Catchment Prediction In Changing Environments


The aim of the CAPICHE experiment, running in conjunction with the Root Zone Storage Experiment, is to gain a comparative understanding of how different models simulate catchment hydrological response to changing environments. These changes may occur, for example, due to land cover change, or changes in climate forcing.

In order to understand how different models simulate change, a time-varying sensitivity analysis is conducted to identify the time-stability of model parameters, and therefore which parameters change to simulate hydrological response. Given a particular change in catchment forcing, such analysis allows us to understand whether the model parameter changes are plausible, and therefore identify potential compensatory parameter effects in simulating change.

The protocol for simulating change, along the necessary code for posterior model analysis, has been developed and evaluated at HJ Andrews Watershed 01. The whole catchment was clear felled over a period of four years (1962-1966), with subsequent planting and re-growth. The way in which several models, including FLEX, Hymod, HBV, PIHM and E-HYPE simulate the changes is being evaluated. 

Science Questions

  1. How do different hydrological models simulate catchments under changing environments (changing in forcing, landscape change, water demand)?
  2. How is the ability of different models to simulate catchments under change affected by data availability?
  3. What hydrological signatures, and therefore what model behaviours are most sensitive to model structure, data availability and parameter uncertainty when simulating under changing conditions?

Final Protocol

1) Partners

University of Bristol, SMHI, TU Delft

2) Data and Method Preparation

Initial results are derived for HJ Andrews Watershed 01.

3) Experiment Execution/Analysis Steps

We present the development of a methodology for model inter-comparison to analyse how models simulate landscape change, built upon time-varying sensitivity analysis of model parameters. First, for a given catchment, hydrologic signatures are calculated over consecutive windows covering the period of forcing change to analyse how the catchment responds hydrologically to change. Then, each model is calibrated to each window, and within each window, to each signature, which allows us to analyse the time-varying relationship between catchment forcing, hydrological response and model parameter metrics (e.g. posterior distribution, parameter interactions).

  • Derive Perceptual Models: based on exploratory data analysis, derive perceptual models of each catchment, depending on data availability and known changes to the system/forcing and hydrological response.
  • Identify Process Constraints: For each catchment identify process constraints that can be applied to constrain model behaviour.
  • Classify Constraints: to help identify the optimal blend of model and data, classify constraints based on different levels of data availability.
  • Identify Parameter Constraints: Identify parameter constraints to be applied to each model.
  • Run Sensitivity Analysis: For each model applied to each catchment perform sensitivity analysis.
  • Evaluate model performance: Based on a set of behavioural models that satisfy process constraints, evaluate time-varying sensitivity of model performance.

4) Result Reporting


Figure 1. Changes in simulated soil moisture capacity in Hymod over the period of tree-felling and recovery for HJ Andrews WS01.

Figure 2. Parallel co-ordinate plot of hymod model parameters. Colours show different behavioural thresholds (blue = best performing models).


Published Papers


Hutton et al. in Preparation