phenology-climatology.yml

#####
#See https://shorturl.at/irMQW for a guide to all the options below
####
team_name: climatology
# ids are optional except for the first author, who is also the 'contact'
team_list:    
  - individualName:
      givenName: Quinn
      surName: Thomas 
      id: rqthomas@vt.edu
#You should not add anything on the metadata and forecast lines (next two lines should only say metadata and forecast)      
metadata:
  forecast:
    timestep: 1 day #time step of model
    forecast_horizon: 35 days #number of time steps in the future
#You should not add anything on the model_description lines (next line should only say model_description)      
    model_description:
      # model identifier:
      forecast_model_id: "climatology"
      name: historical climatology
      type: empirical. #General type of model empirical, machine learning, process
      repository: www.github.com/eco4cast/neon4cast-phenology  # put your GitHub Repository in here
#
#INITIAL CONDITIONS      
#Uncertainty in the initialization of state variables (Y). Initial condition 
#uncertainty will be a common feature of any dynamic model, where the future 
#state depends on the current state, such as population models, process-based 
#biogeochemical pool & flux models, and classic time-series analysis. 
#
#complexity = number of state variables in the model. Examples of this would be the number of
#species in a community model, number of age/size classes in a population model, 
#number of pools in a biogeochemical model.
    initial_conditions:
      status: absent #options: absent, present, data_driven, propagates, assimilates
      complexity:  
      propagation:
        type: 
        size: 
      #Leave everything below blank  UNLESS status = assimilates  
      assimilation:
        type: 
        reference: 
        complexity: 
#        
#DRIVERS
#uncertainty in model drivers, covariates, and exogenous scenarios (X). 
#Driver/covariate uncertainties may come directly from a data product, as a 
#reported error estimate or through driver ensembles, or may be estimated based 
#on sampling theory, cal/val documents, or some other source.
#
#complexity = Number of different driver variables or covariates in a model. For example, in 
#a multiple regression this would be the number of X’s. For a climate-driven 
#model, this would be the number of climate inputs (temperature, precip, solar 
#radiation, etc.).
    drivers:
      status: absent #options: absent, present, data_driven, propagates, assimilates
      complexity: 
      #Leave everything below blank if status = absent, present, or data_driven
      propagation: 
        type: ensemble 
        size: 
      #Leave everything below blank  UNLESS status = assimilates  
      assimilation:
        type: 
        reference: 
        complexity: 
#        
#PARAMETERS
#Uncertainty in model parameters (). For most ecological processes the parameters
#(a.k.a. coefficients) in model equations are not physical constants but need to 
#be estimated from data. 
#
#complexity = number of estimated parameters/coefficients in a model at a single point in 
#space/time. For example, in a regression it would be the number of beta’s.
    parameters:
      status: absent 
      complexity: 
      #Leave everything below blank if status = absent, present, or data_driven
      propagation:
        type: 
        size: 
      #Leave everything below blank  UNLESS status = assimilates  
      assimilation:
        type: 
        reference: 
        complexity: 
#
#RANDOM EFFECTS
#Unexplained variability and heterogeneity in model parameters (). Hierarchical 
#models, random effect models, and meta transfer learning approaches all attempt
#to acknowledge that the ‘best’ model parameters may change across space, time, 
#individual, or other measurement unit.

#complexity = number of random effect terms, which should be equivalent to the number of 
#random effect variances estimated. For example, if you had a hierarchical 
#univariate regression with a random intercept you would have two parameters 
#(slope and intercept) and one random effect (intercept). At the moment, we are 
#not recording the number of distinct observation units that the model was 
#calibrated from. So, in our random intercept regression example, if this model 
#was fit at 50 sites to be able to estimate the random intercept variance, that 
#would affect the uncertainty about the mean and variance but that ‘50’ would 
#not be part of the complexity dimensions.
#
    random_effects:
      status: absent 
      complexity: 
      #Leave everything below blank if status = absent, present, or data_driven
      propagation:
        type: 
        size: 
      #Leave everything below blank  UNLESS status = assimilates  
      assimilation:
        type: 
        reference:
        complexity: 
#
#PROCESS ERROR
#Dynamic uncertainty in the process model () attributable to both model 
#misspecification and stochasticity. Pragmatically, this is the portion of the 
#residual error from one timestep to the next that is not attributable to any of
#the other uncertainties listed above, and which typically propagates into the future. 
#
#complexity =  dimension of the error covariance matrix. So if we had a n x n 
#covariance matrix, n is the value entered for <complexity>. Typically n should 
#match the dimensionality of the initial_conditions unless there are state 
#variables where process error is not being estimated or propagated
    process_error:
      status: data_driven #options: absent, present, data_driven, propagates, assimilates
      complexity: 2 #Leave blank if status = absent
      #Leave everything below blank if status = absent, present, or data_driven
      propagation:
        type: 
        size: 
      #Leave everything below blank  UNLESS status = assimilates
      assimilation:
        type: 
        reference: 
        complexity:
        covariance: 
        localization:
#
#OBSERVATION ERROR
#Uncertainty in the observations of the output variables (g). Note that many 
#statistical modeling approaches do not formally partition errors in observations
#from errors in the modeling process, but simply lump these into a residual error. 
#Because of this we make the pragmatic distinction and ask that residual errors 
#that a forecast model do not directly propagate into the future be recorded as 
#observation errors. Observation errors now may indeed affect the initial condition 
#uncertainty in the next forecast, but we consider this to be indirect. 
#
#complexity = dimension of the error covariance matrix. So if we had a n x n 
#covariance matrix, n is the value entered for <complexity>. Typically n should 
#match the dimensionality of the initial_conditions unless there are state 
#variables where process error is not being estimated or propagated
    obs_error:
      status: absent 
      complexity: 
      #Leave everything below blank if status = absent, present, or data_driven
      propagation:
        type: 
        size: 
      #Leave everything below blank  UNLESS status = assimilates
      assimilation:
        type: 
        reference: 
        complexity:
        covariance: 
        localization: