Visualising seasonal climate forecasts in Google Earth
This is our winning entry to Google's KML in Research Competition and is based on work done with David Stephenson and Rachel Lowe at Exeter University, facilitated through the Willis Research Network. Thanks to Erik Andersson at ECMWF and the EUROBRISA project. The text of our competition entry is below.
Download the KMZ file here (~1.5Mb)
Note: We've moved the position of one of the information panels so it is not obcsured by the timeline in Google Earth v5.
Visualising seasonal climate forecasts
Seasonal climate forecasts have important implications for government, health, agricultural and industrial planning and operations. However, uncertainty inherent in environmental data and in numerical prediction models make these models inherently probabilistic. Good decisions require that this uncertainty be communicated to those who rely upon the forecasts.
We use Google Earth to explore seasonal precipitation forecasts over a 10 year period in South America, using simple but novel graphics that give an indication of the (un)certainty associated with the forecasts and the confidence one might place in them. Forecasts can be compared temporally and geographically - between each year in the period and between areas and regions. Comparisons can also be made with actual precipitation levels.
Description of Research
This research project was set up to develop new ways of visualising seasonal climate forecasts. Google Earth was found to be an appropriate and powerful tool particularly the ability to specify and format geometrical shapes, to separate the graphics into layers and timeslices, to produce information panels and to respond to users' interactions such as zooming behaviour. Our aim is to provide effective means of interpreting and assessing the quality of seasonal climate forecast, within the familiar intuitive and rich interface of Google Earth. Being able to do this in a single sharable KMZ file is very efficient.
'Precipitation anomalies' express rainfall as an amount above or below a long-term mean. 'Observed' and 'mean modelled' precipitation anomalies are shown in 2.5 degree grid cells (the resolution of the forecasts) as red for greater than usual and blue for less than usual using a ColorBrewer diverging scheme.
The 'ensemble forecasts' used here work by running a model multiple times with slightly different starting conditions. The extent to which results agree with each other can indicate the confidence one might place in the forecast, yet the mean result is often issued as a forecast (as above).
A widely-used method for showing agreement between results is to categorise the precipitation anomalies into three groups 'wetter than normal', 'normal' and 'drier than normal'. These groups correspond to 'terciles' (33rd and 66th percentiles). The proportion of ensemble member results that fall into each tercile corresponds to the probability of 'wetter than normal', 'normal' and 'drier than normal' conditions. We show these results in two ways; (1) circles and (2) glyphs.
The circles are coloured by the dominant tercile as follows: red=wetter than normal; yellow=normal; blue=drier than normal. They are sized by the sum of the deviations between the tercile proportions so that the largest circles show the most agreement (e.g. all in 'normal') and the smallest show least agreement (e.g. same proportion in each tercile group). One can be more confident about the large circles because there is a higher probability that conditions will be as indicated, than for smaller circles.
The glyphs have three arms, each of which corresponds to a single tercile. The length of each arm corresponds to the probability that the forecasted conditions will correspond to that tercile where top-left='drier than normal', top-right='wetter than normal', bottom='normal'. One can be more confident about forecasts whose glyphs are least symmetrical.
Both these techniques also use opacity, where the high opacity corresponds to forecast with a high 'skill' (quality). This is defined by the Brier Skill Score - a measure of the consistency of the forecast over the period.
Users are unable to zoom in to an inappropriately high zoom level for the spatial resolution of these data.
These techniques illustrate that Google Earth supports the use of rich highly specified data graphics for conveying complex information through a consistent and intuitive interface. We have developed ColorBrewer colour scheme specifications for widespread adoption in KML cartography: http://gicentre.org/carto/colorbrewer/
The authors acknowledge the help, support and input of:
- David Stephenson and Rachel Lowe (Exeter Climate Systems, School of Engineering, Exeter University, UK)
- Willis Research Network
- ECMWF and the EUROBRISA network project (F/00 144/AT) kindly funded by the Leverhulme Trust
- Erik Andersson (ECMWF)
The dynamical ensemble forecast data were kindly provided by ECMWF as part of the EUROSIP project. Three forecasting centres are the partners in EUROSIP, these are ECMWF, the UK Met Office and Meteo-France.
Citations for resources
- Brier Skill Score
- Existing examples of seasonal climate data maps:
- Slingsby, A., Lowe, R., Dykes, J., Stephenson, D., Wood, J, Jupp, T. 2009. A Pilot Study for the Collaborative Development of New Ways of Visualising Seasonal Climate Forecasts. Proceedings of GISRUK, 1-3 April, Durham, UK [pdf]
- Wood, J, Dykes, J., Slingsby, A. and Clarke K. 2007. Interactive visual exploration of a large spatio-temporal data set: reflections on a geovisualization mashup. IEEE Transactions on Visualization and Computer Graphics 13 (6), pp1176-1183, November/December 2007. [pdf]
Please contact Aidan Slingsby for more information.