Heterogeneity in post-fire thermal responses across Pacific Northwest streams: A multi-site study

Abstract

Over the past century, water temperatures in many streams across the Pacific Northwest (PNW) have steadily risen, shrinking endangered salmonid habitats. The warming of PNW stream reaches can be further accelerated by wildfires burning forest stands that provide shade to streams. However, previous research on the effect of wildfires on stream water temperatures has focused on individual streams or burn events, limiting our understanding of the diversity in post-fire thermal responses across PNW streams. To bridge this knowledge gap, we assessed the impact of wildfires on daily summer water temperatures across 31 PNW stream sites, where 10–100% of their riparian area burned. To ensure robustness of our results, we employed multiple approaches to characterize and quantify fire effects on post-fire stream water temperature changes.

Averaged across the 31 burned sites, wildfires corresponded to a 0.3 – 1°C increase in daily summer water temperatures over the subsequent three years. Nonetheless, post-fire summer thermal responses displayed extensive heterogeneity across burned sites where the likelihood and rate of a post-fire summer water temperature warming was higher for stream sites with greater proportion of their riparian area burned under high severity. Also, watershed features such as basin area, post-fire weather, bedrock permeability, pre-fire riparian forest cover, and winter snowpack depth were identified as strong predictors of the post-fire summer water temperature responses across burned sites. Our study offers a multi-site perspective on the effect of wildfires on summer stream temperatures in the PNW, providing insights that can inform freshwater management efforts beyond individual streams and basins.

1. Introduction

Over the past century, the water temperature of many streams in the Pacific Northwest (PNW) have steadily risen (Kaushal et al., 2010; Isaak et al., 2012) although the warming trend across streams can be highly variable (Arismendi et al., 2012). This warming poses a threat to thermally sensitive cold water aquatic species as higher water temperatures can lead to increased metabolic rates, elevated bioenergetic costs, a higher pervasiveness of diseases, and alterations in predator–prey interaction (McCullough et al., 2009; Beakes et al., 2014; Rosenberger et al., 2015). Additionally, the prevalence of unsuitably warm waters is already limiting the spatial distribution of economically, culturally, and ecologically important cold-water fishes like trout and salmon in the PNW, and further warming is expected to exacerbate habitat loss (Isaak et al., 2012; Steel et al., 2019). Over the past half century, large, high-severity wildfires have become more pervasive in the western US (Westerling, 2016; Reilly et al., 2017), increasing concerns about their impact on stream water temperatures in the PNW (Isaak et al., 2010).

Stream temperature is an integrated response to the thermal energy and moisture exchange occurring locally as well as in the upstream watershed. Thus, by altering terrestrial landscapes, wildfires impact stream temperature via direct riparian influences such as shading and bank stability, as well as indirectly by changing water routing and stream hydrology. The burning of riparian canopies that provide stream shade is likely to raise the amount of sunlight reaching the stream surface, increasing local daytime stream water temperatures (Leach and Moore, 2010; Beakes et al., 2014; Swartz and Warren, 2022). The stream heating effect of a post-fire increase in insolation can be further amplified by the combustion of tree roots that help stabilize stream banks, leading to a wider and shallower flow (Dunham et al., 2007). However, any stream warming effect due to higher post-fire incident solar radiation can be offset by possible increases in sub-surface flow volume (Rey et al., 2023) due to the scorching of riparian vegetation and concomitant decreases in plant-related evapotranspiration (Bart and Tague, 2017; Collar et al., 2022). Moreover, wildfire-related losses of riparian canopies often lead to higher nighttime stream heat losses (to the atmosphere) along affected reaches, stemming from elevated net long wave radiation and turbulent heat fluxes (Leach and Moore, 2010). Finally, post-fire changes in vegetation cover and soil hydraulic properties promote higher upstream surface runoff and stream water volumes (Moody et al., 2013), reducing the sensitivity of local stream temperatures to atmospheric heat input (Kelleher et al., 2012; Leach et al., 2023). The relative contributions of advective, radiative, and other heat fluxes (and their sources) on the stream heat budget and ultimately water temperature vary across time and reach (Leach et al., 2023). Consequently, wildfires can elicit diverse and even contrasting stream thermal responses across streams (Rey et al., 2023), seasons (Beyene et al., 2022), and water temperature metrics (Koontz et al., 2018).

Post-fire warming of summer water temperatures can create energetically stressful conditions for native coldwater fishes (Isaak et al., 2010). Consequently, there have been studies on the effects of wildfires on summer stream water temperatures across the western North America. Reported post-fire summer water temperature responses range from negligible water temperature changes (Mahlum et al., 2011), to modest water temperature warmings of less than 1°C (Beakes et al., 2014; Beyene et al., 2022), to major water temperature warmings by 6–10°C (Dunham et al., 2007; Warren et al., 2022). Few studies have also related potential differences in post-fire summer water temperature responses across burned sites to watershed and riparian features. Post-fire summer stream water temperatures are generally observed to increase with higher riparian burn area and severity (Dunham et al., 2007; Beakes et al., 2014), with the rate of water temperature increase being greater for downstream sites near the burn perimeter than for those further away (Mahlum et al., 2011; Beyene et al., 2022). Koontz et al. (2018) reported that the post-fire response in the frequency of warm summer water temperature events in PNW streams appeared to be mediated by watershed variables such as streamflow, watershed elevation, and precipitation patterns. However, with the exception of Koontz et al. (2018), all studies were based on individual watersheds or burn events, making it unclear how post-fire stream thermal responses vary spatially.

Post-fire water temperatures are expressions of fire effects commingled with weather effects. To parse out the fire-related water temperature changes, past empirical studies have employed multiple approaches ranging from paired watersheds (burned and unburned) comparisons in a before and after control design (e.g., Beakes et al., 2014; Swartz and Warren, 2022) to using statistical models for water temperatures conditioned on meteorological variable(s) (e.g., Beyene et al., 2022; Rey et al., 2023). These variations in the applied approaches for evaluating post-fire effects on summer stream water temperatures further complicate comparisons across studies.

The objective of this study was to assess wildfire effects on daily summer water temperatures of PNW streams for the following 1–3 years, with focus on understanding the spatial variability in the post-fire thermal response across sites. To this end, we utilized daily summer water temperatures and environmental data from 31 unregulated PNW streams draining watersheds with 10–100% of their riparian area burned. Moreover, we applied multiple analytical approaches to characterize and quantify fire effects on daily summer water temperatures. Key questions in the context of the main objective are (a) How do PNW wildfires affect daily summer stream water temperatures? Are wildfires always associated with the warming of daily summer water temperatures?, (b) is there spatial variability in the post-fire daily summer stream water temperature responses? If yes, what landscape parameters correspond to fire-related changes in daily summer water temperatures across sites?, and (c) Is the sensitivity of daily summer water temperature metrics (i.e., minimum, maximum and mean) to a wildfire uniform? If not, which daily summer water temperature metric shows the highest or lowest sensitivity to post-fire effects?.

2. Methodology

2.1. Study Sites

For this study, we utilized daily summer (July to September) stream water temperature data from a total of 31 burned and 19 control (unburned) Northwest Stream Temperature (NorWeST; Isaak et al., 2017) watersheds across the Pacific Northwest (Fig. 1 and Tables 1 and S1). These NorWeST watersheds were delineated based on the location of the NorWeST gages. The selected burned watersheds had 10–100% of their riparian vegetation burned by a wildfire event(s) during any year between 1990–2015 and did not reburn during the following three years. Throughout this study, the term ”riparian area” refers to the strip of land on either side of the stream channel that is within 100 meters from the centre of the National Hydrography plus Dataset (NHDPlus V2; Moore et al., 2019) stream flow line. Moreover, the size and severity of these riparian and watershed burns were computed by overlaying the Monitoring Trends for Burn Severity (MTBS) rasters (Eidenshink et al., 2007) on riparian and watershed delineations. According to the US Army Corps of Engineers’ National Dam inventory dataset ( https://nid.sec.usace.army.mil/-/), there were no major hydrological or hydrothermal structures in the selected burned watersheds. Moreover, all burned watersheds are located in federal public lands and wilderness areas where commercial logging is prohibited. Regarding stream temperature data availability, our selection criterion was that burned watersheds needed to have at least three years of daily summer stream water temperature data during the ten years before the fire years and at least one year of daily temperature data during the first three years after the fire years. However, 85% of selected watersheds had five-years of pre-fire daily summer water temperature data and two years of post-fire summer daily water temperature data.

Fig. 1.

Location and ecoregion map of studied burned and paired unburned stream sites.

Table 1.

Location, ecoregion, morphology, and burn characteristics of studied stream sites and their watersheds.

Site No.	Site Name	State	Long (E)	Lat (N)	Ecoregion	Modified Strahler’s Stream Order	Basin Area (km2)	Fire Year	Riparian Area Burned (%)	Site Distance from Burned Perimeter (m)
1	Castle Rock Fork	OR	−122.59	43.11	Cascades	4	111	2002	32	0
2	Panther Creek	OR	−122.68	43.30	Cascades	5	49	2002	87	0
3	Ginger Creek	OR	−122.55	42.52	Cascades	2	1	2008	100	0
4	East Fork Mill Creek	OR	−120.54	44.46	Blue Mountains	4	40	2000	90	0
5	Tucannon River	OR	−117.71	46.21	Blue Mountains	5	165	2006	54	0
6	Otter Creek	OR	−118.76	44.98	Blue Mountains	2	6	2007	64	0
7	Horse Creek	OR	−116.78	45.69	Blue Mountains	5	139	2007	23	0
8	Middle Fork Eel River	CA	−123.02	39.90	Klamath Mountains	6	469	2012	14	0
9	North Fork Flathead River	MT	−114.13	48.50	Northern Rockies	7	4030	2003	15	0
10	Flat Creek	ID	−115.85	45.27	Idaho Batholiths	2	11	2000	96	0
11	South Fork White Bird Creek	ID	−116.22	45.79	Blue Mountains	4	92	2000	47	0
12	Bear Valley Creek	ID	−115.37	44.39	Idaho Batholiths	5	168	2006	20	0
13	Secesh River	ID	−115.81	45.22	Idaho Batholiths	4	286	2007	26	0
14	Emigrant Creek	OR	−119.39	43.81	Blue Mountain	5	228	2007	34	0
15	Missouri ST 3	MT	−111.75	44.60	Middle Rockies	2	8	2008	17	0
16	Rock Creek	ID	−114.62	46.61	Idaho Batholiths	2	6	2000	100	0
17	Elizabeth Creek	ID	−115.22	46.79	Northern Rockies	3	24	2000	37	0
18	Lavadoure Creek	OR	−123.11	42.95	Klamath Mountains	5	4	2004	100	0
19	Calf Creek	OR	−122.63	43.27	Cascades	5	48	2002	68	19
20	Pataha Creek	WA	−117.52	46.27	Blue Mountains	3	24	2005	30	0
21	South Fork Trinity River	CA	−123.50	40.67	Klamath Mountains	8	2035	2008	16	70
22	Nameless stream	CA	−123.03	40.17	Klamath Mountains	4	16	1996	44	1392
23	Meadow Creek	MT	−113.80	45.83	Idaho Batholiths	3	19	2000	100	0
24	Piquett Creek	MT	−114.17	45.81	Idaho Batholiths	4	36	2007	91	0
25	Wind River	ID	−115.95	45.46	Idaho Batholiths	5	168	2006	21	0
26	Big Mallard Creek	ID	−115.27	45.54	Idaho Batholiths	4	18	2007	40	932
27	Parks Creek	ID	−115.53	44.96	Idaho Batholiths	3	47	2005	37	1138
28	Tamarack Creek	ID	−115.39	44.96	Idaho Batholiths	3	24	2006	12	3370
29	Bargamin Creek	ID	−115.03	45.72	Idaho Batholiths	4	79	2005	11	165
30	Moonlight Creek	CA	−120.80	40.18	Sierra Nevada	5	23	2007	100	0
31	Big Flat Creek	ID	−114.49	46.40	Idaho Batholiths	3	37	2007	45	52

For each burned site, we identified a control site that had no burn record since 1984, daily summer stream water temperature data for the study period, and was located within the same EPA level 3 ecoregion (Omernik and Griffith, 2014). A level 3 ecoregion represents a region that is similar in geology, physiography, vegetation, climate, and soils. Similar criteria have been used in many wildfire studies when pairing a reference site to a burned site (Williams et al., 2022; Giovando and Niemann, 2022; Pennino et al., 2022). In this study, reference sites were within a distance of 100 km for approximately 80% of the burned sites. Pearson’s correlation coefficients, used to measure the similarity in pre-fire daily temperature variability between burned sites and control sites, ranged between 0.6–0.95 (Table S2), indicating that paired burned and unburned sites were reasonably analogous to each other in their pre-fire stream water temperature patterns.

We employed StreamStats ( https://streamstats.usgs.gov/ss/) to derive physiographic, climatic, and land cover characteristics of burned and unburned watersheds. The watershed area of burned sites ranged between 1.4 km² and 4030 km² with the median watershed size being 40 km². The watershed averaged elevation also varied from 527–2356 m. Burned watersheds also had minimal man-made land cover disturbances with forests covering over 70% of the land area and urban and agricultural areas covering less than 5%, based on the 2001 National Land Cover Dataset (Homer et al., 2007). According to the State Geologic Map Compilation Geodatabase (Horton, 2017), the surface lithology of watersheds for burned sites in northern and central Idaho and California was dominated by sedimentary rocks, while watersheds of burned sites in eastern Oregon and California, central Oregon, and southern Idaho were mainly composed of igneous and metamorphic rocks. The watershed averaged annual mean air temperature for the 1985–2015 period ranged between 2.1°C and 12.2°C (PPRISM and P.SM Climate Group and P., 2020). Streams in burned watersheds exhibited a cold or cool thermal regime with the average pre-fire mean summer water temperature varying from 6–21°C. The watershed characteristics of studied burned and unburned sites are provided in Tables 1 and S1.

2.2. Analytical Methods

To evaluate the summer thermal response of streams to wildfires across stream sites and daily summer water temperature metrics (i.e., mean, maximum, minimum), we employed multiple approaches (Fig. 2). The bootstrap method was utilized to calculate the overall (total) post-fire change in daily summer water temperatures at stream sites and determine its statistical significance. We broke down the total post-fire change in daily summer water temperature at burned sites into fire-and weather-related changes using two analytical approaches: paired watersheds comparison and weather-based attribution methods. Subsequently, we employed the results from the paired watersheds comparison to develop random forest regression-based spatial models and identify environmental variables correlated with fire-related daily summer water temperature changes across burned sites. Previous studies have demonstrated these approaches to be useful for characterizing and quantifying forest disturbance effects on stream water quality and quantity (e.g., Gomi et al., 2006; Beyene et al., 2023; Pennino et al., 2022). We present a summary of these approaches in the following sections.

