Developing an Effective Model for Predicting Spatially and Temporally Continuous Stream Temperatures from Remotely Sensed Land Surface Temperatures

Testing a Generalizable Machine Learning Workflow for Aquatic Invasive Species on Rainbow Trout (Oncorhynchus mykiss) in Northwest Montana

Biological invasions are accelerating worldwide, causing major ecological and economic impacts in aquatic ecosystems. The urgent decision-making needs of invasive species managers can be better met by the integration of biodiversity big data with large-domain models and data-driven products. Remotely sensed data products can be combined with existing invasive species occurrence data via machine learning models to provide the proactive spatial risk analysis necessary for implementing coordinated and agile management paradigms across large scales. We present a workflow that generates rapid spatial risk assessments on aquatic invasive species using occurrence data, spatially explicit environmental data, and an ensemble approach to species distribution modeling using five machine learning algorithms. For proof of concept and validation, we tested this workflow using extensive spatial and temporal hybridization and occurrence data from a well-studied, ongoing, and climate-driven species invasion in the upper Flathead River system in northwestern Montana, USA. Rainbow Trout (RBT; Oncorhynchus mykiss), an introduced species in the Flathead River basin, compete and readily hybridize with native Westslope Cutthroat Trout (WCT; O. clarkii lewisii), and the spread of RBT individuals and their alleles has been tracked for decades. We used remotely sensed and other geospatial data as key environmental predictors for projecting resultant habitat suitability to geographic space. The ensemble modeling technique yielded high accuracy predictions relative to 30-fold cross-validated datasets (87% 30-fold cross-validated accuracy score). Both top predictors and model performance relative to these predictors matched current understanding of the drivers of RBT invasion and habitat suitability, indicating that temperature is a major factor influencing the spread of invasive RBT and hybridization with native WCT. The congruence between more time-consuming modeling approaches and our rapid machine-learning approach suggest that this workflow could be applied more broadly to provide data-driven management information for early detection of potential invaders.

The importance of warm habitat to the growth regime of cold-water fishes

A common goal of biological adaptation planning is to identify and prioritize locations that remain suitably cool during summer. This implicitly devalues areas that are ephemerally warm, even if they are suitable most of the year for mobile animals. Here we develop an alternative conceptual framework, the growth regime, which considers seasonal and landscape variation in physiological performance, focusing on riverine fish. Using temperature models for 14 river basins, we show that growth opportunities propagate up and down river networks on a seasonal basis, and that downstream habitats that are suboptimally warm in summer may actually provide the majority of growth potential expressed annually. We demonstrate with an agent-based simulation that shoulder-season use of warmer downstream habitats can fuel annual fish production. Our work reveals a synergy between cold and warm habitats that could be fundamental for supporting coldwater fisheries, highlighting the risk in conservation strategies that underappreciate warm habitats.

Land-Cover and Climatic Controls on Water Temperature, Flow Permanence, and Fragmentation of Great Basin Stream Networks

The seasonal and inter-annual variability of flow presence and water temperature within headwater streams of the Great Basin of the western United States limit the occurrence and distribution of coldwater fish and other aquatic species. To evaluate changes in flow presence and water temperature during seasonal dry periods, we developed spatial stream network (SSN) models from remotely sensed land-cover and climatic data that account for autocovariance within stream networks to predict the May to August flow presence and water temperature between 2015 and 2017 in two arid watersheds within the Great Basin: Willow and Whitehorse Creeks in southeastern Oregon and Willow and Rock Creeks in northern Nevada. The inclusion of spatial autocovariance structures improved the predictive performance of the May water temperature model when the stream networks were most connected, but only marginally improved the August water temperature model when the stream networks were most fragmented. As stream network fragmentation increased from the spring to the summer, the SSN models revealed a shift in the scale of processes affecting flow presence and water temperature from watershed-scale processes like snowmelt during high-runoff seasons to local processes like groundwater discharge during sustained seasonal dry periods.

