How your observations have contributed to better forecasts

Every observation has helped meteorologists build a clearer picture of what’s happening across the country. Your contributions have supported forecasting, scientific research and operational decision-making, helping improve the understanding of local weather conditions and identify developing weather features that might otherwise have gone unnoticed.

As we approach the retirement of WOW later this year, we have put together a few examples of how your measurements have, and will continue to contribute to better forecasts, more accurate warnings, and a deeper understanding of our changing climate.

• Home weather stations improve severe weather monitoring and forecasting
• Citizen science observations help transform UK Climate Services
• Improving urban heat predictions with machine learning and citizen science

---

Home weather stations improve severe weather monitoring and forecasting.

Observations from the Weather Observations Website (WOW) provided a step forward for observing the weather over the UK, providing us with unprecedented levels of localised detail. As we retire the WOW platform and move to a new way of meeting our evolving observational needs through third-party suppliers such as Synoptic Data and Netatmo, we take a look at how the WOW community have contributed to improved operational capabilities.

Challenge

Many weather hazards, including thunderstorms, hailstorms and gust fronts, occur on scales that are often too small to be picked up by our traditional surface observation network. A storm can pass between official weather stations, leaving important features undetected.

The WOW network significantly increased observational coverage across the UK by incorporating data from thousands of home weather stations, providing a much clearer picture of the weather conditions at local scales.

Approach

Following the severe hailstorms that affected northern England on 1 July 2015, researchers combined observations from hundreds of home automatic weather stations with conventional meteorological observations and radar data. After applying rigorous quality control and bias correction to account for the varying instrument quality and siting, they were able to produce high-resolution analyses of pressure, wind, temperature and storm structure.

Results

The study demonstrated that quality-controlled home weather station data could successfully resolve storm-scale features that were largely invisible in the standard observing network, including:

• Gust fronts
• Areas of inflow and outflow
• Mesoscale pressure anomalies
• Fine-scale thunderstorm structure

The analyses closely matched radar observations and established conceptual models of severe thunderstorms, showing that dense citizen-observation networks can provide scientifically robust information despite the variability inherent in-home weather stations.

Operational impact

The success of the 2015 hailstorm analysis has directly influenced the development of operational meteorological systems.

• Quality control and bias-correction methods developed from this work have subsequently been implemented operationally.
• WOW observations of temperature, dew point, wind and pressure became the foundation of the PLUVIA Mesoanalysis which, since 2021, has provided hourly near-surface analyses to operational meteorologists.
• Ensemble Optimal Interpolation was later introduced, combining WOW observations with numerical weather prediction (NWP) model guidance to improve spatial consistency.

This capability is now being integrated into operational workflows through the Enhancing Nowcasting Project, with processing incorporated into the next-generation observations framework, and deployment progressing within the Met Office operational suite.

Why it matters

The 2015 northern England hailstorm study demonstrates how citizen science can move beyond supplementary observations to become part of our operational forecasting infrastructure. Home weather stations can provide the observation density needed to identify localised hazardous weather that conventional observing networks miss, while quality control, bias correction and data assimilation techniques ensure the data can be used with confidence.

With this work subsequently evolving from research into operational capability, underpinning the Met Office's PLUVIA Mesoanalysis it has helped to deliver more detailed, real-time situational awareness for forecasters across the UK.

---

Citizen science observations help transform UK Climate Services

The Met Office's Weather Observations Website (WOW) played a pivotal role in demonstrating how citizen science observations can enhance climate science and climate services. A pilot project focused on Greater Manchester showed that crowdsourced weather observations could provide valuable insights into local climate conditions that are often missed by conventional weather station networks. The success of this work established the scientific credibility of citizen-generated weather data for climate applications and laid the foundations for the development of the national Crowd-Grid dataset.

Challenge

Traditional climate datasets are typically built from weather stations operated to World Meteorological Organization standards. While these stations provide highly accurate measurements, they are relatively sparse and are often located in open, grassy environments that may not fully represent the temperatures people experience in densely populated urban areas. As a result, important local climate variations, particularly urban heat hotspots, can be underrepresented in climate services and risk assessments.

To address this challenge, Met Office scientists explored whether citizen science observations from personal weather stations could be used to create a more detailed picture of urban climate conditions. Greater Manchester was selected as a pilot study area because it is a large metropolitan region with a mix of urban, suburban and upland environments.

The role of WOW

WOW was a core data source for the Manchester pilot. Researchers combined observations from the WOW network with data from other personal weather station networks and official Met Office observing sites to create high-resolution, 1 km gridded maps of daily maximum temperatures across the region. The WOW network provided 62 observation sites across the study area, substantially increasing the density of available weather observations compared with the eight Met Office reference stations available in the wider region. After rigorous quality control, 45 WOW sites were retained for analysis. This dense network of citizen observations enabled scientists to capture local temperature variations at a much finer scale than was possible using traditional observing networks alone.

Key findings

The pilot demonstrated that citizen science observations provide important additional information about urban climate conditions.

Using WOW and other crowdsourced observations, researchers found:

• Greater spatial detail in temperature patterns across Manchester.
• Urban areas that were typically 1-2°C warmer than indicated by the national HadUK-Grid climate dataset.
• Temperature differences of up to 4-5°C on particularly hot days in some parts of Greater Manchester.
• Previously unresolved hotspots, including industrial and highly urbanised areas.
• A significant increase in estimated heat exposure compared with existing climate service baselines.

