Download presentation
Presentation is loading. Please wait.
1
Uncertainty between simulations
SIndex is (meanfocusyears-meanbaseline)/sdbaseline and in this case, its the SIndex for evapotranspiration ET from the earth2observe ensemble. These result from using the JULES model to calculate ET, with the model driven by data from the CMORPH precipitation product.
2
Uncertainty between simulations
SIndex is (meanfocusyears-meanbaseline)/sdbaseline and in this case, its the SIndex for ET. ... and the ORCHIDEE model driven by the CMORPH precipitation product.
3
Uncertainty between simulations
SIndex is (meanfocusyears-meanbaseline)/sdbaseline and in this case, its the SIndex for ET. ... and the HTESSEL-CaMa model driven by the CMORPH precipitation product. Similar variation is visible if we vary the precipitation product rather than the model.
4
The models broadly agree which areas have large-than-normal and smaller-than normal ET, but there are spatial points where they disagree: Note that this calculation means a model configuration that differs from the mean by a little (X, say) in every month will count more than one that differs by a lot in one month (Y, say) and agrees with the ensemble average for 11 months (as long as Y<12X anyway), i.e. it’s more sensitive to persistent differences than to extremes. An alternative would be to score an SD<thresh as a ‘match’ (=0) and anything else a ‘miss’ (=1). If thresh>X then this will highlight the extremes a little more, but for now go with this (simpler) option. Calculations for top row: For each pixel and for each month, calculate the SD of the index across the products, then average over the 12 months to give an index of annual variability across models for this product at each pixel. Calculations for bottom row: For each pixel and for each month, calculate the SD of the index across the models, then average over the 12 months to give an index of annual variability across products for this model at each pixel.
5
Now average the model and product uncertainties (i.e. average along the rows on the last slide gridcell by gridcell) and sum to get total uncertainty (rightmost plot). So: uncertainty in ET prediction by these land surface models seems to be affected much more by changing precipitation product than changing the model in use. However, the same conclusion would be reversed looking at particular subregions (e.g. red circles and note the colour scale hides how widespread this is) or looking at a different continent. There aren’t actually that many places where we have the same predictions irrespective of both model and precipitation product used (e.g. yellow circle), but note that this index of uncertainty is tough: in operational terms, the predictions are acceptable enough at all spatial points.
6
Summary plot for scaled anomalies (SIndex) of evapotranspiration ET:
Summary plot for scaled anomalies (SIndex) of evapotranspiration ET: n.b. Quite a contribution from MSWEP.
7
The same plots looking at scaled anomalies (SIndex) of soil moisture content:
8
The same plots looking at scaled anomalies (SIndex) of soil surface moisture content:
9
Really all except LISFLOOD, WaterGAP and PCR, but I don’t have data from the other two The same plots looking at scaled anomalies (SIndex) of LE (latent heat flux, QE): n.b. For all models except WaterGAP, LE is as given by the model; for WaterGAP LE is assumed to equal the energy equivalent of ET)
10
The same plots looking at scaled anomalies (SIndex) of total water storage:
11
cover fraction for the British Isles:
Because of lack of snow, snow-related variables can’t be plotted for tropical Africa, so plot snow cover fraction for the British Isles: Note: the uncertainty is now much more in the choice of model.
12
The same plots looking at scaled anomalies (SIndex) of snow water equivalent:
13
Note also that the colour scales of these plots are not the same between variables: some variables returned higher uncertainty values than others:
14
Finally, look at ET for the UK: the uncertainty is now much more in the choice of model.
15
Some conclusions: Generally, it seems that there is more uncertainty in the precipitation product we use than our choice of land surface model, although this pattern is not spatially consistent and some regions show the opposite. For some variables, e.g. snow-related variables this is apparently consistenty reversed. We can understand this in more than one way: perhaps there is more disagreement between models on how to simulate snow than other water cycle processes... ... or perhaps the precipitation products simply show more agreement on their statistics for snowfall then for other driving variables. We need to understand much more clearly the interplay between choice of model(s) and product(s) when making predictions of land surface variables.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.