Spatial generalized linear mixed-models, consisting of a linear covariate effect and a Gaussian Process (GP) distributed spatial random effect, are widely used for analyses of geospatial data. We consider the setting where the covariate effect is non-linear and propose modeling it using a flexible machine learning algorithm like random forests or deep neural networks. We propose well-principled extensions of these methods, for estimating non-linear covariate effects in spatial mixed models where the spatial correlation is still modeled using GP. The basic principle is guided by how ordinary least squares extends to generalized least squares for linear models to account for dependence. We demonstrate how the same extension can be done for these machine learning approaches like random forests and neural networks. We provide extensive theoretical and empirical support for the methods and show how they fare better than naïve or brute-force approaches to use machine learning algorithms for spatially correlated data.

(Invited session) This session is organised by the Environmental Statistics Section (ESS) of the Royal Statistical Society (RSS), UK, with the aim of illustrating the work of members of the section, individuals, institutions, and organisations related to the ESS and creating stronger links between the ESS/RSS and TIES communities. The session will particularly focus on statistical and data science perspectives from UK academics, research institutes and public sector on applications related to climate change. Speakers are from the UK Met Office, UK British Geological Survey, the Data Science for the Natural Environment Team at Lancaster and the Alan Turing Institute.

(Organized session) This session will discuss different models that involve model averaging for monitoring suspended solids in water quality in Nigeria the most populous African country. We shall also discuss change-point models for evaluating carbon dioxide pollution using Bayesian approach and also to be considered is the use of log-normal distribution in detecting the concentration of polluted water sample as the concentration reduces along various dilution stages. In this session attempts shall be made to provide solutions to various environmental pollution problems using data that originate from the local communities.

(Invited Session) Data sets, quantitative methods and implementation of these methods now existwhich allow us to investigate the importance of factors and of changing regimes in explainingoutcomes. In this session, two of the papers consider the impact of climate change, one on thevegetation dynamics of four East African climatic and agricultural zones, and the other on themalnutrition status of children in Egypt. To analyze the complex data sets involved, themethods of wavelet and causal discovery analysis are used in the first case, and a multi-levelgeostatistical model fitted under the Bayesian framework using the integrated Laplaceapproximation in the second. Another common characteristic of environmental data sets is thatof regime changes over time. Various change-point methods are considered which are suitablefor the assessment of environmental pollution and climate data in the third paper. The sessionwill be of interest to individuals who are working with one of the types of methods or who wouldlike to learn more about such methods, as well as individuals interested in the issues related tothe impact of climate change or pollution considered in the papers.

Contributed Talks on the topic of Spatio-Temporal Modeling

(Invited Session) Predictive Analytics in Agriculture

Contributed Talks on the topic of Water

Incredible progress has been made in recent years with the vast quantity of accurate and reliable Earth Observations (EO) data and tools to drive innovation and inform complex environmental problems—from climate and socio-economic modeling for disaster risk reduction (DRR) assessments, to ecosystem accounting (EA) that provides a GDP value frame for environmental services. High resolution satellite data allows temporal tracking of environmental trends with multi-terabyte cloud catalogs and AI-supported statistical cloud computing, providing a greater granularity of natural assets for decision-making. Yet, “despite commitments to build resilience, tackle climate change and create sustainable development pathways, current societal, political and economic choices are doing the reverse” (UN DRR, 2022). Creating system of systems—Systems Thinking—is critical, as more data isn’t necessarily moving the lever sufficiently toward better decision-making, achieving international climate goals, nor toward justice and equity. Systems thinking re-frames the perspective: the biggest modeling unknown is the knowledge systems itself that scientists are working within, and how to constellate to other knowledge systems. Bridging barriers between more than just statistics and environmental sciences, but between western and Indigenous sciences, is proving a critical reorientation to action scientific approaches. Working with communities that experience disproportionate effects of climate change and environmental injustice challenges can offer a radically different perspective; when people’s relationship to the land is about more than an essential part of their survival, but as intrinsic to their identities, this offers a shift in worldview that can accommodate a new set point and flourishing innovation. We live in times that demand more than an EO data infrastructure, but a just and relevant knowledge infrastructure, where community-level knowledge guides research and informs policy and decision support tools. Dr. Caudill will talk about Systems Thinking frameworks for de-siloing western sciences, bridging across sectors in society, and co-design principles that constellate worldviews.

