Living European Rivers and Global Dam Watch Workshop2019-09-16T16:41:30+01:00

Project Description

Global Dam Watch and Living European Rivers Workshop

April 2019, Zeist, Netherlands

Workshop summary notes

Overview

WWF and Global Dam Watch hosted a three-day workshop in April of 2019 in Zeist, Netherlands. Over the course of the three days, nearly 30 participants discussed the collection and harmonization of dam and reservoir data, the effects of infrastructure on nature and people, and the considerations involved in carrying out a free-flowing rivers analysis for Europe.

The backbone of Global Dam Watch’s initial data offering is made up of the three core datasets: the Global Georeferenced Database of Dams, the Global Reservoir and Dam Database, and the Future Hydropower Reservoir and Dams. The core team is in the process of harmonizing these datasets so they can be used on their own or in combination with little pre-processing.

Discussion also included new methods of dam and reservoir data collection. These included using different citizen science platforms and using satellite imagery and machine learning. How data derived from these new methods will be incorporated into GDW data is still being explored.

In the near term, data of this nature will be used to support WWF’s Living European Rivers project, which will undertake a free-flowing rivers analysis at the European scale. Discussion here focused on the impacts that dams and reservoirs have on biodiversity and nature’s contributions to people, and on the technical challenges that will come with carrying out what started as a global analysis at a higher resolution.

  • Amadou Crookes, Upstream Tech
  • Arnout Van Soesbergen, King’s College London
  • Bart Geenen, WWF NL
  • Bernhard Lehner, McGill University
  • Christiane Zarfl, Eberhard Karls Universität Tübingen
  • David Gernaat, Netherlands Environment Agency
  • Guenther Grill, McGill University
  • Jan Janse, Netherlands Environment Agency
  • Jonathan Higgins, Nature Conservancy
  • Josh Jones, Swansea University
  • Kate Brauman, University of Minnesota
  • Leonhard Burger, Osnabruck University
  • Lisa Mandle, Natural Capital Project/Stanford University
  • Luca De Felice, EU Joint Research Centre
  • Mark Mulligan, King’s College London
  • Michele Thieme, WWF US
  • Natalie Shahbol, WWF US
  • Oskar de Roos, WWF NL
  • Penny Beames, McGill University
  • Peter van Puijenbroek, Netherlands Environment Agency
  • Qingke Wen, EU Joint Research Centre
  • Stefan Schmutz, Vienna University
  • Stephanie Januchowski-Hartley, Swansea University
  • Ulrich Schwarz, Fluvius Consultant
  • Valerio Barbossa, GLOBIO/DRIP
  • Wouter Van de Bund, AMBER

Quick wins

Census Dataset

The team agreed on the creation of an initial harmonized GDW dataset that would include include the three core datasets, key attributes like construction year, dam purpose, reservoir volume, and the minimum and maximum reservoir extent. The dataset would account for duplicates to make them easy to filter for analysis.

Comment paper

Participants discussed the drafting of a comment paper, the basic structure of which would cover the studies have already made use of our data; the work that has gone into creating these datasets, identification of gaps, and the work we plan to undertake to fill those gaps; the challenges associated with mapping different types of river barriers and the steps to be taken to work through these issues; the methodology to bring together disparate datasets and collection methods (including machine learning, citizen science), and where we are in the sequence; and a diagram that illustrates the framework and working paths.

The paper would also act as a skeleton or agenda for subsequent funding proposals.

Global Dam Watch’s Consensus Dataset

The idea behind the consensus GDW dataset is to harmonize all three of the core datasets (GRanD, GOOD2, and FHRed) to create one global dam and reservoir dataset.

The core datasets would remain available should users prefer these over the consensus data.

Harmonization efforts

The first pressing question for a consensus database is how do we define a dam? At this point, the initial efforts will focus on harmonizing existing data, but the question of dam definition is one we will revisit in subsequent meetings.

When harmonizing the three existing datasets, the most important consideration is to ensure that duplicates between datasets are easily identified and dealt with so as to prevent errors in analysis.

The core datasets would remain available should users prefer these over the consensus data.

