How do you personally and professionally relate to water and/or space technologies?
Water is the common denominator for the vast majority of the environmental and climatological analysis I have been a part of. From analyzing atmospheric rivers, which are massive corridors of water vapor traversing the atmosphere, to modeling annual river ice breakup in Arctic river systems. Growing up, my family would drive to Cayuga Lake near Ithaca NY, where I am originally from, to go swimming almost every day in the summer. Anyone who is familiar with the Finger Lakes region knows that around every turn there is a creek to explore or a lake or pond to swim in. It was easy to take water for granted growing up in an environment like that, knowing now that many communities in the world have to travel several miles to even access fresh drinking water. I’ll also add that given my career path, I now have a far greater appreciation for the anthropogenic interplay between riparian and littoral zones. For example, thinking back to the Finger Lakes, much of the local economy is centered around farming. Pair that with a hilly topography and many of those water bodies are susceptible to potentially harmful run-off. As for spaceborne technologies, I have used a myriad of products in my work, ECMWF (European Centre for Medium-Range Weather Forecasts) Re-Analysis 5 (ERA5), Global Ecosystem Dynamics Investigation (GEDI), Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) just to name a few. I see them as central to my area of work.
Can you tell us about your current position as a Graduate Research Assistant through The University of Tennessee?
Until recently I was in a program called the Bredesen Center through the University of Tennessee, Knoxville, working towards a PhD in Data Science and Engineering. My education primarily focused on machine learning, statistical analysis and computer science, with supplemental courses in geospatial analysis. I successfully defended my dissertation titled “Deep Learning-Based Time Series Methods for Predictive Understanding of River Ice Breakup in North American Cold Regions” on April 1, 2025. The next step in my career will focus on analyzing and modeling environmental disturbance but in the context of grid security and resilience.
Your dissertation revolves around Arctic river ice breakup. What will the Arctic look like in the future? What related challenges will we need to address and how can space technologies and applications help?
My work aims to accurately model annual river ice breakup timing across rivers throughout Alaska USA and Western Canada. While the work is still ongoing, we can safely make a couple key assumptions about Arctic and near Arctic river ice regimes in the future. The consensus from the Coupled Model Intercomparison Project Phase 6 (CMIP6) shows that temperatures are likely to rise globally on average, with high latitude regions showing disproportionately greater increases in warming this century. This generally points to an earlier breakup date for river ice which is consistent with the historical record. However, while we can generally deduce an earlier breakup, preliminary results also indicate that there may be an increase in variability in breakup timing in the future. There are likely many reasons for this but one of them is the anticipated change in both magnitude and frequency of moisture transport. One analysis we published in Geophysical Research Letters last year shows a correlation between increased precipitation during the coldest period of the year and a delay in river ice breakup timing. All the while, research that I am a part of through a collaborative effort – as well as past work across the community – has demonstrated that storm systems, such as atmospheric rivers, are capable of trapping heat over cold regions of both the Arctic and Antarctic. Therefore, modeling such phenomena is incredibly challenging given the complexity of these systems. In the coming years spaceborne technologies will be all the more important to help us parameterize models that can predict regime shifts. Missions like GRACE and ICESat-2 will play a crucial role in providing data to make informed predictions pertaining to this topic.
River ice monitoring is quite a complex task, can you explain why, and how it is done?
River ice monitoring is very important to the communities of colder regions that rely on river systems. Arctic rivers are used for transportation and providing supplies to otherwise relatively isolated communities. In the winter months when the rivers are frozen, trucks, snowmobiles, All-Terrain Vehicles (ATVs), etc., are driven along the surface of the ice while during the warmer months, boats and river barges are used to carry people and supplies. In the US and Canada, data on river ice is maintained by organizations like the Alaska-Pacific River Forecast Center and the Canadian River Ice Database, respectively. Many of the in-situ river ice measurements are collected by citizen science campaigns such as the Fresh Eyes on Ice project hosted by the University of Alaska Fairbanks. I have it pretty easy, since my work only requires data from the annual freeze and breakup records. However, other researchers must go to great lengths to collect images and video taken on the ground to analyze river ice regimes. Other important measurements include ice thickness and the temperature and flow rate of the water underneath the river ice surface. These measurements are often ascertained by drilling holes along the ice surface with an auger. These variables help officials determine whether it is safe to operate vehicles along the surface of the frozen rivers or not and they provide insight as to the structural integrity of the ice.
