Characterizing Climate Regime Changes from Parallel Climate Model (PCM) Predictions

December 3, 2003

Forrest M. Hoffman, William W. Hargrove, David J. Erickson III, Robert J. Oglesby

Contents

• Individual PCM Run Analyses
• Ensemble Cluster Analysis
• Ensemble Average Cluster Analysis
• Time Interval Average of the Ensemble Average
• Time Series Comparison Methodology

This analysis uses Parallel Climate Model (PCM) output in an attempt to understand how land surface conditions affecting terrestrial ecology might change based on model predictions. Three monthly output fields from five 100 year Business-As-Usual (BAU) PCM runs (for years 2000 through 2098) were chosen for inclusion in a cluster analysis. The three fields are surface temperature [K], precipitation [kg m^-2 s^-1], and soil moisture [volume fraction] (LSM root zone volumetric soil water). Ocean and sea ice cells were removed prior to clustering so only the 2,796 land cells (out of a total of 8,192 cells at T42 resolution) were included.

Initially, each of the five BAU runs was clustered independently into 32 regimes. Table 1 contains visualization products resulting from all five cluster analyses. The Map Animations, available in random or similarity colors, show which regime each land cell occupies every month for the entire 100 year period. In addition, a histogram of climate regime or cluster occupancy is shown next to the map for each month. The Regime Evolution graphs, also available in random or similarity colors, show how climate regimes grow or shrink in spatial area over the 100 year simulation period. These curves represent five-year running averages of regime (or climate state) occupancy globally. Regimes with low variance, i.e., those having an occupancy change of <100 grid cells, have been removed from these graphs.

The Climate Phase Space animations show where each climate regime, represented by a small cube, is located in the three-dimensional climate phase space for each BAU run. These are the cluster centroids for each state or climate regime. The Regime Definition tables contain the temperature, precipitation, and soil moisture values for these climate regimes. Finally, two additional animations included in Table 1 show cliamte regime occupancy for a single grid cell from the model output.

The grid cell of interest is located in the Middle East. The Phase Trajectory animation shows how that spatial location (represented by a "spider") moves among climate states as model time progresses. The transitions between favored states are displayed as a black line segments which form a "web." As these transitions are repeatedly traversed, the segments get thicker so that favored transitions among climate states can be observed.

In the Climate Manifold animation, the soil moisture variable on the x-axis has been droped so that time can be shown on this axis. The animations show all the transitions among states for this single point on the globe drawn out through time so that it forms a coil or manifold. Rare transitions to extreme states can be easily seen in this visualization.

Warning: The movies below are large animated GIF files. These animations are best viewed using a browser which streams the images instead of attempting to load the entire file into memory at once. Netscape 4.x will correctly stream these animations, but other browsers may not.

The Regime Evolution graphs in Table 1 show regime occupancy changes on a global basis. The climate changes in some land areas may compensate for climate changes in other land areas globally causing a low overall regime variance through time. As a result, similar evolution curves were produced for each continent. For these graphs, shown in Table 2, only curves which exhibited more than a 1% change in continental spatial area are included. This thresholding eliminates curves for regimes which do not undergo a significant change in spatial area.

In order to compare predictions from the different model runs, climate regimes common to all model runs must be computed. By including all five BAU runs (for a total of 500 years) in a single clustering analysis, we obtain a single common set of climate regimes or states. These common states serves as a basis for head-to-head comparison of the individual model runs. This clustering contained over 16M data points. These points have been plotted in an animation of the three-dimensional climate phase space colored by run number (see Movie).

Since all the model results are displayed with respect to a common set of climate regimes, they can be easily compared. Moreover, the phase trajectories from all five model runs can be plotted together in a single animation to more-easily discern when model predictions diverge and converge (see Movie).

In addition to the global Regime Evolution graphs above, similar evolution curves were produced for each continent. For these graphs, shown in Table 4, only curves which exhibited more than a 1% change in continental spatial area are included. This thresholding eliminates curves for regimes which do not undergo a significant change in spatial area.

A second map animation was generated showing only the regimes which undergo a significant spatial change globally (see Movie). This animation makes it easy to see where and when these environmental conditions occur.

In addition to the global Regime Evolution graph above, similar evolution curves were produced for each continent. For these graphs, shown in Table 6, only curves which exhibited more than a 1% change in continental spatial area are included. This thresholding eliminates curves for regimes which do not undergo a significant change in spatial area.