Desired attributes to be included

Discussion both in the plenary and online supports an understanding that most users need at least the following key attributes:

  • Dam spatial location — This location can be true to dam location in satellite imagery or snapped to a hydrological network; if snapped to a hydrological network, transferring points to a different network is fairly straightforward
  • Dam height and reservoir volume
  • Dam purpose
  • Age of the dam
  • Minimum and maximum reservoir extent — This spatial data will be available soon through the EU Joint Research Commission’s efforts on using satellite imagery to track reservoir extents, however this raises an interesting question: if a reservoir ends at the dam, where does it begin?
  • Reservoir capacity — Remote sensing techniques being developed by the JRC will enable very sophisticated estimates of reservoir volume

An open discussion highlighted a number attributes that would be of use to a variety of studies. The group discussed the possibility of creating a system that would allow users to download the data and select from the following additional attributes:

  • Reservoir or run-of-river — The definition here could be very slippery as some dams are classified as run-of-river but still create a reservoir
  • Dam operation — This may be difficult as privacy or commercial interests may impede inclusion, though at some point remote sensing may give us a sense of operations based on fluctuations in reservoir extent
  • Intra and interannual fluctuations in volume — Is this reservoir always a reservoir, or does it only fill up every few years during a flood event?
  • Megawatts for hydropower dams — This is already in FHReD, could be transferred into consensus GDW dataset
  • Irrigation extent for irrigation dams — This is more difficult to ascertain, but there are methods that allow for modelling of this information
  • Passability — Are there fish passes? If so, what kind? It was noted that this is where dam height becomes really important: there are statistical methods that can be used to extrapolate dam height using associated characteristics (see Januchowski-Hartley SR, Jézéquel C, Tedesco PA. 2019. Modelling built infrastructure heights to evaluate common assumptions in aquatic conservation. J. Environ. Manage. 232:131–137.)
  • Lake control structure — Was this a natural lake that has been modified using a dam or weir? For instance, Lake Superior is considered a reservoir because it is dammed. This is a really important consideration and also very difficult to ascertain
  • Year built — Remote sensing techniques could help to estimate year for dams built after the 1980s; this attribute is more difficult to fill for planned dams as numerous factors to cause delays in construction
  • Year removed — Has a dam been dismantled? Has it fallen down or has the reservoir filled up?
  • Commissioning year — commissioning year may be different from construction year and even reservoir filling year, and could give an indication as to when normal reservoir operations came online
  • Dam purpose — The purpose of a dam will give an indication as to its operations, which ultimately influences its impacts. Where a dam purpose is not known, statistical methods could be used to make an educated guess based on the characteristics of the catchment in which the dam sits
  • Dam name — This can be extremely difficult as dams may have multiple names or transliterations, though crowd-sourcing may prove helpful in naming smaller dams
  • Length of the reservoir — This can be calculated using GRanD polygons and the JRC’s remote sensing techniques
  • Temperature — Finer temperature attributes include the location of the reservoirs intake and/or release point
  • Runoff — Average and seasonal values, if possible; this could be populated by the upcoming HydroATLAS publication that will include runoff
  • Downstream discharge — This could also be populated using HydroATLAS
  • The presence of a spillway — Moreover, is that considered flow loss or is the spillway just part of normal operations?
  • Flood and other types of inundation modelling
  • Atmospheric interactions — This could come from surface area, evaporation
  • Sediment retention — Are there sediment flushing technologies implemented in the structure? As standards exist for this, the data should be available somewhere
  • Flow disturbance, fragmentation — This could come from the new connectivity index, but the dataset would need to include smaller dams if this indicator were to be useful at anything closer than the global scale
  • Species extent data
  • Ecological flow guidelines in the region of the dam
  • Species richness — This may need to be carried out at the catchment scale rather than global, but a global assessment could be used as an initial screening tool
  • Health indicators — Mosquitoes as disease vectors in warm and shallow reservoirs
  • Nutrition, drinking water availability

Plan for updates

We want GDW to be a living dataset, and as such it needs to evolve as dams are built and removed from the landscape.

Citizen science projects like DRIP could also be used to add data to the GDW consensus dataset. DRIP is ongoing, and the tool could be tailored to suit tasks like merging databases. For instance, there are three points that overlap in space, the tool could be used to determine which of the points is “correct”, or to identify points that are catalogued by multiple datasets.

Remote sensing and machine learning can also be called upon to track new and removed dams. There are a few groups already working on this task; notes from them are below.

Future of European and Global Dam Data Creation

The workshop confirmed that there are a number of groups currently working on the dam and reservoir data creation puzzle. Each group uses a slightly different approach to tackle specific aspects of the task, and this variety could produce overlaps that help to resolve pressing issues.

AMBER

A lack of river connectivity is the main cause of rivers in Europe not being classified as rivers of good status according to the EU Water Framework Directive. However, assessing river connectivity is difficult because of inconsistent monitoring across Europe. Most of the national databases only contain large dams, but most of the dams in Europe are very small.

