An official website of the United States government
Water quality is affected by multiple spatial and temporal factors, including the surrounding land characteristics, human activities, and antecedent precipitation amounts. However, identifying the relationships between water quality and spatially and temporally varying environmental variables with a machine learning technique in a heterogeneous urban landscape has been understudied. We explore how seasonal and variable precipitation amounts and other small-scale landscape variables affect E. coli, total suspended solids (TSS), nitrogen-nitrate, orthophosphate, lead, and zinc concentrations in Portland, Oregon, USA. Mann–Whitney tests were used to detect differences in water quality between seasons and COVID-19 periods. Spearman’s rank correlation analysis was used to identify the relationship between water quality and explanatory variables. A Random Forest (RF) model was used to predict water quality using antecedent precipitation amounts and landscape variables as inputs. The performance of RF was compared with that of ordinary least squares (OLS). Mann–Whitney tests identified statistically significant differences in all pollutant concentrations (except TSS) between the wet and dry seasons. Nitrate was the only pollutant to display statistically significant reductions in median concentrations (from 1.5 mg/L to 1.04 mg/L) during the COVID-19 lockdown period, likely associated with reduced traffic volumes. Spearman’s correlation analysis identified the highest correlation coefficients between one-day precipitation amounts and E. coli, lead, zinc, and TSS concentrations. Road length is positively associated with E. coli and zinc. The Random Forest (RF) model best predicts orthophosphate concentrations (R2 = 0.58), followed by TSS (R2 = 0.54) and nitrate (R2 = 0.46). E. coli was the most difficult to model and had the highest RMSE, MAE, and MAPE values. Overall, the Random Forest model outperformed OLS, as evaluated by RMSE, MAE, MAPE, and R2. The Random Forest was an effective approach to modeling pollutant concentrations using both categorical seasonal and COVID data along with continuous rain and landscape variables to predict water quality in urban streams. Implementing optimization techniques can further improve the model’s performance and allow researchers to use a machine learning approach for water quality modeling.
more »
« less
Meteorological and Vegetation Structure Controls on Liquid Throughfall in US Forests and Shrublands
Abstract Throughfall is the dominant input of water to most terrestrial ecosystems and is primarily driven by precipitation quantity, although the relationship varies among sites. A wide range of meteorological and site‐based properties also influence throughfall and may explain this variability, but their importance for accurately predicting throughfall quantities across differing sites remains unknown. Here I develop models to predict daily throughfall quantities at ∼1 m2resolution based on up to 19 environmental parameters using multi‐year data from sites throughout the US. Three random forest models of varying complexity were trained to predict throughfall: a simple model (RF‐1) driven solely by precipitation quantity, and more complex models that incorporated an additional eight (RF‐9) and eighteen (RF‐19) variables. RF‐1 was able to predict throughfall quantities (±28%) and accuracy was modestly improved by including additional model parameters (±24–26%). Improvements in model performance were most apparent for smaller precipitation events (<10 mm), which are less likely to fully saturate the canopy (22% improvement in prediction accuracy for the RF‐19 model). Precipitation quantity, maximum intensity, and duration were consistently identified as the most important drivers of throughfall, whereas variables relating to evaporative potential and canopy water storage capacity were identified as moderately important. These models allow the impacts of environmental changes (e.g., forest regrowth after clearcutting or increased precipitation intensity) to be evaluated, as well as inform decisions about which parameters to include in field‐ and model‐based studies of throughfall and its converse, interception, when resources are limited.