Fig. 2.

Procedural framework applied for assessing wildfire effects on daily stream water temperatures. NSC = Nash–Sutcliffe efficiency coefficient; RMSE = root mean squared error.

2.2.1. Analytical Methods

We used the bootstrap technique (Efron, 1982) to compute the difference in post-and pre-fire daily summer water temperature metrics ($\text{n}=3$) and determine its statistical significance for both burned and their corresponding unburned sites. The advantage of this method is that it does not rely on any distributional assumption when assessing the significance of the post-fire change in daily summer water temperatures. It also accounts for the inequality in the length of the pre-and post-fire years for studied sites when calculating the magnitude of the post-fire change in daily summer stream water temperatures.

In this procedure, pre-fire years were randomly shuffled to generate 1000 subsamples with size equal to the length of the post-fire years ($n=1–3$). For a paired unburned site, pre-and post-fire periods referred to the same periods as for the burned site. We computed the daily summer water temperature for each pre-fire year, and determined the median of the daily summer water temperature for each subsample. Lastly, we computed the standardized difference in post-and pre-fire daily summer water temperatures as the difference between the median post-fire daily summer water temperature and the median (500^th ranked) subsample value, divided by the standard deviation of the pre-fire daily summer water temperatures $\left({\sigma }_{pre}\right)$. Standardizing the difference between pre-and post-fire daily summer water temperature allows for comparability across burned sites. We considered the standardized difference between post-and pre-fire daily summer stream water temperature to be statistically significant at the $p<{0.05}^{th}$ level if the median of the post-fire stream water temperature c was either higher than the 975^th ranked (pre-fire) subsample value or lesser than the 25^th ranked (pre-fire) subsample value.

2.2.2. Paired Watersheds Comparison

To quantify riparian fire effects on post-fire daily summer water temperatures, this approach encompassed the following steps: (a) developing models to establish the pre-fire daily summer water temperature relationship between burned and paired unburned sites for each burned site $(n=31)$ and water temperature metric $(n=3)$, (b) assessing the predictive skills of these models in estimating daily water temperatures under reference (unburned) conditions, to provide an estimate of model error and bias, (c) computing the difference between observed and predicted post-fire water temperatures and dividing it by ${\sigma }_{pre}$, which provides a first order estimate of riparian burn effects on water temperatures (Fig. 3a). For this analysis, we excluded the stream water temperature data during the fire year.

Fig. 3.

Illustrations of (a) paired watershed comparison for Ginger Creek in Idaho, and (b) weather-based attribution approach for Ginger Creek in Idaho.

We applied the generalized least squares (GLS) regression approach to model the pre-fire daily summer water temperatures of a burned site as a function of the corresponding daily water temperatures of a paired unburned site. GLS regression is an extension of the linear regression procedure designed to handle serial autocorrelation (Kariya and Kurata, 2004), a feature inherent in daily water temperature data. Our fitted GLS models for summer daily water temperature at a burned site had the general form:

$$T_w^B(d)={\beta }_0+{\beta }_1\ast T_w^C(d)+{\beta }_2\ast sin(2\ast pi\ast d/T)+{\beta }_2\ast cos\left(2\ast pi\ast d/T\right)+\epsilon (d)$$ (1)

Where $T_w^B(d)$ is the daily summer water temperature at a burned site on day $d$ and $T_w^C(d)$ is the corresponding daily water temperature at the control site. $\beta$ values are the parameters to be estimated during regression. The sin and cos terms are trigonometric covariates describing the seasonality of water temperature where $d$ is the day of the water year ($d=1$ on October 1^st) and $T$ is the number of days in a hydrologic year (October 1^st-September 30^th). $\epsilon (d)$ is the error term which we modeled as an autoregressive process of order k, and we estimated this k for each burned site by examining the partial autocorrelation plot and retaining the terms with statistically significant autocorrelation coefficients. We utilized the gls function in the nlme package (Pinheiro et al., 2017) implementation in the $R$ computing environment to provide the algorithm to develop GLS models for daily summer water temperatures.

We then used the Leave One Out Cross Validation (LOOCV) technique to evaluate the predictive skills of each GLS model in estimating daily summer water temperatures under unburned (reference) conditions. In this technique, we withheld data for each pre-fire year, re-calibrated the GLS model with the remaining data, and predicted daily summer water temperature for the omitted year. We repeated the procedure for all pre-fire years. For each GLS model, we assessed the discrepancy between estimated and observed pre-fire daily summer water temperatures during LOOCV using two common model performance metrics: Root Mean Squared Error (RMSE), and Nash–Sutcliffe Efficiency Coefficient (NSE). RMSE is a measure of the overall error with optimal values being near zero, indicating accurate model prediction. NSE measures the overall goodness of fit between observed and predicted values, with NSE equal to 1 being the optimal value. Model performance is deemed adequate when NSE is greater than 0.5 (Nossent and Bauwens, 2012).

Finally, we input post-fire daily summer water temperature data from the paired unburned site into GLS models for each burned site to predict daily summer water temperatures for the post-fire years. The difference $(\text{Δ}T)$ between observed and model-predicted daily summer water temperatures was computed for each post-fire year. We determined the statistical significance of post-fire $\text{Δ}T\text{s}$ by comparing them to the distribution of pre-fire model residuals. The null hypothesis posited that post-fire $\text{Δ}T\text{s}$ fell within model error ranges because wildfire had negligible contributions on post-fire daily summer water temperatures. Our alternative hypothesis was that post-fire $\text{Δ}T\text{s}$ lay outside pre-fire model error ranges due to substantial effects of wildfire on post-fire daily summer water temperatures. We rejected the null hypothesis if post-fire $\text{Δ}T\text{s}$ were higher or lower than the 95% pre-fire model uncertainty ranges. Additionally, we regarded post-fire $\text{Δ}T\text{s}$ as a first order estimate of fire-related change in daily summer water temperatures.

2.2.3. Weather-based Attribution Approach

In this method, we established the pre-fire local weather-summer water temperature relationship for each daily water temperature metric and burned site, to estimate the offset in post-fire daily summer water temperature not accounted for by local weather-water temperature relations (Fig. 3b). Employing Paired Random Forest Regression-Generalized Least Square (PRFR-GLS) approach, we modelled the pre-fire daily summer water temperatures at a burned site as a function of daily-to-seasonal air temperature and precipitation metrics. A random forest regression (RFR; Breiman, 2001) model is built by constructing an ensemble of regression trees trained using recursive subsets of observations, and predictions are computed as the expected value of all individual predications from each tree in the RFR model (Cutler et al., 2012). RFR models are increasingly applied in ecological and hydrological studies because of their superior predictive performance particularly when there are complex non-linearities and interaction effects in the relationship between the predictors and response variable (Cutler et al., 2007).

