Flood monitoring Drone Guide

# Flood Monitoring Drone Guide

Flood monitoring is one of the most valuable emergency and infrastructure applications for drones because flooding can change rapidly, affect very large areas and make normal ground access difficult or dangerous. Roads can become impassable, bridges may be damaged, rivers can change course and entire communities may become isolated within a short period of time.

Drones provide emergency services, water authorities, municipalities, infrastructure operators and environmental agencies with a flexible way to collect current aerial information. High-resolution cameras, thermal sensors, LiDAR, photogrammetry and mapping software can help teams understand where water has spread, which assets are affected and how conditions are changing.

The strongest use of drones is not simply to record dramatic flood imagery. Their real value is in producing repeatable, georeferenced information that can support decisions before, during and after a flood.

When combined with river gauges, weather forecasts, satellite imagery, hydrological models and GIS, drones can become an important part of a broader flood-management system.

Understanding Flood Monitoring

Flood monitoring covers several different stages.

Before a flood, drones can inspect rivers, drainage channels, embankments and flood-defence structures.

During an event, they can map water extent and identify threatened roads, bridges and communities.

After the water recedes, drones can assess erosion, structural damage, debris and changes to the landscape.

This makes them useful across prevention, response and recovery.

Different sensor combinations suit different stages.

RGB cameras are ideal for rapid visual assessment, while LiDAR and photogrammetry are valuable for terrain and structural mapping. Thermal imaging may support selected search-and-rescue or infrastructure applications.

Why Use Drones for Flood Monitoring?

Floods often restrict the very roads that ground teams would normally use to inspect an area.

A drone can reach locations without driving through floodwater.

This reduces unnecessary exposure of personnel and vehicles.

The aerial perspective also reveals how water connects across the landscape.

From the ground, separate flooded roads may appear to be individual problems. From above, the wider flow path may become obvious.

This broader view supports better emergency planning.

Pre-Flood Baseline Mapping

One of the most useful things an organisation can do is survey flood-prone areas before an event occurs.

Baseline imagery provides a record of normal conditions.

This may include riverbanks, roads, bridges, levees, culverts, drainage channels and nearby buildings.

When flooding occurs, new imagery can be compared directly with the baseline.

Change detection then becomes much easier.

Without a pre-event reference, emergency teams may struggle to determine whether visible damage is new.

River Monitoring

Rivers are natural candidates for drone monitoring.

Aircraft can inspect channel condition, bank erosion, debris accumulation and vegetation.

High-resolution imagery may reveal areas where flow is restricted.

Drones can also document changes in river width or alignment.

Repeat surveys can show how a river is evolving over time.

The data should complement hydrological monitoring rather than replace river gauges or professional flood modelling.

River Level Observation

Drones can visually document river level at bridges, banks and known reference points.

In some cases, image analysis may support approximate water-level estimation.

However, dedicated gauges remain more reliable for continuous quantitative measurement.

The drone is particularly useful when a gauge becomes inaccessible or when visual context is needed around the reading.

For example, a gauge may report a high level, while aerial imagery shows whether nearby roads or properties are already affected.

Flood Extent Mapping

Mapping the extent of floodwater is one of the strongest drone applications.

The aircraft can fly over affected areas and collect overlapping imagery.

This data can be converted into an orthomosaic showing the boundary of inundation.

The resulting map can be integrated into GIS.

Emergency teams can then compare flooded areas with roads, buildings, utilities and population data.

This creates a more operationally useful picture than isolated photographs.

Rapid Flood Mapping

During an active event, speed can be more important than survey-grade precision.

A drone can perform a rapid reconnaissance flight and send imagery back within minutes.

This may be enough to identify blocked roads, flooded houses or damaged bridges.

More detailed mapping can follow later.

A two-stage approach works well: rapid situational awareness first, detailed survey second.

Flood Depth Estimation

Estimating water depth from aerial imagery alone can be difficult.

If accurate terrain data is already available, the visible flood boundary can be combined with water-surface elevation to estimate depth.

LiDAR-derived terrain models may support this process.

Known reference points can also help.

Drone operators should be cautious about presenting depth estimates as exact unless the methodology supports it.

Emergency teams should not assume a road is safe simply because the water appears shallow from the air.

Urban Flood Monitoring

Urban flooding creates complex conditions.