more »
« less
- Award ID(s):
- 2217817
- PAR ID:
- 10593793
- Publisher / Repository:
- DOI PREFIX: 10.1029
- Date Published:
- Journal Name:
- Journal of Geophysical Research: Biogeosciences
- Volume:
- 130
- Issue:
- 5
- ISSN:
- 2169-8953
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Evaporation of precipitation from plant surfaces, or interception, is a major component of the global water budget. Interception has been measured and/or modelled across a wide variety of forest types; however, most studies have focused on mature, second‐growth forests, and few studies have examined interception processes across forest age classes. We present data on two components of interception, total canopy interception (Ei) and litter interception—that is, Oi + Oehorizon layers—(Eff), across a forest age chronosequence, from 2 years since harvest to old growth. We used precipitation, throughfall, and stemflow collectors to measure total rainfall (P) and estimateEi; and collected litter biomass and modelled litter wetting and drying to estimate evaporative loss from litter. CanopyEi,Pminus throughfall, increased rapidly with forest age and then levelled off to a maximum of 21% ofPin an old‐growth site. Stemflow also varied across stands, with the highest stemflow (~8% ofP) observed in a 12‐year‐old stand with high stem density. ModelledEffwas 4–6% ofPand did not vary across sites. Total stand‐level interception losses (Ei + Eff) were best predicted by stand age (R2 = 0.77) rather than structural parameters such as basal area (R2 = 0.49) or leaf area (R2 < 0.01). Forest age appears to be an important driver of interception losses from forested mountain watersheds even when stand‐level structural variables are similar. These results will contribute to our understanding of water budgets across the broader matrix of forest ages that characterize the modern forest landscape.more » « less
-
Precipitation channelled down tree stems (stemflow) or into drip points of ‘throughfall’ beneath trees results in spatially concentrated inputs of water and chemicals to the ground. Currently, these flows are poorly characterised due to uncertainties about which branches redirect rainfall to stemflow or throughfall drip points.We introduce a graph theoretic algorithm that ‘prunes’ quantitative structural models of trees (derived from terrestrial LiDAR) to identify branches contributing to stemflow and those contributing to throughfall drip points. To demonstrate the method's utility, we analysed two trees with similar canopy sizes but contrasting canopy architecture and rainfall partitioning behaviours.For both trees, the branch ‘watershed’ area contributing to stemflow (under conditions assumed to represent moderate precipitation intensity) was found to be only half of the total ground area covered by the canopy. The study also revealed significant variations between trees in the number and median contribution areas of modelled throughfall drip points (69 vs. 94 drip points tree−1, with contributing projected areas of 28.6 vs. 7.8 m2tree−1, respectively). Branch diameter, surface area, volumes and woody area index of components contributing to stemflow and throughfall drip points may play a role in the trees' differing rainfall partitioning behaviours.Our pruning algorithm, enabled by the proliferation of LiDAR observations of canopy structure, promises to enhance studies of canopy hydrology. It offers a novel approach to refine our understanding of how trees interact with rainfall, thereby broadening the utility of existing LiDAR data in environmental research.more » « less
-
Abstract Background Predictive models utilizing social determinants of health (SDH), demographic data, and local weather data were trained to predict missed imaging appointments (MIA) among breast imaging patients at the Boston Medical Center (BMC). Patients were characterized by many different variables, including social needs, demographics, imaging utilization, appointment features, and weather conditions on the date of the appointment. Methods This HIPAA compliant retrospective cohort study was IRB approved. Informed consent was waived. After data preprocessing steps, the dataset contained 9,970 patients and 36,606 appointments from 1/1/2015 to 12/31/2019. We identified 57 potentially impactful variables used in the initial prediction model and assessed each patient for MIA. We then developed a parsimonious model via recursive feature elimination, which identified the 25 most predictive variables. We utilized linear and non-linear models including support vector machines (SVM), logistic regression (LR), and random forest (RF) to predict MIA and compared their performance. Results The highest-performing full model is the nonlinear RF, achieving the highest Area Under the ROC Curve (AUC) of 76% and average F1 score of 85%. Models limited to the most predictive variables were able to attain AUC and F1 scores comparable to models with all variables included. The variables most predictive of missed appointments included timing, prior appointment history, referral department of origin, and socioeconomic factors such as household income and access to caregiving services. Conclusions Prediction of MIA with the data available is inherently limited by the complex, multifactorial nature of MIA. However, the algorithms presented achieved acceptable performance and demonstrated that socioeconomic factors were useful predictors of MIA. In contrast with non-modifiable demographic factors, we can address SDH to decrease the incidence of MIA.more » « less
-
Basal area is a key measure of forest stocking and an important proxy of forest productivity in the face of climate change. Black walnut ( Juglans nigra ) is one of the most valuable timber species in North America. However, little is known about how the stocking of black walnut would change with differed bioclimatic conditions under climate change. In this study, we projected the current and future basal area of black walnut. We trained different machine learning models using more than 1.4 million tree records from 10,162 Forest Inventory and Analysis (FIA) sample plots and 42 spatially explicit bioclimate and other environmental attributes. We selected random forests (RF) as the final model to estimate the basal area of black walnut under climate change because RF had a higher coefficient of determination ( R 2 ), lower root mean square error (RMSE), and lower mean absolute error (MAE) than the other two models (XGBoost and linear regression). The most important variables to predict basal area were the mean annual temperature and precipitation, potential evapotranspiration, topology, and human footprint. Under two emission scenarios (Representative Concentration Pathway 4.5 and 8.5), the RF model projected that black walnut stocking would increase in the northern part of the current range in the USA by 2080, with a potential shift of species distribution range although uncertainty still exists due to unpredictable events, including extreme abiotic (heat, drought) and biotic (pests, disease) occurrences. Our models can be adapted to other hardwood tree species to predict tree changes in basal area based on future climate scenarios.more » « less
