Agricultural productivity is sensitive to temperature and precipitation extremes, which are increasing with climate change. It is well-established that planting and growing season weather affects crop yields, but conditions in other seasons may also be important. We generate a county-level dataset that links yields for six major crops in the US with agriculture-relevant weather variables for four distinct seasons (planting, growing, harvest, and non-growing) over the years 1983 to 2021. The data include binned temperature variables, precipitation, and the Palmer Drought Severity Index (PDSI). We demonstrate that models using weather conditions from all four crop calendar seasons segments outperform those using only growing season or yearlong data at predicting yields, highlighting the importance of considering the impact of non-growing season weather on agricultural productivity.

Agriculture is increasingly threatened by climate change impacts, especially in lower-income nations where traditional paper survey-based data collection for detailed agricultural reporting can be impractical. Leveraging advancements in remote sensing and machine learning, we propose a Convolutional Neural Network-Long Short-Term Memory model designed to infer farm-level crop yields from one year of high-resolution weather data. We also propose assessing the potential of predicting based on forecasted weather data and implementing transfer learning to predict crop yield in regions outside the US with sparse data collection. If successful, this approach would offer a detailed and adaptive tool for countries with all levels of monitoring resources.

Carbon Capture and Storage (CCS) is a pivotal technology for reducing greenhouse gas emissions. While developments have been made in capture and storage capabilities , the planning and development of an optimized transport pipeline network for linking emission sources to storage sites remains understudied. This study aims to extend the capabilities of SimCCS, a widely-used CCS planning tool, to incorporate environmental, social, and cultural considerations alongside economic costs of pipeline networks. Utilizing multi-objective optimization, we introduce an additional objective function that minimizes environmental and social impacts. This function integrates spatial data layers representing critical habitats, protected areas, and other socio-ecological factors. Preliminary results illustrate the model's capacity for multi-objective optimization. The annual expense for maintaining a sample pipeline network increased from $434 million to $622 million, with pipeline lengths of 1986 kilometers and 2878 kilometers, respectively, when shifting focus from cost to environmental and social impacts. This research contributes a more comprehensive framework for the planning of future CCS infrastructure that is both economically and environmentally sustainable.

Accurate forest inventory data are essential for tracking carbon sequestration, estimating carbon emissions from deforestation, assessing plant and animal habitats for biodiversity, and predicting environmental risks such as wildfires. Traditional methods of data collection have faced challenges in either scale or precision. The advent of terrestrial laser scanners addressed some of these issues but faced limitations in cost and mobility. This paper proposes a new approach using mobile LIDAR for forest inventory data collection. By integrating advancements in computer vision, the methodology aims to provide comprehensive individual tree data, including parameters like diameter at breast height, volume estimations, species identification, and temporal tracking of individual trees. This proposed research direction addresses current gaps in the use of LIDAR and camera inference for forestry data where existing work does not generate domain context-aware data by narrowly focusing collection on isolated tree attributes.