Representing Uncertainty Associated with the Land Surface Component of Coupled Atmosphere-Land Models

1
Representing Uncertainty Associated with the Land
Surface Component of Coupled Atmosphere-Land
Models

• Dingchen Hou and Zoltan Toth

2
Motivation

• Under Dispersive forecasts from operational
  ensemble forecast systems, in near surface
  variables.
• Forecast models are Coupled Land-Atmospheric
  Forecast Systems
• Seasonal to Interannual forecast, e.g. CFS
• SST forcing, Land memories
• Short Range Forecast, e.g. NCEP WRF ensemble (Du)
• Heat Fluxes
• Medium Range Forecast, e.g. GFS, GEFS
• ????
• Representing Uncertainty in Ensemble Forecast
• Success in Atmospheric component
• Progress in Ocean Component
• Need to work on Land-surface component
• Sensitivity of Prediction to Land Surface
  Processes.
• Seasonal Forecast
• Short range Forecast

3
DTRA Chesapeake Bay 2001 ExperimentARPS model,
ETA ana/fct as initial/boundary condition
10km-1km one-way nested simulationControl run
Reduced initial soil moisture
4
DTRA Chesapeake Bay 2001 ExperimentARPS model,
ETA ana/fct as initial/boundary condition
10km-1km one-way nested simulationObservation,
control and reduced initial soil moisture

5
DTRA Chesapeake Bay 2001 ExperimentARPS model,
ETA ana/fct as initial/boundary condition
10km-1km one-way nested simulationObservation,
control and reduced initial soil moisture
6
Land-Surface Related Uncertainties

• Initial Conditions of Land Surface Variables
  (limited observation, its accuracy and
  representativeness)
• Soil moisture
• Soil temperature
• Land Surface Model Structures and Parameters
• LSM models (Noah, VIC, Mosaic, Sacramento, etc)
• Empirical parameters
• Land Surface Characteristics (Land/Sea Mask, Soil
  Type)
• Vegetation and its change

7
Scientific Objectives

• Developing a frame work for assessing,
  quantifying and representing the uncertainties.
• Apply the framework in various operational
  forecast systems.

8
Approaches

• Developing and applying ensemble generation
  methods to
• Initial conditions of prognostic variables of LSM
• land surface model(s)

9
Ensemble Generation MethodologyInitial Conditions

• Existing methods
• Lagged Average Forecast (LAF)
• Bred Vectors/ Ensemble Transform (ET)
• Singular Vectors
• Perturbed Observation
• Proposed Method
• Breeding/ET

10
Ensemble Generation MethodologyForecast Model

• Existing methods
• Multi-model Ensemble
• Perturbed Parameters
• Stochastic Perturbations
• Proposed Method
• Perturbed Parameters/Stochastic Perturbations

11
Proposed Modification of EnsembleSystems to
Represent LS-related Uncertainty

• Initiate ET by generating land surface variable
  perturbations with lagged average forecast in
  NLDAS/GLDAS or a similar system.
• Make the LS perturbations consistent with
  Atmospheric perturbations used in the forecasting
  system (GFS or WRF) by empirically re-scaling the
  size of the perturbations to ensure approximate
  balance.
• Control the size of the perturbation by
  estimation of analysis uncertainty.
• Cycle the land surface and atmospheric
  perturbations simultaneously in the operational
  forecasting system, with periodical breeding/ET
  procedure.
• Evaluate the performance of the ensemble system
  and modify the perturbation size.
• Add/replace members with perturbed model
  parameters or add stochastic perturbations to
  represent model related uncertainty.