kalman: WRTDSKalman Bootstrapping

In USGS-R/EGRETci: Exploration and Graphics for RivEr Trends Confidence Intervals

WRTDSKalman Bootstrapping

Description

Function to get multiple bootstrap replicates at a daily time step using the WRTDS_K method. It is done by doing bootstrap resampling of the original Sample data frame. The number of these replicate samples that are created is called nBoot and in each case the WRTDS model is estimated. Then, for each of these models, there are nKalman time series of daily values computed, using all of the sample values in the original Sample data frame. The total number of replicates of the complete process is nBoot * nKalman. For example we might generate 500 replicates by setting nBoot = 20 and nKalman = 25.

Usage

genDailyBoot(eList, nBoot = 10, nKalman = 10, rho = 0.9, setSeed = NA,
  jitterOn = FALSE, V = 0.2)

Arguments

eList	is the data with a fitted model already done. Note that the eList$Sample may have multiple values on a given day and it can also have censored values.
nBoot	number of times the bootstrap resampling and model estimating is done.
nKalman	number of different realizations of the daily time series for each re-estimated model.
rho	numeric the lag one autocorrelation. Default is 0.9.
setSeed	value. Defaults is NA, which will not specify a randomized seed. This can be used to make repeatable output.
jitterOn	logical, if TRUE, adds "jitter" to the data in an attempt to avoid some numerical problems. Default = FALSE. See Details below.
V	numeric a multiplier for addition of jitter to the data, default = 0.2. See Details below.

Details

In some situations numerical problems are encountered in the bootstrap process, resulting in highly unreasonable spikes in the confidence intervals. The use of "jitter" can often prevent these problems, but should only be used when it is clearly needed. It adds a small amount of random "jitter" to the explanatory variables of the WRTDS model. The V parameter sets the scale of variation in the log discharge values. The standard deviation of the added jitter is V * standard deviation of Log Q. The default for V is 0.2. Larger values should generally be avoided, and smaller values may be sufficient.

Value

dailyBootOut a matrix of daily flux values (in kg/day). The number of columns of the matrix is the number of replicates produced which is nBoot * nKalman The number of rows is the number of days in the record. The set of days simulated is the same set of days that are in the eList$Daily data frame.

Examples

eList <- EGRET::Choptank_eList
# Very long running function:
## Not run: 
dailyBootOut <- genDailyBoot(eList,
                             nBoot = 20,
                             nKalman = 25)

## End(Not run)

USGS-R/EGRETci documentation built on April 17, 2023, 8 p.m.