(Invited session) A digital twin has the potential to revolutionise our ability to adapt environmental management strategies by:1. taking advantage of large volumes of new, rich and highly heterogeneous data2. developments in fusing data driven and process modelling, to develop emulations as the engine of a digital twin;3. taking a whole system approach, enabling the sensitivity of different parts of the system to challenges, and management scenarios to be assessed and uncertainty quantified; and enabling a decision pipeline that runs in real- or near real-time (from data acquisition to analysis, ensemble model execution, uncertainty quantification and visualisation) that supports the decision-making process at a variety of temporal and spatial scales.

Contributed Talks on the topic of Time Series Methods.

Contributed Talks on the topics of Forestry and Climate Change.

(Invited Session) Climate Resilience and Natural Hazards

Contributed talks on the topic of Statistical Modeling (1).

Situated on the treaty and traditional territory of the Mississauga (Michi Saagiig) Anishnaabeg, Trent University houses Canada’s first Indigenous Environmental Studies and Sciences program. Centered on the Otonabee River, it is the traditional “place at the end of the rapids” where peoples and ideas have come together since time immemorial. This provides a frame for this session which seeks to go beyond ‘Breaking Down Barriers between Statistics and Environmental Science' to break down barriers between western sciences and Indigenous sciences. Following the two-eyed seeing approach, this session will constellate and weave knowledge systems to meet the grand complexities of biodiversity loss, environmental, and climate change challenges that remain unmet through siloed scientific focal points.

Following the session talks, a sharing circle will be offered in lieu of a traditional panel of speakers to explore just protocols and avenues that action scientific knowledge.

(Invited Session) Environmental and ecological data sets are complex and, for statistical methods to be effective, the methods must account for the nature of the data. Two important areas are temporal-spatial modelling and functional data analysis. For statisticians, the questions considered are 1) how to account for surveys where yearly data are not collected uniformly at sites and extreme abundances occur (Rivest et al.); 2) the advantage of state-space models for population size and growth rates incorporating environmental effects (Hyman et al.); and 3) improved estimates of functional response with estimates of bias and variance (Di Batista et al.). The data sets are cod stocks, seagrass availability and blue crab populations, and Adriatic Sea warming and diversity profiles.

NASA’s Earth observing instruments have provided comprehensive observations of wildfires and aerosol plumes from wildfires. At present, JPL’s Segmentation, Instance Tracking, and data Fusion Using multi-SEnsor imagery (SIT-FUSE), utilizes an unsupervised machine learning (ML) framework that allows users to segment instances of objects like wildfires and smoke plumes in single and multi-sensor scenes from NASA’s satellite instruments with minimal human intervention, in low and no label environments. The output of the ML framework also facilitates the tracking of smoke plumes, allowing users to more easily, but still manually, track plumes across multiple scenes over time. Here, we discuss the approaches and progress being made towards the automation of tracking instances across scenes from the same instrument sets as well as the exploration of techniques like contrastive learning (CL), enhanced by the topological features of the object instances detected, to augment SIT-FUSE with the capability to automatically track wildfire and smoke plume instances across datasets from like and disparate instrument sets.

Contributed Talks on the topics of Atmosphere and Wind

Contributed Talks on the topic of Microplastics and Dispersion (Special Session)

(Invited Session) Water and sanitation for all by 2030 is Goal 6 of the UNEP Sustainable Development Goals, adopted by all United Nations Member States in 2015. Water quality and water quantity are issues pertinent to Goal 6 and they are interconnected with climate change and biodiversity loss, the other two planetary crises pushing nature to the breaking point according to UNEP in 2023.  Several water quality issues, including pollution loads and concentration-critical toxic contaminants, and statistical methods that have been used in addressing these issues in Canada will be discussed. Floods are an aspect of water quantity that is being exacerbated by climate change. An analysis of flood data from across Canada using statistical extreme value methods will be presented. The link to biodiversity will be explored. The session will help to broaden the discussion around climate change and keep these other important topics, and the role of suitable statistical methodology, in the public and scientific mindset.  Water and sanitation for all by 2030 is Goal 6 of the UNEP Sustainable Development Goals, adopted by all United Nations Member States in 2015. Water quality and water quantity are issues pertinent to Goal 6 and they are interconnected with climate change and biodiversity loss, the other two planetary crises pushing nature to the breaking point according to UNEP in 2023.  Several water quality issues, including pollution loads and concentration-critical toxic contaminants, and statistical methods that have been used in addressing these issues in Canada will be discussed. Floods are an aspect of water quantity that is being exacerbated by climate change. An analysis of flood data from across Canada using statistical extreme value methods will be presented. The link to biodiversity will be explored. The session will help to broaden the discussion around climate change and keep these other important topics, and the role of suitable statistical methodology, in the public and scientific mindset.

Contributed talks on the topic of Environmental Modeling

Contributed talks on the topic of Statistical Modeling (2).