東華大學圖書館 |

Unraveling Water Quality Issues in the Colorado River Basin: Utilizing Remote Sensing Satellite Images, Statistical, and Machine Learning for Improved Monitoring.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Unraveling Water Quality Issues in the Colorado River Basin: Utilizing Remote Sensing Satellite Images, Statistical, and Machine Learning for Improved Monitoring./
作者:	Adjovu, Godson Ebenezer.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2023,
面頁冊數:	680 p.
附註:	Source: Dissertations Abstracts International, Volume: 85-12, Section: B.
Contained By:	Dissertations Abstracts International85-12B.
標題:	Environmental engineering. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30818033
ISBN:	9798382787756

Unraveling Water Quality Issues in the Colorado River Basin: Utilizing Remote Sensing Satellite Images, Statistical, and Machine Learning for Improved Monitoring.
Adjovu, Godson Ebenezer.

Unraveling Water Quality Issues in the Colorado River Basin: Utilizing Remote Sensing Satellite Images, Statistical, and Machine Learning for Improved Monitoring. - Ann Arbor : ProQuest Dissertations & Theses, 2023 - 680 p.

Source: Dissertations Abstracts International, Volume: 85-12, Section: B.

Thesis (Ph.D.)--University of Nevada, Las Vegas, 2023.

This research was aimed at exploring innovative and cost-effective tools in understanding the spatiotemporal variability of water quality parameters in the Colorado River Basin (CRB), which includes the Colorado River and major reservoirs and lakes in the USA including Lake Mead. The river which arises in the state of Colorado and empties into the Republic of Mexico at the Gulf of California, is a source of water to seven US states and the Republic of Mexico and provides water to about 40 million people and million acres of farmlands in seven states in the western US and the Republic of Mexico. Monitoring of water quality parameters (WQPs) has been traditionally carried out using field and laboratory methods. Although these methods offer standardized and accurate means of measuring these WQPs, they face some limitations including limited coverages, high labor and laboratory costs, and limited spatiotemporal sampling frequencies. This study applied and integrated statistical techniques including descriptive and inferential statistics, remote sensing (RS) images, and machine learning (ML) models categorized as standalone and ensemble models as cost-effective tools to estimate the varying spatiotemporal concentrations of WQPs in Lake Mead and the Colorado River system. RS spectral signatures from Sentinel-2A/B Multi-Spectral Instrument (MSI) and Landsat 8 Operational Land Imager (OLI) images were retrieved from the Google Earth Engine (GEE) repository for the study. This research used field data collected on Lake Mead by the cities of Las Vegas, Henderson, North Las Vegas, and the Clark County Water Reclamation District to ensure compliance with the Nevada Division of Environmental Protection (NDEP) National Pollutant Discharge Elimination System (NPDES) permits. Water quality data on the Colorado River was also obtained from the United States Geological Survey (USGS) data repository. The approach utilized in this study involved conducting a comprehensive review of WQPs monitoring using RS applications as well as an overview of strengths and limitations of the conventional and RS measurement of WQPs of concern namely total dissolved solids (TDS), total suspended solids (TSS) and the implications of these WQPs on the ecosystem. Statistical analyses were performed to assess the spatiotemporal distributions of TDS and TSS along the Colorado River to understand the variabilities and trends of these parameters. Key WQPs including TDS, TSS, dissolved oxygen (DO), and temperature were also studied in Lake Mead to understand the impact of hydrological variables and lake stratification in the lake on these parameters. Additionally, the study employed machine learning (ML) models as an effective tool to estimate TDS in Lake Mead using electrical conductivity (EC) and temperature which are simpler, inexpensive, and takes less time. Results produced from the analysis showed that the eXtreme Gradient Boosting (XGBoost) algorithm was the best-performing ensemble model on the external validation datasets with R2 of 0.81 and RMSE of 34.19 mg/L. Another objective includes the employment of Landsat 8 OLI and Sentinel-2 MSI images for the estimation of the concentration of optically inactive TDS in Lake Mead. The Gradient Boosting Machine (GBM) and XGBoost algorithms were found to be generally showing extremely efficient capabilities in estimating TDS across all three datasets (training, testing, and external validation), respectively for the images from Sentinel-2 and Landsat 8 OLI sensors with evaluations on the external validation derived as 0.99, 9.07 mg/L, and 7.06 mg/L, respectively for R2, RMSE, and MAE for TDS estimations with Sentinel-2 images using GBM and with that of Landsat 8 OLI also yielding 0.95, 21.15 mg/L, and 15.89 mg/L, respectively for R2, RMSE, and MAE using the XGBoost algorithm. The use of GBM with the respective optimal features and hyperparameters described in this study was therefore proposed for the future estimation of TDS in Lake Mead using the Sentinel-2 images. This research also evaluated the potential of using remote sensing and machine learning to retrieve the concentrations of optically active TSS and optically inactive TDS in the Colorado River. The analysis found that ensemble algorithms such as the XGBoost were the optimal estimators for TDS using images from both Sentinel-2 MSI and Landsat 8 OLI with performance on the external validation derived as 0.99, 26.52 mg/L, and 19.19 mg/L, respectively for R2, RMSE, and MAE for Sentinel-2 images. The XGBoost algorithm yielded R2, RMSE, and MAE of 0.97, 35.82 mg/L, and 27.90 mg/L, respectively. The AdaBoost algorithm was found to be the best model for TSS estimations on the external validation 0.92, 29.48 mg/L, and 24.64 mg/L, respectively, for R2, RMSE, and MAE for the Sentinel-2 image. The RF model was found to be the optimal model for TSS estimations with the Landsat 8 OLI with reported R2, RMSE, and MAE of 0.90, 32.80 mg/L, and 22.91 mg/L, respectively. The study therefore proposed the utilization of XGBoost with its hyperparameters and optimal or best features for the estimation of TDS for images from both sensors. The AdaBoost and RF models, together with their hyperparameters and the optimal features, respectively, for the Sentinel-2 MSI and Landsat 8 OLI. The findings from this study provide key stakeholders, water resources managers, and researchers with perspectives into the spatiotemporal dynamics of WQPs in waterbodies in Lake Mead as well as a cost-effective tool for retrieving water quality data in the basin to improve monitoring efforts of the fast-depleting water resources.