We considered 13 daily-to-seasonal air temperature and precipitation metrics (Table 2) as possible covariates when constructing RFR models for daily summer water temperatures under pre-fire conditions. This was premised on the assumption that pre-fire daily summer water temperature variability of burned sites was primarily driven by prevailing daily-to-seasonal meteorological conditions given the negligible (direct) anthropogenic impact on water temperature in minimally managed streams. Moreover, stream water temperature studies have demonstrated that the meteorological variables considered here can be good proxies of atmospheric and to a lesser extent advective heat inputs into streams, and therefore effective predictors of water temperatures under pre-fire conditions. For instance, Mayer (2012) highlighted the strong correspondence between air and water temperatures by modeling summer water temperatures in many Pacific Northwestern streams using weekly air temperatures. We fitted multiple sets of candidate RFR models using one or more of these weather variables as predictors. To reduce model overfitting, we limited the maximum number of covariates included in any RFR model to two and set the maximum (Pearson’s) correlation between potential predictors in a model at 0.8 to avoid multicollinearity effects. The randomForest (Liaw and Wiener, 2002) implementation in the $R$ computing environment was used to provide the algorithm for developing RFR models for daily summer water temperatures.

Table 2.

List of daily-to-seasonal air temperature and precipitation variables considered as possible predictors when modeling pre-fire daily summer water temperatures at studied sites. All variables were derived from the PRISM dataset.

Variable	Description
Daily mean air temperatures	1-day mean air temperatures
7-day mean air temperatures	7-day rolling mean air temperatures
30-day mean air temperatures	30-day rolling mean air temperatures
60-day mean air temperatures	60-day rolling mean air temperatures
120-day mean air temperatures	120-day rolling mean air temperatures
150-day mean air temperatures	150-day rolling mean air temperatures
180-day mean air temperatures	180-day rolling mean air temperatures
Daily precipitation totals	Sum total of precipitation over each day
7-day precipitation totals	7-day rolling mean precipitation totals
30-day precipitation totals	7-day rolling mean precipitation totals
60-day precipitation totals	60-day rolling precipitation totals
January-June precipitation totals	Sum total of January 1st to June 31st daily precipitation
October-March Precipitation totals	Sum total of October 1st to March 31st daily precipitation

To address the temporal autocorrelation in RFR modeling, we paired all RFR models with a simple GLS function containing an autoregressive component to model RFR model residuals. Hence, the prediction of daily summer stream water temperatures for a PRFR-GLS model is obtained by summing the RFR predictions and GLS prediction of the RFR residuals. Mathematically, a PRFR-GLS prediction of daily stream water temperatures $\left(\widehat{T_w}\left(d_n\right)\right)$

$$\widehat{T_w}\left(d_n\right)=\widehat{f_{RFR}}\left(x\left(d_n\right)\right)+\widehat{\epsilon }\left(d_n\right)$$ (2)

where $\widehat{f_{RFR}}$ is the RFR model predictions with covariates $x\left(d_n\right)$, and $\epsilon \left(d_n\right)$ is the GLS function prediction of the residuals, and the index $d_n$ is the day. Here, the GLS function is described by an autoregressive component of order $k$, and we estimated this $k$ for each burned site by examining the partial autocorrelation plot and retaining the terms with statistically significant autocorrelation coefficients.

We employed the bias-corrected Akaike Information Criterion (AICc; Akaike, 1974) to evaluate the complexity and goodness of fit of candidate PRFR-GLS models for each daily summer water temperature metric and burned site. Best fit PRFR-GLS models were selected as those with Best fit PRFR-GLS models were selected as those with the least AICc score. Subsequently, we utilized the LOOCV technique, as detailed in the previous section, to evaluate the predictive capabilities of each PRFR-GLS model in estimating daily summer water temperatures under unburned (reference) conditions. We also characterized model predictive performance during LOOCV using the two model performance metrics of RMSE and NSE.

In the final step, we input post-fire weather data into the best fit PRFR-GLS models to predict daily summer water temperatures for each post-fire year. The offset $(\text{Δ}T)$ between observed and predicted daily summer water temperatures was computed for each post-fire year. To determine the significance of post-fire $\text{Δ}T\text{s}$, we compared them to pre-fire model residual ranges. We rejected the null hypothesis that post-fire $\text{Δ}T\text{s}$ were within pre-fire model error ranges if post-fire $\text{Δ}T\text{s}$ were higher or lower than the 95% pre-fire model uncertainty ranges. Furthermore, we considered post-fire $\text{Δ}T\text{s}$ as a first order estimate of fire-related change in daily water temperatures.

2.2.4. Spatial model for post-fire stream thermal response

We used the post-fire $\text{Δ}T\text{s}$, obtained from the paired watershed comparison approach across burned sites, to develop best fit RFR models and identify environmental variables that correlated with the inter-site variability in fire-related changes in daily summer stream water temperatures. The development of these best-fit models followed an iterative process, where predictors from a subset of 30 watershed attributes (Table 3) were systematically introduced. These attributes were chosen for their physical interpretability and their ability to yield the most significant improvement in the mean squared error of the model. The selection process ended when the addition of any further predictor variable failed to improve the mean squared error. The importance of a predictor (environmental variable) in the RFR models was evaluated by computing the mean decrease in accuracy (mean squared error) observed between model predictions and actual values by randomly permuting the selected variable (Breiman, 2001).

Table 3.

List of watershed attributes considered in the random forest regression models for fire effects on daily summer water temperatures.

Variable (Unit)	Description	Source Dataset
Total Watershed Area (km2)	Watershed Area of stream	NHD Plus V2 Dataset
Watershed Elevation (m)	Average elevation over watershed	NHD Plus V2 Dataset
Watershed Area Burned (%)	Ratio of the burned watershed area to the total watershed area of stream	MTBS Dataset
Watershed Area Burned under High-Severity (%)	Ratio of the burned watershed area burned under medium-to-high severity to the total watershed area of stream	MTBS Dataset
Riparian Area Burned (%)	Ratio of the burned riparian area to the total riparian area	MTBS Dataset
Riparian Area Burned under High-Severity (%)	Ratio of the burned riparian area under medium-to-high severity to the total riparian area	MTBS Dataset
Mean Watershed Elevation (m)	Watershed elevation averaged over the total watershed area of stream	NHD Plus V2 Dataset
Depth to the Water Table (m)	Depth to the water table (from surface) averaged over the total watershed area of stream	STATSGO
Depth to the Bedrock (m)	Depth to the bedrock (from surface) averaged over the total watershed area of stream	STATSGO
Soil Permeability (mm hr−1)	Permeability of soil averaged over the total watershed area of stream	STATSGO
Compound Topographic Index (−)	Natural logarithm of the ratio of watershed area to the tangent of slope	ENVIROATLAS
Soil Bulk Density (g cm−3)	Bulk density of soil averaged over the total watershed area of stream of stream	STATSGO
Bed Permeability Class (−)	Bedrock permeability class averaged over the total watershed area of stream	STATSGO
Sand Content (%, by weight)	Soil sand content averaged over the total watershed area of stream	STATSGO
Clay Content (%, by weight)	Soil clay content averaged over the total watershed area of stream	STATSGO
Lithology Classified under Non-Carbonate Residual (%)	Ratio of the watershed area whose lithology is classified as non-carbonate residual material to the total watershed area of stream	Cress et al., 2010
Open Water Area (%)	Ratio of the watershed area classified as open water land cover to the total watershed area of stream	NLCD 2001
Ice/Snow Area (%)	Ratio of the watershed area classified as ice/snow land cover to the total watershed area of stream	NLCD 2001
Urbanized Area (%)	Ratio of the watershed area classified as developed, high intensity land use to the total watershed area of stream	NLCD 2001
Barren Area (%)	Ratio of the watershed area classified as barren land cover to the total watershed area of stream	NLCD 2001
Forested Area (%)	Ratio of the watershed area classified as forest land cover to the total watershed area of stream	NLCD 2001
Wetland Area (%)	Ratio of the watershed area classified as wetland land cover to the total watershed area of stream	NLCD 2001
Post-fire weather effects (%)	Post-fire water temperature change in paired unburned sites	NorWeST
Mean annual air temperature (°C)	Mean annual (October-September) air temperature averaged over 1985–2014	PRISM
Mean summer air temperature (°C)	Mean summer (June-August) air temperature averaged over 1985–2014	PRISM
October-March precipitation total (mm)	Total sum of daily precipitation from October 1st (Year-1) to March 31st averaged over 1985–2014	PRISM
October-June precipitation total (mm)	Total sum of daily precipitation from October 1st (Year-1) to June 31st averaged over 1985–2014	PRISM
October-March snow-water equivalent (mm)	Total sum of daily snow-water equivalent from October 1st (Year-1) to March 31st (Year) averaged over 1985–2014	Broxton et al., 2018
October-June snow-water equivalent (mm)	Total sum of daily snow-water equivalent from October 1st (Year-1) to June 31st averaged over 1985–2014	Broxton et al., 2018
Ratio of annual snow-water equivalent to precipitation total (%)	Ratio of annual snow water equivalent to annual precipitation total	PRISM; Broxton et al., 2018
Stream Order	Modified Strahler’s stream order	NHDPlus V2 Dataset
Ecoregions	EPA’s Level III Ecoregion Classification	Omernik and Griffith (2014)