How will local communities in Alaska be affected as impacts on river ice and the surrounding environments increase?
From the perspective of my research, the safe travel time afforded to local communities on frozen river surfaces is likely to decrease in the future. In my opinion, the most significant impact of increased temperature on these communities will be seen in the species these communities depend on or contend with. Changes in river temperature have been shown to alter migratory patterns in salmon which many individuals rely on heavily. This is likely true for many other species of fish as well. This in turn will impact other animals on the food chain that have been traditionally hunted by local communities for thousands of years. In addition, there will likely be an increase in mosquitoes as the breeding and survival season will lengthen. This can lead to an increase in disease spread by these and similar pests. My research shows that we can expect changes in river ice breakup patterns but changes in other landscape structures are also likely. This will be seen with permafrost thaw which increases the probability of erosion particularly along coastlines. There is also a process called thermokarst which is characterized by the thawing of ice-rich permafrost near the surface of the ground leading to the formation of sinkholes. This could damage buildings, roads and other structures across communities, and alter the landscape that people so vitally depend on. Unfortunately, these are just some of the impacts the people of this region are expected to face.
You used meteorological data from sources like Daymet and ERA5 reanalysis for your research. Do you have tips for young professionals on finding reliable and appropriate data for their research?
There are many excellent data products available that are completely open source. If you are working with data from North America from 1980 to present and you only need basic meteorological inputs, I highly recommend using Daymet. Daymet is derived through extrapolation over space, using a wide array of in-situ meteorological instruments. ERA5, along with many other products, can be easily accessed via the European Centre for Medium-Range Weather Forecasts (ECMWF) database. There you can find a much wider range of variables covering the entire globe and over a broader expanse of time. I have two primary pieces of advice on data collection:
- Understand your data thoroughly. I have regretted not taking my own advice on this point several times. More often than you might think there are a number of sources describing the caveats of a data product you are attempting to use or for a similar product.
- Try to avoid GUIs. Sometimes that’s just the way the data is presented in which case - have at it. But whenever possible you should try using tools in the command line (wget, curl, etc.) or APIs in a shell script or your favorite versatile language (such as Python or R) to retrieve data. This will allow you to streamline the flow of collecting and processing your data. Generally speaking, the lower level the tool you are using the more control you have over manipulating your data. In the context of remote sensing, more control is a good thing.
How do your studies in Data Analytics influence your research?
The way I see it, there are two ways to solve an analytical problem, deterministically or statistically. The two are not mutually exclusive and represent a wide breadth of possibilities. Data analytics is pretty broad, but it typically is set in the context of statistical methodologies. When you’re trying to solve a problem that has a great deal of underlying complexity and or insufficient data leading to many assumptions, a background in statistical analysis is very helpful. This is certainly the case in most environmental and Earth system analyses.
This background is helpful when approaching machine learning tasks as well because it allows me to leverage statistical thinking to further inform the model. So, for example, in modeling river ice breakup, I was able to derive at what values the model parameters of the neural network should start training from, given the context of the problem. This was done by biasing the initialization through informing the activation functions within the network, so that the model recognized that it should anticipate only one day out of the year to be the breakup date. This adjustment was essential to the success of the model. Knowing how to derive adjustments to common methodologies such as this is where a foundational understanding of probability and statistics is extremely helpful.
Can you explain the Long Short Term Memory Model (LSTM) in more detail?
