Artificial intelligence and geospatial information to correct watershed model forecasts
Although watershed models are achieving a mature status to estimate water flow and other related properties there are some limitations for their application and forecast. Among these limitations, some are not derived by the uncertainty of the numerical models (meteorological or watershed) but related to the socio-economic aspects that cannot be easily quantified. These socio-economic aspects are related to the water retention in dams, human consumption, irrigation practices, etc. Since these influences are not easily model explicitly, especially for large watersheds such as the Tagus River, and since some of them are related to other independent variables (i.e., population, rain, seasonal agriculture practices, daily and seasonal dams’ operation, previous months climate, etc..) the approach chosen will be to apply and train neural networks and artificial intelligence methods to improve and correct the outcomes of operational watershed models.
The PhD will need to analyse spatial land-use information, rain patterns, population evolution and past and future scenarios.
Candidates
AIR Centre PhD grants are aimed at applicants enrolled or that comply with the requirements to enrol for PhD related studies and who wish to carry out research towards this degree. Candidates holding a master’s degree or a 5-year undergraduate diploma in Information Management, Geographic Information Systems, Environmental Engineering or related fields may apply. Good skills on Fluid Mechanics (from oceanography or engineering) and programming are preferred. Candidates must explain in the motivation letter their vision of the problem to be solved and why they have the skills necessary to carry the job. Candidates must present a tentative workplan for the applied theme.
Working Hypothesis
In recent times, watershed modelling results have emerged as a possibility to provide accurate land-boundary conditions and to substitute traditionally used river climatologies. However, both products, river climatologies and watershed models, have their own limitations. River climatologies main weakness is the incapacity to include the interannual variability compared to watershed model applications which, in general, follow the main river flow trends (Campuzano et al., 2016). On the other hand, watershed models tend to overestimate river flows, especially during dry seasons when due to reduced availability of fresh water, human management activities such as dams’ retention, irrigation, human consumption, etc., alter more intensively the natural river flow.
The PhD candidate will explore the use of a numerical model as a valuable tool for completing the observed data. The main goal of this task is to estimate the amount of freshwater entering the coastal areas of Western-Iberia and to reconstruct the freshwater influx in the coastal area in the last 25 years, comparing against observed flows. The MOHID Land model will be used to calculate river flow and water temperature.
The MOHID Land model, part of the MOHID Modelling System (http://www.mohid.com; Neves, 2013) is the numerical model used to simulate the water contributions in the Iberian Peninsula. MOHID Land is a physically based, spatially distributed, continuous, variable time step model for the water and property cycles in inland waters. The numerical model is open-source, and its code can be accessed in the following GitHub repository (https://github.com/Mohid-Water-Modelling-System/Mohid).
At the watershed level, the MOHID Land model estimates operationally water flow and associated properties (i.e., temperature, oxygen and nutrient concentrations) for the main river catchments discharging in the Western Iberian coast. Two domains with different horizontal resolution, 2 km for the West Iberia region (WI domain) and 10 km for the Iberian Peninsula (IP domain), were designed to represent adequately the Portuguese catchments and to include the spatial scale of the large trans-boundary rivers as the Tagus, Douro and Guadiana rivers (Brito et al., 2015).
Currently, the watershed model applications disregard the effects of human consumption, water reservoirs or dam management that could modify the amount and timing of the water reaching the coastal zone. AI methods to correct these socio-economics interventions will be implemented during the PhD project taking into consideration geographic spatial information.
The main objective is to develop a system that can predict estuarine river flow through the use of Machine Learning models. Estimating river flow is essentially a regression problem. The choice of architecture is dependent on the type of data and its volume. Therefore, several algorithms may be considered, such as Neural Networks, Random Forests and Polynomial Regression. The research subject should include a machine learning model that will allow the estimation of river estuarine flow, using historical values on the multiple properties that seem to control the operational decisions (e.g., reservoir discharge flow, precipitation in the water basin in the previous periods, population and density, changes in agricultural practices, etc.).
