30 Nov Forecasting the impacts of storm events on Lough Feeagh in near real time
This is an entry originally posted in the PROGNOS blog (http://prognoswater.org/), but due to its relevance to WATExR, we want to share it here as well. Used with permission from the authors.
By Elvira de Eyto (Marine Institute), Tadhg Moore (Dundalk Institute of Technology), Mary Dillane (Marine Institute), Eleanor Jennings (Dundalk Institute of Technology), and Martin Rouen (Lakeland Instrumentation)
A central aim of the PROGNOS project is to couple short term weather forecasts with lake models, in order to produce real time forecasts of water quality. To do this, we need to calibrate and parameterise 1D hydrodynamic models (which can predict lake water temperature and thermal structure) and then drive these models with weather forecast data. The forecast of (e.g.) lake thermal structure can then be checked against near-real time observational data, and adjusted if necessary. In this post, we will talk about some of the progress we are making towards setting up the infrastructure that is required to access real time data from lake monitoring stations. There are of course many different ways of doing this, and each solution will be tailored to the lake in question, who is monitoring it, and what sensor hardware is being used. Here we explain the data flow for Lough Feeagh, the Irish PROGNOS case study site. For an insight into the development of lake monitoring at this site, have a look at this previous blog post (http://prognoswater.org/guinness-brown-lakes-and-lter/ ).
The Lough Feeagh AWQMS (automatic water quality monitoring station) was installed in 1996, and originally used VHF telemetry to transfer data from the data logger to a computer in the Marine Institute’s field station in Furnace. More recently, telemetry has been vis GPRS, and data are transmitted at sub hourly frequency (almost real time) to a computer in our office. From there, we use an off-the-shelf software package from Campbell Scientific, LNDB (https://www.campbellsci.com/lndb ) to copy data to a SQL server in the Marine Institute’s Galway office. This happens in near real time, so theoretically we can access data from 2 minutes ago if we want. We then utilise a really nice service called ERDAPP (https://erddap.marine.ie/erddap/information.html ) to access data stored on the SQL server and deliver it (as open access) to whoever wants it (Fig. 1).
ERDAPP was developed by NOAA (National Oceanographic and Atmospheric Administration www.noaa.gov ) to allow easy access to scientific data from a wide range of data providers, and the Marine Institute maintains its own ERDAPP server. The Lough Feeagh data can be accessed through this server at this link: (https://erddap.marine.ie/erddap/tabledap/IMINewportBuoys.html ). Anyone can access this data and either take a look at it through the generic graphing facility on the ERDAPP webpage (red circle in Fig. 2), or else use a data download link to bring the data into e.g. a python or R script for further processing. (Red arrow in Fig.2).
Once this workflow was set up, we were then able to use it to download the previous couple of months of data from Lough Feeagh, use these data as boundary conditions for GOTM (http://gotm.net/ ) and then run GOTM forward for ten days using meteorological forecast data available from the Norwegian meteorological service which has an open access facility (www.yr.no ).
As a demonstration, we did this for the week leading up to Storm Ali, which hit the west coast of Ireland on the 19th of September 2018 (Fig 3 from https://earth.nullschool.net).
The lake was already starting to destratify in the 2 weeks leading up to this event, but we guessed that this storm would lead to the lake being fully mixed, and we wanted to see whether our forecasting process would capture this (spoiler – IT DID! J). We used 7-day weather forecasts with a 3-hour timestep for 6 days and then 6-hour for the last day. This weather forecast data was used to predict the lake thermal structure, and this output was then used to calculate Schmidt stability using the rLakeAnlyzer package (https://cran.r-project.org/web/packages/rLakeAnalyzer/rLakeAnalyzer.pdf ). Model data are shown in red in Fig. 4, while observed data from L. Feeagh AWQMS are shown in blue. The red dashed line shows where the weather forecast begins. We still have a lot of work to do on this process, including integrating the inflow data into the model in real time, but we are moving in the right direction.