My research interests lie in the fields of catchment hydrology, analysis and evaluation of hydrological systems under uncertainty, model identification and evaluation, prediction in ungauged catchments and prediction under environmental change (i.e. climate and land use change). Specifically, my current visions fall under the theme “diagnostic model evaluation in a non-stationary context” and focus on the identification of robust numerical solutions to better characterise the possible evolutions of surface hydrology. In this content, I intend to better quantify the model uncertainties and improve the capacity of hydrological models to predict extreme flood events and their future change. For this research goal, I address two key questions: (a) during which periods does a model structure (not) produce the static and dynamic properties of the data? and (b) can we point towards model components that are responsible for the bad model performance? Outcomes can provide the hydrologic community with a better understanding of individual system components and the dominant mechanisms of change, and hence a better judgement of our models’ ability to predict the flood response given specific resource and data limitations.
One of the main challenges facing hydrological science is to develop a holistic and quantitative understanding of the changing behaviour of hydrological systems (Wheater and Evans, 2009). While spatial variation of soil, land cover, and elevation leads to complex catchment processes that vary from one location to another, changing climatic and/or physical conditions contribute to time varying nature of hydrological responses over different temporal scales (Levesque et al., 2008). However, the temporal dynamics of hydrological systems bring uncertainties into hydrological predictions which are different from uncertainties related to spatial heterogeneity (i.e. precipitation, soil, land cover, and elevation); hence predictions need to additionally allow for adaptive temporal evolution of soils, vegetation and climate.
Hydrological models have been mainly used to assess the impact of climate and land use change on flood magnitudes. While the motivation for such models is thus clear, the challenges – of identifying model structures and estimating parameters, and conducting uncertainty analysis at realistic computational cost – persist (Wilby and Harris, 2006). The process of developing, calibrating and evaluating models carries a significant degree of subjectivity. Models are evaluated using measures which are functions of the residuals between the modelled and observed quantities; hence they emphasise different systematic and/or dynamic behaviours within the hydrological system. Moreover, hydro-climatic time series used for model calibration usually include combined conditions of dry, normal and wet years, whereas the selected time series could additionally correspond to significant alterations of physical system characteristics (Toth, 2009). Nevertheless, current methods identify parameter sets that can only characterise the average hydrological responses under varying climatic and physical conditions, despite the detection of non-stationarities in the optimal range of parameters (de Vos et al., 2010; Perrin et al., 2008).
The hydrologic modelling community has highlighted the importance of moving from a model “calibration” philosophy towards a diagnostic evaluation approach that aims to: i) characterise the information contained in the data and in the model, ii) examine the extent to which a model can be reconciled with observations, and iii) point towards the aspects of the model that need improvement (Gupta et al., 2008; Schaefli et al., 2011). In this regard, several approaches (e.g. multi-objectives, signature measures, use of complementary data, and quantification of uncertainty during various stages of model development) have been applied to reveal significant information about the hydrological system and indicate structural errors (Clark et al., 2011; Hingray et al., 2010; Reichert and Mieleitner, 2009). Other strategies relate model performance to individual model components by measuring the model performance during different stages of the model response (Reusser and Zehe, 2011; Zhang et al., 2011).
Recent work has proposed the concept of a diagnostic evaluation approach rooted in information theory (Weijs et al., 2010). Information theory provides a powerful approach to closing fundamental gaps in our ability to relate multiple interconnected data streams to process understanding. Applications of information theory in hydrology encompass derivation of frequency distributions and estimation of their parameters, monitoring and evaluation of networks and flow forecasting, investigation of the scaling behaviour of hydrological processes in space and time, spatial-temporal precipitation analyses, and introduction as objective functions in hydrological modelling (Jackson et al., Under Review).
Research in Progress
My research has been focusing on disentangling the complex relationship between climate, physical catchment properties and runoff generation using hydrological models. Consequently, I have been dealing with challenges that arise from the application of this type of models. In the context of flood management, I have investigated the impact of the spatial variability of rainfall on flood generation as a function of catchment scale, physical properties (soil, land use) and antecedent conditions (Pechlivanidis, 2009). In addition, I have empirically and numerically investigated the dependencies between spatial-temporal rainfall characteristics (duration, volume, location, spatial variability, maximum and mean intensity) and runoff properties (peak and volume) for several events.
My current research is an extension to my previous work (see Pechlivanidis et al., 2010; 2012) and aims to develop a diagnostic evaluation framework rooted in information theory to characterise the temporally varying information structures (due to climate and land use change) within our catchment systems. Application of information-based tools in hydrological modelling may assist in the generation of more robust frameworks allowing us to better diagnose fundamental inconsistencies between data, system understanding and our models, and reduce further the uncertainty in our model predictions. To address this key point, I seek for an answer in the following two questions: (a) during which periods does a model structure (not) produce the static and dynamic properties of the data? and (b) can we point towards model components that are responsible for the bad model performance? This in turn will provide a guide to complexity of models, the number of parameters appropriate for different parts of the system and how to interpret natural phenomena (i.e. flood generation).
