astrid project results webinar
TRANSCRIPT
![Page 1: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/1.jpg)
Assessing Statistical models of Temporary River Intermittence for Decision makers
ASTRID project results webinar
Michael Eastman, statistical modelling Cecilia Svensson, statistical hydrologySimon Parry, hydrological extremesCatherine Sefton, hydrological intermittence
with thanks to Eric Sauquet, INRAE, Franceand the NERC Landscape Decisions Programme
NERC Grant nos. NE/T004215/1 & NE/T004215/2
![Page 2: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/2.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 3: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/3.jpg)
Why the focus on intermittent rivers?
3
Paraleptophlebia werneri
Hydrologically dynamic
Widespread?
Ecologically diverse
Misrepresented
Under pressure
a.k.a temporary or non-perennial
![Page 4: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/4.jpg)
Why Assess Statistical models of Temporary River Intermittence for Decision makers?
1Why temporary river
hydrology?
under-represented in hydrological monitoring and regulation, despite being under pressure and biodiverse
Why modelling?provides scope for simulation, investigation of drivers and infilling
Why statistical modelling?
processes are complex, varied and not well understood
2
3
What decisions?future-fit monitoring, water resources, ecosystem services, climate change4
![Page 5: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/5.jpg)
5
What are the needs for enhanced decision making on intermittent rivers in Great Britain?
Based on perennial
rivers
Metrics of intermittence
Equip for adaptation to
temporary rivers
Water resource assessment methodologies
Current ASTRID EnhancedElement of decision making for temporary rivers
Under-protection in
law
Engagement with senior regulators
Influence future
legislation and policy
Protective legislation
Propagation of opinion on social media
Robust science
Better informed
debate Public debate
Spatial extent unknown
Intermittence models and
mapping
Need for bespoke methods
established
Awareness of temporary rivers amongst decision makers
Lack of temporary river data
Promotion of citizen science
An improving evidence-base for decisions
Evidence-base for decisions
![Page 6: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/6.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 7: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/7.jpg)
The wider view…
Grouping of hydrologically similar intermittent rivers Sauquet et al., 2021
…increasing recognition of the prevalence and importance of intermittent rivers.
![Page 8: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/8.jpg)
The catchment-scale view…
0 5 10 km
Trained hydrologistsBegan in 1997Approximately monthly
River Thames
Superficial hydrogeology
Generally high permeability
Generally low permeability
Mixed permeability
Unmapped / Ignored
Bedrock hydrogeology
High Permeability (Fissured and Intergranular)
Mixed Permeability
River Thames
Superficial hydrogeology
Generally high permeability
Generally low permeability
Mixed permeability
Unmapped / Ignored
Bedrock hydrogeology
High Permeability (Fissured and Intergranular)
Mixed Permeability
Rivers
Observation sites
Regime influenced by superficial deposits
Regime dominated by groundwater
DRY
PONDED
FLOWING
…what is possible given observations of intermittent behaviour at sufficient spatiotemporal resolution.
![Page 9: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/9.jpg)
Eastman et al. 2021. Reconstructing Spatiotemporal Dynamics in Hydrological State Along Intermittent Rivers. Water 13 (4), 493.
Simulation of hydrological state - at a monthly time step and along the river
Misbourne Ash
![Page 10: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/10.jpg)
The story so far…
1. Modelled intermittence at catchment-scale with good spatiotemporal resolution data;
2. Needing to scale up to national, with limited data availability in Great Britain;
3. Spotting a decent resolution dataset with national coverage in France.
?
Images: www.pixabay.com
![Page 11: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/11.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 12: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/12.jpg)
ASTRID project overview
1 Oct 2019 – 30 Sept 2021 (extended by 12 months)
1. Engaging stakeholders in knowledge exchange on hydrological intermittence in Great Britain;
2. Statistical modelling of intermittence in Great Britain’s temporary rivers through training and validating models;
3. Mapping hydrological intermittence in temporary rivers across Great Britain.
When:
What:
Assessing Statistical Models of Temporary River Intermittence for Decision Makers
![Page 13: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/13.jpg)
Launch workshop October 2019
EA Lincs and Northants
EA Thames
• Knowledge exchange on previous work;
• Discussion on metrics that are useful for decision makers;
• Discussion on useful spatial units, resolution and extent of simulations.
