Challenges in Hydrological Modeling in a Data-Limited Catchment: The case of the barrage de l’Amitié, using HEC-HMS and GR4H

Djibouti, located in the Horn of Africa, experiences limited and sporadic rainfall events, typical of arid climates. Although infrequent, these rainfall episodes can be intense and trigger flash floods, resulting in significant damage to infrastructure and livelihoods. With more than 70% of the country’s population concentrated in Djibouti-ville, mitigating flood risks and ensuring sufficient water resources are key priorities for sustainable development. In response to recurring flood threats, the barrage de l’Amitié was constructed on Oued Oueah to protect Djibouti-ville from catastrophic flood events, a role it appears to fulfill effectively. Beyond its primary function of flood control, the dam also supports agricultural irrigation and potentially improves groundwater recharge, as infiltration in the reservoir area can help replenish Djibouti's aquifers

The barrage de l’Amitié located in a catchment of about 494 km², an area typified by arid conditions and subject to only a few of intense rainfall events each year. These irregular yet powerful events are essential for recharging the local water table (Djibouti aquifer), which is used to supply drinking water to the city of Djibouti. However, monitoring efforts are constrained by the single hydrometric station located roughly seven kilometers upstream from the dam. Because it does not measure flows below one meter, smaller or moderate runoff events go unrecorded. This gap introduces notable uncertainty into hydrological models, which depend on accurate data to represent the full range of runoff processes.

To address these challenges, we used five rainfall–runoff events for calibration and validation of two recognized hydrological models: HEC-HMS and GR4H. HEC-HMS employs the Curve Number (CN) loss method and Clark’s Unit Hydrograph, whereas GR4H applies a reservoir-based conceptual approach to capture surface and subsurface flow processes. In addition to standard calibration, a cross-validation procedure tested the transferability of parameters from one event to another, providing a stricter measure of model robustness given the limited dataset.

In HEC-HMS, several events produced high Nash-Sutcliffe (NASH) coefficients during calibration, demonstrating accurate hydrograph simulations under those specific conditions. Yet, validation runs often returned negative NASH values, suggesting that parameter sets calibrated for one event did not translate well to others in this arid environment. Meanwhile, GR4H also calibrated effectively for most events but showed vulnerabilities when confronted with multi-peak storms or complicated runoff patterns, again reflected by negative NASH values in certain cross-validation scenarios.

Overall, both models highlight the need for enhanced data collection particularly measurements capturing low-flow conditions essential for groundwater recharge and irrigation. Improved rainfall monitoring and stage discharge measurements below one meter would significantly enhance model reliability and better inform water resource strategies for Djibouti-ville. By refining model structures, expanding the observational network, and exploring infiltration processes more thoroughly, water managers could achieve a more holistic understanding of the hydrology in Oued Oueah, ultimately reinforcing flood protection and supporting the region’s agricultural and economic goals.