• Energy Research
• 2025-2025close
• IT close
• FI close

Buildings account for 40% of global energy consumption. A considerable portion of building energy consumption stems from heating, ventilation, and air conditioning (HVAC), and thus implementing smart, energy-efficient HVAC systems has the potential to significantly impact the course of climate change. In recent years, model-free reinforcement learning algorithms have been increasingly assessed for this purpose due to their ability to learn and adapt purely from experience. They have been shown to outperform classical controllers in terms of energy cost and consumption, as well as thermal comfort. However, their weakness lies in their relatively poor data efficiency, requiring long periods of training to reach acceptable policies, making them inapplicable to real-world controllers directly. In this paper, we demonstrate that using federated learning to train the reinforcement learning controller of HVAC systems can improve the learning speed, as well as improve their ability to generalize, which in turn facilitates transfer learning to unseen building environments. In our setting, a global control policy is learned by aggregating local policies trained on multiple data centers located in different climate zones. The goal of the policy is to minimize energy consumption and maximize thermal comfort. We perform experiments evaluating three different optimizers for local policy training, as well as three different federated learning algorithms against two alternative baselines. Our experiments show that these effects lead to a faster learning speed, as well as greater generalization capabilities in the federated policy compared to any individually trained policy. Furthermore, the learning stability is significantly improved, with the learning process and performance of the federated policy being less sensitive to the choice of parameters and the inherent randomness of reinforcement learning.

Precise and reliable climate projections are required for climate adaptation and mitigation, but Earth system models still exhibit great uncertainties. Several approaches have been developed to reduce the spread of climate projections and feedbacks, yet those methods cannot capture the nonlinear complexity inherent in the climate system. Using a Transfer Learning approach, we show that Machine Learning can be used to optimally leverage and merge the knowledge gained from global temperature maps simulated by Earth system models and observed in the historical period to reduce the spread of global surface air temperature fields projected in the 21st century. We reach an uncertainty reduction of more than 50% with respect to state-of-the-art approaches while giving evidence that our method provides improved regional temperature patterns together with narrower projections uncertainty, urgently required for climate adaptation.

Global warming may induce significant alterations to the rainfall regimes, especially in the Mediterranean basin, which can be considered as a hot-spot for climate change. Several previous studies focused on the variations in annual rainfall and extreme values, while rainfall seasonal variations were less explored. Rainfall seasonality is a critical climate factor affecting the evolution of natural vegetation, water resource availability, and water security. Rainfall seasonality anomalies may have a high impact, especially in areas of the Mediterranean basin where water supplied during the wet season is used to offset rainfall shortages in the dry season. In southern Italy, the occurrence of long water deficit periods and extremely concentrated rainy seasons could limit water uses and cause serious effects on crop yield and, consequently, on food production.This study aims at exploring potential variations in rainfall seasonality over the last 100 years across three regions of southern Italy (Campania, Sardinia, and Sicily) through a dynamic approach proposed by Feng et al. (2013). The study area is characterized by a Mediterranean climate, where the hydrological year consists of a net alternation of two seasons: a cold-rainy period (wet season), usually including fall-winter months, and a hot-dry period (dry season), typically including spring-summer months. The analysis proposed involves the determination of time-variant values of rainfall magnitude and frequency of the two seasons (wet and dry).Daily rainfall values, recorded between 1916 and 2023, are gathered from hundreds of rain gauge stations distributed over the three regions. A pre-processing procedure was applied for data quality check, data reconstruction in years with less than 80% of missing data, and rain gauge selection; then, only rain gauge datasets with adequate data availability (i.e., more than 70 complete years, with at least 15 years in the last two decades, 15 years in the pre-World War II period, and without significant data interruptions) were retained and used for data analyses. Rainfall depth over each season is idealized as an exponentially distributed independent random variable with mean values h (mm), whereas the seasonal rainfall occurrence is modelled as a Poisson process with rate l (d-1). Rainfall seasonality at each rain gauge was defined annually, considering different indices: the Dimensionless Seasonality Index (DSI); the seasonal rainfall depth and the seasonal values of h and l; the wet season timing (i.e., centroid of the season) and duration. The reference period was divided into different equal-size and non-overlapping subperiods.Differences in the various rainfall seasonality indices and their distributions among the various gauges, regions, and subperiods were analyzed, also investigating the influence of some climatic and topographic factors (i.e., temperature, gauge distance from the sea and elevation). A trend analysis based on Mann-Kendall's and Sen's Slope Method with statistical significance at 95% level of confidence, was also carried out considering a limited subset of gauges with the largest data availability for each region.

Transferring climate change physical knowledge Repository including the data needed to reproduce the workflow and the results presented in Immorlano et al., Transferring climate change physical knowledge PNAS, Volume 122, Issue 14, 2025. Related Publication---------------------F. Immorlano, V. Eyring, T. le Monnier de Gouville, G. Accarino, D. Elia, S. Mandt, G. Aloisio, P. Gentine, Transferring climate change physical knowledge, Proc. Natl. Acad. Sci. U.S.A. 122 (15) e2413503122, https://doi.org/10.1073/pnas.2413503122 (2025). Related code repository-------------------------- GitHub Contributors--------------- Francesco Immorlano (francesco.immorlano@cmcc.it)- Veronika Eyring (veronika.eyring@dlr.de)- Thomas le Monnier de Gouville (thomas.le-monnier-de-gouville@polytechnique.edu)- Gabriele Accarino (gabriele.accarino@cmcc.it)- Donatello Elia (donatello.elia@cmcc.it)- Stephan Mandt (mandt@uci.edu)- Giovanni Aloisio (giovanni.aloisio@cmcc.it)- Pierre Gentine (pg2328@columbia.edu) Acknowledgements and References---------------------------------------If you use the resource in your research, please cite our paper. At the moment, we offer the bibliography of the arXiv preprint version.

