Model Structures

Basic model types in streamMetabolizer

Many model structures are available in the streamMetabolizer package. These fall into three general types:

bayes - Inverse Bayesian modeling of GPP, ER, and K600
mle - Inverse modeling by maximum likelihood estimation of GPP, ER, and optionally K600
night - Nighttime regression for estimation of K600 and ER

Two additional model types can be useful in specific situations:

Kmodel - Builds a multi-day relationship between K600 and predictor variables; can be used in conjunction with mle and/or night
sim - Simulates dissolved oxygen “data” for model testing

Specific model structure

A complete specification of a model structure begins with a basic type but also includes other information about the model error assumptions, any hierarchical components, the numerical integration method, the process equations used to describe GPP and ER, and more. A model structure is fully specified by a concise if cryptic model name. Examples are “b_np_oipi_tr_plrckm.stan” and “m_np_oi_tr_plrckm.nlm”. You never need to remember these concise names; instead, you can construct and interpret such names using the mm_name and mm_parse_name functions, respectively.

Setting a model name

The mm_name() function creates a cryptic (but complete and concise) model name from several options that you can specify. For example:

mm_name(type='bayes') # the default Bayesian model

## [1] "b_np_oipi_tr_plrckm.stan"

mm_name(type='bayes', pool_K600='normal') # a Bayesian model with simple pooling of K600 to a shared mean

## [1] "b_Kn_oipi_tr_plrckm.stan"

mm_name(type='mle') # the default MLE model

## [1] "m_np_oi_tr_plrckm.nlm"

mm_name(type='mle', ode_method='euler') # an MLE model with a simpler ode_method

## [1] "m_np_oi_eu_plrckm.nlm"

See the help file at ?mm_name to learn about the structural options that you can adjust.

Interpreting a model name

The mm_parse_name() function splits a cryptic model name into its component features. For example:

mm_parse_name(c('b_np_oipi_tr_plrckm.stan', 'm_np_pi_tr_psrqkm.nlm'))

##    type pool_K600 err_obs_iid err_proc_acor err_proc_iid err_proc_GPP
## 1 bayes      none        TRUE         FALSE         TRUE        FALSE
## 2   mle      none       FALSE         FALSE         TRUE        FALSE
##   ode_method  GPP_fun   ER_fun deficit_src engine
## 1  trapezoid linlight constant      DO_mod   stan
## 2  trapezoid satlight  q10temp      DO_mod    nlm

See the help file at ?mm_parse_name, and refer again to ?mm_name for definitions of the column names in the output of mm_parse_name.

Listing all the options

You can see the full list of available model structures by calling mm_valid_names() (here we’ll just print a sampling of the full list):

valid_names <- mm_valid_names(type=c('bayes','mle','night'))
length(valid_names)

## [1] 882

c(valid_names[seq(1,length(valid_names),length.out=20)], '...')

##  [1] "b_np_oipi_tr_plrckm.stan"  "b_Kb0_oi_eu_psrckm.stan"  
##  [3] "b_Kbx_oipi_eu_plrckm.stan" "b_Kl_oipi_tr_psrcko.stan" 
##  [5] "b_Kl0_oipp_eu_plrcko.stan" "b_Klx_pi_eu_psrcko.stan"  
##  [7] "b_Kn_pipp_tr_plrckm.stan"  "b_Knx_oi_eu_plrckm.stan"  
##  [9] "b_np_oi_tr_psrcko.stan"    "m_np_oi_o6_plrckm.nlm"    
## [11] "m_np_oi_o10_psrckm.nlm"    "m_np_pi_o13_pqrckm.nlm"   
## [13] "m_np_pi_o16_plrqkm.nlm"    "m_np_pi_tr_pqrqkm.nlm"    
## [15] "m_np_oi_o2_plrcko.nlm"     "m_np_oi_o5_psrcko.nlm"    
## [17] "m_np_oi_o9_pqrcko.nlm"     "m_np_pi_o12_plrqko.nlm"   
## [19] "m_np_pi_o15_psrqko.nlm"    "n_np_pi_eu_rckf.lm"       
## [21] "..."

Using a model name

Once you have a model name in hand, pass that name to metab() to fit a model with the selected structure. See the Quickstart tutorial for more.