Browsing by Author "Barnes, Elizabeth, advisor"
Now showing 1 - 6 of 6
Results Per Page
Sort Options
Item Open Access Application of an interpretable prototypical-part network to subseasonal-to-seasonal climate prediction over North America(Colorado State University. Libraries, 2024) Gordillo, Nicolas J., author; Barnes, Elizabeth, advisor; Schumacher, Russ, committee member; Anderson, Chuck, committee memberIn recent years, the use of neural networks for weather and climate prediction has greatly increased. In order to explain the decision-making process of machine learning "black-box" models, most research has focused on the use of machine learning explainability methods (XAI). These methods attempt to explain the decision-making process of the black box networks after they have been trained. An alternative approach is to build neural network architectures that are inherently interpretable. That is, construct networks that can be understood by a human throughout the entire decision-making process, rather than explained post-hoc. Here, we apply such a neural network architecture, named ProtoLNet, in a subseasonal-to-seasonal climate prediction setting. ProtoLNet identifies predictive patterns in the training data that can be used as prototypes to classify the input, while also accounting for the absolute location of the prototype in the input field. In our application, we use data from the Community Earth System Model version 2 (CESM2) pre-industrial long control simulation and train ProtoLNet to identify prototypes in precipitation anomalies over the Indian and North Pacific Oceans to forecast 2-meter temperature anomalies across the western coast of North America on subseasonal-to-seasonal timescales. These identified CESM2 prototypes are then projected onto fifth-generation ECMWF Reanalysis (ERA5) data to predict temperature anomalies in the observations several weeks ahead. We compare the performance of ProtoLNet between using CESM2 and ERA5 data. We then demonstrate a novel approach for performing transfer learning between CESM2 and ERA5 data which allows us to identify skillful prototypes in the observations. We show that the predictions by ProtoLNet using both datasets have skill while also being interpretable, sensible, and useful for drawing conclusions about what the model has learned.Item Open Access Detecting forced change within combined climate fields using explainable neural networks(Colorado State University. Libraries, 2021) Rader, Jamin, author; Barnes, Elizabeth, advisor; Rugenstein, Maria, committee member; Witt, Jessica, committee memberAssessing forced climate change requires the extraction of the forced signal from the background of climate noise. Traditionally, tools for extracting forced climate change signals have focused on one atmospheric variable at a time, however, using multiple variables can reduce noise and allow for easier detection of the forced response. Following previous work, we train artificial neural networks to predict the year of single- and multi-variable maps from forced climate model simulations. To perform this task, the neural networks learn patterns that allow them to discriminate between maps from different yearsāthat is, the neural networks learn the patterns of the forced signal amidst the shroud of internal variability and climate model disagreement. When presented with combined input fields (multiple seasons, variables, or both), the neural networks are able to detect the signal of forced change earlier than when given single fields alone by utilizing complex, nonlinear relationships between multiple variables and seasons. We use layer-wise relevance propagation, a neural network visualization tool, to identify the multivariate patterns learned by the neural networks that serve as reliable indicators of the forced response. These "indicator patterns" vary in time and between climate models, providing a template for investigating inter-model differences in the time evolution of the forced response. This work demonstrates how neural networks and their visualization tools can be harnessed to identify patterns of the forced signal within combined fields.Item Open Access Planning for an unknown future: incorporating meteorological uncertainty into predictions of the impact of fires and dust on US particulate matter(Colorado State University. Libraries, 2019) Brey, Steven, author; Fischer, Emily, advisor; Barnes, Elizabeth, advisor; Pierce, Jeffrey, committee member; Rocca, Monique, committee memberExposure to particulate matter (PM) pollution has well documented health impacts and is regulated by the United States (U.S.) Environmental Protection Agency (EPA). In the U.S. wildfire smoke and wind-blown dust are significant natural sources of PM pollution. This dissertation shows how the environmental conditions that drive wildfires and wind-blown dust are likely to change in the future and what these changes imply for future PM concentrations. The first component of this dissertation shows how human ignitions and environmental conditions influence U.S. wildfire activity. Using wildfire burn area and ignition data, I find that in both the western and southeastern U.S., annual lightning- and human-ignited wildfire burn area have