Correspondence Between Model Structures and Hydrological Signatures: A Large‐Sample Case Study Using 508 Brazilian Catchments

Abstract

Daily streamflow dynamics can be accurately simulated by conceptual models as simple as a single bucket in some catchments, while they require more complex configurations in other catchments. However, without resorting to calibration, anticipating where and why a given model structure may be appropriate remains difficult. In this work, we explored the feasibility of relating suitable model structures to the climate and streamflow characteristics of 508 catchments in Brazil. Specifically, we tested four model structures using up to three reservoirs, where each reservoir is intended to represent a catchment function: the rainfall‐runoff threshold, the fast, and the slow hydrograph response. We hypothesized a relationship between suitable model structures and hydrological signatures of aridity (IA) and baseflow index (IB). Our results show that different classes of signatures resulted in distinct patterns of model performance. Wet catchments (IA < 0.9) with low baseflow (IB < 0.4) were the easiest to model, with a single‐reservoir model presenting a relatively good performance. In the case of low baseflow, adding a rainfall‐runoff threshold reservoir resulted in better performance than adding a slow response reservoir, whereas in the case of high baseflow (IB < 0.6) the opposite occurred. In the case of low baseflow, the inclusion of a slow response reservoir helped the simulation of dry catchments (IA < 1.1), but not of wet ones, which we attributed to the impact of permeability in dry catchments. These results indicate a path toward model structure identification from streamflow signatures and potentially from landscape features.

Investigating the Model Hypothesis Space: Benchmarking Automatic Model Structure Identification With a Large Model Ensemble

“…Again, these are the catchments with the lowest rainfall runoff ratios and small mean flow values, indicating a lack of capability of KGE calibrated models to represent low flow behavior. This supports previous findings stating that dryer catchments are more difficult to simulate than wetter catchments (e.g., David et al, 2022;Esse et al, 2013;Gan et al, 1997;Li et al, 2015;Poncelet et al, 2017). We show this in terms of model accuracy (Figure 7) as well as model adequacy (Table 4).…”

“…These structures can have between zero and 12 parameters, depending on which processes are active and which equations are used. The simple two (soil) storage assumption allowing for a quick and slow response has been chosen because it has often been perceived as a “best‐compromise model” in the literature (David et al., 2022). The SD‐MHS also includes a rain‐snow‐routine that can consider interception and canopy evaporation processes if the according process is turned on.…”

CLIMBra - Climate Change Dataset for Brazil

“…Six different interpolation techniques were evaluated: arithmetic averaging, thin plate spline, natural neighbor, inverse distance weighting, angular distance weighting, and ordinary point kriging. In general, the dataset shows a good performance in describing the weather station observations and is widely applied in hydrological and climatological studies in Brazil 31 , 32 , and also as the ground truth to GCMs/RCMs projections assessment and impact studies 33 , 34 .…”

A hydrologic signature approach to analysing wildfire impacts on overland flow

“…Hydrologic signatures are indices calculated from streamflow and occasionally precipitation time series that quantify aspects of the hydrologic regime of a catchment (Gupta et al, 2008; McMillan, 2021). Signatures can be used as objective functions in hydrologic models (Shafii & Tolson, 2015) or can inform decisions about model structure (David et al, 2022) to improve similarity between simulated and observed runoff patterns. Hydrologic signatures facilitate further understanding of hydrologic processes when carefully selected, as different signatures represent the occurrence and/or magnitude of specific processes including evapotranspiration, baseflow and overland flow (McMillan, 2020).…”