Integrating the flow regime and water quality effects into a niche-based metacommunity dynamics model for river ecosystems.

Aquatic community dynamics are closely dominated by flow regime and water quality conditions, which are increasingly threatened by dam regulation, water diversion, and nutrition pollution. However, further understanding of the ecological impacts of flow regime and water quality conditions on aquatic multi-population dynamics has rarely been integrated into existing ecological models. To address this issue, a new niche-based metacommunity dynamics model (MDM) is proposed. The MDM aims to simulate the coevolution processes of multiple populations under changing abiotic environments, pioneeringly applied to the mid-lower Han River, China. The quantile regression method was used for the first time to derive ecological niches and competition coefficients of the MDM, which are demonstrated to be reasonable by comparing them with the empirical evidence. Simulation results show that the Nash efficiency coefficients for fish, zooplankton, zoobenthos, and macrophytes are more than 0.64, while the Pearson correlation coefficients for them are no less than 0.71. Overall, the MDM performs effectively in simulating metacommunity dynamics. For all river stations, the average contributions of biological interaction, flow regime effects, and water quality effects to multi-population dynamics are 64%, 21%, and 15%, respectively, suggesting that the population dynamics are dominated by biological interaction. For upstream stations, the fish population is 8%-22% more responsive to flow regime alteration than other populations, while other populations are 9%-26% more responsive to changes in water quality conditions than fish. For downstream stations, flow regime effects on each population account for less than 1% due to more stable hydrological conditions. The innovative contribution of this study lies in proposing a multi-population model to quantify the effects of flow regime and water quality on aquatic community dynamics by incorporating multiple indicators of water quantity, water quality, and biomass. This work has potential for the ecological restoration of rivers at the ecosystem level. This study also highlights the importance of considering threshold and tipping point issues when analyzing the "water quantity-water quality-aquatic ecology" nexus in future works.

Characterisation of the dynamics of past droughts.

Drought is a complex natural phenomenon. The description of the way in which drought changes (moves) in space may help to acquire knowledge on its drivers and processes to improve its monitoring and prediction. This research presents the application of an approach to characterise the dynamics of drought. Tracks, severity, duration, as well as localisation (onset and end position), and rotation of droughts were calculated. Results of calculated droughts were compared with documented information. Data from the Standardized Precipitation Evaporation Index (SPEI) Global Drought Monitor was used to identify droughts in India as an example for the period 1901-2013. Results show regions where droughts with considerable coverage tend to occur. Paths, i.e. consecutive spatial tracks, of six of the most severe reported droughts were analysed. In all of them, areas overlap considerably over time, which suggest that drought remains in the same region for a period of time. Results of this research are being used to build a model to predict the spatial drought tracks, incl. India (https://www.researchgate.net/project/STAND-Spatio-Temporal-ANalysis-of-Drought).

Assimilating flow and level data into an urban drainage surrogate model for forecasting flows and overflows.

It is crucial to be able to forecast flows and overflows in urban drainage systems to build good and effective real-time control and warning systems. Due to computational constraints, it may often be unfeasible to employ detailed 1D hydrodynamic models for real-time purposes, and surrogate models can be used instead. In rural hydrology, forecast models are usually built or calibrated using long historical time series of, for example, flow or level observations, but such series are typically not available for the ever-changing urban drainage systems. In the current study, we therefore used a fast, reservoir-based surrogate forecast model constructed from a 1D hydrodynamic urban drainage model. Thus, we did not rely directly on historical time series data. Forecast models should preferably be able to update their internal states based on observations to ensure the best initial conditions for each forecast. We therefore used the Ensemble Kalman filter to update the surrogate model before each forecast. Water level or flow observations were assimilated into the model either directly, or indirectly using rating curves. The model forecasts were validated against observed flows and overflows. The results showed that model updating improved the forecasts up to 2 h ahead, but also that updating using water level observations resulted in better flow forecasts than assimilation based on flow data. Furthermore, updating with water level observations was insensitive to changes in the noise formulation used for the Ensemble Kalman filter, meaning that the method is suitable for operational settings where there is often little time and data for fine-tuning.

Knowledge-based modularization and global optimization of artificial neural network models in hydrological forecasting.

Natural phenomena are multistationary and are composed of a number of interacting processes, so one single model handling all processes often suffers from inaccuracies. A solution is to partition data in relation to such processes using the available domain knowledge or expert judgment, to train separate models for each of the processes, and to merge them in a modular model (committee). In this paper a problem of water flow forecast in watershed hydrology is considered where the flow process can be presented as consisting of two subprocesses -- base flow and excess flow, so that these two processes can be separated. Several approaches to data separation techniques are studied. Two case studies with different forecast horizons are considered. Parameters of the algorithms responsible for data partitioning are optimized using genetic algorithms and global pattern search. It was found that modularization of ANN models using domain knowledge makes models more accurate, if compared with a global model trained on the whole data set, especially when forecast horizon (and hence the complexity of the modelled processes) is increased.

Machine learning approaches for estimation of prediction interval for the model output.

A novel method for estimating prediction uncertainty using machine learning techniques is presented. Uncertainty is expressed in the form of the two quantiles (constituting the prediction interval) of the underlying distribution of prediction errors. The idea is to partition the input space into different zones or clusters having similar model errors using fuzzy c-means clustering. The prediction interval is constructed for each cluster on the basis of empirical distributions of the errors associated with all instances belonging to the cluster under consideration and propagated from each cluster to the examples according to their membership grades in each cluster. Then a regression model is built for in-sample data using computed prediction limits as targets, and finally, this model is applied to estimate the prediction intervals (limits) for out-of-sample data. The method was tested on artificial and real hydrologic data sets using various machine learning techniques. Preliminary results show that the method is superior to other methods estimating the prediction interval. A new method for evaluating performance for estimating prediction interval is proposed as well.