Comparing simple and complex hydrological models in regions with scarce data: a case study in the upper Vilcanota basin, Peru

Glaciers in the tropical Andes play an important role for both water supply, economic activities and cultural beliefs. Their importance is particularly high during the dry season (May – September) when glaciers can contribute more than 50% to total streamflow. People are used to take advantage of the important buffering function of the glaciers, but in a climate change context with rapid glacier retreat, this dependence could poses a considerable risk.

Several studies have focused on understanding and simulating the glacio-hydrological patterns for historical and future periods by using different hydrological models with different levels of complexity. However, most existing studies focus on regions with high availability of data, while data scarce regions are studied poorly.

The Vilcanota basin is located in such a data scarce region and encompasses the second largest glaciated mountain range in Peru and the tropics worldwide. As in many other mountainous regions, high-mountain conditions with complex topography and related variations of climatic variables are contrasted with poor availability of data. In view of this challenge, a key question is what level of model complexity would be most appropriate to achieve robust simulations of the hydrological cycle for historical and future climate conditions?

To answer this question, we simulated the hydrological conditions in the Sibinacocha catchment (area: 132 km2; glacier extent: 15 km2) that is part of the upper Vilcanota basin. The simulation was performed with three different hydrological models of different complexity on a monthly time scale from 1981 to 1996. Input data like precipitation and temperature were obtained from the Peruvian gridded precipitation and temperature data set PISCO2.1 (SENAMHI). Streamflow records for calibration were obtained from a hydropower company in the area. Finally, glacier outlines were obtained for three different periods from satellite images in order to incorporate glacier change.

The selected models include a lumped hydrological model based on equations by Temez (6-parameters), and two implementations of the HBV model (HBV Light and RS Minverve with 15 and 14 parameters respectively). Each model is capable of simulating groundwater and glacier contribution. For the simulations with HBV, the catchment was divided into 10 elevation bands. For the simulation with RS Minerve an additional Glacier and Snow model was performed with its own pool of parameters (10-parameters) and own elevation bands. Calibration was performed in two ways: 1) comparing observed and simulated flows, and 2) comparing the simulated and expected glacier and snow contribution to streamflow.

Results show that each of the models examined can reach high efficiencies when using only streamflow records for calibration. By contrast, multicriteria calibration provides more robust results than using one single indicator, even when efficiency indicators are in the same range of values.

In the context of the study region, we found that increasing complexity for hydrological simulation is only feasible if adequate input data are available. In cases with scarce data, lumped or simple semi-distributed models provide robust results. These simulations can be used later to implement more complex models and tools.