similar relationships with key environmental conditions (temperature, relative humidity, and precipitation). These results suggest that burn area for human- and lightning-ignited wildfires will be similarly impacted by climate change. Next, I quantify how the environmental conditions that drive wildfire activity are likely to change in the future under different climate scenarios. Coupled Model Intercomparison Project phase 5 (CMIP5) models agree that western U.S. temperatures will increase in the 21st century for representative concentration pathways (RCPs) 4.5 and 8.5. I find that averaged over seasonal and regional scales, other environmental variables demonstrated to be relevant to fuel flammability and aridity, such as precipitation, evaporation, relative humidity, root zone soil moisture, and wind speed, can be used to explain historical variability in wildfire burn area as well or better than temperature. My work demonstrates that when objectively selecting environmental predictors using Lasso regression, temperature is not always selected, but that this varies by western U.S. ecoregion. When temperature is not selected, the sign and magnitude of future changes in burn area become less certain, highlighting that predicted changes in burn area are sensitive to the environmental predictors chosen to predict burn area. Smaller increases in future wildfire burn area are estimated whenever and wherever the importance of temperature as a predictor is reduced. The second component of this dissertation examines how environmental conditions that drive fine dust emissions and concentrations in the southwestern U.S. change in the future. I examine environmental conditions that influence dust emissions including, temperature, vapor pressure deficit, relative humidity, precipitation, soil moisture, wind speed, and leaf area index (LAI). My work quantifies fine dust concentrations in the U.S. southwest dust season, March through July, using fine iron as a dust proxy, quantified with measurements from the Interagency Monitoring of PROtected Visual Environments (IMPROVE) network between 1995 and 2015. I show that the largest contribution to the spread in future dust concentration estimates comes from the choice of environmental predictor used to explain observed variability. The spread between different environmental predictor estimates can be larger than the spread between climate scenarios or intermodel spread. Based on linear estimates of how dust concentrations respond to changes in LAI, CMIP5 estimated increases in LAI would result in reduced dust concentrations in the future. However, when I objectively select environmental predictors of dust concentrations using Lasso regression, LAI is not selected in favor of other variables. When using a linear combination of objectively selected environmental variables, I estimate that future southwest dust season mean concentrations will increase by 0.24 Ī¼g mā3 (12%) by the end of the 21st century for RCP 8.5. This estimated increase in fine dust concentration is driven by decreases in relative humidity, precipitation, soil moisture, and buffered by decreased wind speeds.Item Open Access Seasonal sensitivity of the eddy-driven jet to tropospheric heating in an idealized atmospheric general circulation model(Colorado State University. Libraries, 2015) McGraw, Marie C., author; Barnes, Elizabeth, advisor; Birner, Thomas, committee member; Venayagamoorthy, Karan, committee memberA dry dynamical core is used to investigate the seasonal sensitivity of the circulation to two idealized thermal forcingsāa tropical upper tropospheric forcing, and a polar lower tropospheric forcing. The circulation is modified using a set of perpetual simulations to simulate each month of the year, while the thermal forcings are held constant. The circu- lation responses to tropical warming and polar warming are studied separately, and then the response to the simultaneously applied forcings is analyzed. Finally, the seasonality of the internal variability of the circulation is explored as a possible mechanism to explain the seasonality of the responses. The primary results of these experiments are: 1) There is a seasonal sensitivity in the circulation response to both the tropical and polar forcings. 2) The jet position response to each forcing is greatest in the transition seasons, and the jet speed response exhibits a seasonal sensitivity to both forcings although the seasonal sensi- tivities are not the same. 3) The circulation response is nonlinear in the transition seasons, but approximately linear in the summer and winter months. 