Random Forest models to estimate bankfull and low flow channel widths and depths across the conterminous United States.

Channel dimensions (width and depth) at varying flows influence a host of instream ecological processes, as well as habitat and biotic features; they are a major consideration in stream habitat restoration and instream flow assessments. Models of widths and depths are often used to assess climate change vulnerability, develop endangered species recovery plans, and model water quality. However, development and application of such models require specific skillsets and resources. To facilitate acquisition of such estimates, we created a dataset of modeled channel dimensions for perennial stream segments across the conterminous U.S. We used random forest models to predict wetted width, thalweg depth, bankfull width, and bankfull depth from several thousand field measurements of the National Rivers and Streams Assessment. Observed channel widths varied from <5 m to >2000 m and depths varied from <2 m to >125 m. Metrics of watershed area, runoff, slope, land use, and more were used as model predictors. The models had high pseudo R-squared values (0.70 to 0.91) and median absolute errors within ±6% to ±21% of the interquartile range of measured values across ten stream orders. Predicted channel dimensions can be joined to 1.1 million stream segments of the 1:100K resolution National Hydrography Dataset Plus (version 2.1). These predictions, combined with a rapidly growing body of nationally available data, will further enhance our ability to study and protect aquatic resources.

A multiscale landscape approach for prioritizing river and stream protection and restoration actions.

River and stream conservation programs have historically focused on a single spatial scale, for example, a watershed or stream site. Recently, the use of landscape information (e.g., land use and land cover) at multiple spatial scales and over large spatial extents has highlighted the importance of incorporating a landscape perspective into stream protection and restoration activities. Previously, we developed a novel framework that links information about watershed-, catchment-, and reach-scale integrity with stream biological condition using scatterplots and a landscape integrity map. Here we examined an application of this approach for streams in urban and other settings in King County, Washington State, United States, where we related stream macroinvertebrate condition to two indices of landscape integrity, the US Environmental Protection Agency's (USEPA) nationally available Index of Watershed Integrity (IWI) and Index of Catchment Integrity (ICI). We generated a scatterplot of IWI versus ICI for sample sites, where points represented site macroinvertebrate condition from poor to good. The same data were also visualized as a landscape integrity map that displayed catchments of King County according to the level of watershed and catchment integrity (high or low IWI/ICI). Almost three-quarters of poor-condition sites were associated with high-integrity watersheds and catchments (i.e., underperforming sites), which suggested that either one or both national indicators were insufficient for this area, and that sites underperformed because of local-scale factors. In response, we used a catchment-scale indicator related to forest condition (PctForestCat) after examining several GIS-based dispersal indicators from the National Hydrography Dataset and other candidates from the USEPA's StreamCat dataset. We then compared the results of the scatterplots and maps based on the current and original analyses and found that many of the sites previously classified as underperforming now performed as expected, that is, they were poor-condition sites in poor-condition catchments. This analysis demonstrates how results based on a national dataset can be improved by developing an alternative that represents regionally important stressors. The methods used to develop an effective landscape indicator based on StreamCat datasets, and the utility of the multiscale approach, could provide important tools for prioritizing, optimizing, and communicating stream conservation actions.

A landscape assessment and associated dataset of stream confluences for the conterminous U.S.

Stream confluences are important components of fluvial networks. Hydraulic forces meeting at stream confluences often produce changes in streambed morphology and sediment distribution. These changes often increase habitat heterogeneity relative to upstream and downstream locations, which have led some to identify them as biological hotspots. Despite their potential ecological importance, there are relatively few empirical studies documenting ecological patterns upstream and downstream of confluences. We have produced a publicly available dataset of stream confluences and associated watershed attributes for the conterminous USA. The dataset includes 1,085,629 stream confluences and 383 attributes for each confluence organized into 15 dataset tables for both tributary and mainstem upstream catchments and watersheds. Themes in the dataset include hydrology (e.g., stream order), land cover, land cover change, geology (e.g., calcium content of underlying lithosphere), physical condition (e.g., precipitation), measures of ecological integrity, and stressors (e.g., impaired streams). Additionally, we used measures of ecological integrity to assess the condition of the stream confluences. Aside from a generally positive east-to-west gradient in ecological condition, we found that approximately one-third of the confluences had markedly contrasting ecological conditions between mainstem and tributary, catchment and watershed, or both. The dataset should support many, multifaceted studies of stream confluence ecology.

Revising the index of watershed integrity national maps.

