PROJECT: Developing an Energy Efficient Deep Learning Model for Onsite Detection of Forest Fire and its Severity
PI: Arghya Das, Assistant Professor, UAF
Starting from saving lives to saving our environment and climate from toxic chemicals and greenhouse gasses, quick and accurate detection of forest fires is of paramount need. It is especially important to save the natural and human resources of Alaska; the state which has reported a five-fold increase in the area burned due to forest fires in the last fifteen years (2004-2021) compared to the previous fifteen-year period (1990-2003).
This research will accelerate forest fire detection by developing an energy-efficient, deep neural network model and evaluating its efficacy when used on low-power embedded devices such as the Nvidia Jetson Nano. We will develop our model utilizing the power of TinyML [1], a set of strategies that aims to shift deep learning from traditional high-end systems to low-end embedded computing devices. We will develop and compare the accuracy and power consumption of two different types of models built with TensorFlow and Pytorch trained with an optical image dataset FLAME [2] openly available at the IEEE data portal and NASA website, respectively. The resultant model, with higher accuracy and lower power consumption upon deployment, will be tuned using transfer learning and validated utilizing real datasets available at ACUASI, AFSC, and IARC. We will also develop a set of metrics called Forest Fire Impact Index (F2I2) to quantify the severity of a detected fire to help develop better mitigation strategies, resilience plans, and rescue missions.
PROJECT: Evaluating the Performance of MicroChannel Plate Detectors in the Presence of Plastic
Components
PI: Don Hampton, Research Faculty, UAF
Alaskans know, cherish, and always wonder at the Aurora Borealis, that is ultimately driven by the sun in its roiling bombardment of Earth with photonic, magnetic, and kinetic energy. The aurora is primarily caused by beams of energetic electrons that impinge on Earth’s high-latitude upper atmosphere near 100 km altitude from deep in the geomagnetic tail, a turbulent wake-like structure that forms the downstream portion of Earth’s magnetosphere, the magnetic cavity that presents an electromagnetic obstacle to the impinging solar wind. These electrons cause the visible aurora by collisional excitation of atmospheric atomic and molecular constituents and cause enhanced charged particle density in Earth’s ionosphere. It is the changes in plasma density that are of most interest since they can cause disruptions to radio communications, induce ground currents by magnetic fluctuations, and affect radar operations for military operations. For example, GPS signals can be degraded by auroral ionization. In several cases this disruption would cause a typical GPS receiver to drop the signal and could cause disruption in the navigation solution. Thus, auroral activity has impacts on research, navigation and security in the arctic.
One of the primary methods in understanding the aurora, and its impacts, is satellite missions. The primary tool on these missions has been a class of scientific instruments known as electron spectrometers, which reveal details of the electrons that cause the aurora. At the heart of most of these instruments are MicroChannel Plate (MCP) electron multipliers that detect and amplify individual electrons into millions, producing signals sufficient to be processed and recorded with conventional electronics components. MCPs, however, are subject to degradation and failure during flight, owing to accumulation of molecular contaminants on their sensitive surfaces by outgassing from other instrument components in the vacuum of space. Materials selection and use is therefore crucial in these instruments. The advent of additive manufacturing (AM, aka 3D printing) capabilities, has introduced new classes of materials that require testing with MCPs to ensure that they won’t ruin MCP performance in flight. Therefore, we propose to perform MCP life testing with one such material – a Carbon NanoTube (CNT) infused plastic that has very promising instrument applications future electron spectrometers. We will develop and perform this testing in vacuum facilities at the University of Alaska’s Geophysical Institute in Fairbanks, thereby qualifying this material for use in space flight instruments and growing the technological capacity at UAF to develop and fly 3D-printed MCPbased instruments on future space missions.
PROJECT: Mapping Agricultural Fields in Alaska: Creating a Baseline Dataset to Understand and Monitor the Environmental and Socio-Economic Effects of a Growing Sector
PI: Melissa Ward-Jones, Research Assistant Professor, UAF
Climate change is simultaneously driving positive and negative impacts to crop yields that may spatially alter the geographic distribution of food production globally. In the US, increasing temperatures are projected to decrease yields within the contiguous US while high latitude regions, including Alaska, are predicted to extend those areas with suitable climate to grow globally important crops such as maize.
This geographic shift may already be underway as the number of farms has increased by 30 % in Alaska between 2012 and 2017 while decreasing by 3.2 % in the contiguous US. As agricultural activity continues to increase within Alaska, it is essential that datasets are created to both understand and monitor the potential environmental and socio-economic impacts of this growing industry. Currently, there are no known geospatial datasets of agricultural fields in Alaska.
This project will map farm fields across the state using machine-learning object-based mapping techniques. Mapping will be conducted using 2017 imagery to allow preliminary analysis with the 2017 Alaska Census of Agriculture as well as other environmental geospatial datasets including US Departmental of Agriculture hardiness zones, ecoregions, and permafrost zones. This dataset will provide a baseline to monitor increasing agricultural activities in the state, be utilized in modelling studies, be used to strengthen management capabilities at the landscape and regional scale, while also providing opportunities to increase Alaska’s research capabilities for this emerging sector.