Water may accumulate along roads, underpasses, car parks and low-lying neighbourhoods.

Drainage systems can overflow.

Drones provide a clear overview of which streets remain accessible.

They can also identify isolated buildings and blocked intersections.

Urban operations require careful management of people, buildings and aviation restrictions.

Residential Area Assessment

After severe rainfall or river flooding, drones can survey residential districts.

The imagery can show which properties have standing water around them.

This supports emergency prioritisation and later damage assessment.

Aerial data can also help identify safe access routes for responders.

Privacy should be considered carefully when surveying occupied residential areas.

Road Flooding

Flooded roads are a major operational concern.

Drones can identify sections covered by water and show where traffic is being diverted.

They can also reveal whether water is still rising or beginning to recede.

The aerial perspective helps authorities avoid sending vehicles into uncertain conditions.

Drone imagery should not be used alone to determine whether a flooded road is physically safe to cross.

Bridge Monitoring

Bridges can be vulnerable during floods.

Fast-moving water may carry debris into piers.

Scour can affect foundations.

Approach roads may also be washed away.

Drones can provide rapid visual inspection of the deck, piers, approaches and surrounding channel.

This allows engineers to prioritise which bridges need closer investigation.

Aerial imagery cannot confirm hidden foundation condition.

Culvert Monitoring

Blocked culverts can cause localised flooding.

Drones can inspect entrances and exits where they are visible.

Debris, vegetation and sediment accumulation may be identified.

This is useful both before and during flood events.

Culvert interiors usually require other inspection methods.

Drainage Channel Inspection

Open drainage channels can become obstructed by vegetation, debris or sediment.

Drones can inspect long sections efficiently.

This helps maintenance teams identify locations where flow may be restricted.

Repeat inspection before seasonal storms can reduce flood risk.

The same imagery can also support erosion monitoring.

Stormwater System Monitoring

Urban stormwater infrastructure includes drains, retention basins, channels and outfalls.

Drones can assess open sections of these systems.

They can document whether detention ponds are filling normally or overflowing.

During extreme rainfall, this helps authorities understand how the network is performing.

Underground drainage still requires dedicated inspection methods.

Levee and Embankment Monitoring

Flood-defence embankments need regular inspection.

Drones can identify visible erosion, cracking, vegetation problems or seepage indicators.

During a flood, they can monitor exposed sections without requiring personnel to walk the crest.

Thermal imagery may support selected seepage investigations.

Any suspected structural weakness should be assessed by qualified engineers.

Seepage Monitoring

Water may begin passing through or beneath an embankment before visible failure occurs.

Surface wetness or unusual vegetation may provide clues.

Thermal differences may also appear under suitable conditions.

A drone can help map these indicators.

It should be treated as a screening tool rather than a definitive diagnosis of embankment integrity.

Dam Flood Monitoring

Dams and reservoirs can experience extreme inflow during major storms.

Drones can provide an overview of spillways, downstream channels and surrounding terrain.

They may also document debris or erosion.

This can support dam operators during emergency response.

Flight activity should be coordinated carefully with dam safety procedures.

Reservoir Monitoring

Reservoir water levels can rise quickly.

A drone can document shoreline changes and flooding around adjacent infrastructure.

It can also inspect access roads and nearby slopes.

The imagery may help identify areas at risk of erosion or landslide.

Water-level gauges and hydrological models remain essential for quantitative reservoir management.

Coastal Flood Monitoring

Storm surge can affect large coastal areas.

Drones can map inundation along beaches, roads, ports and low-lying communities.

They can also document coastal erosion.

This is valuable after storms because access may be restricted.

Salt spray and strong wind create additional operational challenges.

Storm Surge Assessment

Storm surge can push seawater far inland.

Aerial imagery can show how far the water travelled.

This helps authorities understand which neighbourhoods and infrastructure were exposed.

Post-event mapping also supports future flood-risk models.

The data can be combined with tide and weather records.

Coastal Erosion

Flooding and storm waves can remove significant amounts of beach or cliff material.

Drones can create 3D models before and after an event.

The difference can be measured.

This supports coastal management.

Repeat photogrammetry is particularly valuable for long-term erosion monitoring.

Flash Flood Monitoring

Flash floods develop rapidly.

Small rivers and normally dry channels can become dangerous within minutes.