3. Results

3.1. Postfire Changes in Daily Summer Stream Water Temperatures

Post-fire daily mean summer water temperatures were generally warmer than pre-fire in nearly two-thirds of burned watersheds (Fig. 4a). Except for the Sierra Nevada ecoregion, which had only one burned site, each ecoregion featured at least one burned site with a statistically significant $(p<0.05)$ increase in the daily mean summer water temperature following a wildfire. Among the 11 burned sites displaying significantly higher post-fire daily mean summer water temperatures, the median post-fire increase in daily mean summer water temperature ranged between 0.12 – 9 times ${\sigma }_{pre}$ (or 0.4–7.3°C) (Fig. 4a). As illustrated by the post-fire change in paired unburned sites (Fig. 4a), about half of burned sites that observed post-fire warming of daily mean water temperatures also had climate condition conditions during the post-fire periods that favored warmer daily mean summer water temperatures. However, in the Cascades, Middle Rockies, and Klamath ecoregions, the majority of burned sites recorded post-fire warming in their daily mean summer water temperatures even under post-fire climate conditions promoting cooler daily mean summer water temperatures (Fig. 4a).Conversely, six burned sites showed a significant $(p<0.05)$ cooling of post-fire daily mean summer water temperatures by up to 8.5 times ${\sigma }_{pre}$ (or 1.9°C) with half of these sites situated in the Blue Mountains ecoregion. In addition, majority (60%) of sites with significantly cooler post-fire daily mean summer water temperatures also had post-fire weather conditions that corresponded to the cooling of daily mean summer water temperatures.

Fig. 4.

Comparison of post-fire change in daily summer water temperature in burned and paired unburned site for (a) daily mean, (b) daily minimum, and (c) daily maximum. Studied sites are located in the following ecoregions: CS- Cascades, NR- Northern Rockies, BM- Blue Mountains, MR- Middle Rockies, KM- Klamath Mountains, SN- Sierra Nevada, and IB- Idaho Batholiths.

At burned sites, the direction of the post-fire change in daily mean summer water temperatures often corresponded to that of the post-fire change in daily maximum and minimum summer water temperatures (Figs. 4a–c). Statistically significant $(p<0.05)$ post-fire increases in daily maximum and minimum summer water temperatures were recorded at 13 and 10 burned sites, respectively (Figs. 4b and c). In burned sites with significant post-fire increase in daily maximum and minimum summer water temperatures, the median post-fire daily maximum summer water temperatures were up to 11.1 times ${\sigma }_{pre}$ (or 8.5°C) higher than pre-fire whereas post-fire daily minimum summer water temperatures were 0.8 – 6.5 times ${\sigma }_{pre}$ (or 0.3–5.6°C) higher than pre-fire. Despite weather conditions favoring cooler daily maximum and minimum summer water temperatures post fire, most burned sites in the Cascades, Middle Rockies, and Klamath ecoregions recorded significantly warmer post-fire daily maximum and minimum summer water temperatures. In contrast, post-fire daily summer water temperatures were significantly cooler than pre-fire at five sites for daily maximum and five sites for daily minimum. 60% of burned sites with a significant post-fire decrease in daily maximum summer water temperatures were in the Blue Mountains ecoregion. Finally, three out of the five sites with a significant decrease in their post-fire daily maximum summer water temperatures also had post-fire meteorological conditions corresponding to cooler summer water temperatures.

3.2. Quantifying Fire Effects on Water Temperatures

In the paired watershed comparison approach, the GLS model developed for predicting daily summer water temperatures under unburned conditions had an NSE coefficient greater than 65% for 90% of sites (Table S3). Moreover, RMSE of these model predictions for the pre-fire period across burned sites ranged from 0.36 to 4.8 $\text{times}{\sigma }_{pre}$ with a median of 1.05 times ${\sigma }_{pre}$. Similarly, the combined PRFR-GLS models constructed during the weather-based attribution approach showed similar predictive performances during LOOCV procedures although model NSE coefficients were slightly lower and RMSE values were somewhat higher as compared to GLS models for most burned sites (Table S4). These results indicate that both approaches can indirectly provide an estimate of possible fire contributions on the post-fire daily summer stream water temperature.

In the paired watershed comparison approach, a positive post-fire $\text{Δ}T$ at a burned site indicates that the observed post-fire increase in daily summer water temperatures were higher than predicted based on the pre-fire daily summer water temperature relationship between burned and paired unburned sites. A significantly positive $\text{Δ}T$ represents post-fire daily summer water temperatures warming not accounted for by burned-unburned daily water temperature relations or within model error ranges, and therefore is most likely related to wildfire effects. The median post-fire $\text{Δ}T$ in daily mean, minimum, and maximum summer water temperatures were significantly $(p<0.05)$ positive at 10, 9, and 16 out of the 31 burned sites, respectively (Fig. 5a). In burned sites with significantly positive post-fire $\text{Δ}T$, the median post-fire $\text{Δ}T$ in daily mean summer water temperature ranged from 0.2 – 8 times ${\sigma }_{pre}$ (or 0.6–5.3°C) whereas the median post-fire $\text{Δ}T$ in daily minimum and maximum summer water temperature varied from 0.11 – 6.7 times ${\sigma }_{pre}$ (or 0.4–4.4°C) and 0.5 – 8 times ${\sigma }_{pre}$ (or 0.6–6.4°C), respectively (Figs. 5 and S1). All ecoregions except for the Sierra Nevada ecoregion had at least one burned site with significantly positive $\text{Δ}T$. Moreover, half or more of burned sites in the Cascades, Klamath, Middle Rockies, and Idaho Batholith ecoregions had a significantly positive post-fire $\text{Δ}T$ in their daily mean, minimum, and maximum summer water temperatures. In contrast, the median post-fire $\text{Δ}T$ in daily summer water temperature was significantly negative (cooling) only at three burned sites for daily mean, four burned sites for daily minimum, and three burned sites for daily maximum. Burned watersheds with significantly negative $\text{Δ}T\text{s}$ in daily summer water temperatures were found in the Blue Mountain and Sierra Nevada ecoregions. In burned sites with significantly negative post-fire $\text{Δ}T$, the median post-fire $\text{Δ}T$ in daily summer water temperature ranged from −0.9 – ₋5 times ${\sigma }_{pre}$ (or −0.5 - −2°C) for daily mean, −0.5 – ₋5.1 times ${\sigma }_{pre}$ (or −0.2–1.6°C) for daily minimum, and −1.8 – −3.2 times ${\sigma }_{pre}$ (or −1 - −1.4°C) for daily maximum.

