There is a relatively small but very robust set of literature on this so a little lit search will yield some good results - too many to identify my favorites. There's been lots of work with soil moisture assimilation for very large-scale (continental, regional) hydrology, less at smaller (basin, subbasin, field) scales, though examples do exist.

The extent to which it improves forecasts of course depends on a number of things, including:

* the quality of the soil moisture products (if they're not very good, they may actually degrade the forecasts or estimates, rather than improve them);

* the degree of "match" between the scale and/or resolution of the observed data to be assimilated, and the resolution of the model (observed soil moisture at a point, for example, may not be very useful when the model runs at a course resolution);

* the accuracy of the baseline model-only forecasts (if the models are doing a pretty good job already, then the assimilation may be more confusing than helpful), which itself is a function of many things, including, the uncertainties in all the input parameters (assimilation may be most useful for situations where it's especially difficult to parameterize or drive the model with high-quality data).

Thanks a lot for your reply. I agree with you about the different issues to be addressed for improving runoff prediction through soil moisture assimilation. However, my experience showed that, on one hand, there are a lot of synthetic studies investigating this issue but, on the other hand, very few authors employed actual observations (from in situ or satellite sensors).

I'm asking why there is a so strong bias to synthetic, and not real, studies, that, in my opinion, should be much more interesting?

Using synthetic observations is the smartest first step when trying a data assimilation scheme in a new model. You can tightly control the characteristics of the synthetic data, and you can eliminate the mismatch of scale between observed soil moisture and modeled. That wayyou can then really scrutinize the interaction between the assimilation scheme and the model itself. If you start your testing using actual observations right away, it's harder to assess how much of the resulting change in the model output is due to the assimilation scheme versus how much is due to the observations. Ideally, you use synthetic data so you can isolate the impact of the assimilation scheme, then once you're sure the assimilation is functioning and helpful, you move to using real data.

The scales issue is an important one indeed. Flood forecasting suggests catchment scale while soil moisture, in particular in situ measurements, suggests point-scales.

Then on one side you increase the amount of information content using supplementary data such as soil moisture. On another side you have to increase the model complexity to let it able to simulate BOTH large scale variable (discharges, river heights, etc) and small scale variables (soil moisture).

No evidence then that assimilation of supplementary soil moisture data in more complex models improves reliability and relevancy of simulations results compared to more simple models (despite already complex in case of flood forecasting!) without soil moisture assimilation.

My personal opinion is that assimilation aims to improve models simulations results using supplementary informations. Then the model complexity should not have to increase when assimilating supplementary data in order to maximize the benefit of using supplementary data. Supplementary data should be then at the relevant scales compared to the model variable scales.

For rainfall-runoff models I was never totally convinced so far (and very often disappointed I must say) when assimilating highly spatially variable soil moisture (I mean using real data. Synthetic experience are much more pretty and accurate as you control everything but with low robustness when testing real catchments and real measurements). At the exception of very large-scale flood forecasting (> 5000-10000km2) where SVAT models can be relevant for both flood forecasting and soil moisture satellite data assimilation as (i) flood at these scales are less dependent on hydrological processes but are very controlled by meteorological variables (rainfall, evapotranspiration, etc) and as (ii) SVAT models are often used to calculate soil moisture estimates from satellite measurements: a complete agreement then between data scales and model scales.

Cheers

I would like to add to the discussion that I fully agree that soil moisture exhibits high spatial variability, also at very small scale. However, the "temporal stability" concept introduced by Vachaud et al. (1985, SSSAJ) allows to infer soil moisture estimates for large areas from a few point measurements ( http://dx.doi.org/10.2136/sssaj1985.03615995004900040006x ). Therefore, the scale issue could be not so significant.

For instance, see the very recent paper by Chen et al. (2012, HYP) who concluded: "This paper demonstrates the potential usefulness of continuous time, point scale soil moisture data and simulations for predicting the soil wetness status over a catchment of significant size (up to 1000 km²)" ( http://dx.doi.org/10.1002/hyp.9518 ).

Luca's totally right that there are several ways to try to upscale point measurements to try to make something that's more representative of the larger scale. For me this is still a hot research area, though. A former student and I found that in our region, the temporal stability approach was not really any better than random sampling (manuscript in progress) so now we're trying some different approaches (another manuscript in progress). I'm happy to share some posters or conference papers if anybody is really interested in the issue.

Yes, I also believe that much work has to be done for this research activity and ...

... Yes, any new material for upscaling (and, hence, for using) of point soil moisture measurements is welcome.

Large scale estimate of soil moisture from local measurements necessary need assumptions: e.g. spatial relative variability of soil moisture, spatial variability of soil characteristics, etc which relevancy may be questionned. Take care also about temporal variability at large scales which in the case of your interest - flood forecasting - may be of smaller importance than, for example, rainfall temporal and spatial variabilities. No doubt that in "drier" conditions upscaling (both spatial and temporal variabilities) is reasonnably feasible (see also for example all the SMOS litterature). But in the case of flood forecasting, except spatial variability of initial soil moisture prioir rainfall event, i have doubt about the interest of assimilating field soil moisture measurements compared to all other spatial and temporal variabilities that may infuence flood genesis: preferential subsurface flows dynamic, hydraulic,propagation within the river network, spatial and temporal rainfall variabilites, etc.