The first question has been addressed exploring the seasonal dynamics in model calibration and analysis of model performance. This involves investigation of either intra-annual patterns (e.g. warm-dry, rainy and cold-dry periods; winter and summer periods) or application of temporal clustering approaches (self-organising maps; K-means clustering; Fuzzy C-means) to identify periods of hydrologic similarity and further explore the model sensitivity and performance; hence identify and understand model structural deficiencies. To address the second question, I have focused on the development and investigation of diagnostic measures based on entropy that emphasise on the systematic and dynamic behaviours within the hydrological system. As entropy provides a measure-of-fit and/or a diagnostic measure with very different sensitivities and insensitivities to those currently in use, it also has potential in combination with other measures in a multi-objective calibration framework to decouple timing and other errors (Pechlivanidis et al., In preparation). This pioneering approach not only can point towards the aspects of the model that need improvement and explore the seasonal dynamics, but can overcome, to a certain extent, uncertainty into model’s predictability derived from inconsistencies of hydroclimatic conditions and/or land use signals between different periods.
Clark, M.P., McMillan, H.K., Collins, D.B.G., Kavetski, D. and Woods, R.A., 2011. Hydrological field data from a modeller's perspective: Part 2: process-based evaluation of model hypotheses. Hydrological Processes, 25(4): 523-543.
de Vos, N.J., Rientjes T.H.M. and Gupta, H.V.., 2010. Diagnostic evaluation of conceptual rainfall-runoff models using temporal clustering. Hydrological Processes, 24(20): 2840–2850.
Gupta, H., Wagener, T. and Liu, Y., 2008. Reconciling theory with observations: elements of a diagnostic approach to model evaluation. Hydrological Processes, 22(18): 3802-3813.
Hingray, B., Schaefli, B., Mezghani, A. and Hamdi, Y., 2010. Signature-based model calibration for hydrological prediction in mesoscale Alpine catchments. Hydrological Sciences Journal, 55(6): 1002-1016.
Jackson, B., Schaefli, B., Gupta, H.V. and others. What is the information content of hydrologic data? Advances in Water Resources, Under Review.
Levesque, E., Anctil, F., van Griensven, A. and Beauchamp, N., 2008. Evaluation of streamflow simulation by SWAT model for two small watersheds under snowmelt and rainfall. Hydrological Sciences Journal, 53: 961–976.
Pechlivanidis, I.G., 2009. The significance of spatial variability of rainfall on runoff generation. PhD Thesis, Department of Civil and Environmental Engineering, Imperial College London, 1-391 pp.
Pechlivanidis, I.G., Jackson, B. and McMillan, H., 2010. The use of entropy as a model diagnostic in rainfall-runoff modelling. Paper presented at iEMSs 2010: International Congress on Environmental Modelling and Software, 5-8 July, Ottawa, Canada, 5-8 July.
Pechlivanidis I.G., Jackson, B., McMillan, H. and Gupta, H., 2012. Using an informational entropy-based metric as a diagnostic of flow duration to drive model parameter identification. In the special issue of the Global NEST Journal on Hydrology and Water Resources (In Press).
Pechlivanidis I.G., Jackson B., McMillan H. and Gupta H., Use of an entropy-based metric in multi-objective calibration to improve model performance. In preparation for the Water Resources Research.
Perrin, C., Andreassian, V., Rojas Serna, C., Mathevet, T. and Le Moine, N., 2008. Discrete parameterization of hydrological models: Evaluating the use of parameter sets libraries over 900 catchments. Water Resources Research, 44(W08447).
Reichert, P. and Mieleitner, J., 2009. Analyzing input and structural uncertainty of nonlinear dynamic models with stochastic, time-dependent parameters. Water Resources Research, 45(W10402).
Reusser, D.E. and Zehe, E., 2011. Inferring model structural deficits by analyzing temporal dynamics of model performance and parameter sensitivity. Water Resources Research, 47(W07550).
Schaefli, B., Harman, C.J., Sivapalan, M. and Schymanski, S.J., 2011. HESS Opinions: Hydrologic predictions in a changing environment: behavioral modeling. Hydrology and Earth System Sciences, 15: 635-646.
Toth E., 2009. Classification of hydro-meteorological conditions and multiple artificial neural networks for streamflow forecasting. Hydrology and Earth System Sciences, 13: 1555–66.
Weijs, S.V., Schoups, G. and van de Giesen, N., 2010. Why hydrological predictions should be evaluated using information theory. Hydrology and Earth System Sciences, 14(12): 2545-2558.
Wheater H.S. and Evans E.P., 2009, Land use, water management and future flood risk, Land Use Policy, 26S, S251-S264.
Wilby R.L. and Harris I., 2006, A framework for assessing uncertainties in climate change impacts: Low-flow scenarios for the River Thames, UK, Water Resources Research, 42(W02419).