Fig. 5.

Median post-fire $\text{Δ}T$ for daily mean (top panels), minimum (middle panels), and maximum (bottom panels) at burned sites using (a) paired watershed comparison (a), and (b) weather-water temperature analysis (b). Blue bars represent the median offset $\left({\sigma }_{\text{pre}}\right)$ between observed and modeled post-fire daily water temperature at each burned site and an asterisk represents an offset that is outside the 95% pre-fire model error range. Studied sites are located in the following ecoregions: CS- Cascades, NR- Northern Rockies, BM- Blue Mountains, MR- Middle Rockies, KM- Klamath Mountains, SN- Sierra Nevada, and IB- Idaho Batholiths.

Results from the weather-based attribution approach were largely similar to those of the paired watershed comparisons (Fig. 5b). The majority (> 60%) of burned sites with a significant post-fire $\text{Δ}T$ from the paired watershed comparison also had a significant post-fire $\text{Δ}T$ from the weather-based attribution approach. Moreover, all ecoregions apart from the Sierra Nevada, had significantly positive post-fire $\text{Δ}T$. Nonetheless, post-fire $\text{Δ}T$ estimates from weather-related attribution approach were often higher than that from paired watershed approach. Also, a lesser number of burned sites had statistically significant post-fire $\text{Δ}T$ using the weather-related attribution approach.

The offset between observed and model-predicted daily summer water temperatures varied among pre-and post-fire years and burned basins. Nevertheless, it would be highly unlikely for a multi-basin averaged $\text{Δ}T\text{s}$ to be consistently positive or negative for the first three post fire years unless wildfire had altered the stream thermal regime. Averaging the 31-burned site mean difference between observed and modeled daily summer water temperature $\left(\text{Δ}T\right)$ from the paired watershed comparison approach demonstrated that post-fire $\text{Δ}T\text{s}$ in daily summer water temperatures were positive for the first three post-fire years and ranged from 0.29 – 0.81 times ${\sigma }_{pre}$ (or 0.4 – 0.54 °C) for daily mean, 0.25 – 1 times ${\sigma }_{pre}$ (or 0.30 – 0.39 °C) for daily minimum, and 1.79 – 2.11 times ${\sigma }_{pre}$ (or 0.72 – 1 °C) for daily maximum (Figs. 6a–c). Moreover, post-fire $\text{Δ}T\text{s}$ in daily mean and maximum summer water temperature were outside the 95% pre-fire $\text{Δ}T$ variability range. Similar findings were also observed when averaging the 31-burned site mean $\text{Δ}T\text{s}$ from the weather-based attribution approach (Figs. S2a–c). These imply that the overall post-fire daily mean and maximum summer water temperature warming across burned sites was distinctly outside the range that could be explained by model errors, paired unburned and burned watersheds comparison or weather-water temperature relations.

Fig. 6.

Averaged $\text{Δ}T$ for daily mean (a), minimum (b), and maximum (c) water temperature among studied (31) burned sites in years prior (blue) and after (orange) each basin’s fire year. The blue dashed lines represent the 95% pre-fire variability range.

3.3. Environmental Correlates of Postfire Water Temperature Changes

Three RFR models with environmental variables as covariates were developed to explain the variability in fire-related changes (post-fire $\text{Δ}T$) in daily mean, minimum, and maximum summer water temperatures across studied stream sites. These models represented 27 – 40% of the inter-site variability in median post-fire $\text{Δ}T$ for daily mean (38.2%), daily minimum (25.4%), and daily maximum (40%) summer water temperatures. They showed that the best predictors of the post-fire $\text{Δ}T$ in daily summer water temperatures across studied sites includes % riparian and watershed area burned, %watershed area burned under medium-to-high severity, bedrock permeability, watershed area, post-fire weather, %watershed area classed as forested and barren (Figs. 7a–c). The partial dependence plots for these models indicated that the likelihood and magnitude of a positive post-fire $\text{Δ}T$ for daily mean, minimum, and maximum summer water temperatures increased with higher %watershed and riparian burn area and %watershed area burned under medium-to-high severity (Figs. 8a–c). Post-fire weather was also positively associated with post-fire $\text{Δ}T$ for daily maximum and mean water temperatures, whereas bedrock permeability and percent forest cover were inversely correlated with post-fire $\text{Δ}T$ for daily minimum and mean summer water temperatures (Figs. 8a and b). Finally, post-fire $\text{Δ}T$ for daily summer maximum water temperature had a positive relation with the percent watershed area classed as barren and a negative relation with the climatological October-March Snow -water Equivalent (SWE) and watershed area (Figs. 8c).

Fig. 7.

Ranked importance of predictor variables for post-fire daily mean summer stream temperature response models. These models represented 38.2%, 25.4% and 40% of the inter-site variability in median post fire $\text{Δ}T$ for daily mean, minimum, and maximum stream water temperature, respectively.

Fig. 8.

Partial dependence plots showing how post-fire daily summer stream temperature responded to predictors selected for (a) daily mean, (b) daily minimum, and (c) daily maximum. These models represented 38.2%, 25.4% and 40% of the inter-site variability in median post fire $\text{Δ}T$ for daily mean, minimum, and maximum stream water temperature, respectively.

4. Discussion

4.1. Postfire Stream Thermal Responses across Sites

Results from our analyses on post-fire water temperature responses indicated that wildfires had spatially variable impact on the daily summer water temperature of 31 Pacific Northwest stream sites draining watersheds with 10–100% of their riparian area burned. A significant $(p<0.05)$ fire-related warming of daily summer stream water temperatures was detected at over a third of burned sites for the following 1–3 years. Additionally, the fire-related increase in summer daily water temperature at these sites ranged between 0.20 – 9 times ${\sigma }_{pre}$ (or about 0.40 – 8 °C). This finding is consistent with previous studies that have reported post-fire warmings of summer stream water temperatures (Dunham et al., 2007; Rosenberger et al., 2015; Beyene et al., 2022; Warren et al., 2022; Swartz and Warren, 2022). The inter-site variability in the post-fire summer water temperature warmings across sites also corresponds with the broad range of reported post-fire water temperature increases in prior studies: modest increases by 0.3 – 1 °C (Beakes et al., 2014; Beyene et al., 2022) to 6–10 °C increases (Dunham et al., 2007; Warren et al., 2022).