Accurate spatiotemporal predictions of daily stream temperature from statistical models accounting for interactions between climate and landscape

Spatial and temporal patterns in stream temperature are primary factors determining species composition, diversity and productivity in stream ecosystems. The availability of spatially and temporally continuous estimates of stream temperature would improve the ability of biologists to fully explore the effects of stream temperature on biota. Most statistical stream temperature modeling techniques are limited in their ability to account for the influence of variables changing across spatial and temporal gradients. We identified and described important interactions between climate and spatial variables that approximate mechanistic controls on spatiotemporal patterns in stream temperature. With identified relationships we formed models to generate reach-scale basin-wide spatially and temporally continuous predictions of daily mean stream temperature in four Columbia River tributaries watersheds of the Pacific Northwest, USA. Models were validated with a testing dataset composed of completely distinct sites and measurements from different years. While some patterns in residuals remained, testing dataset predictions of selected models demonstrated high accuracy and precision (averaged RMSE for each watershed ranged from 0.85–1.54 °C) and was only 17% higher on average than training dataset prediction error. Aggregating daily predictions to monthly predictions of mean stream temperature reduced prediction error by an average of 23%. The accuracy of predictions was largely consistent across diverse climate years, demonstrating the ability of the models to capture the influences of interannual climatic variability and extend predictions to timeframes with limited temperature logger data. Results suggest that the inclusion of a range of interactions between spatial and climatic variables can approximate dynamic mechanistic controls on stream temperatures.

What are the Conditions of Riparian Ecosystems? Identifying Impaired Floodplain Ecosystems across the Western U.S. Using the Riparian Condition Assessment (RCA) Tool

Environmental stressors associated with human land and water-use activities have degraded many riparian ecosystems across the western United States. These stressors include (i) the widespread expansion of invasive plant species that displace native vegetation and exacerbate streamflow and sediment regime alteration; (ii) agricultural and urban development in valley bottoms that decouple streams and rivers from their floodplains and reduce instream wood recruitment and retention; and (iii) flow modification that reduces water quantity and quality, degrading aquatic habitats. Here we apply a novel drainage network model to assess the impacts of multiple stressors on reach-scale riparian condition across two large U.S. regions. In this application, we performed a riparian condition assessment evaluating three dominant stressors: (1) riparian vegetation departure from historical condition; (2) land-use intensity within valley bottoms; and (3) floodplain fragmentation caused by infrastructure within valley bottoms, combining these stressors in a fuzzy inference system. We used freely available, geospatial data to estimate reach-scale (500 m) riparian condition for 52,800 km of perennial streams and rivers, 25,600 km in Utah, and 27,200 km in 12 watersheds of the interior Columbia River Basin (CRB). Model outputs showed that riparian condition has been at least moderately impaired across ≈70% of the streams and rivers in Utah and ≈49% in the CRB. We found 84% agreement (Cohen's ĸ = 0.79) between modeled reaches and field plots, indicating that modeled riparian condition reasonably approximates on-the-ground conditions. Our approach to assessing riparian condition can be used to prioritize watershed-scale floodplain conservation and restoration by providing network-scale data on the extent and severity of riparian degradation. The approach that we applied here is flexible and can be expanded to run with additional riparian stressor data and/or finer resolution input data.

A spatio-temporal statistical model of maximum daily river temperatures to inform the management of Scotland's Atlantic salmon rivers under climate change

