Baseline Conditions and Projected Future Hydro-Climatic Change in National Parks in the Conterminous United States

: The National Park Service (NPS) manages hundreds of parks in the United States, and many contain important aquatic ecosystems and / or threatened and endangered aquatic species vulnerable to hydro-climatic change. More e ﬀ ective management of park resources under future hydro-climatic uncertainty requires information on both baseline conditions and the range of projected future conditions. A monthly water balance model was used to assess baseline (1981–1999) conditions and a range of projected future hydro-climatic conditions in 374 NPS parks. General circulation model outputs representing 214 future climate simulations were used to drive the model. Projected future changes in air temperature (T), precipitation ( p ), and runo ﬀ (R) are expressed as departures from historical baselines. Climate simulations indicate increasing T by 2030 for all parks with 50th percentile simulations projecting increases of 1.67 ◦ C or more in 50% of parks. Departures in 2030 p indicate a mix of mostly increases and some decreases, with 50th percentile simulations projecting increases in p in more than 70% of parks. Departures in R for 2030 are mostly decreases, with the 50th percentile simulations projecting decreases in R in more than 50% of parks in all seasons except winter. Hence, in many NPS parks, R is projected to decrease even when p is projected to increase because of increasing T in all parks. Projected changes in future hydro-climatic conditions can also be assessed for individual parks, and Rocky Mountain National Park and Congaree National Park are used as examples. These ratios indicate that the uncertainty in projections of 2030 p are small relative to baseline p values for all seasons.


Introduction
The National Park Service (NPS) mission is to preserve resources unimpaired for future generations [1]. Many NPS parks contain important aquatic ecosystems that may have threatened and/or endangered aquatic species vulnerable to hydro-climatic change. In addition to air temperature (T) and precipitation (p), runoff (R; streamflow per unit area in a drainage basin) is critical for sustaining these aquatic ecosystems. Projected future T and p may result in changes to hydrologic conditions including R that may alter park aquatic conditions and associated ecosystems [2]. Additionally, many parks rely on surface and subsurface water for park facilities and to provide water to visitors. Changes to water supplies resulting from hydro-climatic change could fundamentally challenge key park operations, as well as increase potential water-rights conflicts in water scarce regions. Identifying

Future Conditions
The MWBM application of the NHM was driven with statistically downscaled T and p from 95 GCMs (38 CMIP3 GCMs and 57 CMIP5 GCMs), representing 214 climate simulations [22] from 1950 to 2099. Statistical downscaling is a technique for deriving fine-scale interpolations from coarse-scale GCMs based on statistical relations between observed local-scale climate data, such as meteorological observations or gridded station data derived from historical climate observations, and the coarsescale GCM variables [33]. The GCM simulations used in this analysis were statistically downscaled using the bias-correction and spatial disaggregation (BCSD) methodology [33][34][35][36]. The 214 climate simulations were summarized using the GDP for the 109,951 HRUs from the geospatial fabric across the CONUS [29].
The 214 GCM climate simulations from CMIP3 and CMIP5 are composed of the Special Report on Emission Scenarios (SRES) B1, A1B, and A2 for CMIP3, and the Representative Concentration Pathways (RCPs) 4.5, 6, and 8.5 for CMIP5 [35,36]. These scenarios/pathways represent assumptions about future greenhouse gas emissions, accounting for short-and long-term climate cycles, and anthropogenic drivers such as changes in demographics and economic and technological development [37,38]. Climatic conditions represented by these scenarios range from stabilized populations after 2050, coupled with rapid development of more efficient technological systems across the globe (A1B, RCP4.5), to globally increasing populations and regionally oriented economic development (A2, RCP8.5). For a full description of the differences between the climate scenarios, Figure 1. Baseline (1981Baseline ( -1999 climatic conditions for seasonal average monthly temperature in degrees Celsius ( • C), precipitation in millimeters (mm), and runoff in mm. JFM, January-March; AMJ, April-June; JAS, July-September; OND, October-December.

Future Conditions
The MWBM application of the NHM was driven with statistically downscaled T and p from 95 GCMs (38 CMIP3 GCMs and 57 CMIP5 GCMs), representing 214 climate simulations [22] from 1950 to 2099. Statistical downscaling is a technique for deriving fine-scale interpolations from coarse-scale GCMs based on statistical relations between observed local-scale climate data, such as meteorological observations or gridded station data derived from historical climate observations, and the coarse-scale GCM variables [32]. The GCM simulations used in this analysis were statistically downscaled using the bias-correction and spatial disaggregation (BCSD) methodology [32][33][34][35]. The 214 climate simulations were summarized using the GDP for the 109,951 HRUs from the geospatial fabric across the CONUS [28].
The 214 GCM climate simulations from CMIP3 and CMIP5 are composed of the Special Report on Emission Scenarios (SRES) B1, A1B, and A2 for CMIP3, and the Representative Concentration Pathways (RCPs) 4.5, 6, and 8.5 for CMIP5 [34,35]. These scenarios/pathways represent assumptions about future greenhouse gas emissions, accounting for short-and long-term climate cycles, and anthropogenic drivers such as changes in demographics and economic and technological development [36,37]. Climatic conditions represented by these scenarios range from stabilized populations after 2050, coupled with rapid development of more efficient technological systems across the globe (A1B, RCP4.5), to globally increasing populations and regionally oriented economic development (A2, RCP8.5). For a full description of the differences between the climate scenarios, consult IPCC [36] for CMIP3, and Taylor et al. [37] for CMIP5. Appendices 1 and 2 from Bock et al. [22] list the 214 GCM climate simulations.