ISBN: 9798382787756Subjects--Topical Terms:

548583
Environmental engineering.
Subjects--Index Terms:

Google earth engine

Unraveling Water Quality Issues in the Colorado River Basin: Utilizing Remote Sensing Satellite Images, Statistical, and Machine Learning for Improved Monitoring.
LDR:06959nmm a2200397 4500 001 2399443
005 20240916065443.5
006 m o d
007 cr#unu||||||||
008 251215s2023 ||||||||||||||||| ||eng d
020 $a 9798382787756
035 $a (MiAaPQ)AAI30818033
035 $a AAI30818033
040 $a MiAaPQ $c MiAaPQ
100 1 $a Adjovu, Godson Ebenezer. $3 3769411
245 1 0 $a Unraveling Water Quality Issues in the Colorado River Basin: Utilizing Remote Sensing Satellite Images, Statistical, and Machine Learning for Improved Monitoring.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2023
300 $a 680 p.
500 $a Source: Dissertations Abstracts International, Volume: 85-12, Section: B.
500 $a Advisor: Ahmad, Sajjad;Stephen, Haroon.
502 $a Thesis (Ph.D.)--University of Nevada, Las Vegas, 2023.
520 $a This research was aimed at exploring innovative and cost-effective tools in understanding the spatiotemporal variability of water quality parameters in the Colorado River Basin (CRB), which includes the Colorado River and major reservoirs and lakes in the USA including Lake Mead. The river which arises in the state of Colorado and empties into the Republic of Mexico at the Gulf of California, is a source of water to seven US states and the Republic of Mexico and provides water to about 40 million people and million acres of farmlands in seven states in the western US and the Republic of Mexico. Monitoring of water quality parameters (WQPs) has been traditionally carried out using field and laboratory methods. Although these methods offer standardized and accurate means of measuring these WQPs, they face some limitations including limited coverages, high labor and laboratory costs, and limited spatiotemporal sampling frequencies. This study applied and integrated statistical techniques including descriptive and inferential statistics, remote sensing (RS) images, and machine learning (ML) models categorized as standalone and ensemble models as cost-effective tools to estimate the varying spatiotemporal concentrations of WQPs in Lake Mead and the Colorado River system. RS spectral signatures from Sentinel-2A/B Multi-Spectral Instrument (MSI) and Landsat 8 Operational Land Imager (OLI) images were retrieved from the Google Earth Engine (GEE) repository for the study. This research used field data collected on Lake Mead by the cities of Las Vegas, Henderson, North Las Vegas, and the Clark County Water Reclamation District to ensure compliance with the Nevada Division of Environmental Protection (NDEP) National Pollutant Discharge Elimination System (NPDES) permits. Water quality data on the Colorado River was also obtained from the United States Geological Survey (USGS) data repository. The approach utilized in this study involved conducting a comprehensive review of WQPs monitoring using RS applications as well as an overview of strengths and limitations of the conventional and RS measurement of WQPs of concern namely total dissolved solids (TDS), total suspended solids (TSS) and the implications of these WQPs on the ecosystem. Statistical analyses were performed to assess the spatiotemporal distributions of TDS and TSS along the Colorado River to understand the variabilities and trends of these parameters. Key WQPs including TDS, TSS, dissolved oxygen (DO), and temperature were also studied in Lake Mead to understand the impact of hydrological variables and lake stratification