Comparison of machine learning algorithms to predict dissolved oxygen in an urban stream.

Water quality monitoring for urban watersheds is critical to identify the negative urbanization impacts. This study sought to identify a successful predictive machine learning model with minimal parameters from easy-to-deploy, low-cost sensors to create a monitoring system for the urban stream network, Hunnicutt Creek, in Clemson, SC, USA. A multiple linear regression model was compared to machine learning algorithms k-nearest neighbor, decision tree, random forest, and gradient boosting. These algorithms were evaluated to understand which best predicted dissolved oxygen (DO) from water temperature, conductivity, turbidity, and water level change at four locations along the urban stream. The random forest algorithm had the highest performance in predicting DO for all four sites, with Nash-Sutcliffe model efficiency coefficient (NSE) scores > 0.9 at three sites and > 0.598 at the fourth site. The random forest model was further examined using explainable artificial intelligence (XAI) and found that temperature influenced the DO predictions for three of the four sites, but there were different water quality interactions depending on site location. Calculating the land cover type in each site's sub-watershed revealed that different amounts of impervious surface and vegetation influenced water quality and the resulting DO predictions. Overall, machine learning combined with land cover data helps decision-makers better understand the nuances of urban watersheds and the relationships between urban land cover and water quality.

Evaluation of a new low-cost particle sensor as an internet-of-things device for outdoor air quality monitoring.

Many low-cost particle sensors are available for routine air quality monitoring of PM2.5, but there are concerns about the accuracy and precision of the reported data, particularly in humid conditions. The objectives of this study are to evaluate the Sensirion SPS30 particulate matter (PM) sensor against regulatory methods for measurement of real-time particulate matter concentrations and to evaluate the effectiveness of the Intelligent AirTM sensor pack for remote deployment and monitoring. To achieve this, we co-located the Intelligent AirTM sensor pack, developed at Clemson University and built around the Sensirion SPS30, to collect data from July 29, 2019, to December 12, 2019, at a regulatory site in Columbia, South Carolina. When compared to the Federal Equivalent Methods, the SPS30 showed an average bias adjusted R2 = 0.75, mean bias error of -1.59, and a root mean square error of 2.10 for 24-hour average trimmed measurements over 93 days, and R2 = 0.57, mean bias error of -1.61, and a root mean square error of 3.029, for 1-hr average trimmed measurements over 2300 hours when the central 99% of data was retained with a data completeness of 75% or greater. The Intelligent AirTM sensor pack is designed to promote long-term deployment and includes a solar panel and battery backup, protection from the elements, and the ability to upload data via a cellular network. Overall, we conclude that the SPS30 PM sensor and the Intelligent AirTM sensor pack have the potential for greatly increasing the spatial density of particulate matter measurements, but more work is needed to understand and calibrate sensor measurements.Implications: This work adds to the growing body of research that indicates that low-cost sensors of particulate matter (PM) for air quality monitoring has a promising future, and yet much work is left to be done. This work shows that the level of data processing and filtering effects how the low-cost sensors compare to existing federal reference and equivalence methods: more data filtering at low PM levels worsens the data comparison, while longer time averaging improves the measurement comparisons. Improvements must be made to how we handle, calibrate, and correct PM data from low-cost sensors before the data can be reliably used for air quality monitoring and attainment.

Evaluating the integrity of forested riparian buffers over a large area using LiDAR data and Google Earth Engine.

Spatial and temporal changes in land cover have direct impacts on the hydrological cycle and stream quality. Techniques for accurately and efficiently mapping these changes are evolving quickly, and it is important to evaluate how useful these techniques are to address the environmental impact of land cover on riparian buffer areas. The objectives of this study were to: (1) determine the classes and distribution of land cover in the riparian areas of streams; (2) examine the discrepancies within the existing land cover data from National Land Cover Database (NLCD) using high-resolution imagery of the National Agriculture Imagery Program (NAIP) and a LiDAR canopy height model; and (3) develop a technique using LiDAR data to help characterize riparian buffers over large spatial extents. One-meter canopy height models were constructed in a high-throughput computing environment. The machine learning algorithm Support Vector Machine (SVM) was trained to perform supervised land cover classification at a 1-m resolution on the Google Earth Engine (GEE) platform using NAIP imagery and LiDAR-derived canopy height models. This integrated approach to land cover classification provided a substantial improvement in the resolution and accuracy of classifications with F1 Score of each land cover classification ranging from 64.88 to 95.32%. The resulting 1-m land cover map is a highly detailed representation of land cover in the study area. Forests (evergreen and deciduous) and wetlands are by far the dominant land cover classes in riparian zones of the Lower Savannah River Basin, followed by cultivated crops and pasture/hay. Stress from urbanization in the riparian zones appears to be localized. This study demonstrates a method to create accurate high-resolution riparian buffer maps which can be used to improve water management and provide future prospects for improving buffer zones monitoring to assess stream health.

