Researchers introduce crowd forecasting at Pulpit Rock to help manage visitor numbers

Around 400,000 people make the trip to one of Norway's most popular natural attractions every year. When lots of tourists arrive at the same time, it puts pressure on roads, car parks, hiking trails, and the viewpoint itself.

Large variations in visits to Pulpit Rock throughout the day place high demands on visitor management, staffing, and infrastructure.

(Stock photo: Ivars Utināns / Unsplash)

Kathrine Nitter Kathrine Nitter Senior Communications Adviser

Presented by: Presented by: SINTEF

Published 20 October 2026 - 00:01

Preikestolen, also known as Pulpit Rock, and the Lysefjord area are among Norway’s most visited natural attractions. The number of visitors have grown steadily for years.

This puts increased pressure on both nature and local infrastructure.

Local tourism operators are required to document their environmental and social impact through schemes such as 'Sustainable Destination' and 'National Tourist Trails.'

Researchers at SINTEF and Lysefjord Development have now shown that existing data can be used to predict when and where visitor numbers will peak.

Their model addresses a large and shared need of many who work in the area, without the need to collect new data.

Collects data from various sources

Data on visitor numbers, weather, traffic, and cruise ship arrivals already exist. But it is spread across different organisations and rarely used to predict future visitor numbers.

By combining this information, researchers can provide more accurate forecasts of when the busiest periods are likely to occur.

“The value is not just in knowing how many people come per day, but also at what time of day they come,” says Johannes C. Apon.

He is a tourism developer at Lysefjorden Development.

“An hourly forecast gives us a better basis for staffing, traffic management, and information. This will both provide a better experience for the guests and a better day for the employees,” he says.

A group of hikers with backpacks climbs a rocky trail through a dense forest towards Preikestolen. — Tourists on their way up to Preikestolen. Peak visitor times can now be announced sooner.

Visitor peaks predicted a day in advance

The researchers trained a machine learning model called LightGBM using two years of data collected from the trail counters at Preikestolen.

They combined this with weather data, traffic figures from the Norwegian Public Roads Administration, and information on cruise ship arrivals in Stavanger.

The model was designed to predict visitor numbers 24 hours into the future. The researchers then compared its forecasts with actual visitor numbers from 2025 and 2026.

The results were better than expected. The model successfully identified patterns in visitor behaviour, showing that the season, day of the week, weather, and cruise ship arrivals explain much of the variation in hourly visitor numbers.

“We really just wanted to see if this could be done at all. That a simple model produced such clear results shows that it has great potential,” says Nathalie Labonnote.

She is head of research and innovation at SINTEF and manages the research project VisitorLAB.

“This means that we don’t need new data. We just need to structure and connect what already exists,” she explains.

From reacting to crowds to planning ahead

Researchers at SINTEF also interviewed local organisations and businesses in the Lysefjord area, including tour operators, boat companies, municipalities, and Norwegian People’s Aid.

Nearly everyone expressed the same need: better visitor forecasts and a shared overview of what happens in the fjord from day to day.

“What surprised us the most was the strong consensus among the various actors. Large and small, private and public. Everyone pointed to the same need for better forecasts and a common picture of the situation. It suggests that this actually solves a problem that the tourism industry is familiar with on a daily basis,” says Labonnote.

With better forecasts, tour operators could arrange extra shuttle buses, adjust staffing levels, and encourage visitors to choose less busy days.

Next step: improving the model

The model still needs further development, and data currently has to be collected manually from various systems. Before it can be used in daily operations, this process needs to be automated.

The researchers also plan to improve the model's accuracy by including more data sources.

One proposed solution is a shared digital dashboard that would give businesses and organisations in Lysefjord an up-to-date overview of visitor activity.

The aim is to help local operators move beyond day-to-day management and short-term forecasts towards better seasonal planning and more sustainable long-term development.