(TurbineModel:windSpeedSD ) %>%
  dplyr::relocate(RotorSpeedAndPitch_SimOption, .after = 1) %>%
  tidyr::pivot_longer(RotorRadius:windSpeedSD) %>%
  dplyr::mutate(
    par = dplyr::if_else(grepl("SD|sd|Sd", name), "sd", "mean"),
    feature = gsub("(SD|sd|Sd)|(Mean|MEAN|mean)","", name)
  ) %>%
  dplyr::select(-name) %>%
  tidyr::pivot_wider(names_from = par, values_from = value) %>%
  tidyr::nest(pars = c(mean, sd)) %>%
  tidyr::pivot_wider(names_from = feature, values_from = pars) %>%
  dplyr::left_join(., trb_opr_pars)

# --- windspeed, rotation speed and blade pitch relationship
wndspd_rtn_ptch_example

# --- windfarm parameters
wf_pars <- data.frame(
  wf_id = c("wf_1", "wf_2"),
  n_turbs = c(200, 400),
  wf_width = c(4, 10),
  wf_lat = c(55.8, 55.0),
  td_off = c(2.5, 2),
  large_array_corr = c(FALSE, TRUE)
)


# -------------------------------------------------------------- #
# ----      Run stoch_crm() for multiple scenarios           ----
# -------------------------------------------------------------- #

# --- Set up scenario combinations
scenarios_specs <- tidyr::expand_grid(
  spp = bird_pars$Species,
  turb_id = trb_pars$TurbineModel,
  wf_id = wf_pars$wf_id) %>%
  tibble::add_column(
    scenario_id = paste0("scenario_", 1:nrow(.)),
    .before = 1)

# --- Set up progress bar for the upcoming iterative mapping step
pb <- progress::progress_bar$new(
  format = "Running Scenario: :what [:bar] :percent eta: :eta",
  width = 100,
  total = nrow(scenarios_specs))

# --- Map stoch_crm() to each scenario specification via purrr::pmap
outputs <- scenarios_specs %>%
  purrr::pmap(function(scenario_id, spp, turb_id, wf_id, ...){

    pb$tick(tokens = list(what = scenario_id))

    # params for current species
    c_spec <- bird_pars %>%
      dplyr::filter(Species == {{spp}}) 

    # density for current species
    c_dens <- dens_pars %>%
      dplyr::filter(Species == {{spp}})

    # params for current turbine scenario
    c_turb <- trb_pars %>%
      dplyr::filter(TurbineModel == {{turb_id}})

    # params for current windfarm scenario
    c_wf <- wf_pars %>%
      dplyr::filter(wf_id == {{wf_id}})

    # inputs in list-columns need to be unlisted, either via `unlist()` or
    # indexing `[[1]]`
    # switching off `verbose`, otherwise console will be 
    # cramped with log messages

    stoch_crm(
      model_options = c(1, 2, 3),
      n_iter = 1000,
      flt_speed_pars = c_spec$Flight_Speed[[1]],
      body_lt_pars = c_spec$Body_Length[[1]],
      wing_span_pars = c_spec$Wingspan[[1]],
      avoid_bsc_pars = c_spec$AvoidanceBasic[[1]],
      avoid_ext_pars = c_spec$AvoidanceExtended[[1]],
      noct_act_pars = c_spec$Nocturnal_Activity[[1]],
      prop_crh_pars = c_spec$Prop_CRH_Obs[[1]],
      bird_dens_opt = c_dens$dens_opt,
      bird_dens_dt = c_dens$mth_dens[[1]],
      flight_type = c_spec$Flight,
      prop_upwind = c_spec$prop_upwind,
      gen_fhd_boots = gen_fhd_boots[[spp]],
      n_blades = c_turb$Blades,
      rtr_radius_pars = c_turb$RotorRadius[[1]],
      air_gap_pars = c_turb$HubHeightAdd[[1]],
      bld_width_pars = c_turb$BladeWidth[[1]],
      rtn_pitch_opt = c_turb$RotorSpeedAndPitch_SimOption,
      bld_pitch_pars = c_turb$Pitch[[1]],
      rtn_speed_pars = c_turb$RotationSpeed[[1]],
      windspd_pars = c_turb$windSpeed[[1]],
      rtn_pitch_windspd_dt = wndspd_rtn_ptch_example,
      trb_wind_avbl = c_turb$wind_avbl[[1]],
      trb_downtime_pars = c_turb$trb_dwntm[[1]],
      wf_n_trbs = c_wf$n_turbs,
      wf_width = c_wf$wf_width,
      wf_latitude = c_wf$wf_lat,
      tidal_offset = c_wf$td_off,
      lrg_arr_corr = c_wf$large_array_corr,
      verbose = FALSE,
      seed = 1234,
      out_format = "summaries",
      out_sampled_pars = FALSE,
      out_period = "seasons",
      season_specs = season_dt[[spp]],
      log_file = NULL
    )
  })

