Session
2023 poster session
Location
Weber State University
Start Date
5-8-2023 10:00 AM
Description
Remote sensing data (data contained in satellite imagery) are used extensively to monitor waterquality parameters such as clarity, temperature, and chlorophyll-a (chl-a). This is generally done bycollecting in situ data coincident with satellite data collections, then creating empirical water qualitymodels using approaches such as multi-linear regression or step-wise linear regression. Theseapproaches, which require modelers to select model parameters, may not be well-suited for opticallycomplex waters, where interference from suspended solids, dissolved organic matter, or otherconstituents may act as “confusers”. For these waters, it may be useful to include non-standard terms,which might not be considered using traditional methods. Recent machine learning work hasdemonstrated the ability to explore large feature spaces and generate accurate empirical models thatdo not require parameter selection. However, these methods, because of the large number of includedterms, result in models that are not explainable and cannot be analyzed. We explore the use of LeastAbsolute Shrinkage and Select Operator (LASSO), or L1, regularization to fit linear regressionmodels and produce parsimonious models with limited terms to allow interpretation andexplainability. We demonstrate this approach with a case study developing chl-a models for UtahLake, Utah, USA., an optically complex freshwater body, and compare the resulting model’sperformance to the Utah Lake chl-a models in the literature.
LASSO (L1) Regularization for the Development of Sparse Remote Sensing Models of Water Quality
Weber State University
Remote sensing data (data contained in satellite imagery) are used extensively to monitor waterquality parameters such as clarity, temperature, and chlorophyll-a (chl-a). This is generally done bycollecting in situ data coincident with satellite data collections, then creating empirical water qualitymodels using approaches such as multi-linear regression or step-wise linear regression. Theseapproaches, which require modelers to select model parameters, may not be well-suited for opticallycomplex waters, where interference from suspended solids, dissolved organic matter, or otherconstituents may act as “confusers”. For these waters, it may be useful to include non-standard terms,which might not be considered using traditional methods. Recent machine learning work hasdemonstrated the ability to explore large feature spaces and generate accurate empirical models thatdo not require parameter selection. However, these methods, because of the large number of includedterms, result in models that are not explainable and cannot be analyzed. We explore the use of LeastAbsolute Shrinkage and Select Operator (LASSO), or L1, regularization to fit linear regressionmodels and produce parsimonious models with limited terms to allow interpretation andexplainability. We demonstrate this approach with a case study developing chl-a models for UtahLake, Utah, USA., an optically complex freshwater body, and compare the resulting model’sperformance to the Utah Lake chl-a models in the literature.