Regionalising rainfall-runoff modelling for predicting daily runoff: Comparing gridded spatial proximity and gridded integrated similarity approaches against their lumped counterparts

• Gridded regionalisation studies are carried out in the Australian continent.
• Overall the four regionalisation approaches are marginally different.
• Gridded integrated similarity performs best for data–sparse Australia.
• Regionalisation results obtained from two rainfall–runoff models are consistent.

Rainfall–runoff models are widely used for regionalisation studies to predict daily runoff time series in ungauged catchments. Most studies focus mainly on a particular region or a small scale, and are applied in a lumped way. It is not clear how grid–based regionalisation methods perform at continental or global scale, particularly for data–sparse region. This study uses 605 unregulated catchments widely distributed across Australia to evaluate two grid–based regionalisation approaches—gridded spatial proximity (SP_g) and gridded integrated similarity (IS_g) — and their lumped counterparts (SP_l and IS_l). To test robustness of the regionalisation methods, each was tested using two rainfall–runoff models: SIMHYD and Xinanjiang. We found that overall the gridded and lumped regionalisation approaches are marginally different and the two models show consistent regionalisation results. However, the IS_g approach outperforms the others in the dry and sparsely located catchments, and it overcomes the unnatural tessellated effect obtained from the SP_g approach. It is promising to use the IS_g approach for runoff estimates and water accounts in the Australian continent and possibly in other parts of world.