FrostGardens & farming
Frost unlikely
2% chance of temperatures below 0 °C
Berlin
SunshineSolar energy
About usual
1.7 kWh per m² expected; 1.6 is usual in October
WindOutdoors
Gentle breeze
Around 12 km/h on average through the day
Tomorrow's frost risk, sunshine and wind for 12 German cities, forecast by machine learning from German Weather Service observations.
Based on DWD weather data up to Fri 16 Oct 2026
Frost unlikely
2% chance of temperatures below 0 °C
About usual
1.7 kWh per m² expected; 1.6 is usual in October
Gentle breeze
Around 12 km/h on average through the day
ClimateLens forecasts tomorrow's weather conditions for 12 of Germany's largest cities from three everyday angles: the risk of frost for gardens and farming, the sunshine energy available to solar panels, and how windy it will be for time spent outdoors.
Instead of simulating the atmosphere like a national weather service, ClimateLens learns from history. Years of daily measurements from the German Weather Service (DWD) teach a machine-learning model how today's weather tends to turn into the weather of the next days. It is built as a modern cloud service: data collection, retraining and publishing run automatically on Google Cloud, and every forecast can be traced back to the data and the model version that produced it.
Every afternoon, ClimateLens downloads the newest measurements from 18 DWD weather stations serving 12 cities: temperature, wind, humidity, air pressure, cloud cover, rain and sunshine. So far it holds 13,656 daily records covering 3.1 years of weather, since Tue 05 Sep 2023.
A machine-learning model called LightGBM studies this history. It combines hundreds of small decision rules, each capturing a pattern such as “a clear, calm evening after a cold day often brings frost”.
DWD publishes each day's measurements on the following day, so the newest data is usually from yesterday. ClimateLens therefore looks two days ahead of it to reach tomorrow, using the recent days and the time of year. Before use, every forecast was tested on past days it had never seen and beat simple rules such as “tomorrow will be like today” and “the usual weather for this time of year”.
New measurements arrive every day, so the forecast is refreshed daily. Once a month the model is retrained on everything collected so far, and the most accurate method is kept. Last trained on Sat 17 Oct 2026.