Advantages of our dataset are that (1) all sites have a record spanning 51 years (1960–2010); (2) it covers a broad geographical area; and (3) it is data rich (51,012 AT values; 48,960 WT values). By leveraging existing information in databases, we found statistically significant increasing AT and WT trends across the mid-Atlantic region for 1960–2010. Because the data are spatially and temporally abundant, our results are statistically robust and provide trend estimates that are unlikely to be confounded by short-term climate cycles. In addition, we were able to compare trend results across the study area because all sites have a 51-year record.
The irregular-interval WT data are not ideal but also are not unique to our study (see Kaushal et al. 2010 on-line materials). We analyzed the limitations of such data by two methods and the results were consistent. The bootstrapping results indicated that the irregular-interval data did not affect the direction of the WT trends (Fig. 3). Analysis of the 51-year synthetic dataset derived from continuous WT data indicated that our trend results were conservative, i.e., for both increasing and decreasing trends, we detected fewer significant trends than were actually present (ESM Fig. 2c). Most importantly, analysis of the synthetic dataset corroborated findings of the bootstrap analysis that the irregular-interval data did not bias the results such that the direction of the detected trend was incorrect.
There was a significant shift in the mean anomaly toward warmer temperatures for the latter period (1986–2010) of 0.54 and 0.39 °C for AT and WT, respectively, relative to the former period (1961–1985). These results are notable because they indicate how quickly both AT and WT have increased in the region (Fig. 2).
Watershed characteristics, sources and volume of water, and anthropogenic influences, such as urbanization, dams, and thermal pollution, can affect WT regimes (e.g., Webb et al. 2008). Although correlation does not imply causation, we examined correlations between trends in AT-WT relations and landscape factors to evaluate whether AT-WT relation trends were synchronous with patterns in likely drivers.
Principal Components Analysis results indicated that major dams were an important factor affecting changes in the AT-WT relation in the region, specifically a decreased influence of AT on WT over time. Metrics representing major dams were inversely correlated with AT-WT relation trends, presumably because many dams release water from the hypolimnion, an effect observed by Kelleher et al. (2012) for streams in Pennsylvania, and because of a smaller surface-area:water-volume ratio (SA:Wvol), which reduces exposure to increasing AT. Further supporting the effect of SA:Wvol is that woody wetlands were directly correlated with WT increasing faster than AT over time, which we ascribe to the typically shallow, expansive waters of woody wetlands that have greater SA:Wvol than do dammed reservoirs.
Principal Components Analysis results also indicated that landscape factors representative of shading were drivers of AT-WT relation trends, as watersheds with greater agricultural land uses (less shading) were directly correlated with WT increasing faster than AT over time. In contrast, watersheds with greater deciduous forest cover were correlated with WT increasing slower than AT over time. The damping effect of major dams, however, was found to outweigh the loss of shading, as watersheds with both agriculture and major dams remained inversely correlated with positive AT-WT relation trends.
Our findings also indicated that WT increased faster than AT over time in larger watersheds, and in watersheds with greater numbers of dams present in the later decades (1970s– 1990s) represented by the GAGESII dataset. The watershed-size correlation is interpreted as an exposure-time effect, similar to the exposure-volume effect of woody wetlands (or, inversely, with major dams). The correlation with more recently constructed dams also may represent the exposure time and/or volume effect. As construction of large reservoirs was less common in recent decades, these metrics may represent smaller impoundments, such as farm ponds and storm-water control basins, which likely are shallower than major impoundments and have different outlet structures (i.e., draining from the epilimnion). Such small impoundments increase exposure time to the air by increasing the time of transport to receiving waters and likely have greater SA:Wvol than large impoundments.
The first component of the PCA on landscape factors represented developed or urbanized land uses. This component was not significantly correlated with AT-WT relation trends, indicating that the AT-WT relation was static in the urbanized streams within our dataset. Both AT and WT, however, may have changed at rates greater than occurred in other land uses in these urbanized areas.
Increased streamflow can limit WT increases or even cause a cooling (e.g., Webb et al. 2003; Mayer 2012); therefore, we assessed the influence of streamflow on the calculated WT trends. Annual mean streamflow in the CBW increased from 1930 to 2010, with a larger increase in streams north of latitude 40.25°N (Rice and Hirsch 2012) as a result of increased precipitation (Karl and Knight 1998). Our log10 streamflow annual trend results, computed from instantaneous streamflow values, showed a larger trend in streams north of latitude 40.25°N (Fig. 4), consistent with results of Rice and Hirsch (2012) for annual mean streamflow. Inverse relations of WT trends with latitude indicate that rising WT was damped at greater latitude. The increase in streamflow has damped, but not stopped or reversed, the warming trend in streams in the northern part of the study area. Our finding that WT increased despite increased streamflow is consistent with analyses of the Hudson River Estuary (Seekell and Pace 2011).
For freshwaters, our results have implications for potential shifts in floral and faunal species distributions. Streams at the upper end of the WT distribution may become unsuitable habitat for certain cool-water fish species (Eaton and Scheller 1996; Isaak et al. 2012). Increasing WT also may make some streams suitable for species not currently present, allowing warm-water species, including invasive species and pathogens, to move into previously cool-water habitats. Streams draining forested watersheds with major dams warmed more slowly than other watersheds and are likely to become even more important as refugia for cool-water species in a warming world.
For Chesapeake Bay, our results have implications for potential shifts in floral and faunal species distributions and for changes in density, stratification, and eutrophication within the bay and its contributing estuaries. As eutrophic conditions have improved little over the past two decades in the Potomac River Estuary flowing into the bay (Bricker et al. 2014), our results are of particular significance with respect to nutrient fluxes to the bay. Rising WT of streams increases soluble reactive phosphorus (SRP) and to a lesser extent nitrate concentrations (Duan and Kaushal 2013). In particular, higher WT enhances reduction of iron and manganese oxides, causing release of SRP from sediments to the stream (Duan and Kaushal 2013). As WT continue to increase, the flux of SRP to the bay likely will increase, further thwarting land-based management measures to reduce eutrophication. Although we found a median rate of temperature increase of streams that contribute to the bay of 0.28 °C per decade, Preston (2004) reports bay water temperature increasing at rates of 0.16 and 0.21 °C per decade at the surface and subsurface, respectively. The difference in rates suggests that bay water-residence times and mixing with ocean water help buffer the bay from the warmer water supplied from its watershed.
Despite the wide variability of the streams with respect to watershed area, channel geometry, aspect, elevation, thermal capacity, the presence or absence of riparian buffers, microclimate conditions, and land cover, on the whole, WT increased from 1960 to 2010. For sites with significantly increased WT, 85 % of the variability could be explained by increased AT, despite increased streamflow at some sites. Our statistically robust results from a large dataset are consistent across the broad region, consistent with what one might expect from a physical basis, and consistent with previous analyses of AT and WT in the region.