Disruptive reservoirs hold back an enormous amount of water, hike evaporation loss and alter the magnitude and timing of streamflow at all scales. Thus, any hydrological model (HM) must correctly represent reservoirs in the simulation. There are two issues while representing reservoirs in the river network of a gridded system. First issue is the error in reservoir catchment area where grids containing a part of the catchment results in under-/overestimation of reservoir inflow. Hence, it is impossible to conserve the catchment area correctly as long as a grid has only one outflow (D8 routing scheme) which is the case for the state-of-the-art gridded HMs. Secondly, when multiple dams are located in the same grid only one dam can be represented to lie on the major stream and be part of the stream network. Currently, gridded HMs either i) classify reservoirs into groups ("global/local","major/minor") simulating them differently in order to conserve the reservoir set, or ii) treat all reservoirs equally but are unable to conserve the reservoir set with smaller reservoirs disappearing at coarser resolutions. So either the set of reservoirs simulated or the reservoir simulation itself gets compromised while modeling across scales. This lack of scalability in space is a prominent source of model uncertainty in HMs (Samaniego et al. 2010, Kumar et al. 2013). We introduce subgrid catchment conservation (SCC), a novel scheme for routing that conserves reservoir catchment at all scales. We hypothesise that the conservation of reservoir catchment paves the way for a scalable hydrological system for reservoir modeling.

To test this hypothesis, we developed a reservoir module in the mesoscale hydrological model (mHM, https://mhm-ufz.org). mHM is tested across seven model resolutions ranging from 1 km to 100 km. The experiment set is the GRanD database, wherein the scalability of the reservoir set is tested for the whole set (7320 reservoirs) and the scalability of reservoir inflow simulation is tested at the headwater reservoirs (approx. 1500 reservoirs). Preliminary results in 70+ headwater reservoirs show that SCC routing preserves the full reservoir set across all scales. In comparison, the classic D8 routing scheme loses 15%, 25% and 50% reservoirs at 0.125 degree, 0.25 degree and 0.50 degree model resolutions, respectively. This indicates the potential of SCC in regulating interscale discrepancies in reservoir states and fluxes, leading to virtually seamless model performance.

The dilemma for modellers using distributed HMs is to compromise in resolution  (i.e., runtime) or to compromise on the number of reservoirs to model. Based on the preliminary results, SCC is poised to solve this long-standing dilemma and complete the quest for scalable hydrological modeling with reservoirs. The findings of this study would contribute to the contemporary effort of hydrological modeling society towards improved global water balance closure, where a good representation of reservoirs and lakes is a crucial element.