Browsing by Subject "time series"

Now showing 1 - 4 of 4
  • Thumbnail Image
    Item
    The Effect of Annual and Seasonal Variation in Precipitation on Temporal Water Storage Dynamics in Six Headwater Peatland Catchments: Marcell Experimental Forest, Minnesota
    (2023-06) Adams, David
    Using data collected from six headwater peatland catchments at the Marcell Experimental Forest in northern Minnesota, I assessed the relationship between variability in annual precipitation and annual changes in catchment water storage. Three hypotheses are addressed; (1) annual variability in precipitation is a primary driver of catchment storage change, (2) years of below average precipitation drive the relationship between precipitation and catchment water storage change, and (3) winter and fall precipitation variability are significant seasonal drivers of the annual catchment water storage change. The above relationships were analyzed via cross-correlation lag analysis and linear regression analysis of long-term precipitation, peatland water table elevation (WTE), and upland soil moisture (SM) time series, where WTE and SM served to quantify catchment water storage. Results indicate strong correlations between annual water storage change and annual precipitation variability, both in contemporaneous and antecedent years. Concurrent fall precipitation and antecedent winter precipitation were found to have the most influence on a given year’s water storage change. Years in which precipitation fell below the catchment average (dry years) exhibited a moderately significant linear relationship with annual catchment water storage change. Results of the above analysis were used to create a series of multivariate linear regression models, both with and without moving-average (MA) errors; these models were able to explain between approximately 50% and 70% of the variance found in the annual water storage change time series. Boreal peatlands play a vital role in the planet’s carbon cycle; developing a better understanding of the hydrologic function of these environments will likely prove important to future climate management practices.
  • Thumbnail Image
    Item
    Groundwater chemistry data, real-time temperature, elevation, and specific conductance for MN Wells 668848 and 668849 (May 2021 - May 2024)
    (2024-05-07) McDaris, John; Feinberg, Joshua; Wiest, Nicholas; mcda0030@umn.edu; McDaris, John; University of Minnesota Department of Earth and Environmental Sciences
    Full water chemistry analysis data from two wells (668848, 668849) adjacent to Williamson Hall on the East Bank Campus of the University of Minnesota, Twin Cities. Water samples were taken ~quarterly in conjunction with a PhD project led by McDaris. Feinberg and Wiest assisted in sample collection and interpretation. The wells were also instrumented with in situ sensors measuring groundwater temperature, elevation, and specific conductance. Measurements from these sensors were taken multiple times per hour for three years.
  • Thumbnail Image
    Item
    Making national forest inventory data relevant for local forest management
    (2018-07) Wilson, Barry
    The national forest inventory conducted by the United States Forest Service Forest Inventory and Analysis (FIA) program provides information for strategic level decisions regarding national and regional management of forest ecosystem goods and services. However, the sampling intensity typically limits the application of traditional direct estimators to areas the size of a large county, if not larger. This dissertation describes methods for combining FIA data with auxiliary information to enhance its relevance for local forest management. Background information is provided on the way population estimates are currently produced, and how precision can be improved via satellite imagery. A study is described that uses features extracted from dense time series of Landsat imagery with a model-assisted direct estimator. The study examined the relative predictive power of land cover models incorporating extracted spectro-temporal features versus composite imagery alone. Non-parametric models were fitted for multiple attributes measured on FIA plots using all archived Landsat scenes for Minnesota from 2009-2013. The estimated coefficients developed by harmonic regression of the time series imagery were shown to be moderately to highly correlated with tree-level and land cover attributes. When comparing results for spectro-temporal features to monthly image composites, regression models had greater explained variance and classification models had greater overall and individual class accuracies. Finally, a study is presented that tested the performance of a proposed variant of the k-nearest neighbors algorithm for areas too small to use a direct estimator. Spectro-temporal features were extracted for one ecological unit in Minnesota. A simulated population of tree canopy cover was sampled at FIA plot locations. The proposed algorithm was used to fit a non-parametric model to predict tree canopy cover that incorporates the spectro-temporal features. The model was used to construct predictive intervals for spatial domains over a range of domain sizes, and the resultant tests showed the coverage probability approached the theoretical value for areas as small as 1200 hectares. The study suggests that, given good auxiliary data and models, the scale of valid inference using FIA data can approach what is needed for local decision makers.
  • Thumbnail Image
    Item
    Models for Limited Labeled Time Series Data with Applications in Sleep Science
    (2023-04) Aggarwal, Karan
    Time series are encountered universally in any natural or man-made phenomenon. Time-series analysis has applications in critical domains like healthcare, meteorology, and finance. Recently, there has been a big shift in the nature of collected time-series data, with the popularity of cheaper consumer-grade sensors, e.g., smartwatches. This has provided us with a plethora of lower-quality but high-volume data. Modeling time-varying data is challenging owing to its high dimensionality and complex patterns. These challenges are compounded by issues like missing data which have detrimental effects on downstream tasks like classification. Feature engineering has been an important part of time series analysis, with the use of features like seasonality or frequency transforms. Time-series data's complexity makes feature engineering quite challenging, and hence, deep learning is quite promising. Recently, there has been a lot of work on the time-series using deep learning architectures, which requires access to labeled examples. Labeling is an expensive operation,  especially in areas requiring specialized knowledge like healthcare. In this thesis, we focus on utilizing the limited labeled data efficiently. We propose solutions that leverage: 1) unlabeled data; 2) data with missing time-series observations; and 3) effective use of scarce labels. We primarily focus on showcasing these techniques for applications in sleep science, with data from consumer-grade devices like smart watches becoming available. First, we present a method for unsupervised representation learning to create representations for human activity and sleep data. We exploit the context and content, and reduce subject-specific noise using adversarial training. These representations can be exploited to boost the performance of supervised learning models in low-labeled data settings, unlike the traditional time-series models. Empirical evaluation demonstrates that our proposed method performs better than many strong baseline methods, and adversarial learning helps improve the generalizability of our representations. Second, we use conditional random fields (CRFs) with deep neural networks to capture longer-term dependencies in the dynamics of output labels for time series segmentation tasks. This allows us to capture longer-term context while performing the segmentation labeling, allowing for more efficient usage of limited labels. Our method shows significant improvement over the baseline methods. We apply the proposed method for the detection of sleep stages from  Continuous Positive Air Pressure (CPAP) signals, an at-home therapy device for sleep apnea. Ours is the first work to detect a patient's sleep stages based on the CPAP collected data with reasonable accuracy.   Third, we present a novel semi-supervised method for time series data imputation. Observing missing data in time series is common because of issues like data drops or sensor malfunctioning. Imputation methods are used to fill in these values, with the quality of imputation having a significant impact on downstream tasks like classification. Our proposed semi-supervised approach uses unlabeled data as well as downstream task's labeled data. Our results indicate that the proposed method outperforms the existing supervised and unsupervised time series imputation methods measured on the imputation quality as well as on the downstream tasks ingesting imputed time series. Last, we adapt MixUp, a simple data augmentation technique for time series data. We show that a simple modification in the training process can improve the performance of time series classification methods. We perform data augmentation in both raw time series as well as latent space from time series classification models. The improvement in performance is observed consistently in low labeled data regimes as well as higher data regimes.