4) The internal variability of the unforced circulation exhibits a seasonal sensitivity that may partly explain the seasonal sensitivity of the forced response. The seasonality of the internal variability of daily MERRA reanalysis data is compared to that of the model, demonstrating that the broad conclusions drawn from this idealized modeling study may be useful for understanding the jet response to anthropogenic forcing.Item Open Access The impact of tropical intraseasonal variability on subseasonal-to-seasonal predictability(Colorado State University. Libraries, 2021) Hsiao, Wei-Ting, author; Maloney, Eric, advisor; Barnes, Elizabeth, advisor; Mueller, Nathan, committee memberSubseasonal-to-seasonal (S2S) timescales have been identified as a gap in weather forecast skill at 2 weeks to 2 months lead times. This timescale is set by midlatitude synoptic predictability limits, and sits between the typical weather timescale and the longer annual to interannual periods that may have skill due to knowledge of low-frequency phenomena such as El NiƱo-Southern Oscillation (ENSO). Previous studies have shown that tropical intraseasonal variability serves as an important source of S2S predictability in the midlatitudes based on a linear Rossby wave theory. The theory suggests that consistent weather patterns are excited by tropical divergence and associated teleconnections to the extratropics on S2S timescales that influence predictability. However, those physical processes that provide sources of S2S forecast skill have yet to be fully characterized. This thesis examines aspects of tropical intraseasonal variability that are important for S2S prediction, including how tropical intraseasonal variability has changed with warming over the last century and how the misrepresentation of such variability in a weather forecast model leads to errors in midlatitude precipitation S2S forecasts. In the first part of this thesis, three reanalyses datasets (ERA5, MERRA-2, and ERA 20-C) are examined to quantify the amplitude changes in a dominant mode of intraseasonal tropical variability, the Madden-Julian oscillation (MJO), over the last century. MJO-associated precipitation and vertical velocity amplitude are found to exhibit a complex evolution over the observational record, where the precipitation has larger increases than the vertical velocity. A decrease in the ratio of MJO circulation to precipitation anomaly amplitude is detected over the observational period. Tropical weak temperature gradient theory is used to show that this decrease is consistent with the change in tropical dry static stability that has occurred under climate warming. The weakening MJO circulation per unit precipitation over the past century may have modified associated teleconnections and has implications for S2S prediction in the tropics and midlatitudes. In the second part of the thesis, emphasis is placed on understanding S2S precipitation forecast errors for the western United States (U.S.) in an operational weather model. A set of hindcasts during boreal winter, where the tropics are nudged toward reanalysis, is compared to hindcasts without nudging. The western U.S. precipitation forecasts are found to improve with nudging at 3-4 week lead times. Using a multivariate k-means clustering method, hindcasts are grouped by their initial states and one cluster that exhibits an initially strong Aleutian Low is found to provide better forecast improvement. The improvement originates from the poor representation in the non-nudged hindcasts of the destructive interference between (1) the anomalous Aleutian Low and (2) the teleconnection pattern generated by certain phases of the MJO during non-cold ENSO conditions. These results suggest that improving the simulation of tropical intraseasonal precipitation during the early MJO phases under non-cold ENSO may lead to better 3-4 week precipitation forecasts in the western U.S.Item Open Access Using neural networks to learn the forced response of the jet-stream to tropospheric temperature tendencies(Colorado State University. Libraries, 2023) Connolly, Charlotte, author; Barnes, Elizabeth, advisor; Randall, David, committee member; Anderson, Chuck, committee memberTwo distinct features of anthropogenic climate change, warming in the tropical upper troposphere and warming at the Arctic surface, have competing effects on the midlatitude jet-stream's latitudinal position, often referred to as a "tug-of-war". Many previous studies have investigated the strength of the jet response to these thermal forcings, as well as many others, and have shown that the jet response is sensitive to model type, season, initial atmospheric conditions, and the shape and magnitude of the forcing. Here, we explore the potential for training a convolutional neural network (CNN) on internal variability alone to examine possible nonlinear jet responses to a variety of thermal forcings. Our approach thus makes use of the fluctuation-dissipation theorem, which relates the internal variability of a system to its forced response. We train a CNN on data from a long control run of the CESM dry dynamical core, thereby providing it with ample data to learn relationships between the temperature forcing and the jet movement over the coming days. Then, we use the CNN to explore the jet response to a wide range of tropospheric temperature tendencies. Despite being trained on the jet-stream response to internal variability alone, we show that the trained CNN is able to skillfully predict the nonlinear response of the jet-stream to sustained external forcing. The trained CNN provides a quick method for exploring the jet-stream sensitivity to a wide range of tropospheric temperature tendencies and, considering that this method can likely be applied to any model with a long control run, could lend itself useful for early stage experiment design.