The main obstacle to the development of a general model is the existence of dams: since each river has its socioeconomic regional context, the usage of the dams’ reservoirs varies, and so does the response of the river flow to environmental conditions. If the results are successful, the methodology will be replicated and integrated with the watershed model, contributing to a smarter and more reliable estimation of fresh waters reaching the coastal area that will correct non-monitored watersheds and will contribute for more precise forecasts.
Name(s) of supervisor(s)
Supervisor- Prof. Dr. Marco Painho, painho@novaims.unl.pt, Universidade Nova de Lisboa, Portugal
Co-Supervisor – Dr. Francisco Campuzano, Francisco.campuzano@colabatlantic.com, +ATLANTIC CoLAB, Portugal.
Name(s) of hosting institution(s)
MagIC is the R&D center of NOVA IMS, one of the worldwide leading schools in the area of Information Management. NovaIMS is also ranked in the TOP 5 of the Best Masters and Postgraduate Programs in the world by the Eduniversal and the first European Institution to obtain the ABET accreditation. With over 50 integrated researchers and more than 1000 published articles since 2011, MagIC is organized around 4 research streams: Data Science, Information Systems, Geoinformatics, and Data-Driven Marketing. MagIC is continuously contributing for the advancement of Information Management and Data Science fields, as the ability to find insights in our data-intensive society translates into new and clever solutions for pressing societal challenges.
Identification of PhD program
PhD Program in Information Management offered by Nova Information Management of the Universidade Nova de Lisboa
Please check program’s website for PhD application deadlines: https://www.novaims.unl.pt/en/education/programs/doctoral-program-in-information-management/doctoral-program-in-information-management/
Jury Composition
Prof. Dr. Marco Painho, painho@novaims.unl.pt, (Chairman)
Dr. Marta Lopes, mib.lopes@fct.unl.pt, Universidade Nova de Lisboa, Portugal
Dr. Pedro Montero, pmontero@intecmar.gal, INTECMAR, Spain
Dr. Pablo Carracedo García, pablo.enrique.carracedo.garcia@xunta.gal, MeteoGalicia, Spain
Dra. Susana Vinga susanavinga@tecnico.ulisboa.pt, Instituto Superior Técnico, Portugal
References
Awwad, Haitham M., Juan B. Valdés, and Pedro J. Restrepo. “Streamflow forecasting for Han River basin, Korea.” Journal of Water Resources Planning and Management 120.5 (1994): 651-673.
Brito D., Campuzano F.J., Sobrinho J., Fernandes R., Neves R. Integrating operational watershed and coastal models for the Iberian Coast: Watershed model implementation – A first approach (2015). Estuarine, Coastal and Shelf Science; 167, Part A: 138-146.
Campuzano F, Brito D, Juliano M, Fernandes R, de Pablo H, Neves R. (2016) Coupling watersheds, estuaries and regional ocean through numerical modelling for Western Iberia: a novel methodology. Ocean Dynamics, 66(12): 1745–1756.
Cheng, Yuan. Evaluating an autoregressive model for stream flow forecasting. Hydraulic Engineering. ASCE, 1994.
Dibike, Yonas B., and Dimitri P. Solomatine. River flow forecasting using artificial neural networks. Physics and Chemistry of the Earth, Part B: Hydrology, Oceans and Atmosphere 26.1 (2001): 1-7.
Firat, M. Comparison of artificial intelligence techniques for river flow forecasting. Hydrology and Earth System Sciences 12.1 (2008): 123-139.
Imrie, C. E., S. Durucan, and A. Korre. River flow prediction using artificial neural networks: generalisation beyond the calibration range. Journal of hydrology 233.1-4 (2000): 138-153.
Kumar, D. Nagesh, K. Srinivasa Raju, and T. Sathish. “River flow forecasting using recurrent neural networks.” Water resources management 18.2 (2004): 143-161.
Neves R (2013). The Mohid concept. Case studies with MOHID, M. Mateus & R. Neves (eds.), IST Press. pp 1-11
Yaseen, Zaher Mundher, et al. “An enhanced extreme learning machine model for river flow forecasting: State-of-the-art, practical applications in water resource engineering area and future research direction.” Journal of Hydrology 569 (2019): 387-408.