Spatial correlation of rainfall data

E. Linacre and B. Geerts


If rainfall occurred at my location yesterday, then what is the probabilty that rain also fell at places in the vicinity? Frontal disturbances may produce continuous precipitation in an area that may be longer than 1000 km, and has a typical comma-shaped appearance. Even for daily precipitation totals, the shape of the affected domain is not circular. The area of 24h rainfall from tropical cyclones is several 100s of km in diameter and in length depends on the stormís motion. Thunderstorms generally produce highly localised rainfall, and even the accumulated rainfall over 24h is highly discontinuous. Topography has an important impact on the areal extent of precipitation.

One can ignore the driving synoptic conditions and approach the question of rain area size from a purely statistical vantage point, and ask what the probability of rain is at some location, given that so much rain fell at a neighbouring reference location. A study of rainfalls in eastern Queensland (Australia) related measurements of daily rainfalls at each of 830 places to the simultaneous values at the other places, at various distances away (1). Thus the correlation coefficient of daily rainfall amounts was unity for places zero distance apart, roughly 0.7 for places separated by 200km, 0.4 if the separation is 400km. However, the rate of correlation decline with distance was highly asymmetric: for a few pairs of places 400km apart the coefficient is zero, and for other pairs up to 0.8. To gain a better understanding of the rainfall dynamics, it is useful to examine the spatial structure of the correlation coefficients.

Another study (mentioned in (1)) concerned rainfalls in the smaller region of the Australian Capital Territory. There the coefficient varies from about 0.9 for places 50km apart, 0.75 for 100km and 0.65 at 150km.



(1) Hutchinson, M.F. 1995. Stochastic space-time weather models from ground-based data. Agric. & Forest Meteor. 73, 237-64.