Watersheds provide a range of services valued by society, incorporating biotic and abiotic functions within their boundaries. Recently, an operational definition of watershed integrity was applied and indices of watershed integrity (IWI) and catchment integrity (ICI) were developed and mapped for the conterminous United States. However, these indices were originally derived using equally-weighted first-order approximations of relationships between anthropogenic stressors (obtained from the U.S. EPA's StreamCat dataset) and six watershed functions. In addition, the original calculations of the IWI and ICI did not standardize metrics across these differing scales, resulting in IWI and ICI values that are not directly comparable. We provide an example of how to iteratively update the stressor-watershed function relationships using random forest models and a nationwide response metric representative of one of the six watershed functions. Specifically, we focused on the chemical regulation function (CHEM) of IWI and ICI by relating a composite metric of chemical water quality from 1914 samples to land use metrics explicit to CHEM to refine the nature of these relationships (e.g., non-linear versus linear). The rate of nitrogen fertilizer, agricultural land use, and urban land use were found to be the three most important stressors predicting the national water quality response metric. Revision of CHEM values improved the prediction of several regional- to national-scale water quality indicators. In all cases, exponential decay curves replaced the original negative linear relationship for CHEM. Therefore, the original IWI and ICI values are probably over-estimates of the actual integrity of the Nation's watersheds and catchments. With these revisions, we provide updated national maps of IWI and ICI. The methods outlined here can be implemented iteratively as more and better data become available for all six of the watershed functions to elevate the accuracy and applicability of these indices to various land management issues.

Performance of National Maps of Watershed Integrity at Watershed Scales.

Watershed integrity, the capacity of a watershed to support and maintain ecological processes essential to the sustainability of services provided to society, can be influenced by a range of landscape and in-stream factors. Ecological response data from four intensively monitored case study watersheds exhibiting a range of environmental conditions and landscape characteristics across the United States were used to evaluate the performance of a national level Index of Watershed Integrity (IWI) at regional and local watershed scales. Using Pearson's correlation coefficient (r), and Spearman's rank correlation coefficient (rs ), response variables displayed highly significant relationships and were significantly correlated with IWI and ICI (Index of Catchment Integrity) values at all watersheds. Nitrogen concentration and flux-related watershed response metrics exhibited significantly strong negative correlations across case study watersheds, with absolute correlations (|r|) ranging from 0.48 to 0.97 for IWI values, and 0.31 to 0.96 for ICI values. Nitrogen-stable isotope ratios measured in chironomids and periphyton from streams and benthic organic matter from lake sediments also demonstrated strong negative correlations with IWI values, with |r| ranging from 0.47 to 0.92, and 0.35 to 0.89 for correlations with ICI values. This evaluation of the performance of national watershed and catchment integrity metrics and their strong relationship with site level responses provides weight-of-evidence support for their use in state, local and regionally focused applications.

The Lake-Catchment (LakeCat) Dataset: characterizing landscape features for lake basins within the conterminous USA.

Natural and human-related landscape features influence the ecology and water quality of lakes. Summarizing these features in a hydrologically meaningful way is critical to understanding and managing lake ecosystems. Such summaries are often done by delineating watershed boundaries of individual lakes. However, many technical challenges are associated with delineating hundreds or thousands of lake watersheds at broad spatial extents. These challenges can limit the application of analyses and models to new, unsampled locations. We present the Lake-Catchment (LakeCat) Dataset (https://www.epa.gov/national-aquatic-resource-surveys/lakecat) of watershed features for 378,088 lakes within the conterminous USA. We describe the methods we used to: 1) delineate lake catchments, 2) hydrologically connect nested lake catchments, and 3) generate several hundred watershed-level metrics that summarize both natural (e.g., soils, geology, climate, and land cover) and anthropogenic (e.g., urbanization, agriculture, and mines) features. We illustrate how this data set can be used with a random forest model to predict the probability of lake eutrophication by combining LakeCat with data from US Environmental Protection Agency's National Lakes Assessment (NLA). This model correctly predicted the trophic state of 72% of NLA lakes, and we applied the model to predict the probability of eutrophication at 297,071 unsampled lakes across the conterminous USA. The large suite of LakeCat metrics could be used to improve analyses of lakes at broad spatial extents, improve the applicability of analyses to unsampled lakes, and ultimately improve the management of these important ecosystems.

Mapping watershed integrity for the conterminous United States.