Monitoring spatial and temporal variation of dissolved oxygen and water temperature in the Savannah River using a sensor network.

Dissolved oxygen is a critical component of river water quality. This study investigated average weekly dissolved oxygen (AWDO) and average weekly water temperature (AWT) in the Savannah River during 2015 and 2016 using data from the Intelligent River® sensor network. Weekly data and seasonal summary statistics revealed distinct seasonal patterns that impact both AWDO and AWT regardless of location along the river. Within seasons, spatial patterns of AWDO and AWT along the river are also evident. Linear mixed effects models indicate that AWT and low and high river flow conditions had a significant impact on AWDO, but added little predictive information to the models. Low and high river flow conditions had a significant impact on AWT, but also added little predictive information to the models. Spatial linear mixed effects models yielded parameter estimates that were effectively the same as non-spatial linear mixed effects models. However, components of variance from spatial linear mixed effects models indicate that 23-32% of the total variance in AWDO and that 12-18% of total variance in AWT can be apportioned to the effect of spatial covariance. These results indicate that location, week, and flow-directional spatial relationships are critically important considerations for investigating relationships between space- and time-varying water quality metrics.

An integrated WebGIS framework for volunteered geographic information and social media in soil and water conservation.

Volunteered geographic information and social networking in a WebGIS has the potential to increase public participation in soil and water conservation, promote environmental awareness and change, and provide timely data that may be otherwise unavailable to policymakers in soil and water conservation management. The objectives of this study were: (1) to develop a framework for combining current technologies, computing advances, data sources, and social media; and (2) develop and test an online web mapping interface. The mapping interface integrates Microsoft Silverlight, Bing Maps, ArcGIS Server, Google Picasa Web Albums Data API, RSS, Google Analytics, and Facebook to create a rich user experience. The website allows the public to upload photos and attributes of their own subdivisions or sites they have identified and explore other submissions. The website was made available to the public in early February 2011 at http://www.AbandonedDevelopments.com and evaluated for its potential long-term success in a pilot study.

Impacts of land disturbance on aquatic ecosystem health: quantifying the cascade of events.

The impacts of land disturbance on streams have been studied extensively, but a quantitative mechanism of stream degradation is still lacking. Small changes in land use result in changes in physical and chemical characteristics in the stream, which significantly alter biotic integrity. The objective of this study was to quantify the mechanisms of aquatic ecosystem degradation in streams impacted by watershed urbanization. By quantifying the development level and the changes in the physical parameters of receiving streams, the effects of land use change can be illustrated in a conceptual model and evaluated using a traditional ecological risk assessment framework. Three 1st-order streams draining catchments undergoing varying stages of land development were examined in the upper Piedmont physiographic province of South Carolina, U.S.A. A disturbance index was developed to quantify the changes in land use on a monthly basis. This normalized disturbance index (NDI) was quantitatively linked to an increase in the percentage of impervious cover, stormwater runoff, storm-event total suspended solid (TSS) concentrations, and the North Carolina biotic index (NCBI). The NDI was inversely related to a decline in habitat, median bed-sediment particle size, and benthic index of biotic integrity (BIBI). Unlike the percentage of impervious cover, the NDI facilitated the development of strategies for multiple scales of regulation. Predictive multivariate regressions were developed for storm-event TSS concentrations, the BIBI, and the NCBI. These regressions can be used to develop improved regulations for the effects of development and can lead to better implementation of best management practices, improved monitoring of land use change, and more sustainable development.

Effects of land use on soil inorganic carbon stocks in the Russian Chernozem.

Little is known about changes in soil inorganic carbon (SIC) stocks with depth and with land use in grassland ecosystems. This study was conducted to determine SIC stocks under different management regimes in the Mollisol, one of the typical soils in grasslands. Four sites were sampled: a native grassland field (not cultivated for at least 300 yr), an adjacent 50-yr continuous fallow field, a yearly cut hay field in the V.V. Alekhin Central-Chernozem Biosphere State Reserve in the Kursk region of Russia, and a continuously cropped field in the Experimental Station of the Kursk Institute of Agronomy and Soil Erosion Control. All sampled soils were classified as fine-silty, mixed, frigid Pachic Hapludolls. Significant differences occurred in SIC stocks between cultivated and grassland soil. The inorganic carbon stocks in the top 2 m were 107 Mg ha(-1) for the native grassland, 91 Mg ha(-1) for the yearly cut hay field, 242 Mg ha(-1) for the continuously cropped field, and 196 Mg ha(-1) for the 50-yr continuous fallow. The SIC was in the form of calcium carbonate and was mostly stored below the 1-m depth. The largest difference between inorganic carbon stocks was observed between the continuously cropped field and native grassland. The increase in inorganic carbon in the continuously cropped field and continuous fallow was attributed to initial cultivation and fertilization. Soil inorganic carbon in Mollisols is not accounted for in the current global carbon estimates.