in the lake on these parameters. Additionally, the study employed machine learning (ML) models as an effective tool to estimate TDS in Lake Mead using electrical conductivity (EC) and temperature which are simpler, inexpensive, and takes less time. Results produced from the analysis showed that the eXtreme Gradient Boosting (XGBoost) algorithm was the best-performing ensemble model on the external validation datasets with R2 of 0.81 and RMSE of 34.19 mg/L. Another objective includes the employment of Landsat 8 OLI and Sentinel-2 MSI images for the estimation of the concentration of optically inactive TDS in Lake Mead. The Gradient Boosting Machine (GBM) and XGBoost algorithms were found to be generally showing extremely efficient capabilities in estimating TDS across all three datasets (training, testing, and external validation), respectively for the images from Sentinel-2 and Landsat 8 OLI sensors with evaluations on the external validation derived as 0.99, 9.07 mg/L, and 7.06 mg/L, respectively for R2, RMSE, and MAE for TDS estimations with Sentinel-2 images using GBM and with that of Landsat 8 OLI also yielding 0.95, 21.15 mg/L, and 15.89 mg/L, respectively for R2, RMSE, and MAE using the XGBoost algorithm. The use of GBM with the respective optimal features and hyperparameters described in this study was therefore proposed for the future estimation of TDS in Lake Mead using the Sentinel-2 images. This research also evaluated the potential of using remote sensing and machine learning to retrieve the concentrations of optically active TSS and optically inactive TDS in the Colorado River. The analysis found that ensemble algorithms such as the XGBoost were the optimal estimators for TDS using images from both Sentinel-2 MSI and Landsat 8 OLI with performance on the external validation derived as 0.99, 26.52 mg/L, and 19.19 mg/L, respectively for R2, RMSE, and MAE for Sentinel-2 images. The XGBoost algorithm yielded R2, RMSE, and MAE of 0.97, 35.82 mg/L, and 27.90 mg/L, respectively. The AdaBoost algorithm was found to be the best model for TSS estimations on the external validation 0.92, 29.48 mg/L, and 24.64 mg/L, respectively, for R2, RMSE, and MAE for the Sentinel-2 image. The RF model was found to be the optimal model for TSS estimations with the Landsat 8 OLI with reported R2, RMSE, and MAE of 0.90, 32.80 mg/L, and 22.91 mg/L, respectively. The study therefore proposed the utilization of XGBoost with its hyperparameters and optimal or best features for the estimation of TDS for images from both sensors. The AdaBoost and RF models, together with their hyperparameters and the optimal features, respectively, for the Sentinel-2 MSI and Landsat 8 OLI. The findings from this study provide key stakeholders, water resources managers, and researchers with perspectives into the spatiotemporal dynamics of WQPs in waterbodies in Lake Mead as well as a cost-effective tool for retrieving water quality data in the basin to improve monitoring efforts of the fast-depleting water resources.
590 $a School code: 0506.
650 4 $a Environmental engineering. $3 548583
650 4 $a Remote sensing. $3 535394
653 $a Google earth engine
653 $a Machine learning
653 $a Total dissolved solids
653 $a Total suspended solids
653 $a Water quality parameters
690 $a 0543
690 $a 0775
690 $a 0799
690 $a 0800
710 2 $a University of Nevada, Las Vegas. $b Civil and Environmental Engineering and Construction. $3 2104569
773 0 $t Dissertations Abstracts International $g 85-12B.
790 $a 0506
791 $a Ph.D.
792 $a 2023
793 $a English
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30818033