The Potential of Causal Inference for examining the connection of Snowmelt and River Discharge – A Northern Hemisphere Analysis from 1980 - 2022

Kurzfassung

Snow is the main freshwater source for 20% of the global population and a crucial economic and environmental factor. However, climate change is reducing snow reliability, with shrinking snowpacks and increasing variability in snow cover and melt patterns. Sectors like agriculture and hydropower must adapt to these shifts, and are therefore relying on robust data for informed decision-making. Our previous research has shown that although the snowpack itself is well studied, the connection between snow and water availability, particularly on a large scale, remains underexplored. To address this gap, we performed a causal inference analysis across the Northern Hemisphere from 1980 to 2022, linking snow to river discharge while accounting for seven additional influencing variables. The chosen causal inference approach, PCMCI, consists of two steps: causal discovery and causal effect estimation. In the causal discovery phase, a graph connecting the various variables is derived through multiple conditional independence tests. The causal effect estimation then fits a random forest regressor model to each time lag examined, allowing us to determine the effect of an intervention on the system, known as the causal effect. Using a moving window design, we evaluated how the causal effect of snowmelt on river discharge changed over the period from 1980 to 2022. This causal effect trend was then compared to other variable trends. Additionally, we examined how different snow data sources influence the causal inference approach. To this end, we utilized passive microwave (PMW) data (SnowCCI v3.1), reanalysis data (ERA5-Land), and a combined dataset in which we used reanalysis data to fill spatial and temporal gaps in the PMW dataset. At the Eastern Snow Conference, we will present our methodology and results, showing how the causal effect of snowmelt on river discharge has changed in several river basins across the Northern Hemisphere. Furthermore, we will use the results to demonstrate the potential of causal inference in snow hydrological analyses, as well as the influence of different snow data sources on analysis outcomes.