Some time ago, I wrote about the need for remote sensing experts to collaborate more closely with hydrologists to improve both hydrological forecasting and remote sensing applications. While the communities have largely continued working separately (and who am I to challenge such a fruitful paradigm?), my involvement in several projects with ASI (the Italian Space Agency) and ESA (the European Space Agency) has made this integration an unavoidable priority for me. This means I must become an Earth Observation (EO) scientist myself to cover the gap. First question actually would be:, is the EO data resolution enough for hydrologic simulations ? (The answer here is usually no, it is usually not, but we have, as usual, to survive with this and understand how the coarser resolution can be useful. Paraphrasing a well know motto: EO are usually wrong but can be useful).
Two Categories of Earth Observation Data
Earth Observation (EO) data can be broadly divided into two categories: quantities we routinely observe on the ground and those we typically do not (or could but usually don't) measure directly.
Ground-Measured Quantities
For simplicity, let's assume proximal data are correct, subject to verification and cross-checking with EO and modeled data. The first category includes rainfall, air temperature, point discharge measurements, surface water levels, groundwater storage (though less frequently measured), solar radiation, soil moisture (rarely measured), wind speed and evapotranspiration if we fully trust eddy covariance towers, which I don't take for granted.
EO-Derived Quantities
The second category encompasses snow cover, depth, and water content, along with soil moisture, vegetation water content, and evapotranspiration. It also includes various ancillary data such as albedo, leaf area index, land cover and land use classifications, surface temperature (which is distinct from the air temperature we typically measure), groundwater storage variations, precipitable water content, cloud cover, atmospheric liquid water content, and other quantities I'm not recalling at the moment. (Yes evapotranspiration is also on this side)
Challenges with Ground-Measured Quantities
For ground-measured quantities, comparing data sources is essential, acknowledging that ground measurements are generally more accurate but come with important caveats. There exists a fundamental scale mismatch where ground measurements are point observations, while EO measurements represent areal averages derived from radiation or gravity observations. Following Rodriguez-Iturbe and Mejia (1974), connecting point values to areal values requires careful assumptions about spatial scaling. Additionally, any systematic differences between ground and EO measurements must be identified and eliminated through proper bias correction procedures. How EO data are biased if compared to ground stations ?
However, ground data are spatially sparse. Obtaining geographic information requires interpolation, typically using methods like Kriging that preserve spatial correlation structure. Yet hydrological processes may be spatially nonlinear, raising the question: Do EO data, even if biased, better reproduce the spatial structure of hydrological variables?
If so, merging ground and EO data could yield superior spatial fields—an approach we pursued in Azimi et al. (2025). These quantities share another important characteristic: they can be produced by hydrological models and serve as inputs, validation data, or assimilation targets to improve model performance (which raises questions about how we evaluate "improvement").
Challenges with EO-Derived Quantities
The second category presents different challenges. These are primarily ancillary quantities that help estimate hydrological variables—leaf area index being the classic example for estimating transpiration, since plants without leaves don't transpire. Each quantity must find its proper role in hydrological models to be truly useful, requiring them to become model parameters.
The key question becomes: How can we effectively use these data in models?
While leaf area index integration seems straightforward, the data become useless if transpiration is estimated using temperature-only formulas. Similarly, how do we use land use data in models? What model characteristics connect to "land use" if it's not an explicit parameter?
Dal Molin et al. (2020) provide an interesting answer: geological setting relates to model structure within the framework of Hydrological Dynamical Systems (reservoir-based hydrological models). This expands possibilities but complicates the connections.
Another crucial question: Are these data actually necessary for the modeler's purpose?
Perhaps not, if the goal is simply discharge forecasting, since calibration procedures have proven effective at compensating for missing data and model's structure (as shown in the 4DHydro project outcomes). However, if the objective is reproducing realistic water budgets, these data may be essential, making assessment of all intermediate steps necessary, as shown in the 4DHydro Science Case 5.
Conclusions
This blogpost does not aim to be comprehensive but just an aid to start the discussion. Two clear conclusions emerge from this analysis. First, EO ancillary data are required only if models have mechanisms to utilize them effectively. Second, they may not be necessary for all modeling purposes, particularly when the primary objective is discharge forecasting rather than comprehensive water budget reproduction.
Ultimately, new data should either increase the precision and accuracy of predictions or enhance our understanding of hydrological process mechanisms. The challenge lies in determining when and how to integrate these diverse data streams effectively. The final goal is what stated in the last paragraph of this post, which I invite you to read again.
No comments:
Post a Comment