 Global re-analyses allow us to reconstruct the past weather with a resolution of around 80 to 200 kilometres. This means that they cannot resolve small-scale weather variability or small-scale processes such as thunderstorms and convection. To address this challenge and reconstruct the weather on a local scale, so-called downscaling techniques are used. We differentiate between numerical and statistical downscaling. With numerical, often also called dynamical downscaling, a regional weather forecasting model is used for instance, capturing only the Atlantic European sector. At the model boundaries, the re-analysis is fed in, but internally this model considers high-resolution topography and land cover and hence their effects on precipitation patterns or wind flow. Into that reconstruction, another, more detailed forecasting model is embedded again. This is referred to as nesting. After repeating this step multiple times, the resolution is 2 to 3 kilometres and shows local weather. In a sense, numerical downscaling is like zooming in and creating the most reasonable weather according to the model and the inputted re-analysis data. In the dynamical downscaling approach, the local information is generated by the model, but we can also use local observations and generate the information statistically. One approach is the so-called analog method, which presumes that similar large-scale conditions lead to similar local weather. We need a sufficiently large pool of analogs that is a period with both local and large-scale information. For a weather reconstruction, we may simply search this pool for the large-scale conditions most similar to those that we are interested in. This day is called the best analog and the local information for that day serves as a reconstruction. These reconstructions can still be corrected and improved based on historical station measurements. Numerical and statistical downscaling produce very similar and most importantly plausible local weather of the past. Numerical downscaling has the advantage of providing us with a three-dimensional state of the atmosphere. It is, however, much more computationally intensive than the efficient analog method and it carries a model bias. From historical measurements, we have generated global and local weather information on different timescales. This information can tell us much about processes in the atmosphere. It is also increasingly used for model applications such as hydrological modelling, crop modelling or others. Let us return to the 1868 case. To understand a flood event like 1868, not only the weather but also its effects on the surface are important. In a first step, we can use the downscaled weather data in a hydrological model to reconstruct river runoff or lake levels at one place over time. But whether a large runoff actually leads to flooding and which areas would be affected also depends on further factors such as flow characteristics and the surrounding terrain. Therefore, our next step is to feed the results of hydrological modelling into hydraulic models where past terrain and buildings are included. We can see how the Ticino River reacted during the 1868 flood. By modifying parameters in the model, we can experiment with many what-if scenarios. For example, if the 1868 event happened today, this allows us to study the effect of a changing landscape and assess the effectiveness of measures such as river corrections. For 1868, we find that the flood eroded much of the Ticino Riverbed and the now increased runoff makes a repeated flooding of that size impossible. These results now finally allow for a detailed risk assessment. We can generate hazard maps by grouping areas by their risk of flood and visualise floods in three dimensions. Furthermore, we can study how climatic changes affect extreme precipitation and flood risk. Information like this is very useful to policy and decision makers in infrastructure projects and a wide array of other areas. As we have seen, historical climate data has great value in reconstructing the past weather. In combination with models, we can create global re-analyses which can then be downscaled to a regional level. This can reveal much about atmospheric mechanisms and provides us with the necessary information for a wide range of applications.