Catchment classification by runoff behaviour with self-organizing maps (SOM)

Abstract. The baffling diversity of runoff generation processes, alongside our sketchy understanding of how physiographic characteristics control fundamental hydrological functions of water collection, storage, and release, continue to pose major research challenges in catchment hydrology. Here, we propose innovative data-driven diagnostic signatures for overcoming the prevailing status quo in catchment inter-comparison. More specifically, we present dimensionless double mass curves (dDMC) which allow inference of information on runoff generation and the water balance at the seasonal and annual timescales. By separating the vegetation and winter periods, dDMC furthermore provide information on the role of biotic and abiotic controls in seasonal runoff formation.A key aspect we address in this paper is the derivation of dimensionless expressions of fluxes which ensure the comparability of the signatures in space and time. We achieve this by using the limiting factors of a hydrological process as a scaling reference. We show that different references result in different diagnostics. As such we define two kinds of dDMC which allow us to derive seasonal runoff coefficients and to characterize dimensionless streamflow release as a function of the potential renewal rate of the soil storage. We expect these signatures for storage controlled seasonal runoff formation to remain invariant, as long as the ratios of release over supply and supply over storage capacity develop similarly in different catchments.We test the proposed methods by applying them to an operational data set comprising 22 catchments (12-166 km 2 ) from different environments in southern Germany and hydrometeorological data from 4 hydrological years. The diagnostics are used to compare the sites and to reveal the dominant controls on runoff formation.The key findings are that dDMC are meaningful signatures for catchment runoff formation at the seasonal to annual scale and that the type of scaling strongly influences the diagnostic potential of the dDMC. Adding discrimination between growing season and winter period was of fundamental importance and easy to implement by means of a temperatureindex model. More specifically, temperature aggregates explain over 70 % of the variability of the seasonal summer runoff coefficients. The results also show that the soil topographic index, i.e. the product of topographic gradient and saturated hydraulic conductivity, is significantly correlated with winter runoff coefficients, whereas the topographic gradient and the hydraulic conductivity alone are not. We conclude that proxies for gradients and resistances should be interpreted as a pair. Lastly, the dDMC concept reveals memory effects between summer and winter runoff regimes that are not relevant in spring between the transition from winter to summer.

supporting

confidence: 70%

Unravelling abiotic and biotic controls on the seasonal water balance using data-driven dimensionless diagnostics

Seibert

Jackisch

Ehret

et al. 2017

“…Approaches include the use of physical and climatic characteristics (e.g., Winter, 2001;Brown et al, 2013;Buttle, 2006;Leibowitz et al, 2016), the use of hydrologic signatures (e.g., Ley et al, 2011;Olden et al, 2012;Sawicz et al, 2011;Singh et al, 2016), or the inclusion of water quality (Arheimer et al, 1996;Arheimer and Lidén, 2000). The advantage of the first approach is that physical characteristics such as topography and land cover are now available for any location on earth (though with varying quality of the data available), while the second approach groups catchments directly by the characteristic we mainly care about, i.e., their hydrologic behavior (see the discussion in Wagener et al, 2007).…”

Section: Introductionmentioning

confidence: 99%

“…They condense hydrologic information derived from streamflow observations (alone or in combination with other variables) (Sivapalan, 2005). The choice of the specific signatures used for classification can be guided by (i) the attempt to describe basic hydrological behavior (e.g., Ley et al, 2011;Sawicz et al, 2011;Trancoso et al, 2016); (ii) the need to relate to societally relevant issues such as floods and droughts (Wagener et al, 2008); (iii) the objective to characterize ecologically relevant characteristics of the catchment response (e.g., Olden et al, 2012); or (iv) in relation to subsequent hydrologic modeling (Euser et al, 2013;Hrachowitz et al, 2014;Donnelly et al, 2016). Studying differences and similarities in flow signatures as well as in catchment characteristics can also improve our understanding of hydrological processes under potential future conditions (Sawicz et al, 2014;Berghuijs et al, 2014;Pechlivanidis and Arheimer, 2015;Rice et al, 2015).…”

Section: Introductionmentioning

confidence: 99%

Understanding hydrologic variability across Europe through catchment classification

Kuentz

Arheimer

Hundecha

et al. 2017

125

Abstract. This study contributes to better understanding the physical controls on spatial patterns of pan-European flow signatures -taking advantage of large open datasets for catchment classification and comparative hydrology. Similarities in 16 flow signatures and 35 catchment descriptors were explored for 35 215 catchments and 1366 river gauges across Europe. Correlation analyses and stepwise regressions were used to identify the best explanatory variables for each signature. Catchments were clustered and analyzed for similarities in flow signature values, physiography and the combination of the two. We found the following. (i) A 15 to 33 % (depending on the classification used) improvement in regression model skills when combined with catchment classification versus simply using all catchments at once. (ii) Twelve out of 16 flow signatures were mainly controlled by climatic characteristics, especially those related to average and high flows. For the baseflow index, geology was more important and topography was the main control for the flashiness of flow. For most of the flow signatures, the second most important descriptor is generally land cover (mean flow, high flows, runoff coefficient, ET, variability of reversals). (iii) Using a classification and regression tree (CART), we further show that Europe can be divided into 10 classes with both similar flow signatures and physiography. The most dominant separation found was between energy-limited and moisturelimited catchments. The CART analyses also separated different explanatory variables for the same class of catchments. For example, the damped peak response for one class was explained by the presence of large water bodies for some catchments, while large flatland areas explained it for other catchments in the same class. In conclusion, we find that this type of comparative hydrology is a helpful tool for understanding hydrological variability, but is constrained by unknown human impacts on the water cycle and by relatively crude explanatory variables.

“…A very interesting and promising approach to classification makes use of an innovative and data-driven classification method based on unsupervised artificial neural networks (ANNs), known as Self Organising Maps (SOM, Kohonen, 1982, Toth, 2009Ley et al, 2011).…”

Section: Introductionmentioning

confidence: 99%

Data-driven catchment classification: application to the pub problem

Prinzio

Castellarin

Toth

2011

Abstract.A promising approach to catchment classification makes use of unsupervised neural networks (Self Organising Maps, SOM's), which organise input data through non-linear techniques depending on the intrinsic similarity of the data themselves. Our study considers ∼300 Italian catchments scattered nationwide, for which several descriptors of the streamflow regime and geomorphoclimatic characteristics are available. We compare a reference classification, identified by using indices of the streamflow regime as input to SOM, with four alternative classifications, which were identified on the basis of catchment descriptors that can be derived for ungauged basins. One alternative classification adopts the available catchment descriptors as input to SOM, the remaining classifications are identified by applying SOM to sets of derived variables obtained by applying Principal Component Analysis (PCA) and Canonical Correlation Analysis (CCA) to the available catchment descriptors. The comparison is performed relative to a PUB problem, that is for predicting several streamflow indices in ungauged basins. We perform an extensive cross-validation to quantify nationwide the accuracy of predictions of mean annual runoff, mean annual flood, and flood quantiles associated with given exceedance probabilities. Results of the study indicate that performing PCA and, in particular, CCA on the available set of catchment descriptors before applying SOM significantly improves the effectiveness of SOM classifications by reducing the uncertainty of hydrological predictions in ungauged sites.