A Long Short Term Memory Model (LSTM) is a type of neural network designed for sequential pattern recognition. In actuality, an LSTM is a certain type of neuron in the neural network but people often refer to the entire model as an LSTM. The LSTM sees a sequence of inputs that is referred to as a lookback window. In the context of modeling river ice breakup, that sequence is ordered over time in days and contains meteorological variables such as daily minimum and maximum temperature, precipitation, etc. This means that each input to the LSTM is a matrix with the dimensions: batch size - the number of observations considered at each iteration; lookback window - how far back in time relative to the day being assessed we are looking; and the meteorological features. Those input matrices are then sent to each LSTM neuron whereby the same input is seen by each neuron but due to a process called backpropagation, neurons only investigate certain information. This is actually what causes the black-box effect of neural networks as our understanding of this partitioning is unknown. Within each LSTM neuron there is a series of gates, the input, output and forget gates. These weigh how much information from that sequence should be retained and sent to the next layer and where along the lookback to prioritize. The result is then flattened as the new vector no longer needs information about the lookback and is sent to a layer of regular neurons. These are much simpler as they don’t have gates and instead only one vector of weights and one of biases. The result of that layer is output through a single neuron being compared to the actual result of that day. The actual or ground truth, is 0 for a non-breakup day and 1 for a breakup. The output of the final layer of the LSTM is compared in a loss function, in this case binary cross entropy, to inform the previous weights and biases during training. The final output of the model is kept as a probability to mimic a likelihood function showing where the LSTM thinks the most probable breakup event is over the year. Using maximum likelihood estimation, the peak in likelihood is selected and then compared on test data to the actual breakup date by mapping it to a time axis. This is how the model is able to predict when river ice breakup will take place over time using only meteorological inputs. Sounds complicated but it is actually fairly simple!
How do you use Python in your research? How can Python be useful in space applications?
I use Python in my research to analyze large-scale Earth system and hydrological data, focusing on river ice breakup, atmospheric river events and other forms of environmental anomaly detection. Python’s scientific libraries like Xarray, pandas, and NumPy allow me to efficiently process spatiotemporal datasets, while APIs like Pytorch, TensorFlow, SciPy and scikit-learn support the development of machine learning models and statistical analyses for event prediction. I also use mpi4py heavily for distributing calculationson HPC clusters, and tools like matplotlib and Cartopy for visualizing patterns and trends. In space applications, Python is valuable for processing satellite data - integrating with tools like Google Earth Engine and many other request utilities. I regularly leverage Python for modeling Earth system dynamics, thanks to its flexibility, scalability, and rich ecosystem. The versatility of Python allows researchers to import, process and analyze their data, all in one framework. It has been an invaluable tool for my work.
You investigated how seagrass communities in Mosquito Lagoon, Florida, can be identified using a deep neural network and Landsat aerial imagery. What are your main findings?
That was a publication I co-authored; the corresponding author is Stephanie Insalaco, Assistant Professor at Southwestern University. My contribution was writing the machine learning code for the analysis where we found that a deep neural network can effectively identify and map seagrass communities in Mosquito Lagoon Florida using Landsat imagery from 2000 to 2020. This revealed a dramatic and ongoing decline in seagrass coverage most notably after the 2011–2013 super algal bloom, with near-total loss by 2020. The neural network outperformed many previous classification and regression tree-based methods and offers an automated, scalable tool for ecosystem monitoring. It was, however, limited by training data and spectral resolution, so future work along other coastlines would be insightful. These findings highlight both the utility of deep learning for habitat monitoring and the urgent need for conservation of ecosystems at high risk. Studies like this can inform and improve local practices that directly affect ecosystem health.
What do you need to innovate?
To be a great innovator, I think you need to be curious. If you don’t have genuine curiosity for something, it is going to be very difficult to pursue innovation. Along with that same vein, curiosity allows individuals to be thorough. If you’re really curious to know how something works, then you’ll be willing – even eager – to dissect it down to its finer components. Having that foundational understanding is where the interesting part happens because that’s what provides the raw material to build or rebuild something new.
What is your favourite aggregate state of water?
My favorite aggregate state of water is snow. Snow is unique because it carries so much meaning scientifically, culturally, and ecologically. Each snowflake forms through a delicate combination of temperature and humidity, resulting in microscopic crystalline structures that are mathematically complex. Snow acts as a natural insulator, regulating soil temperatures, and its high albedo reflects solar radiation, helping to stabilize regional climates. There's something remarkable about snow’s ability to soften a landscape, preserve stories in the layers and disappear silently with time, feeding rivers and reservoirs as it melts. It’s a state of water that embodies both stillness and transition, and…it tastes great with maple syrup ;)
