A Ranking of Hydrological Signatures Based on Their Predictability in Space

Abstract

We used machine learning (random forests) and a hydrological model to simulate 15 hydrological signatures over 671 catchments in the US. The predictability of the signatures is highly correlated with the smoothness of their spatial pattern, which we quantified using Moran's I. Poorly-predicted signatures vary abruptly in space, are sensitive to streamflow errors and their links to catchment attributes are elusive. AbstractHydrological signatures are now used for a wide range of purposes, including catchment classification, process exploration and hydrological model calibration. The recent boost in the popularity and number of signatures has however not been accompanied by the development of clear guidance on signature selection. Here we propose that exploring the predictability of signatures in space provides important insights into their drivers, their sensitivity to data uncertainties, and is hence useful for signature selection. We use three complementary approaches to compare and rank 15 commonly-used signatures, which we evaluate in 671 US catchments from the CAMELS data set (Catchment Attributes and MEteorology for Largesample Studies). Firstly, we employ machine learning (random forests) to explore how attributes characterizing the climatic conditions, topography, land cover, soil and geology influence (or not) the signatures. Secondly, we use simulations of a conceptual hydrological model (Sacramento) to benchmark the random forest predictions. Thirdly, we take advantage of the large sample of CAMELS catchments to characterize the spatial auto-correlation (using Moran's I) of the signature field. These three approaches lead to remarkably similar rankings of the signatures. We show i) that signatures with the noisiest spatial pattern tend to be poorly captured by hydrological simulations, ii) that their relationship to catchments attributes are

Deriving hydrological signatures from soil moisture data

“…The Soil Moisture Network sites are located in five different climate regions (CD, CW, CX, WD, WW; see Section 2 and Figure ). We make the following hypotheses, based on physical reasoning: Climate: Addor et al () used a large‐scale in the United States to show that climate metrics are a strong predictor for hydrologic signature values. In this study, we hypothesise that climate is the dominant factor for PDF type and seasonal transitions dates, both of which depend on the annual pattern of rainfall and evapotranspiration.…”

“…1. Climate:Addor et al (2018) used a large-scale in the United States to show that climate metrics are a strong predictor for hydrologic signature values. In this study, we hypothesise that climate is the dominant factor for PDF type and seasonal transitions dates, both of which depend on the annual pattern of rainfall and evap-…”

Linking hydrologic signatures to hydrologic processes: A review

“…Yilmaz, Gupta, and Wagener (2008) and McDonnell et al (2007) look more generally for spatialtemporal patterns in the data that could be explained by differences in processes. Large-scale techniques, such as machine learning, random forests, and stepwise regressions, can elucidate the relationship of signature values with climatic and physical characteristics of watersheds, for example, Addor et al (2018) for the United States, Kuentz, Arheimer, Hundecha, and Wagener (2017) in Europe, and Trancoso, Phinn, McVicar, Larsen, and McAlpine (2017) in Australia. These studies found that climate was a dominant control on signatures, with Kuentz et al (2017) also finding that geology (for base flow) and land cover were important characteristics.…”

In Quest of Calibration Density and Consistency in Hydrologic Modeling: Distributed Parameter Calibration against Streamflow Characteristics

“…There are a variety of typical streamflow characteristics proven to be useful for model calibration, including the FDC, baseflow indices, runoff ratio, master recession curve (Addor et al, 2018;Fenicia et al, 2018;Troy et al, 2008;Winsemius et al, 2009;Yilmaz et al, 2008). Moreover, the experiments conducted by Fenicia et al (2018) showed that calibration against several FDC quantiles and baseflow index can produce virtually indistinguishable streamflow prediction from calibration operated directly on the streamflow time series.…”