![Page 14: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/14.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 15: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/15.jpg)
Modelling approach
France intermittence
data
France catchment
data
Statistical model of incidence of
intermittence
GB catchment and river
network data
3222 sites across France1km resolution points along
the network
GB intermittence
data
![Page 16: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/16.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 17: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/17.jpg)
Observatoire National Des Étiages (ONDE)
https://onde.eaufrance.fr/
Observations of hydrological state at >3222 sites in France• Flow visible• Flow not visible• Dry
Temporal extent:• 2012-2019• May - October
![Page 18: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/18.jpg)
Match point data with ONDE sites:• Gauging
stations• Boreholes
The Data
https://onde.eaufrance.fr/
![Page 19: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/19.jpg)
Match point data with ONDE sites:• Gauging
stations• Boreholes
Derive catchment boundaries at
ONDE sites
Summarise variables across catchments:• Aquifer type• Land cover• Precipitation• Actual Evapotranspiration• Soil moisture
The Data
https://onde.eaufrance.fr/
![Page 20: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/20.jpg)
Match point data with ONDE sites:• Gauging
stations• Boreholes
Derive metrics that summarise magnitude
and variation of variables
Derive catchment boundaries at ONDE
sites
Summarise variables across catchments:• Aquifer type• Land cover• Precipitation• Actual
Evapotranspiration• Soil moisture
The Data
https://onde.eaufrance.fr/
![Page 21: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/21.jpg)
The Data• Flow responsiveness
• Magnitude• Variation
• Groundwater responsiveness• Magnitude• Variation
• Actual evapotranspiration• Magnitude• Variation
• Precipitation• Magnitude• Variation
• Soil moisture• Magnitude• Variation
• Aquifer type• Land cover• Catchment area
Number of variables
Ev
alu
atio
n m
etri
c
However, not all variables are created equal…
![Page 22: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/22.jpg)
The Data
![Page 23: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/23.jpg)
The DataRainfallSoil moistureCatchment areaLand coverFlowGroundwater
![Page 24: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/24.jpg)
Random Forests
Data is split into partitions that maximise their homogeneity
This continues until stopping criteria are fulfilled, such as the maximum
number of splits is reached, or homogeneity gains are below
the threshold
Random forests consist of many
trees…
Flashiness
> 0.5
Woodland
> 0.2
PerennialNon-
perennial
GW Q5/Mean
> 0.8
Catchment area
> 180 km2
PerennialNon-
perennial
Catchment area
> 30 km2
PerennialNon-
perennial
Y N
NY
Y N Y N
Y N
Flashiness
> 0.5
Woodland
> 0.2
PerennialNon-
perennial
GW Q5/Mean
> 0.8
Catchment area
> 180 km2
PerennialNon-
perennial
Catchment area
> 30 km2
PerennialNon-
perennial
Y N
NY
Y N Y N
Y N
Flashiness
> 0.5
Woodland
> 0.2
PerennialNon-
perennial
GW Q5/Mean
> 0.8
Catchment area
> 180 km2
PerennialNon-
perennial
Catchment area
> 30 km2
PerennialNon-
perennial
Y N
NY
Y N Y N
Y N
Flashiness
> 0.5
Woodland
> 0.2
PerennialNon-
perennial
GW Q5/Mean
> 0.8
Catchment area
> 180 km2
PerennialNon-
perennial
Catchment area
> 30 km2
PerennialNon-
perennial
Y N
NY
Y N Y N
Y N
Flashiness
> 0.5
Woodland
> 0.2
PerennialNon-
perennial
GW Q5/Mean
> 0.8
Catchment area
> 180 km2
PerennialNon-
perennial
Catchment area
> 30 km2
PerennialNon-
perennial
Y N
NY
Y N Y N
Y N
Flashiness
> 0.5
Woodland
> 0.2
PerennialNon-
perennial
GW Q5/Mean
> 0.8
Catchment area
> 180 km2
PerennialNon-
perennial
Catchment area
> 30 km2
PerennialNon-
perennial
Y N
NY
Y N Y N
Y N
Flashiness
> 0.5
Woodland
> 0.2
PerennialNon-
perennial
GW Q5/Mean
> 0.8
Catchment area
> 180 km2
PerennialNon-
perennial
Catchment area
> 30 km2
PerennialNon-
perennial
Y N
NY
Y N Y N
Y N
Y N
![Page 25: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/25.jpg)
Model tuning
Hypothesis Baseline error Modelled error
1. Perennial or non-perennial 0.495 0.352
2. Permanence category 0.238 0.256
3. Permanence 0.142 0.139
1. What can we model?
Simulating whether a reach is intermittent exhibits the largest skill improvement
Error using an automatically tuned random
forest
Error using simplest model
e.g. all reaches = perennial
![Page 26: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/26.jpg)
Model tuning
2. What model
should we use?
3. How can we improve the model?
Number of trees
200, 400, …, 2000, 2500, 5000, 10000, 25000, 100000
Maximum splits
100, 140, …, 500
10 different models were trained, but the three standouts all belong to the decision tree family of models
After evaluating a number of options it was concluded that 400 trees, with a maximum split of 140 was the most appropriate fit in France, accounting for efficiency and Skill.