One striking result was the assessment of "summer days" (days with maximum temperatures above 25°C), a metric used in heat-health and climate adaptation planning. The crowdsourced dataset identified substantially more summer days across many parts of Manchester than the standard national dataset, with some locations showing increases equivalent to as much as three additional weeks of hot days during 2020.

Building confidence in crowdsourced climate data

A major outcome of the Manchester project was demonstrating that citizen science observations can be used robustly in climate research when supported by appropriate quality assurance procedures. Researchers developed methods to identify and remove faulty observations, assess station quality, and account for differences in instrument exposure and siting conditions. The resulting datasets showed strong spatial coherence and produced realistic representations of both urban and rural temperature patterns.

This work provided critical evidence that crowdsourced weather observations could form part of operational climate datasets and services. As the study concluded, citizen science observations have the potential to create climatology’s that are more representative of people's lived experiences, particularly in cities where conventional monitoring networks are sparse.

From Manchester to national impact: Crowd-Grid

The success of the Manchester pilot directly informed the subsequent development of the national Crowd-Grid dataset. WOW became an essential component of this project, providing the observational foundation for one of the first national-scale, multi-year climate datasets to incorporate crowdsourced weather observations.

By combining the extensive archive of observations contributed by WOW users with official Met Office observations and partner datasets from UK highways agencies, Met Office scientists were able to create high-resolution daily temperature maps covering the UK. These maps reveal local variations in heat that would otherwise be difficult to detect, particularly in urban environments where traditional observing networks are relatively sparse.

The Crowd-Grid dataset represents a significant step forward in integrating citizen science into operational climate science, turning thousands of public weather observations into a nationally consistent climate resource.

Lasting benefits

The value of Crowd-Grid is already being realised through applications that support climate resilience and adaptation planning. The dataset contributes to products such as the Met Office Recent Heat Packs, which provide locally relevant climate information for local authorities and other decision-makers. By offering a more detailed understanding of heat exposure and local climate variability, these products help organisations identify vulnerabilities and plan for future climate impacts.

Together, the Manchester pilot and the subsequent Crowd-Grid dataset demonstrate the long-term scientific value of citizen weather observations. They show how contributions from WOW participants have helped to advance climate science, improve understanding of urban heat, and support the next generation of climate services across the UK.

---

Improving urban heat predictions with machine learning and citizen science 

The Met Office successfully demonstrated how machine learning, combined with weather forecasts and citizen-generated weather observations, can significantly improve the accuracy and detail of urban heatwave predictions. The research focused on London and provided a blueprint for delivering more localised temperature forecasts that help communities, businesses and public services prepare for extreme heat.

Challenge

Accurately forecasting temperatures in cities has become increasingly important as urban populations grow and heatwaves become more frequent. Urban areas experience the ‘urban heat island’ effect, where built environments such as roads and building absorb and retain heat, causing temperatures to be higher than in surrounding rural areas. This can impact public health, energy consumption and critical infrastructure.

However, forecasting temperature variations within cities is challenging. Existing weather models typically operate at a resolution of around 1.5 kilometres, making it difficult to capture neighbourhood-level differences in temperature. Simulating these complex urban processes at a finer scale is also computationally expensive.

The approach

A Met Office led study, explored how machine learning could enhance urban temperature forecasts by combining:

• Operational weather forecasts
• Citizen science observations from the Weather Observations Website (WOW)
• Detailed land-use information at much higher resolution than current operational forecasts use

This research led by Dr Lewis Blunn from the Met Office urban-scale modelling group, working alongside PhD researchers from the University of Reading fused these datasets and developed a machine learning approach capable of downscaling forecast temperature from a 1.5km grid to a more detailed 100m grid. This allowed temperature variations across different neighbourhoods to be represented far more effectively.

The role of citizen science

A key element of the project was the use of observations submitted by the public through WOW. These home weather stations provided a dense network of observations across urban areas, often filling the gaps where professional monitoring stations were sparse.

The study analysed eight heatwave events in London between 2019 and 2021, drawing on this extensive observational dataset to train and test several machine learning algorithms.

Results

The findings showed that machine learning could improve temperature prediction errors by up to 11% compared with uncorrected weather forecasts. The models also delivered a more accurate representation of the urban heat island effect, improving understanding of how temperatures vary between urban and rural areas throughout the day. Three machine learning approaches were evaluated:

• Random Forest Regression (RFR)
• Extreme Gradient Boosting (XGB)
• Multi-Layer Perceptron (MLP)

These approaches consistently outperformed the original UK weather forecast model in representing urban temperature patterns.

Impact

This project demonstrates a practical pathway to improving weather forecasts from kilometre scale predictions to neighbourhood-scale insights. This capability has the potential to:

• Improve heat-health warnings and public safety information
• Support local authorities in climate adaptation planning
• Help infrastructure operators prepare for heat-related impacts
• Provide more detailed information for urban resilience and sustainability planning

Looking ahead

This work provides a strong foundation for future research and operational improvements, with the potential next steps being to expand the approach across the UK, incorporating a larger volume of historical observations, and exploring additional factors such as building characteristics alongside land-use data.

Researchers are also investigating how these techniques could be integrated into the Met Office's operational forecasting systems, while ensuring robust quality control of citizen weather station data. Longer term, advances in urban-scale modelling and machine learning could enable even more accurate, hyper-local forecasts for cities across the UK.

This project highlights how combining traditional weather forecasting expertise with emerging technologies and citizen science can deliver more accurate, localised climate intelligence to support safer and more resilient cities.