Drones can provide rapid observation where launching is safe.

The aircraft can help emergency teams understand which areas are being affected.

Operations should not delay evacuation or other urgent response activities.

Mountain and Valley Flooding

Narrow valleys can concentrate floodwater.

Roads and bridges may be damaged quickly.

Drones can inspect isolated stretches without requiring ground access.

This is especially useful after landslides or debris flows.

GNSS and communications may be more challenging in steep terrain.

Flooded Agricultural Land

Flooding can affect large areas of farmland.

Drones can map the extent of inundation.

This supports farmers, insurers and agricultural agencies.

Multispectral imagery collected after the water recedes may help assess crop stress.

Flood duration is often as important as the initial extent.

Repeat flights can therefore provide useful information.

Crop Damage Assessment

Flooded crops may suffer from waterlogging, sediment or physical damage.

RGB imagery provides visual evidence.

Multispectral data can show changes in plant condition.

The strongest assessment usually occurs after enough time has passed for stress to become visible.

Drone data should complement agricultural field inspection.

Livestock and Farm Access

Flooding can isolate farms or livestock.

Drones can help identify accessible routes and general conditions.

They may also support visual checks of isolated areas.

Any animal-rescue operation requires appropriate ground personnel.

The drone provides situational awareness rather than direct intervention.

Industrial Site Flooding

Factories, warehouses and industrial facilities can experience serious flood damage.

Drones can inspect access roads, storage areas and external infrastructure.

This helps managers understand whether personnel can safely enter.

Thermal imagery may also support selected electrical or equipment assessments.

Contaminated floodwater requires additional safety precautions.

Chemical and Hazardous Material Sites

Flooding at industrial facilities can create environmental concerns.

Drones can provide stand-off imagery without placing personnel directly into the affected area.

They can document damaged tanks, containers or flooded storage zones.

Specialist gas or environmental sensors may be carried where appropriate.

Any suspected hazardous-material incident should remain under specialist response procedures.

Power Infrastructure

Flooding can affect substations, power lines and generation facilities.

Drones can inspect external condition without requiring personnel to enter standing water.

Thermal cameras may support electrical inspection once conditions are suitable.

The drone can also identify blocked access routes.

Electrical safety decisions should remain with qualified utility personnel.

Water and Wastewater Infrastructure

Floods can damage pumping stations, treatment plants and pipelines.

Drones can rapidly assess the exterior of these facilities.

This helps utilities understand which assets may require immediate attention.

They can also map access conditions.

This application links flood monitoring directly with emergency water-infrastructure assessment.

Telecom Infrastructure

Flooding can isolate towers, cabinets and communication sites.

Drones can inspect external condition.

This may support restoration planning.

Communications infrastructure is particularly important during emergencies.

Aerial assessment can help operators prioritise repair teams.

Railway Flooding

Rail lines can be affected by standing water, washouts and embankment damage.

Drones can inspect long sections quickly.

They can identify submerged track and damaged drainage.

LiDAR or photogrammetry may help document embankment deformation.

Rail operators must determine whether the line is safe for traffic.

Highway Flood Monitoring

Major roads and motorways are vulnerable to flooding at low points and river crossings.

Drones can provide traffic-management teams with a wide-area view.

They can identify stranded vehicles, blocked junctions and alternative routes.

Repeat flights can show how quickly water is receding.

This supports decisions about closures and reopening.

Airport Flood Monitoring

Airports contain large paved areas and complex drainage systems.

Heavy rainfall can affect runways, taxiways and service roads.

Drones can support assessment where aviation operations allow.

The aircraft can also inspect perimeter drainage and detention areas.

Coordination with airport authorities is essential.

Port Flood Monitoring

Ports may experience storm surge, river flooding or heavy rainfall.

Drones can survey quays, storage yards and access roads.

This helps operators understand which areas remain usable.

The same mission may document vessel berths and drainage conditions.

Strong wind and active marine operations require careful planning.

Search and Rescue

Drones can support search and rescue during floods by scanning areas that are difficult to reach.

High-resolution cameras can identify people on roofs, vehicles or isolated ground.

Thermal cameras may assist under suitable conditions.

Water, wet surfaces and warm weather can reduce thermal contrast.

AI may help highlight possible detections, but human verification is essential.

People on Rooftops

During severe flooding, people may move to upper floors or roofs.