![Page 27: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/27.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 28: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/28.jpg)
Model performance
• Model accurately simulates whether ONDE sites are intermittent across France
• No immediately obvious spatial pattern in performance
• However, accuracy is higher on catchments dominated by:• Arable & horticulture• Grassland• Aquifer classes 4 & 6
(relatively unproductive)
![Page 29: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/29.jpg)
Tree Agreement
2/3 tree agreement
Welch’s independent t-test:t statistic = -96.68 p-value <0.01
More confident in estimates that have higher tree agreement
![Page 30: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/30.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 31: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/31.jpg)
Modelling approach
France intermittence
data
France catchment
data
Statistical model of incidence of
intermittence
GB catchment and river
network data
3222 sites across France1km resolution points along
the network
GB intermittence
data
![Page 32: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/32.jpg)
Match point data with ONDE sites:• Gauging
stations• Boreholes
Derive metrics that summarise magnitude and
variation of variables
Derive catchment boundaries at ONDE
sites
Summarise variables across catchments:• Aquifer type• Land cover• Precipitation• Actual
Evapotranspiration• Soil moisture
GB Data
![Page 33: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/33.jpg)
GB Data
But first we split the river network into 1 km reaches
Then delineate the catchments and estimate metrics!
![Page 34: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/34.jpg)
GB results
Model simulated 17% of points as non-perennial
Model used 60:40 tree agreement threshold
for perennial:non-perennial estimates
![Page 35: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/35.jpg)
GB results
![Page 36: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/36.jpg)
GB results
![Page 37: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/37.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 38: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/38.jpg)
Performance variationHigher confidence
Lower confidence
![Page 39: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/39.jpg)
Performance variationHigher confidence
Lower confidence
![Page 40: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/40.jpg)
Summary of ASTRID results
17% of points were simulated as non-perennial
Intermittence is distributed throughout Britain, with many simulations overlapping current understanding of occurrence of intermittence
Model performance varies with environmental variables
![Page 41: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/41.jpg)
• The motivation for ASTRID• The global picture on intermittent river research• Hydrological state simulation at catchment scale
Motivation and context
• Project overview• Launch workshop• Modelling approach
ASTRID
introduction
• Data• Model• French results
Model training
• Data• GB results• Performance variation
Model application
• Stakeholder consultation• ASTRID conclusions• Data collection
Conclusions and next steps
Webinar overview
![Page 42: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/42.jpg)
Consultation on ASTRID results• Two weeks, from Monday 6th to Friday 17th September 2021;
• Maps and this webinar downloadable from a dedicated webpage https://www.ceh.ac.uk/astrid-project-stakeholder-consultation
• Plus a short questionnaire to collate views on:
• Your interest in intermittent and ephemeral river hydrology;
• Your knowledge on where and how the model is right/wrong;
• Questions about intermittent and ephemeral rivers you would like to see addressed;
• Interest in a Hydrological Intermittence Forum;
• The responses will be anonymously summarised, circulated to stakeholders and made available via the ASTRID and Landscape Decisions Programme websites.
![Page 43: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/43.jpg)
Conclusions
• Statistical modelling of national hydrological intermittence for Great Britain using available data has proven possible but challenging;
• Maps of hydrological intermittence for Great Britain are useful for raising awareness at national level, but highlight the need for data at regional scale;
• Year-round monitoring of intermittent rivers and streams is needed to support their effective management and prevent deterioration;
• Local knowledge will improve understanding of how model performance varies across Great Britain;
• Progress towards consistent and robust decision making in the management of intermittent rivers is enhanced by collaboration.
![Page 44: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/44.jpg)
Citizen science - CrowdWater appCreate a new spot or… …update an existing spot,
then…
DRY STREAMBED: no visible water and the streambed is dry WET STREAMBED: no visible water, but the streambed is wet (for at least 2 cm depth below the surface) ISOLATED POOLS: separated pools of water that are not visibly flowing are present on the streambed STANDING WATER: water but no visible flow TRICKLING WATER: very small flow, but clearly visible flowing water FLOWING WATER: a continuous pathway of water that is flowing
…upload a photo and select a flow state:
“We call on hydrologists and citizens to observe, sense and report the hydrological state of the aqua temporaria incognita... without these data, it is as if we are trying to complete a puzzle on how headwater catchments function and how water affects ecological processes, while the majority of the puzzle pieces are hidden under the carpet.”
van Meerveld et al., Aqua Temporaria Incognita, 2021
![Page 45: ASTRID project results webinar](https://reader033.vdocuments.mx/reader033/viewer/2022051319/627b16f0511d1050dc6f07a7/html5/thumbnails/45.jpg)
Cath Sefton [email protected]
Michael Eastman [email protected]
Simon Parry [email protected]
Cecilia Svensson [email protected]
Thank you
Any questions?
NERC Grant nos. NE/T004215/1 & NE/T004215/2Photo: River Og at Ogbourne St Andrew Cath Sefton, June 2019