In contrast, in three (out of the 31) burned sites, wildfires were associated with a significant cooling of summer daily water temperatures for the following 1–3 years. The decrease in summer daily water temperature at these sites varied from 0.1–6 times ${\sigma }_{pre}$ (or about 0.5 – 2 °C). Post-fire cooling of daily summer water temperature at a burned site could be attributed to a few mechanisms. Large wildfires often lead to increases in surface and sub-surface flow volumes (Bart and Tague, 2017; Beyene et al., 2021) that tend to reduce stream thermal sensitivity to atmospheric conditions post-fire (Rey et al., 2023). The burning of riparian stands also elevates nighttime summer stream heat loss due to increases in net long wave radiation and turbulent heat fluxes (Leach and Moore, 2010). Additionally, the burning of forest canopies favors higher winter snowpack (Maxwell and St Clair, 2019), whose melt water cool summer stream water temperatures. Nonetheless, a significant post-fire cooling of daily summer water temperatures was found at only 10% of studied stream sites. Moreover, when averaging the fire-related changes in daily summer water temperatures across the 31 burned sites, wildfires were associated with a 0.4 – 1 times ${\sigma }_{pre}$ (or about 0.3 – 0.9 °C) increase in daily summer water temperatures for the three subsequent years. Additionally, most studies on post-fire summer water temperature changes do not show a decrease in summer stream water temperatures following a wildfire. Taken together, these findings indicate that the burning of riparian vegetation by large, high-severity wildfires rarely corresponds to a significant cooling of daily summer stream water temperatures during the first 1–3 years.

Post-fire changes in stream water temperatures at burned reaches are caused by shifts in the heat and moisture transfer locally as well as in upstream areas due to wildfire-related disturbance on the terrestrial landscape. Consequently, we found that the inter-site differences in the post-fire daily summer water temperature responses across the 31 burned sites corresponded to a range of watershed and riparian parameters. According to our RFR model based analyses, watershed burn characteristics generally were reliable indicators of the magnitude and direct of change in post-fire summer daily water temperatures across studied stream sites. 11 out of the 12 sites with a %watershed area burned greater than 60% recorded post-fire warming of daily summer water temperatures (Figs. S3–S5). Higher %watershed and riparian areas burned under medium-to-high severity were often associated with higher post-fire warmings in daily mean, minimum and maximum water temperatures. In contrast, unlike previous studies (Mahlum et al., 2011; Beyene et al., 2022), we did not find that the downstream site’s distance from the burn perimeter was a good predictor of the rate of summer water temperature warming at burned sites. We conceptualize that this was because 27 out of the 31 studied burned sites were found within the burn perimeter 1.

The post-fire daily mean and maximum summer water temperature responses across sites was found to be strongly related to prevailing post-fire weather conditions. Moreover, warmer post-fire weather conditions corresponded to higher post-fire warmings in the daily mean and maximum summer water temperatures at burned sites. Warmer post-fire weather conditions exacerbate the warming of daily summer water temperatures following a wildfire through one or more of the following: (a) reducing the summer streamflow volume and heat capacity, which raises the stream thermal sensitivity to atmospheric conditions, (b) increasing the upstream surface water temperatures and shallow subsurface water temperatures, which promotes warmer stream water temperatures, and (c) increasing the local atmospheric heat input into stream.

Often, groundwater is a source of cooler water to streams during the summer (Mayer, 2012). Moreover, groundwater temperatures can exhibit vertical variation with depth, with groundwater temperatures at lower depth less sensitive to prevailing surface conditions (Maxwell and Kollet, 2008). Consequently, streams primarily fed by subsurface flow originating from high permeability bedrocks, tend to maintain cooler and more stable summer water temperatures compared to streams with shallow and low permeability bedrocks within the same region (Tague et al., 2007; Mayer, 2012; Hare et al., 2021; McGill et al., 2023). Our analysis using RFR models indicated that bedrock permeability was a strong predictor of post-fire daily mean and minimum summer stream water temperature responses across burned sites. Moreover, burned sites with higher bedrock permeability were generally associated with smaller post-fire daily mean and minimum summer water temperature responses across studied sites. This inverse relationship between bedrock permeability and post-fire daily summer water temperature increases could be related to the difference in the contribution of deep ground water on summer stream water volume; burned sites with higher bedrock permeability were likely to buffer fire-related water temperature changes due to higher connectivity to deep aquifers. This result aligns with the (Rey et al., 2023), which noted that wildfires did not have a significant effect on the annual water temperature of streams with deep groundwater connectivity.

Landscape controls including basin area, elevation, slope, and riparian area influence water temperatures through their effect on local climate, water sources, mixing, flow paths, and residence time (Mayer, 2012). Our results showed that burned sites with larger watershed areas were associated with little or no post-fire changes in their daily mean summer water temperatures. This observation could be ascribed to three mechanisms. Firstly, PNW streams with larger drainage areas often feature higher flow volumes and rates (Mayer, 2012), requiring greater amounts of heat input to change the water temperature than those with smaller drainage areas. Secondly, wider streams draining large watersheds are insensitive to changes in riparian forest cover as the canopy overhang is insufficient to provide shade for the channel (Coats and Jackson, 2020). Thirdly, in this study, watershed area was inversely related to percent watershed and riparian area burned, meaning that the riparian area affected by wildfire events was lesser in streams with large drainage areas.

The efficacy of riparian canopies in reducing the amount of solar radiation reaching streams depends various factors including vegetation density, canopy structure, and tree heights; forests with dense, tall trees and closed canopies can provide fuller shading to streams than those with sparse, short trees and open canopies (Dugdale et al., 2018; Jackson et al., 2017). Moreover, the loss of riparian vegetation that offers little or no shades has been observed to have marginal effect on stream water temperatures (Hrachowitz et al., 2010). Similarly here, burned sites with higher pre-fire percent watershed area forested and lower percent watershed area barren corresponded to a higher likelihood of post-fire daily summer water temperature increases. This was expected because, vegetation is likely to offer an incomplete shading and a lower moderating influence on water temperatures in streams with smaller percent watershed area forested or higher percent watershed area barren. As such, any disturbance that removes these vegetation has negligible effect.

Runoff from snowmelt may decrease summer stream thermal sensitivity to atmospheric heat input as advective fluxes of cold snowmelt water can overwhelm surface energy exchanges associated with increasing air temperature (Lisi et al., 2015; Winfree et al., 2018). Moreover, Cline et al. (2020) demonstrated that the buffering effect of snowmelt on summer stream thermal sensitivity was more pronounced during high snow years. We found that the watershed averaged October-March SWE total was negatively associated with post-fire daily maximum summer water temperature changes; burned sites with higher October-March SWE total had a reduced likelihood of post-fire warming in their daily maximum summer water temperatures, relative to sites with lower October-March SWE total. This finding is perhaps because in watersheds with higher winter snow depth, snowpack can persist until summer and provide cold water to mute any stream water heating effect of a post-fire increase in insolation.

We did not find marked differences in post-fire summer thermal responses of streams across ecoregions, despite our expectation that ecoregion type would modulate post-fire stream thermal responses through differences in riparian vegetation type, land cover, weather patterns, wildfire type and severity, flow source and volume, and other factors. One possible explanation is the limited number of study sites for most ecoregions; Middle Rockies and Sierra Nevada ecoregions were represented by only one site each whereas the Northern Rockies ecoregion included just two sites. Another reason could be that not all ecoregions in the PNW were represented in this study. Additionally, stream and watershed size of studied burned sites across ecoregions were not similar, which may mute potential influence of ecoregion type on post-fire daily summer temperature responses.

Previous studies have found that wildfire impacts on stream water temperature varied across different thermal metrics (Koontz et al., 2018). Our results also showed that the wildfire-related changes in daily summer stream water temperatures were pronounced for daily maximum and mean than for daily minimum. This finding aligns with the physical expectation that maximum daily water temperatures, which always occur during the day, are more influenced by the loss of riparian shade and rise in insolation due to wildfire burns. Also, the burning of canopy covers has been shown to enhance heat loss at nighttime (Leach and Moore, 2010) when daily minimum water temperatures occur.