The global goal to mitigate climate change (CC) is to achieve net zero greenhouse gas emissions (GHGE) by 2050; the European Union (EU) aim is to cut GHGE at least by 55% already by 2030. These ambition targets require new GHGE mitigation measures across all land use sectors (LULUCF), where wetlands, as carbon (C) rich ecosystem, can effectively contribute to climate targets, biodiversity, and water-related ecosystem services. Natural peatlands accumulate C effectively due to water-logged conditions. However, they can turn into high GHG sources if they are drained, therefore there is still need to enhance knowledge regarding how and/or how much C is sequestered or released by peatlands after their restoration, as well as the socioeconomic effects.“ALFAwetlands - Restoration for the future” (www.alfawetlands.eu) is a Horizon Europe funded project (2022-2026), which is coordinated by Luke and carried out at local to EU levels with 15 partners across Europe. It’s main goal, in short, is to mitigate CC while supporting biodiversity and ecosystem services (BES) and being socially just and rewarding. This includes, e.g., increasing the knowledge about C storage and release in peatlands, specifically after restoration. While, in terms of C fluxes, focussing on peatlands, the project scope is larger and includes additionally floodplains, coastal wetlands and few artificial wetlands. ALFAwetlands will develop and indicate management alternatives for wetlands including such that have been or will be restored during this project. Measures under this project are not restricted to ecological restoration but include rehabilitation and re-vegetation action to improve ecosystem conditions (e.g., peatland forest: continuous-cover-forestry, cultivated peatlands: paludiculture). Studies are conducted in 9 Living Labs (LL’s) including 30 sites, which are located in wetlands in different parts of Europe (north-south gradient). At the local level, LL’s support and integrate interdisciplinary and multi-actor research on ecological, environmental, economic, and social issues. Experimental data from local sites are scaled-up and will be utilized e.g., by models to gain and understanding the potential impacts of upscaled wetland restoration measures. To achieve ALFAwetlands goals, 5 research workpackages are being implemented, namely: 1)improve geospatial knowledge base of wetlands, 2)co-create socially fair and rewarding pathways for wetland restoration, 3)estimate effects of restoration on GHGE and BES, with the data achieved from field experiments, 4)develop policy relevant scenarios for CC and BES, and 5)study societal impacts of wetland restoration. The project will also encourage stakeholders to utilise outputs and support their active participation in wetland management.

During the next few decades, changes in rainfall frequency and magnitude are expected to have major impacts on landscape evolution, social, and economic aspects of human society.We focus on seasonal rainfall variations by the end of the 21st century to define affected landslide-prone areas, future landslide alerts and the impact of shllow and deep-seated landslides on landscape development in the juncture of the Alpine, Pannonian, and Mediterranean region. For this work, we selected the six regional climate models (RCMs) from the EURO-CORDEX project, with the global climate simulations from CMIP5 (Coupled Model Intercomparison Project phase) driven by the six global circulation models (GCMs).  Of the two available spatial resolutions, i.e., 0.11° (12.5 km) and 0.44° (50 km), we considered the 0.11° spatial resolution with a regular 12.5 km grid with spacing between computational points. Six models were selected from 14 combinations of GCMs and RCMs that differ as much as possible from each other while reflecting as closely as possible the measured values of past climate variables. For this study, we considered climate scenarios variable: the daily rainfall datasets of two Representative Concentration Pathways (RCP), namely RCP4.5 (mid-way) and RCP8.5 (worst-case) for the time window from 1981 to 2100. Daily rainfall data were downscaled from 12.5 km resolution to 1 km. The downscaling of the data was performed daily for all six RCMs. To analyse future climate impact on landslides, the calculated models were divided into three 30-year projection periods: 1st period (near-term) between 2011-2040, 2nd period (mid-century) between 2041-2070, 3rd period (end of the century) between 2071-2100. To show the characteristics of seasonal variations, shorter periods within a year were considered, namely four meteorological seasons: winter (December, January, February), spring (March, April, May), summer (June, July, August), and autumn (September, October, November). Future projections represent a 30-year maximum rainfall from the 30-year baseline period in the past (1981-2010).The observed changes in the occurrence of shallow landslides are significant, especially in the winter months, where we can expect more landslide-prone areas compared to the baseline period. Shallow landslides will have a greater impact on the landscape in spring and summer than deep-seated landslides, especially in vineyards.FundingThis work was supported by the by the Slovenian Research and Innovation Agency (the research project J1-3024). Additional financial support was provided by the project “Development of research infrastructure for the international competitiveness of the Slovenian RRI space – RI-SI-EPOS” (co-financed by the Republic of Slovenia, Ministry of Education, Science and Sport and the European Union from the European Regional Development Fund).ReferenceJemec Auflič, M., Bezak, N., Šegina, E. et al. Climate change increases the number of landslides at the juncture of the Alpine, Pannonian and Mediterranean regions. Sci Rep 13, 23085 (2023). https://doi.org/10.1038/s41598-023-50314-x

Public procurement emerges as an instrument to promote sustainability and innovation, facilitating the implementation of policy goals. Adaptive procurement leverages sustainable innovation available in the market. Developmental procurement creates demand for solutions that do not exist yet but need to be developed first. Real-world applications from sectors such as construction, transportation, and waste management illustrate the transformative role of procurement. However, challenges persist, including the inconsistent implementation of sustainability and innovation-oriented procurement practices across organizations and sectors and the presence of multiple potentially conflicting objectives.