Species distribution models (SDMs) are crucial tools for understanding and predicting biodiversity patterns, yet they often struggle with limited data, biased sampling, and complex species-environment relationships. Here I present NicheFlow, a novel foundation model for SDMs that leverages generative AI to address these challenges and advance our ability to model and predict species distributions across taxa and environments. NicheFlow employs a two-stage generative approach, combining species embeddings with two chained generative models, one to generate a distribution in environmental space, and a second to generate a distribution in geographic space. This architecture allows for the sharing of information across species and captures complex, non-linear relationships in environmental space. I trained NicheFlow on a comprehensive dataset of reptile distributions and evaluated its performance using both standard SDM metrics and zero-shot prediction tasks. NicheFlow demonstrates good predictive performance, particularly for rare and data-deficient species. The model successfully generated plausible distributions for species not seen during training, showcasing its potential for zero-shot prediction. The learned species embeddings captured meaningful ecological information, revealing patterns in niche structure across taxa, latitude and range sizes. As a proof-of-principle foundation model, NicheFlow represents a significant advance in species distribution modeling, offering a powerful tool for addressing pressing questions in ecology, evolution, and conservation biology. Its ability to model joint species distributions and generate hypothetical niches opens new avenues for exploring ecological and evolutionary questions, including ancestral niche reconstruction and community assembly processes. This approach has the potential to transform our understanding of biodiversity patterns and improve our capacity to predict and manage species distributions in the face of global change.

Over the past decade, AI techniques have revolutionized computer vision and natural language processing. There is growing anticipation for similar breakthroughs in scientific domains, driven by pressing societal challenges such as national water resource management, energy and food security, and climate change mitigation and adaptation. However, spatiotemporal data present unique challenges for existing AI models, including spatiotemporal autocorrelation and heterogeneity, long-range and multi-scale dependencies, the existence of physical knowledge and constraints, and the paucity of ground truth. In this talk, I will present my recent research aimed at addressing these challenges, including (1) geospatial AI for observation-based flood inundation mapping in hydrology, and (2) AI surrogates to accelerate coastal ocean circulation modeling in physical oceanography.

This research explores how Unmanned Aerial Vehicles (UAVs), the Internet of Things (IoT), and
Artificial Intelligence (AI) can revolutionize modern agriculture. By harnessing deep learning
technologies, the approach aims to improve crop health management, optimize resource
utilization, and tackle key challenges like pest infestations, soil health and crop diseases through
advanced data-driven solutions that combine AI, UAVs, and IoT systems.
For instance, models like 3D CNN combined with RNN architectures—such as 3D CNN+LSTM,
3D CNN+GRU, 3D CNN+Bi-LSTM, and 3D CNN+Bi-GRU—will be employed to process both
spatial data from UAVs and time-series data from sensors. This dual data input enables more
precise monitoring, decision-making, and resource optimization for precision farming tasks such
as crop monitoring, weed detection, and pesticide application. The combination of 3D CNN and
RNN architectures allows for more accurate predictions and more informed decision-making,
resulting in robust solutions for agricultural challenges. Furthermore, Deep Reinforcement
Learning (DRL) offers adaptability to dynamic environmental changes, providing a scalable
framework for disaster management by optimizing response strategies. DRL enhances resilience
and preparedness for natural disasters, such as hurricanes, droughts, and floods, by offering site-
specific early warning systems and enabling more targeted, efficient responses.
The primary goal of this research is to reduce resource waste, particularly in the use of fertilizers
and pesticides, while maximizing crop yields and minimizing environmental impact. By promoting
sustainable farming practices, AI-driven models will not only enhance agricultural productivity
but also mitigate the effects of climate change and address resource constraints, securing the future
of food production. As AI continues to evolve, these models will play an increasingly critical role
in shaping the future of agriculture, ensuring a balance between efficiency, sustainability, and
resilience against future challenges.