Numerical Weather Prediction and Relative Economic Value framework to improve Integrated Urban Drainage- Wastewater management

Abstract

Integrated urban drainage-wastewater systems (IUDWSs) are challenged by the need for higher environmental and health standards and the increased fre-quency of heavy rain storms caused by climatic change. Real-time control (RTC) offers an alternative to the construction of costly facilities in cities where space is scarce and large-scale construction work a nuisance. This the-sis focuses on flow domain predictions of IUDWS from numerical weather prediction (NWP) to select relevant control objectives for the IUDWS and develops a framework based on the relative economic value (REV) approach to evaluate when acting on the forecast is beneficial or not. Rainfall forecasts are extremely valuable for estimating near future storm-water-related impacts on the IUDWS. Therefore, weather radar extrapolation “nowcasts” provide valuable predictions for RTC. However, radar nowcasts are limited by their prediction horizon of 1 to 2 hours and RTC of IUDWS could benefit from longer forecast horizons. The development of NWP models in parallel to the increase in computational power has led to limited area models (LAM) with increasingly finer spatial-temporal resolution, opening the possibility to use such weather forecast products in urban water management. NWPs are complementary to radar forecasts, providing predictions on a longer time scale (days). However, atmospheric motions are chaotic and highly nonlinear. Applying NWP to urban catchments, which often have a similar size to a single NWP grid cell, is limited by scientific gaps on how to deal with this poor spatial and temporal resolution for urban hydrology application, its predictive skills and uncertainty, etc. Forecast uncertainty is commonly described by meteorologist using ensemble prediction systems (EPS). This thesis used the outcome of the DMI-HIRLAM-S05 model which generates an EPS of 25 members. Each forecast ensemble provides hourly time step predictions over a forecast horizon of 54 hours with a horizontal resolu-tion of 0.05° (approx. 5.6 km). In order to evaluate the predictions based on the end-user perspective, namely the flow in the IUWDS, a case study was established for the urban catchment of Damhuså in Copenhagen, Denmark, and a rainfall-runoff model associated to it was developed. Hence, the predictions were therefore not assessed against observed rainfall but against observed flow at the end of the coupled hydro-meteorological model chain. The predictions were assessed based on flow domain prediction, distinguishing between high and low flow events. The combination of the different possibilities between observations and fore-casts can be summarised in the 2x2 contingency table containing the four possible pairs of forecast-observation: hits (correct positives), false alarms (false positives), misses (false negatives) and correct negatives. The outcome of the contingency table were used to calculate skill scores like the probability of detection (PoD) and the probability of false detection (PoFD), and to plot the relative operating characteristic (ROC) diagram illustrating the discrimination skill of the NWP EPS prediction. Using verification methods from meteorology on flow predictions showed that NWPs have poor precision at such fine resolution. Therefore, NWP post-processing methods are necessary to cope with this limitation, getting the most out of the NWP to enhance its EPS, especially towards reducing the oc-currence of missed events. This thesis investigates two NWP post-processing approaches: The neighbourhood method which includes predicted rainfall from nearby grid cells, and the time lagged ensemble (TLE) which utilises the overlap between consecutive NWP generations to expand the EPS. Both ap-proaches have shown to be beneficial to enhance the NWP EPS, reducing the occurrence of missed events. However these approaches can lead to a large increase of the EPS size. The maximal threat neighbourhood method devel-oped in this thesis improved the EPS discrimination skill with a limited in-crease of the EPS size. Despite these improvements, the high uncertainty embedded in NWP prevents the use of quantitative rainfall values directly for an urban catchment. NWP should be used, instead, in connection with a domain-based decision frame-work, predicting for which domain the IUDWS should be optimized. The following domain-based framework was suggested, distinguishing between 4 operational domains: (i) dry weather conditions with storage basins empty, (ii) dry weather conditions with stored water, (iii) wet weather conditions within the system capacity and (iv) wet weather conditions that exceed the system capacity. Handling uncertainty is challenging for decision makers. Tools are necessary to provide insight on when acting based on uncertain forecast data is beneficial or not. The REV framework developed in this thesis provides a tool to evaluate the added benefit of using control strategies based on uncertain forecast information, to select the most relevant control parameters and to com-pare different control strategies. This REV framework was applied on two case studies. The first one used NWP EPS to predict low flow domains during which the IUDWS can be coupled with the electrical smart grid to optimise its energy consumption. The REV framework was used to determine which decision threshold of the EPS (i.e. number of ensemble members predicting an event) provides the highest benefit for a given situation. In the second case study, the REV framework was used to evaluate the current control strategy switching the WWTP to wet weather operation and to assess other control parameters and strategies. The analysis of the current control showed a significant number of false alarms, and the REV framework was used to calibrate the threshold on radar flow prognosis and to test new control strategies to solve this problem. Uncertainty communication to end-users is a critical and challenging part of forecast usage. It can be achieved through the REV framework, which in-cludes the possibility of considering management mistakes due to erroneous forecast. The REV framework assesses the overall benefit including these potential mismanagements, hence, maintaining the operator trust in the control.