Interview between the International Water Association (IWA) and Dr. Ilias G. Pechlivanidis, Scientific Leader in Forecasts of Water Variables at the Swedish Meteorological and Hydrological Institute (SMHI), Project Manager of the EFAS Dissemination Centre, and Co-Chair of the HEPEX scientific experiment. He further represents SMHI as member of the SPACE-O project consortium.
The Space-O project provides digital solutions for water reservoir managers and water utility operators. By combining satellite technology, hydrodynamic and hydrology modelling software and in-situ monitoring, SPACE-O translates state-of-the-art technology into digital tools tailored for end-users. The forecasting of water quality parameters is one key functionality of those SPACE-O tools, which we asked Ilias to elaborate on.
IWA: How do you address the uncertainty in forecasting? What is done to ensure confidence in the forecasted information?
In SPACE-O, we forecast water quantity and quality in lakes, rivers and reservoirs up to 10 days into the future. However, uncertainty is present in every step of the forecasting chain, given that every model (meteorological and hydrological), and every forecasting service is always prone to uncertainty. The challenge is to reduce uncertainty as much as possible to enhance the confidence in decision-making. To address the SPACE-O needs, we set a long forecasting chain, with the meteorological forecasts, the initial conditions of the hydrological and hydrodynamic model, the hydrological model structure, parameters and data used for setting up the models being sources of uncertainty. Although a service that accounts for all these sources of uncertainty simultaneously is scientifically possible, operational challenges (i.e. computational expense) limit such potential. Therefore, in SPACE-O we concentrate on the sources which are expected to have significant impact on the hydrological forecasts.
We are working with state-of-the-art meteorological forecasts of precipitation and temperature, based on the deterministic forecasting system of the European Centre for Medium-Range Weather Forecasts (ECMWF). To address the meteorological uncertainty and its propagation into the hydrological forecasts, we upgraded our service using the ECMWF probabilistic meteorological forecasts that further force the hydrological and the hydrodynamic models.
“The probabilistic ECMWF system consists of 51 members and hence 51 forecasts, to account for the uncertainty in the meteorological input. From the large set of model forecasts (51), we define the level of confidence for each forecasted day, hence resulting into decision-making based on quantified uncertainty.“
In addition to the probabilistic forecasting, in one of the case studies (see next question), we have been using meteorological forecasts from the high-resolution Numerical Weather Prediction (NWP) model named HARMONIE-AROME. Although this is a deterministic meteorological forecasting system, it operates at a high spatial resolution (2×2 km), meaning that we can represent the precipitation and temperature fields more accurately than other state-of-the-art systems. Regarding the initial conditions of the hydrological and hydrodynamic models, we have been assimilating observations from in-situ as well as from earth observations in order to accurately represent the near real-time conditions of the physical catchment system.
Moreover, with regard to the confidence of the forecasts, we firstly analyse the performance of our hydrological and hydrodynamic systems. This is based on a comparison of the model results to the observations in a historical time period. The variable of interest is based on the user request, for instance users could be interested in when river flooding will occur. Note that accurate and timely provision of forecasts would have a big influence on decision making. Another example would consider a user interested in discharge information of minimum volumes in water bodies, which again affects decision-making in terms of water management.
“There are different interests in hydrological variables, and in order to harmonize the way of communicating model forecasting skill, we use a skill metric which ranges from 0 to 1 showing poor and good performance respectively. The forecasting skill is generally high on the first day (reliable forecasts), but deteriorates with increasing lead time (as we forecast further into the future).”
We conducted evaluation exercises to analyse the predicted confidence in our service for different variables, and how this changes with the lead day.
It is therefore important for our service to identify the critical day before which we are confident in providing good forecast, and after that day the confidence is actually dropping.
IWA: What is the purpose of the scientific case studies in SPACE-O? How are the results being integrated?
In SPACE-O we have the pilot case studies where we are implementing and presenting a complete service line with state-of-the-art techniques that are scientifically sound and with assured quality. However, in an operational service line there is not much flexibility for scientific experiments which allow for deeper understanding. In SPACE-O, we were having two scientific case studies; one in Umeälven River in Sweden, and one in Lake Garda in Italy. These two basins were selected because of extensive available in-situ data, as well as earth observations; together allowing for in-depth analysis. We have been using snow information, specifically fractional snow cover and snow water equivalent and actual and potential evapotranspiration from different EO products and compared them with the hydrological model outputs. In the case of model-observations agreement, which would mean small model biases, we can allow assimilation in order to improve the initial conditions of the hydrological model; and accurate initial conditions result in reliable forecasts.
“We therefore investigated individually the added value of the EO products in hydrological modelling, and we further investigated as part of a sensitivity analysis, the optimum combination of the EO products in order to maximise the beneficial potential from using EOs.”
Moreover, we extended the investigation on the forecasting mode. We hence compared the hydrological forecast based on the state-of-the-art ECMWF meteorological forecasts and on the high-resolution Numerical Weather Prediction (NWP). The technical difference between the two meteorological products is mainly based on the spatial resolution, which results into more accurate representation of localised meteorological events. However, the use of a finer spatial resolution NWP does not necessarily link to improved hydrological forecasts, because the river system is generally acting as a filter of the meteorological information, and hence the forcing signal can be damped from the physical local/regional processes and conditions (e.g. the catchment size, dry or wet conditions, urbanised or rural physical properties).
“The SPACE-O scientific case studies therefore aim to identify where and under which conditions we get an added forecasting value by introducing a high-resolution, yet computationally expensive, NWP.“
IWA: How can modelled and forecasted information be effectively interpreted by water utilities? What approaches are being used?
In operational services, visualisation is an important task. In SPACE-O we visualize results with techniques based on experience in setting and operating similar services in our institute in Sweden as part of national or international projects. User training is also an important component in the uptake of the SPACE-O service. We act as knowledge brokers and run workshops for users to understand their needs and tackle those through the service. Language is therefore a very important tool for communication since the information that we provide needs to be easily understood.
“We believe that the better we communicate the forecasting information to the users, the smoother and more accurate the decision making will be.”
In addition, in the SPACE-O platform we introduce an option to provide information and metadata about the variables of interest, e.g. definition, units etc. In this way, the users can have a better understanding of how forecasted information can further be used for decision-making. Since communication and visualization methods are always upgraded, we constantly work on narrowing the gap between service providers, users and decision-makers.
This interview was conducted on September 07, 2018, between Ilias Pechlivanidis and Hanno Führen (on behalf of the International Water Association).