This study indirectly estimated fire effects on daily summer water temperatures utilizing two general approaches: paired burned and unburned watersheds comparison and weather-based attribution approach. The results of the leave-one-out cross validation methods revealed that predictions from paired-watersheds comparison were only slightly more accurate than those from the paired weather-water temperature analysis for studied sites. Moreover, the estimated fire effects on water temperatures did not significantly differ between the two approaches, particularly when the offset between observed and model post-fire daily water temperature estimates were large. The similarity in performance of the two approaches could be attributed to the strong influence of atmospheric heat exchanges on the daily summer water temperatures at these streams. Given the limited number of gaged unregulated watersheds in the US and elsewhere, these findings indicate that the weather-based attribution approach could be useful for detecting and quantifying the impact of wildfires on summer water temperatures for burned stream sites that are sensitive to changes in atmospheric heat input.

4.2. Implications of Post-fire Summer Water Temperature Warming for Coldwater Fishes

The effect of wildfire-related water temperature changes in the western US on thermally sensitive fishes such as salmonids is still unclear. This is due to (a) the opportunistic nature of most studies severely limiting their temporal and spatial implications, (b) findings that vary widely across locations (Rieman et al., 2012), and (c) the diverse pathways through which wildfires can affect salmonid habitats, including flow velocity and sediment concentrations (Reale et al., 2021). Generally, stream water temperatures control the metabolic rate of salmonids (Lusardi et al., 2020). However, consequences of any post-fire water temperature warming and related increases in the metabolic rate on salmonid populations depend on background water temperature, stream connectivity, focal species, competition and prey availability, and other factors.

Some studies have shown that post-fire summer stream water temperature warmings led to population declines in coldwater trout in burned sites due to greater competition and increased metabolic demand (Beakes et al., 2014; Rosenberger et al., 2015). In contrast, others have found increased or no changes in coldwater salmonid abundance or growth in burned areas following elevated post-fire stream water temperatures (Dunham et al., 2007; Swartz and Warren, 2022; Warren et al., 2022). They hypothesized that the resilience of coldwater salmonid population following wildfires was linked to factors such as increased availability of food resources resulting from loss of shade and higher sunlight exposure, diel fluctuations with cooler water temperatures that allow fish to recover physiologically, evolutionary physiological and behavioral traits, post-fire increases in cold subsurface water, and presence of unburned tributary refugia (Sestrich et al., 2011; Rieman et al., 2012; Jager et al., 2021; Swartz and Warren, 2022; Warren et al., 2022).

Nonetheless, as large wildfires become increasingly common in the Pacific Northwest (PNW), the ability to categorize streams based on their summer thermal responsiveness to post-fire effects could prove to be a powerful tool for conservation and mitigation. Coldwater fish inhabiting streams with heightened post-fire thermal sensitivity face elevated risks due to wildfire occurrence, particularly as water temperatures approach a species thermal threshold as is the case with many salmonid species in the PNW. Similarly, streams exhibiting lower thermal responsiveness to post-fire effects may be more resilient to wildfire occurrence in their watershed and potentially less stressful to coldwater fishes because of smaller post-fire water temperature changes. However, our finding that prevailing post-fire weather conditions influence post-fire summer water temperature changes indicates that the thermal sensitivity of streams to wildfires may not be constant over time. Nevertheless, the relative disparity in the post-fire summer water temperature response among streams based on hydrogeology, climatology, and landcover variables could provide a preliminary stream sensitivity rankings. Such rankings could provide a basis for assigning levels of risk or for prioritizing sites for conservation or mitigation.

4.3. Study Limitations

During the attribution analysis, we assumed that anthropogenic influences on the pre- and post-fire stream thermal regime of studied watersheds were negligible due to (a) the absence of dams or hydropower plants upstream of our sites, and (b) urbanized and agricultural areas within selected watersheds were small (< 4% of watershed area). However, there are only a few US watersheds where human influences would be considered insignificant (Falcone et al., 2010). Additionally, human activities such as post-fire logging and other land cover modifications, stream water withdrawal, and discharging of effluents into streams were not accounted for in our analysis, and these might have affected stream water temperatures at some sites during the study period. Moreover, despite utilizing different approaches and models to characterize the weather-stream water temperature relationship and estimate wildfire contribution on post-fire water temperature changes, it is possible that the actual relations were still not captured for studied sites due to short study period, model limitations, covariates used, and non-stationarity in weather and stream thermal sensitivity. GLS models can only characterize linear relationships. Meteorological variables such as air temperature reflect large-scale atmospheric heat energy and not the local-scale atmospheric heat flux that affects stream water temperatures at the reach-scale. Nevertheless, this study is consistent with previous findings on post-fire effects on daily summer water temperatures as detailed above, and likely captures at least in part the central tendencies.

The PNW region covers about 100,000 km² of an area with extreme heterogeneity in climate, geology, land cover, and physiography. Moreover, wildfires burned over 4,000 km² of PNW wildland area just between 2010–2015 (Reilly et al., 2017). Clearly, not all fire affected stream sites were represented, nor all possible wildfire attributes captured as our inferences about the post-fire daily summer water temperature effects in the PNW relied on 31 burned sites. Moreover, given the opportunistic nature of our analysis, the post-fire period studied here only included the first 1–3 years post-fire. As such, our results only capture the initial daily summer water temperature responses after wildfire occurrences. Overall, our inferences about wildfire effects on summer stream water temperatures in the PNW have limitations both in their spatial and temporal scope. Nonetheless, this study represents the only other regional investigation, aside from Koontz et al. (2018), focused on wildfire effects on summer stream water temperatures in western North America. It includes more sites than any other regional study and employs more than one approach to quantify stream thermal response to wildfires both at individual and multi-site scales. As such, its findings can serve as the basis for future multi-site studies on wildfire impacts on stream water temperatures.

5. Conclusion

With the increased prevalence of high-severity wildfires and their potential to exacerbate stream warming trends across the Pacific Northwest, characterization of the spatial variability in the post-fire stream thermal responses across the region is critical for advancing stream ecosystem management beyond individual streams and basins. Here, we conducted an empirical study that evaluated the effect of wildfires in the PNW on daily summer water temperatures of 31 sites draining basins with 10–100% of their riparian area burned. Averaging across the 31 sites, wildfires were related to a 0.3 – 0.9°C increase in daily summer water temperatures for the subsequent 1–3 years. However, post-fire daily summer stream temperature responses were highly variable across burned sites. The likelihood of a post-fire warming of daily summer stream water temperatures increased with higher burned riparian area and severity. Moreover, direction and magnitude of post-fire water temperature responses across sites were strongly correlated to basin area, post-fire weather, winter snow-water equivalent, bedrock permeability, pre-fire forested and barren watershed area. In contrast, wildfire-related changes in daily stream water temperatures did not show distinct variation across ecoregions in the PNW.

The structured post-fire responses across burned sites observed in this study has important implications for better understanding the heterogeneity in post-fire stream temperature changes and water temperature modeling in the region. Improved predictions of post-fire stream thermal response will ultimately assist regional ecosystem managers in identifying streams whose temperatures are highly sensitive to wildfire effects. Additional research into the role of wildfires in contributing to observed water temperature warming trends will be useful for effective management and conservation of stream ecosystems in the region.

Data availability

Data will be made available on request.