floybix / hydromad Goto Github PK

show independent effect of each parameter in SMA example()

eventapply should drop() result

zoo 1d matrices are problematic

findThresh is ambiguous: may be multiple local solutions with target number of events.

eventsExplorer: include distributions of event lengths and sums

replace eventapply() with aggregate.zoo()

several issues to consider...

absorbScale should be usable directly

Should width be scaled down, or is current error acceptable?
runoffratio.sim is called with default arguments by armax.inverse.fit without the possibility of overriding width

profiling.

sim functions: compiled_code = TRUE

remove reference to hydromad.getOption("pure.R.code")
because the sim functions should ideally stand alone and not depend on the rest of the package.

CMD 2-segment form

dU/dP=(1−n)[1−f1(M)]+n[1−f2(M)]
where
f(x) = min[1, x/d] and f2(x) = min[1, x/d2]

defaults
d2 = d * 2, n = 0.1

vignette describing how to write new models (or adapt/combine existing ones)

also should describe how to write new routing models.

wrapper for DREAM

implement gr4j

hydromad.options("summary.stats")

add more stats e.g.
RMSE
MSLE
MSDE
IRMSE
(Dawson et al, 2009, EMS)

a mechanism to define new stats?

simplify xyplot cases for hydromad and runlist

can use xyplot.list(runlist) for juxtapose

can use
xyplot(residuals(runlist), superpose = TRUE)
for superposed residuals -- but loses zoo time frame

can use
xyplot(fitted(runlist), superpose = TRUE) +
layer(llines(observed(runlist[[1]])))
for superposed fits with observed

and
c(..., xyplot(observed(hmad, item = "P"), type = "h"))
for rainfall
(but need to fix up auto.key)

• expuh fitting
• lambda simulation and fitting
• runoffratio
• cmd, cwi
• sacramento
• events...
• etc

would be good to keep original zoo attributes of DATA in results from residuals / fitted / predict / observed

but might this sometimes cause problems?

probably need to call coredata() or as.ts() in summary, during fitting, etc.

defineFeasibleSet() with options to keep all sims within x% of values / events

glue quantiles using Hmisc::wtd.quantile as an option in predict()

allow state variables to be used in objective function

remove:
byDays() etc
numericSummary?
makePrefilter?
autocorrTime?
tsCleanup?

drop "coef<-" in favour of update(newpars=)

sma.intensity

U <- (P ^ power) ?
should limit U <= P

how to do box-cox transformation in objective function?

'objective' probably needs to be a list, specifying what it needs to have access to. This also relates to the issue of allowing access to state variables.

window.hydromad() ?

not just update(, newdata = window(...)): want to take a subset of a larger simulation without resetting the initial conditions.

for plotting: should be able to just use xlim.

allow multiple thresholds in eventseq -- or is interaction() enough?

allow a 'ulow' level for event separation / termination

skip = FALSE or extend = TRUE to change behaviour of mindur...

buildTsLikelihood()

aggregation + transformation +
specify assumed error distribution?

rethink errormasscurve -- smoothed residuals better?

implement pdm

IHACRES CWI warning - non-physical process representation - beware climate change simulation

Kirchner method

generic SMA with custom drainage / storage / evap functions

This is suitable for n=1, and doesn't matter if warmup>0

armax.sim works fine, but hydromad treats length(init)>1 as an unspecified parameter.

armax.sim(U=c(1,rep(0,20)),a_1=0.1,a_2=0.1,b_0=1,init=as.numeric(Murrindindi$Q[1:2]))
armax.sim(U=c(1,rep(0,20)),a_1=0.1,a_2=0.1,b_0=1,init=as.numeric(Murrindindi$Q[1]))

finish help page for SRIV

example of using complementary data, e.g. multiple rain gauges

progress bars for simulate(), fitBy...(), glue?

stats / objfuns should be functions not formulae?

want to be able to easily combine them on the fly.

allow multiple outputs from SMA, not just a single U series.
(multivariate time series?)
e.g. a groundwater pathway.

vignette for coding new models, including compiling C code

• inline package?

multibucket model(s), following on from bucket.sim(), maybe Xinanjiang type model?

With short and sub-daily datasets, "r.sq.monthly" cannot be calculated and results in cryptic errors, e.g.
In nseStat(aggrQ, doaggr(X), ref = aggrRef, ...) :
"merged time series have no data; incompatible times?"

hydrograph?

anything non-numeric, or "AsIs" objects, should be passed through and not included in coef().

write bucket model in C

quickstart on website
(example session code)

estimateDelay should pick up the first significant CCF, not the largest (in case of components in series)

cache objective in simulate()

r.squared --> NSE
r.sq.sqrt --> NSE_sqrt
r.sq.log --> NSE_log
fitStat --> NSE
ARMAX --> ARMAX type

fix up IHACRES vignette
including description of third-order expuh routing

Currently, only routing functions that return only X will work.
Would ideally like to allow multiple (related) outputs, e.g.

• Xs and Xq from return_components=T in cmd.sim
• groundwater levels
• stage

This could be done by providing an additional OUTPUT variable, which could be referred to in the objective function.

This would allow hydromad to be used without producing X at all. This might break some other functions?