Drones can provide rapid visual confirmation.

This helps responders prioritise rescue resources.

The aircraft can also observe whether the surrounding water is rising.

Any communication with affected people should follow emergency-service procedures.

Vehicle Detection

Floodwater can trap vehicles.

Drones can identify visible vehicles and help determine their location.

The aerial view can also show whether an access route remains available.

The presence of a vehicle does not necessarily mean someone is inside.

Human responders must verify the situation.

Thermal Search

Thermal cameras may help locate people during low-light conditions.

Their effectiveness depends on temperature contrast.

Rain and wet surfaces can influence the image.

Thermal imagery should therefore be used alongside visible cameras.

False positives can occur.

Emergency Operations Centre Support

Drone video can be streamed to an emergency operations centre.

This provides decision-makers with current visual information.

GIS layers can display the drone location and mapped flood extent.

The resulting common operating picture can support coordination between police, fire, utilities and local government.

Good information management is critical because excessive video can overwhelm operators.

Live Mapping

Modern drone platforms can transmit imagery while the aircraft is still flying.

This makes it possible to build preliminary maps during the mission.

Emergency teams do not always need to wait for full photogrammetric processing.

Live mapping is especially useful when conditions are changing rapidly.

A detailed final dataset can be produced later.

GIS Integration

Flood maps become more useful when integrated with GIS.

Water extent can be compared against roads, properties, hospitals, substations and other assets.

Emergency planners can identify which infrastructure lies inside the affected area.

Historical flood maps can also be compared.

This supports both immediate response and long-term planning.

AI Flood Boundary Detection

AI can help identify the boundary between floodwater and dry land.

Computer vision can process large aerial datasets quickly.

This may accelerate mapping across large regions.

Water reflections, shadows and dark surfaces can cause errors.

Human review remains important.

AI Change Detection

Change detection can compare pre-flood and current imagery.

The software highlights areas where roads, buildings or terrain have changed.

This reduces manual review time.

It can be especially useful during recovery.

Important findings should be verified before repair decisions are made.

Building Damage Assessment

Floodwater can damage building exteriors, foundations and surrounding ground.

Drones can document visible condition.

They can also show debris accumulation and access restrictions.

This helps authorities and insurers prioritise physical inspection.

Aerial imagery cannot determine internal structural integrity.

Insurance Assessment

Drone imagery can provide a dated record of flood impact.

Orthomosaics show the extent of water.

3D models can document erosion or structural damage.

This can support insurance claims.

The survey methodology should be documented if the data may be used formally.

Debris Mapping

Floods can move large quantities of debris.

Trees, vehicles and construction material may block roads or waterways.

Drones can map these locations.

This supports cleanup planning.

Debris near bridges and culverts may require particular attention.

Landslide Monitoring

Heavy rainfall can trigger landslides.

Drones can identify new slope failures and blocked roads.

Photogrammetry and LiDAR can measure the affected terrain.

Repeat surveys can monitor whether the slope continues to move.

Geotechnical specialists should interpret instability.

Erosion Monitoring

Floodwater can remove soil from riverbanks, roads and embankments.

A drone can document the extent of erosion.

3D models allow volume loss to be estimated.

This is useful for repair planning.

Repeated surveys help determine whether erosion is continuing.

Scour Assessment Support

Fast-moving water can erode material around bridge piers and other structures.

Much of this scour may occur underwater and cannot be seen directly from a drone.

However, aerial imagery can document surrounding erosion and exposed features.

Underwater sonar or specialist inspection may still be required.

Drone data should therefore support rather than replace scour assessment.

LiDAR for Flood Monitoring

LiDAR provides accurate terrain information.

This is particularly useful before a flood.

A high-quality digital terrain model can later be combined with water-level data to estimate inundation.

LiDAR can also map terrain through gaps in vegetation.

This supports flood modelling and drainage analysis.

Photogrammetry

Photogrammetry creates orthomosaics, point clouds and 3D models.

It is particularly useful for documenting flood damage after water levels fall.

Road washouts, erosion and landslides can be measured.

Repeat surveys enable quantitative change detection.

RTK, PPK or ground control may be used where higher accuracy is required.

Multispectral Imaging

Multispectral cameras can support post-flood environmental assessment.

Vegetation stress may become visible before it is obvious to the eye.