AMBER is being funded to help fill the data gap by creating a publicly available database of dams in Europe, even those that are very small. They are currently streamlining national datasets to including only certain data attributes from each dataset, and are removing duplicates.

The project has encountered some issues. For instance, countries are inconsistent in their coverage; France and Ireland have good coverage, but Italy only maps its larger dams. Some datasets are incomplete, and there are cases where GRanD and Ecrins held dam data that weren’t included in national datasets. Countries also collect data for different purposes, which introduces biases in the data that aren’t easily resolved. Moreover, important attribute information like dam height and construction year are often missing.

AMBER is using statistical models to produce dam density maps for all of Europe by using dam data from countries where the data is reliably complete to estimate dam density in places where data is incomplete. The first test of this used the UK as a case study (Jones J, Börger L, Tummers J, Jones P, Lucas M, Kerr J, Kemp P, Bizzi S, Consuegra S, Marcello L, et al. 2019. A comprehensive assessment of stream fragmentation in Great Britain. Sci. Total Environ. 673:756–762.). The study combined datasets to compile 24,000 barriers and build a dam density map of the UK, but if culverts and road crossings were to be considered the number would be more like 80,000 interruptions to river connectivity.

Citizen Science

Citizen science offers methods through which the general public can assist with projects such as dam and reservoir identification and classification.

Citizen science projects raise questions on longevity. The DRIP project (under Netherlands PBL) has stable funding and a stable team to support the tool. And, in order to generate and maintain participation in the project, clear tasks and guidelines must be communicated. Tasks need to be planned very clearly or else a lot of time can be wasted collecting things that aren’t needed, which can dissuade contributors. Citizen science can be used to validate rather than identify, and this is more motivating because contributors feel like they’re useful. Finally, engagement needs to be fun or no one will participate. Prizes and celebrations work!

Global River Obstructions Database (GROD)

GROD uses Google Earth Engine to map barriers a global dataset of river widths (Allen GH, Pavelsky TM. 2018. Global extent of rivers and streams. Science (80-. ). 361:585–588.) to map river obstructions on streams wider than 30m. Data developed as part of GROD is intended to seed imagery and projects associated with the upcoming SWOT satellite.

The GROD project also developed a training approach that uses a decision tree to step citizen scientists through the decision making process while they are mapping and identifying barriers. GROD users are 70-80% accurate in their mapping and identification, but difficulties arise when mapping low or permeable dams, or channel dams. Data is validated using French and American dam data.

GROD has generated a larger number of obstructions than those currently catalogued in GRanD or GOOD2, but the GROD project also included smaller barriers that weren’t considered by GRanD or GOOD2.

Dam and Reservoir Inventory Project (DRIP)

DRIP is an online tool that uses Google Earth to facilitate dam and reservoir identification. Data from the platform will be used to improve the Globio biodiversity modelling system. The platform uses jumps and plateaus in HydroSHEDS elevation to identify areas that have a high chance of being a dam/reservoir, and citizen scientists go through these ranked areas and mark whether they see a dam or not. Areas that are often falsely identified as probable reservoirs include waterfalls, steep mountain slopes, and data errors.

Harmonizing with GDW

The definitions of what types of obstructions are to be included is a problem that may never be solved entirely. Any hybrid of the core datasets and citizen science data would need to be combed for duplicates.

Remote Sensing

Many teams are now experimenting with remote sensing and machine learning to identify dams and reservoirs in satellite imagery. Three joined the workshop.

Upstream

Upstream is a private consulting firm that is using satellite imagery and machine learning to try to identify dams along river networks. Rather than seek out reservoirs as other similar efforts have done, Upstream’s team looks specifically for signs of built infrastructure. The team started by looking at satellite imagery of smaller dams from the NID. They tested different imagery at different resolutions, and of course high resolution imagery works best for identifying the smaller dams. System accuracy in the United States is roughly 65-70%, but their process includes a human check to reinforce correct identifications and therefore the system is improving.

The team is experimenting with using land use around dams as an additional indicator. They are also experimenting with calculating water temperature and sediment loads.

Natural Capital Project

The Natural Capital Project out of Stanford University is working on an object detection algorithm that would allow researchers to identify dams and reservoirs in satellite imagery.

The team used seed data from Brazil, Myanmar, Mekong, and the Volta and in initial tests have identified ~40,000 dams. They use RapidEye ~5m satellite imagery and are trying to incorporate river networks to filter out the industrial settling ponds that their algorithm detects. Manual validation will still be required.

The team has funding to develop the algorithm by the end of 2019 and will maintain whatever they develop in perpetuity.