# --- close progress bar
pb$terminate()

# --- identify elements of output list
names(outputs) <- scenarios_specs$scenario_id

outputs

Band model example

This is an example usage of band_crm(). This is for a single species and single set of turbine parameters. This replicates the Band (2012) worksheet. The stoch_crm() function wraps around this function, where band_crm() acts in essence as a single draw of stoch_crm().

Please note the example runs on fictional data.

library(stochLAB)
# ------------------------------------------------------
# Run with arbitrary parameter values, for illustration
# ------------------------------------------------------

# Setting a dataframe of parameters to draw from
params <- data.frame(
  flight_speed = 13.1,         # Flight speed in m/s
  body_lt = 0.85,              # Body length in m
  wing_span = 1.01,            # Wing span in m
  flight_type = "flapping",    # flapping or gliding flight
  avoid_rt_basic = 0.989,      # avoidance rate for option 1 and 2
  avoid_rt_ext = 0.981,        # extended avoidance rate for option 3 and 4
  noct_activity = 0.5,         # proportion of day birds are inactive
  prop_crh_surv = 0.13,        # proportion of birds at collision risk height (option 1 only)
  prop_upwind = 0.5,           # proportion of flights that are upwind
  rotor_speed = 15,            # rotor speed in m/s
  rotor_radius = 120,          # radius of turbine in m
  blade_width = 5,             # width of turbine blades at thickest point in m
  blade_pitch = 15,            # mean radius pitch in Radians
  n_blades = 3,                # total number of blades per turbine
  hub_height = 150,            # height of hub in m above HAT
  n_turbines = 100,            # number of turbines in the wind farm
  wf_width = 52,               # width across longest section of wind farm
  wf_latitude = 56,            # latitude of centroid of wind farm
  tidal_offset = 2.5,          # mean tidal offset from HAT of the wind farm
  lrg_arr_corr = TRUE          # apply a large array correction?
)

# Monthly bird densities
b_dens <- data.frame(
  month = month.abb,
  dens = runif(12, 0.8, 1.5)
)

# flight height distribution from Johnston et al
gen_fhd_dat <- Johnston_Flight_heights_SOSS %>%
  dplyr::filter(variable=="Gannet.est") %>%
  dplyr::select(height,prop)

# monthly operational time of the wind farm
turb_oper <- data.frame(
  month = month.abb,
  prop_oper = runif(12,0.5,0.8)
)


stochLAB::band_crm(
  model_options = c(1,2,3),
  flight_speed = params$flight_speed,
  body_lt = params$body_lt,
  wing_span = params$wing_span,
  flight_type = params$flight_type,
  avoid_rt_basic = params$avoid_rt_basic,
  avoid_rt_ext = params$avoid_rt_ext,
  noct_activity = params$noct_activity,
  prop_crh_surv = params$prop_crh_surv,
  dens_month = b_dens,
  prop_upwind = params$prop_upwind,
  gen_fhd = gen_fhd_dat,
  site_fhd = NULL,  # Option 4 only
  rotor_speed = params$rotor_speed,
  rotor_radius = params$rotor_radius,
  blade_width = params$blade_width,
  blade_pitch = params$blade_pitch,
  n_blades = params$n_blades,
  hub_height = params$hub_height,
  chord_prof = chord_prof_5MW,
  n_turbines = params$n_turbines,
  turb_oper_month = turb_oper,
  wf_width = params$wf_width,
  wf_latitude = params$wf_latitude,
  tidal_offset = params$tidal_offset,
  lrg_arr_corr = params$lrg_arr_corr
  )