Watershed integrity is the capacity of a watershed to support and maintain the full range of ecological processes and functions essential to sustainability. Using information from EPA's StreamCat dataset, we calculated and mapped an Index of Watershed Integrity (IWI) for 2.6 million watersheds in the conterminous US with first-order approximations of relationships between stressors and six watershed functions: hydrologic regulation, regulation of water chemistry, sediment regulation, hydrologic connectivity, temperature regulation, and habitat provision. Results show high integrity in the western US, intermediate integrity in the southern and eastern US, and the lowest integrity in the temperate plains and lower Mississippi Valley. Correlation between the six functional components was high (r = 0.85-0.98). A related Index of Catchment Integrity (ICI) was developed using local drainages of individual stream segments (i.e., excluding upstream information). We evaluated the ability of the IWI and ICI to predict six continuous site-level indicators with regression analyses - three biological indicators and principal components derived from water quality, habitat, and combined water quality and habitat variables - using data from EPA's National Rivers and Streams Assessment. Relationships were highly significant, but the IWI only accounted for 1-12% of the variation in the four biological and habitat variables. The IWI accounted for over 25% of the variation in the water quality and combined principal components nationally, and 32-39% in the Northern and Southern Appalachians. We also used multinomial logistic regression to compare the IWI with the categorical forms of the three biological indicators. Results were consistent: we found positive associations but modest results. We compared how the IWI and ICI predicted the water quality PC relative to agricultural and urban land use. The IWI or ICI are the best predictors of the water quality PC for the CONUS and six of the nine ecoregions, but they only perform marginally better than agriculture in most instances. However, results suggest that agriculture would not be appropriate in all parts of the country, and the index is meant to be responsive to all stressors. The IWI in its present form (available through the StreamCat website; https://www.epa.gov/national-aquatic-resource-surveys/streamcat) could be useful for management efforts at multiple scales, especially when combined with information on site condition. The IWI could be improved by incorporating empirical or literature-derived relationships between functional components and stressors. However, limitations concerning the absence of data for certain stressors should be considered.

Predictive mapping of the biotic condition of conterminous U.S. rivers and streams.

Understanding and mapping the spatial variation in stream biological condition could provide an important tool for conservation, assessment, and restoration of stream ecosystems. The USEPA's 2008-2009 National Rivers and Streams Assessment (NRSA) summarizes the percentage of stream lengths within the conterminous United States that are in good, fair, or poor biological condition based on a multimetric index of benthic invertebrate assemblages. However, condition is usually summarized at regional or national scales, and these assessments do not provide substantial insight into the spatial distribution of conditions at unsampled locations. We used random forests to model and predict the probable condition of several million kilometers of streams across the conterminous United States based on nearby and upstream landscape features, including human-related alterations to watersheds. To do so, we linked NRSA sample sites to the USEPA's StreamCat Dataset; a database of several hundred landscape metrics for all 1:100,000-scale streams and their associated watersheds within the conterminous United States. The StreamCat data provided geospatial indicators of nearby and upstream land use, land cover, climate, and other landscape features for modeling. Nationally, the model correctly predicted the biological condition class of 75% of NRSA sites. Although model evaluations suggested good discrimination among condition classes, we present maps as predicted probabilities of good condition, given upstream and nearby landscape settings. Inversely, the maps can be interpreted as the probability of a stream being in poor condition, given human-related watershed alterations. These predictions are available for download from the USEPA's StreamCat website. Finally, we illustrate how these predictions could be used to prioritize streams for conservation or restoration.

Assessing the accuracy and stability of variable selection methods for random forest modeling in ecology.

Random forest (RF) modeling has emerged as an important statistical learning method in ecology due to its exceptional predictive performance. However, for large and complex ecological data sets, there is limited guidance on variable selection methods for RF modeling. Typically, either a preselected set of predictor variables are used or stepwise procedures are employed which iteratively remove variables according to their importance measures. This paper investigates the application of variable selection methods to RF models for predicting probable biological stream condition. Our motivating data set consists of the good/poor condition of n = 1365 stream survey sites from the 2008/2009 National Rivers and Stream Assessment, and a large set (p = 212) of landscape features from the StreamCat data set as potential predictors. We compare two types of RF models: a full variable set model with all 212 predictors and a reduced variable set model selected using a backward elimination approach. We assess model accuracy using RF's internal out-of-bag estimate, and a cross-validation procedure with validation folds external to the variable selection process. We also assess the stability of the spatial predictions generated by the RF models to changes in the number of predictors and argue that model selection needs to consider both accuracy and stability. The results suggest that RF modeling is robust to the inclusion of many variables of moderate to low importance. We found no substantial improvement in cross-validated accuracy as a result of variable reduction. Moreover, the backward elimination procedure tended to select too few variables and exhibited numerous issues such as upwardly biased out-of-bag accuracy estimates and instabilities in the spatial predictions. We use simulations to further support and generalize results from the analysis of real data. A main purpose of this work is to elucidate issues of model selection bias and instability to ecologists interested in using RF to develop predictive models with large environmental data sets.