Environmental models and software are essential tools for understanding the complex interactions of the natural world. They empower us to foresee potential futures, unravel intricate trends and expand our scientific knowledge, ensuring we make informed decisions for a sustainable future. This includes environmental emissions and exposure models, which tell us how chemicals and other potential pollutants enter, move around and behave in the environment. To achieve accurate, efficient, collaborative and integrative insights into this pollution and its sources, it is imperative that our models and software keep pace with scientific and technological advances, the increasing availability of data and a heightened importance on assessing complex, interconnected systems under a changing climate. But often our models follow outmoded programming paradigms and technological setups that makes this difficult. They are often monolithic codebases rather than flexible, standalone modules, making them difficult to adapt to emerging risks or integrate with other models to predict, e.g., the societal drivers and One Health impacts of pollution. Furthermore, modelling efforts often overlook ethical and sustainability issues, like the carbon footprint of running complex simulations or the societal impacts of using model predictions to inform policy. These considerations framed a workshop that took place in October 2024, bringing together interdisciplinary environmental modellers from around the world to discuss the question: what does the next generation of environmental models look like?
The focus was interactive sessions, where participants discussed this question by referering to six pillars: Software engineering and collaborative platforms; Interdisciplinary learning; Cloud-based and exascale computing; Citizen science; Artificial intelligence, and; Big data and better monitoring. In this presentation, we reflect on the outcomes, placing them in the context of emissions and exposure modelling.
The workshop was part of broader efforts to build an international community of practice around environmental modelling. A priority identified is that training, education and knowledge transfer are vital to ensuring that we empower the next generation of environmental modellers, as well as the models themselves, and we hope this community will provide a space to enable this.

Deforestation is associated with a decrease in slope stability through the alteration of hydrological and geotechnical conditions. As such, deforestation increases landslide activity over short, decadal timescales. However, over longer timescales (0.1-10 Myr) the location and timing of landsliding is controlled by the interaction between uplift and fluvial incision. Yet, the interaction between (human-induced) deforestation and landscape evolution has hitherto not been explicitly considered. We address this issue in the North Tanganyika-Kivu rift region (East African Rift). In recent decades, the regional population has grown exponentially, and the associated expansion of cultivated and urban land has resulted in widespread deforestation. In the past 11 Myr, active continental rifting and tectonic processes have forged two parallel mountainous rift shoulders that are continuously rejuvenated (i.e., actively incised) through knickpoint retreat, enforcing topographic steepening. In order to link deforestation and rejuvenation to landslide erosion, we compiled an inventory of nearly 8000 recent shallow landslides in To accurately calculate landslide erosion rates, we developed a new methodology to remediate inventory biases linked to the spatial and temporal inconsistency of this satellite imagery. Moreover, to account for the impact of rock strength on both landslide occurrence and knickpoint retreat, we limit our analysis to rock types with threshold angles of 24-28g. Rejuvenated landscapes were defined as the areas draining towards Lake Kivu or Lake Tanganyika and downstream of retreating knickpoints. We find that shallow landslide erosion rates in these rejuvenated landscapes are roughly 40 % higher than in the surrounding relict landscapes. In contrast, we find that slope exerts a stronger control on landslide erosion in relict landscapes. These two results are reconciled by the observation that landslide erosion generally increases with slope gradient and that the relief is on average steeper in rejuvenated landscapes. The weaker effect of slope steepness on landslide erosion rates in the rejuvenated landscapes could be the result of three factors: the absence of earthquake-induced landslide events in our landslide inventory, a thinner regolith mantle, and a drier climate. More frequent extreme rainfall events in the relict landscapes, and the presence of a thicker regolith, may explain a stronger landslide response to deforestation compared to rejuvenated landscapes. Overall, deforestation initiates a landslide peak that lasts approximately 15 years and increases landslide erosion by a factor 2 to 8. Eventually, landslide erosion in deforested land falls back to a level similar to that observed under forest conditions, most likely due to the depletion of the most unstable regolith. Landslides are not only more abundant in rejuvenated landscapes but are also smaller in size, which may again be a consequence of a thinner regolith mantle and/or seismic activity that fractures the bedrock and reduces the minimal critical area for slope failure. With this paper, we highlight the importance of considering the geomorphological context when studying the impact of recent land use changes on landslide activity.