##
I have been interested to this topic since two years ago, and I realised now I will not find time to do it . So let me put my thought here.

The spatial rainfall variability is well recognised field of study and it is also the one that needs further improvement in hydrology (Syed et al., 2003, Woolhiser, 1996). In the proper estimation of spatial rainfall data for modelling, there are efforts on the characterisation of rainfall field at different scale. In the context of semi-distributed hydrological models, the rainfall inputs are usually required at some level of aggregated (i.e sub basin, HRU, hillslope). All the efforts of rainfall estimation and their spatial variability is focused mainly in the correct estimation of the 1st moment of these hydrological modelling units. Bloschl and Sivapalan (1995) formalised the upscaling activities as two steps: distributing the point measurement to the large area (units), and aggregation of spatially distributed rainfall into a single value. In the series of papers, Foufoula-Georgiou and her co-authors (Kumar and Foufoula-Georgiou, 1990, Foufoula-Georgiou and Lettenmaier, 1986, Foufoula-Georgiou and Georgakakos, 1991, Kumar and Foufoula-Georgiou, 1993b,a), in effort to understand this rainfall spatial organisation at die rent scales, they point out that spatial rainfall characterisation at larger scale provides average (smoothed process) and the "detail lost" information. While the smoothed process is representation of the areal mean rainfall at the sub basin scale, the "detail lost" is the rainfall information that is lost during the smoothing processes, which can be inferred from other statistical moments.

It is note that in the semi-distributed model, the modelling units are based on the "hydrological similarity concepts". This mean that the units (subbasin) are statistically characterized. So far all the studies on semi-distributed models, or those uses large scale grid inputs, focused on the proper representation of the 1st statistical moment, and as to my experience, there is gaps in understanding and incorporating the sub basin variability in to the modelling environments.

With all research eorts in understanding the spatial representation of rainfall at all level of scales, however, it is clearly apparent that the effort to incorporate the rainfall spatial variability in the semi-distributed modelling solution is missed. The epistemic sources of uncertainty, i.e, uncertainty caused by the system representation and lack of skills to treat them, in hydrological model are the one we need to be concerned (Beven et al. 2011; Merz and Thieken, 2005).

What I think important is the use of probability distribution of rainfall at each time steps for each HRU(hydro logical response units) to obtain the probably of discharge time series instead of a single deterministic estimation. I have seen the use of various models to incorporate the model parameters to generate the probability band of simulation, but never seen due to the use of forcing input data. Here this figure is to show the effects of the mean (centre), the maximum and minimum (the blue band) of rainfall of each HRU of a basin in JGrass-NewAge modelling.
Fig 1: JGrass-NewAge simulation using different HRU rainfall representation.

References
Keith Beven, PJ Smith, and Andrew Wood. On the colour and spin of epistemic error (and what we might do about it). Hydrology and Earth System Sciences, 15(10):3123

Gunter Bloschl and M Sivapalan. Scale issues in hydrological modelling: a review. Hydrological processes, 9(3-4):251{290, 1995.

E Foufoula-Georgiou and Konstantine P Georgakakos. Hydrologic advances in space-time precipitation modelling and forecasting. In Recent advances in the modelling of hydrologic systems, pages 47{65. Springer, 1991.

E Foufoula-Georgiou and Dennis P Lettenmaier. Continuous-time versus discrete-time point process models for rainfall occurrence series. Water Resources Research, 22(4):531

Praveen Kumar and E Foufoula-Georgiou. Fourier domain shape analysis methods: A brief review and an illustrative application to rainfall area evolution. Water Resources Research, 26(9):2219

Praveen Kumar and E Foufoula-Georgiou. A multicomponent decomposition of spatial rainfall elds: 2. self-similarity in fluctuations. Water Resources Research, 29(8):2533{2544, 1993a.

Praveen Kumar and E Foufoula-Georgiou. A new look at rainfall fluctuations and scaling properties of spatial rainfall using orthogonal wavelets. Journal of Applied Meteorology, 32(2):209

Bruno Merz and Annegret H Thieken. Separating natural and epistemic uncertainty in food frequency
analysis. Journal of Hydrology, 309(1):114

Kamran H Syed, David C Goodrich, Donald E Myers, and Soroosh Sorooshian. Spatial characteristics of thunderstorm rainfall elds and their relation to runo. Journal of Hydrology, 271(1):1{21, 2003.

The spatial rainfall variability is well recognised field of study and it is also the one that needs further improvement in hydrology (Syed et al., 2003, Woolhiser, 1996). In the proper estimation of spatial rainfall data for modelling, there are efforts on the characterisation of rainfall field at different scale. In the context of semi-distributed hydrological models, the rainfall inputs are usually required at some level of aggregated (i.e sub basin, HRU, hillslope). All the efforts of rainfall estimation and their spatial variability is focused mainly in the correct estimation of the 1st moment of these hydrological modelling units. Bloschl and Sivapalan (1995) formalised the upscaling activities as two steps: distributing the point measurement to the large area (units), and aggregation of spatially distributed rainfall into a single value. In the series of papers, Foufoula-Georgiou and her co-authors (Kumar and Foufoula-Georgiou, 1990, Foufoula-Georgiou and Lettenmaier, 1986, Foufoula-Georgiou and Georgakakos, 1991, Kumar and Foufoula-Georgiou, 1993b,a), in effort to understand this rainfall spatial organisation at die rent scales, they point out that spatial rainfall characterisation at larger scale provides average (smoothed process) and the "detail lost" information. While the smoothed process is representation of the areal mean rainfall at the sub basin scale, the "detail lost" is the rainfall information that is lost during the smoothing processes, which can be inferred from other statistical moments.

It is note that in the semi-distributed model, the modelling units are based on the "hydrological similarity concepts". This mean that the units (subbasin) are statistically characterized. So far all the studies on semi-distributed models, or those uses large scale grid inputs, focused on the proper representation of the 1st statistical moment, and as to my experience, there is gaps in understanding and incorporating the sub basin variability in to the modelling environments.

With all research eorts in understanding the spatial representation of rainfall at all level of scales, however, it is clearly apparent that the effort to incorporate the rainfall spatial variability in the semi-distributed modelling solution is missed. The epistemic sources of uncertainty, i.e, uncertainty caused by the system representation and lack of skills to treat them, in hydrological model are the one we need to be concerned (Beven et al. 2011; Merz and Thieken, 2005).

What I think important is the use of probability distribution of rainfall at each time steps for each HRU(hydro logical response units) to obtain the probably of discharge time series instead of a single deterministic estimation. I have seen the use of various models to incorporate the model parameters to generate the probability band of simulation, but never seen due to the use of forcing input data. Here this figure is to show the effects of the mean (centre), the maximum and minimum (the blue band) of rainfall of each HRU of a basin in JGrass-NewAge modelling.

Fig 1: JGrass-NewAge simulation using different HRU rainfall representation.