The thermal suitability of riverine habitats for cold water adapted species may be reduced under climate change. Riparian tree planting is a practical climate change mitigation measure, but it is often unclear where to focus effort for maximum benefit. Recent developments in data collection, monitoring and statistical methods have facilitated the development of increasingly sophisticated river temperature models capable of predicting spatial variability at large scales appropriate to management. In parallel, improvements in temporal river temperature models have increased the accuracy of temperature predictions at individual sites. This study developed a novel large scale spatio-temporal model of maximum daily river temperature (Tw_max) for Scotland that predicts variability in both river temperature and climate sensitivity. Tw_max was modelled as a linear function of maximum daily air temperature (Ta_max), with the slope and intercept allowed to vary as a smooth function of day of the year (DoY) and further modified by landscape covariates including elevation, channel orientation and riparian woodland. Spatial correlation in Tw_max was modelled at two scales; (1) river network (2) regional. Temporal correlation was addressed through an autoregressive (AR1) error structure for observations within sites. Additional site level variability was modelled with random effects. The resulting model was used to map (1) spatial variability in predicted Tw_max under current (but extreme) climate conditions (2) the sensitivity of rivers to climate variability and (3) the effects of riparian tree planting. These visualisations provide innovative tools for informing fisheries and land-use management under current and future climate.

Can brook trout survive climate change in large rivers? If it rains

We provide an assessment of thermal characteristics and climate change vulnerability for brook trout (Salvelinus fontinalis) habitats in the upper Shavers Fork sub-watershed, West Virginia. Spatial and temporal (2001-2015) variability in observed summer (6/1-8/31) stream temperatures was quantified in 23 (9 tributary, 14 main-stem) reaches. We developed a mixed effects model to predict site-specific mean daily stream temperature from air temperature and discharge and coupled this model with a hydrologic model to predict future (2016-2100) changes in stream temperature under low (RCP 4.5) and high (RCP 8.5) emissions scenarios. Observed mean daily stream temperature exceeded the 21°C brook trout physiological threshold in all but one main-stem site, and 3 sites exceeded proposed thermal limits for either 63- and 7-day mean stream temperature. We modeled mean daily stream temperature with a high degree of certainty (R²=0.93; RMSE=0.76°C). Predicted increases in mean daily stream temperature in main-stem and tributary reaches ranged from 0.2°C (RCP 4.5) to 1.2°C (RCP 8.5). Between 2091 and 2100, the average number of days with mean daily stream temperature>21°C increased within main-stem sites under the RCP 4.5 (0-1.2days) and 8.5 (0-13) scenarios; however, no site is expected to exceed 63- or 7-day thermal limits. During the warmest 10years, ≥5 main-stem sites exceeded the 63- or 7-day thermal tolerance limits under both climate emissions scenarios. Years with the greatest increases in stream temperature were characterized by low mean daily discharge. Main-stem reaches below major tributaries never exceed thermal limits, despite neighboring reaches having among the highest observed and predicted stream temperatures. Persistence of thermal refugia within upper Shavers Fork would enable persistence of metapopulation structure and life history processes. However, this will only be possible if projected increases in discharge are realized and offset expected increases in air temperature.

Big biology meets microclimatology: defining thermal niches of ectotherms at landscape scales for conservation planning

Temperature profoundly affects ecology, a fact ever more evident as the ability to measure thermal environments increases and global changes alter these environments. The spatial structure of thermalscapes is especially relevant to the distribution and abundance of ectothermic organisms, but the ability to describe biothermal relationships at extents and grains relevant to conservation planning has been limited by small or sparse data sets. Here, we combine a large occurrence database of >23 000 aquatic species surveys with stream microclimate scenarios supported by an equally large temperature database for a 149 000-km mountain stream network to describe thermal relationships for 14 fish and amphibian species. Species occurrence probabilities peaked across a wide range of temperatures (7.0-18.8°C) but distinct warm- or cold-edge distribution boundaries were apparent for all species and represented environments where populations may be most sensitive to thermal changes. Warm-edge boundary temperatures for a native species of conservation concern were used with geospatial data sets and a habitat occupancy model to highlight subsets of the network where conservation measures could benefit local populations by maintaining cool temperatures. Linking that strategic approach to local estimates of habitat impairment remains a key challenge but is also an opportunity to build relationships and develop synergies between the research, management, and regulatory communities. As with any data mining or species distribution modeling exercise, care is required in analysis and interpretation of results, but the use of large biological data sets with accurate microclimate scenarios can provide valuable information about the thermal ecology of many ectotherms and a spatially explicit way of guiding conservation investments.