EU’s Joint Research Commission

The JRC team is using satellite imagery to tackle the question: how do we capture the dynamics of water? The Global Surface Water Explorer (GSWE) is the first product of that process. The data set presented in app uses Google Earth Engine and the entire multi-temporal orthorectified Landsat 5, 7 and 8 archive, spanning the past 35 years, to map surface water change. Its layers include monthly and yearly water presence over 35 years, seasonality, intensity of water change over time, intra-annual distribution of water, water presence frequency and transitions that are broken into 10 classes. Clicking on a pixel brings up graphs of monthly water recurrence and water yearly history, and the app even allows users to drill down to monthly data from a specific year in a specific pixel.

The fully validated GSWE dataset (Pekel J.-F., et al. 2016. High-resolution mapping of global surface water and its long-term changes. Nature 540: 418) has been officially endorsed as the official indicator for monitoring progress towards target 6.6 (Ecosystems) of the Sustainable Development Goals (SDGs), of which UN Environment is the custodian agency. This indicator – SDG indicator 6.6.1 seeks to track changes in the extent of water-related ecosystems over time, contribute to halting the degradation and destruction of water-related ecosystems, and to assist the recovery of those already degraded.

Through the GSWE, the JRC is committed to providing continuous monitoring of global surface water dynamics for the next 10 years at least. The team is currently extending the work done with GSWE by combining post-2015 Landsat 7 and 8 data with imagery from Copernicus program’s Sentinel 1 and 2 satellites.

Using these data in combination improves the spatial and temporal resolution. The improved geographic and temporal completeness of the combined Landsat / Sentinel dataset also offers new opportunities for the identification and characterization of seasonally occurring waterbodies.

Using GSWE data as proxy, the team is working on reservoirs’ spatial and temporal characterization which will produce, as output, a dataset of reservoirs’ spatial extent and volumes called Global Reservoir Dynamics Explorer (GRDE).

Moving forward with Global Dam Watch

The core team will continue building the consensus database with the attributes that were considered of high demand by workshop participants. The core team will also lead the development of the comment paper.

Opportunities for collaboration

The team discussed applying for SESYNC Pursuits funding, deadline May 15 2019.

The team also discussed crowdfunding options, linking work to SDGs to attract World Bank and/or UN funding, and using events like World Water Week and EGU as opportunities to hold coinciding workshops.

Living European Rivers Project

Living European Rivers offers most pressing need for clean, up-to-date, and high resolution dam and reservoir data at the European scale. This WWF initiative emerges from the European Water Framework Directive and focuses on identifying areas of high conservation value. Identifying these areas will allow WWF to focus its efforts on protecting particular regions from impacts like hydropower developments.

The free-flowing rivers analysis used six pressure indicators to calculate a Connectivity Status Index (CSI) for each river reach. Connectivity here extends beyond longitudinal to include lateral, vertical, and temporal connectivity.

Downscaling from the global study creates its own set of challenges. For instance, incorporating associated data, i.e. roads and channelization infrastructure, can be incredibly difficult, but it remains important for a full understanding of river connectivity. Removing or preventing dams may improve longitudinal connectivity, but this does not bring the full suite of benefits if species are still unable to reach habitats that exist across the river basin or floodplain because of impediments to latitudinal connectivity.

The team settled on using HydroSHEDS as the river network on which to conduct analysis. However, the question of how to define Europe as a geographical region requires more thought. Some projects involve studies that fall outside of strictly EU boundaries, and some of the most important work to be done occurs in places like Turkey, but data is often most easily found in EU states.

An interesting question that arose from the discussion whether something like the free-flowing rivers analysis could be used to identify which dams could be removed to improve connectivity in the most impactful way. It was decided that this would not be possible at the global scale, but it could be used to identify areas where dams are the real pressure, and then a subsequent local analysis could be carried out to determine which dams could be removed to achieve the highest benefit.

The team considered how there are ways to improve flow through dams for water and organisms that don’t actually affect CSI because the dam remains, i.e. fish passes and sediment flushing techniques. Whether this can be included in the European FFR analysis is still under discussion, and JRC’s work on reservoir dynamics could be incorporated to inform recommendations on efforts like mimicking natural flow regimes to improve connectivity where dams can’t be removed.

The team also discussed whether the European scale version of the project could be linked to the Water Framework Directive. Tailoring messaging around the project to fit the Water Framework Directive could make it more attractive and improve access to funding and data. While monitoring data on fish, macroinvertebrates, etc exists for countries in Europe, the data is inconsistent and scattered across national databases.