This is valuable for agriculture and wetlands.

Multispectral data may also help distinguish water and vegetation in certain mapping applications.

It is usually a secondary sensor rather than the primary emergency-response tool.

Thermal Imaging

Thermal cameras can support search and rescue and selected infrastructure inspections.

They may also identify differences between moving water and surrounding surfaces.

Environmental conditions strongly affect performance.

Thermal imagery should therefore be interpreted carefully.

It is most useful when combined with RGB data.

RTK and PPK

Accurate geolocation is important for repeat mapping.

RTK and PPK can improve alignment between surveys.

This helps when measuring erosion or comparing flood boundaries.

Precise location also allows emergency teams to transfer drone findings directly into GIS.

Accuracy should still be validated where survey-grade measurements are required.

Satellite Integration

Satellites are excellent for regional flood mapping.

Their weakness can be cloud cover, revisit time or resolution.

Drones provide much more detailed local information.

The two technologies complement each other.

Satellite imagery identifies the wider affected region.

Drones provide high-resolution assessment of priority locations.

River Gauge Integration

River gauges provide continuous numerical measurements.

Drones provide visual context.

Combining the two is powerful.

If a river reaches a critical level, a drone mission can inspect vulnerable areas.

The imagery can show whether flood defences are functioning as expected.

This creates a more responsive monitoring system.

Weather Data Integration

Weather forecasts help determine whether conditions are likely to worsen.

Drone findings can then be interpreted alongside predicted rainfall.

This is useful during prolonged flood events.

The flight itself must also account for wind, rain and visibility.

Many drones cannot operate safely in severe weather.

Hydrological Model Integration

Hydrological models predict where water may flow.

Drone data can validate these predictions.

Actual flood boundaries can be compared with the model.

This helps improve future forecasting.

Accurate post-event data is therefore valuable long after the emergency has ended.

Digital Flood Twin

A digital twin can combine terrain, infrastructure, river data and flood models.

Drone imagery updates the real-world condition.

Emergency teams can compare predicted and observed flooding.

Over time, this creates a detailed flood history.

Such systems may become increasingly important for climate adaptation.

Drone-in-a-Box

Drone-in-a-Box systems could support automated flood monitoring around rivers, dams and critical infrastructure.

A docking station can house the aircraft permanently.

If a river gauge reaches a threshold, the system may launch a predefined mission.

The drone collects imagery and returns automatically.

This can reduce response time.

Regulatory and weather limitations still apply.

Automated River Patrol

Routine drone flights can inspect the same river sections repeatedly.

AI can compare each survey.

Debris, bank erosion or vegetation growth may be flagged.

This helps authorities identify problems before major flooding occurs.

Automation is most useful when combined with established maintenance programmes.

Long-Endurance VTOL Drones

Large flood events may cover extensive areas.

Fixed-wing or VTOL drones can cover greater distances than small multirotors.

They may be useful for rivers, coastlines or large rural regions.

Multirotors remain better for detailed local inspection.

A mixed fleet can provide both broad coverage and close observation.

BVLOS Operations

Flood monitoring over long river corridors may benefit from beyond-visual-line-of-sight operations.

BVLOS can make large-area monitoring much more practical.

It usually requires additional regulatory approval and technical safeguards.

Communications and emergency procedures become especially important.

The requirements depend on jurisdiction.

Communications Relay

Flooding can damage terrestrial communications.

Drones may also be used as temporary communications relays.

An elevated aircraft can improve coverage between emergency teams.

Tethered drones may provide persistent support.

This is separate from flood mapping but can be highly valuable during disaster response.

Tethered Drones

Tethered drones can remain airborne for long periods.

They may provide continuous observation over a flooded town, bridge or emergency command post.

Because power is supplied through a cable, endurance is much greater than with normal battery flight.

Their limited mobility makes them best suited to fixed-area monitoring.

Data Processing

Flood operations generate large amounts of imagery.

Rapid processing is essential.

Emergency teams may need a simple map more than a perfect 3D model.

Processing workflows should therefore match the operational need.

Automated cloud platforms can accelerate mapping.

Local processing may be preferable where connectivity is limited.

Edge Computing

Edge computing allows some analysis to happen on the aircraft or field computer.

Flood boundaries or possible people can be detected locally.

Only important information may need to be transmitted.

This reduces bandwidth.

It can also speed up emergency response.

Data Sharing

Flood response often involves multiple organisations.

Maps and imagery should be shared in formats that emergency services, utilities and municipalities can use.

GIS-compatible outputs are particularly useful.

Clear naming and timestamps reduce confusion.

Sensitive imagery should still be protected appropriately.

Public Information

Aerial flood maps may support public communication.

Authorities can show which roads are closed or which districts are affected.

Information should be verified before release.

Poorly interpreted imagery can create unnecessary concern.

Public communication should remain under the responsible authority.

Privacy

Flood monitoring can capture private homes, gardens and people.

Emergency need does not remove the importance of responsible data handling.

Flights should collect the information required for the response.

Unnecessary close imagery should be avoided.

Data retention and access should follow applicable requirements.

Weather Limitations

Floods often occur during exactly the weather conditions that make drone flying difficult.

Heavy rain, strong wind and low cloud can prevent safe operations.

Operators should not take unnecessary risks simply because the information is urgent.

Drones should be one of several observation tools.

Satellites, helicopters, ground teams and fixed sensors may continue providing information when drones cannot fly.

Water and Battery Safety

Emergency operations may require take-off from wet environments.

Operators should keep batteries and charging equipment protected from water.

Landing areas should be selected carefully.

Cold and wet conditions can reduce battery performance.

Equipment should be inspected after exposure.

GNSS and Communications

Urban buildings, valleys and remote locations can affect GNSS and radio links.

Flood response should not assume perfect connectivity.

Operators need procedures for communication loss.

Aircraft behaviour during GNSS degradation should be understood before deployment.

Airspace Coordination

Major floods may involve helicopters and other emergency aircraft.

Drone operations must be coordinated carefully.

An uncoordinated drone can create a serious aviation hazard.

Emergency-service airspace procedures should always take priority.

Operators should be prepared to land immediately if required.

Benefits of Drone-Based Flood Monitoring

The main advantage is rapid access to current information.

Drones can map areas that ground teams cannot reach safely.

They provide wide-area context and detailed local imagery.

They can support flood mapping, search and rescue, infrastructure inspection and damage assessment.

Repeat flights show how conditions change.

GIS integration makes the information useful for multiple agencies.

The same dataset can support both emergency response and long-term flood-risk planning.

Challenges and Limitations

Drones cannot operate in every weather condition.

They may have limited endurance.

Water depth is difficult to determine accurately from imagery alone.

Vegetation and buildings can hide important features.

Thermal imaging has environmental limitations.

Large disasters may generate more data than teams can process immediately.

Airspace can become complex when crewed emergency aircraft are operating.

For these reasons, drones should complement other flood-monitoring systems rather than replace them.

The Future of Flood Monitoring

Flood monitoring is likely to become increasingly automated.

River gauges, rainfall sensors and weather models will identify developing risk.

Drone-in-a-Box systems may launch automatically when thresholds are reached.

AI will map flood boundaries and compare them with infrastructure databases.

Long-endurance aircraft will monitor river corridors.

Satellite and drone imagery will be combined automatically.

Digital twins will model predicted flooding and compare it with real observations.

Emergency operations centres will receive updated maps rather than individual video feeds.

The role of the drone operator will increasingly shift from manual flying toward supervising automated data-collection systems.

The long-term direction is toward continuous flood intelligence, where fixed sensors, models, satellites and drones work together to provide a constantly updated picture of flood risk and impact.

Conclusion

Flood monitoring is one of the strongest emergency-management applications for drones because floods are dynamic, widespread and often difficult to assess safely from the ground.

Drones can map flood extent, inspect roads and bridges, monitor rivers and embankments, support search and rescue and document infrastructure damage.

Photogrammetry and LiDAR can create accurate terrain and damage models, while thermal cameras can support selected night-time and emergency applications.

AI can help identify flood boundaries and changes, while GIS integration allows drone information to be combined with roads, buildings, utilities and population data.

The greatest value comes when drone monitoring is integrated with river gauges, weather forecasts, hydrological models, satellites and established emergency-response systems.

Drones do not replace emergency services, engineers, hydrologists or fixed monitoring infrastructure. Their role is to provide fast, detailed and flexible aerial information that helps those professionals understand where flooding is occurring, how conditions are changing and which people or infrastructure may require attention first.