NPS Study Areas
The NPS manages 419 parks (as of 2019) [1]. Of these, 374 parks (89%) were selected for this study ( Figure 2 and Table S1). NPS parks were excluded if they were located outside of the CONUS, located offshore, included a substantial coastal component, or straddled an international border because the Geospatial Fabric incompletely covered these areas. Unfortunately, this excludes some important parks such as Everglades National Park and Isle Royale National Park for which more specialized analysis would be required.  [17] summarized projections of future T, p, and R from the MWBM as seasonal changes (departures) from each climate simulation's historical GCM baseline (1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999). The GCM baselines are unique for each GCM simulation and are not the same as the MWBM baseline values as described above in Section 2.2. Thus, all projected values of T, p, and R from the MWBM were expressed as a departure from the T, p, and R GCM baselines for each respective climate simulation for each of the 109,951 HRUs. The seasons were defined as winter (January-March (JFM)), spring (April-June (AMJ)), summer (July-September (JAS)), and fall (October-December (OND)). Departures of T, p, and R from the historical GCM baseline were expressed as average monthly seasonal departures for three future periods, 19 years in length, centered on 2030 (2021-2039), 2060 (2051-2069), and 2090 (2081-2099).

NPS Study Areas
The NPS manages 419 parks (as of 2019) [1]. Of these, 374 parks (89%) were selected for this study ( Figure 2 and Table S1). NPS parks were excluded if they were located outside of the CONUS, located offshore, included a substantial coastal component, or straddled an international border because the Geospatial Fabric incompletely covered these areas. Unfortunately, this excludes some important parks such as Everglades National Park and Isle Royale National Park for which more specialized analysis would be required. Outputs from the MWBM application of the NHM (baseline conditions based on the Maurer data [32] and future conditions) were summarized for selected NPS parks located within the CONUS. Area-weighted transfer functions were used to calculate park summaries of baseline and projected future hydro-climatic conditions (T, p, and R) for all HRUs that are wholly, or in part, within park boundaries for the 374 NPS parks in this study. Some parks fall completely within a single HRU and were represented by that single HRU in this study. Outputs from the MWBM application of the NHM (baseline conditions based on the Maurer data [31] and future conditions) were summarized for selected NPS parks located within the CONUS. Area-weighted transfer functions were used to calculate park summaries of baseline and projected future hydro-climatic conditions (T, p, and R) for all HRUs that are wholly, or in part, within park boundaries for the 374 NPS parks in this study. Some parks fall completely within a single HRU and were represented by that single HRU in this study.

Results
The following sections examine the seasonal changes in average monthly T, p, and R for the CONUS and also assess baseline and seasonal changes for selected NPS parks. The changes in T are presented in degrees Celsius ( • C). The changes in p and R are presented in millimeters (mm). Section 3.1 reviews the hydro-climatic changes for the CONUS based on the results presented in the Hay [17] data release. These results are presented for 2030, 2060, and 2090 from 214 climate simulations as departures (by model output percentile) and ranges of departures. Section 3.2 describes how these hydro-climatic changes can be assessed locally using the NPS parks as an example.

Hydro-Climatic Change in the CONUS
Figures 3-5 show the departures of T, p, and R from the historical GCM baselines [17]. These departures as shown by model output percentile for 2030, 2060, and 2090 and are mapped to illustrate the spatial distributions of changes from the 214 climate simulations. T increases for all percentiles and for all periods (2030-2090, Figure 3). For most seasons and periods, there also seems generally to be greater warming in the interior CONUS compared with coastal regions. By 2030, the 50th percentile T departures are between 1 and 3 • C, with the 5th percentile departures between 0 and 2 • C, and the 95th percentile departures between 3 and 5 • C. By 2060, the 50th percentile T departures are larger than 3 • C, with the 5th percentile departures between 0 and 3 • C, and the 95th percentile departures larger than 5 • C for most of the CONUS for all seasons. By 2090, the 50th percentile T departures are generally 4 • C or larger, with the 5th percentile departures generally between 1 and 3 • C, and the 95th percentile departures larger than 6 • C for much of the CONUS for all seasons.     In contrast to the consistently positive T departures (Figure 3), the patterns in p departures indicate large uncertainty in future p, showing a wide range in future projections. The 5th percentile shows decreases in p for most of the CONUS that shift to increases in p for the 95th percentile projections ( Figure 4). In 2030, 2060, and 2090, the 50th percentile p departures are positive in all four seasons for most of the CONUS (Figure 4), which means that more than half of the climate simulations project increases in p in all seasons over most of the CONUS.
In general, projected departures in R ( Figure 5) are smaller in magnitude than the projected departures in p ( Figure 4). Similar to the patterns of departures in p, the R departure patterns indicate large uncertainty in future R. The 5th percentile shows decreases in R for most of the CONUS, whereas the 95th percentile projections of R are increases ( Figure 5). In 2030, 2060, and 2090, the 50th percentile R departures are negative in three of four seasons for most of the CONUS, with the exception of the JFM (winter) season, which means that more than half of the climate simulations project decreases in R over most of the CONUS except in winter, which is in contrast to the projected increases in p for most of the CONUS.  The range in MWBM outputs produced using the 214 GCM climate simulations provides an indication of the uncertainty or variation in plausible future climatic projections. Figure 6 shows the range (95th percentile − 5th percentile) of departures in T, p, and R projected for 2030, 2060, and 2090 from 214 climate simulations mapped for the CONUS.
By 2030, the range of projected departures in T is generally less than 4 • C. The largest ranges in T departures are for the JFM season in the north central CONUS and the smallest ranges in T departures are for the eastern and western portions of the CONUS. The range in T departures increases from 2030 to 2060 and 2090, with the smallest ranges remaining in the eastern and western portions of the CONUS. The increasing range in T departures between 2030 and 2090 indicates increasing uncertainty in projected T into the future ( Figure 6).
Compared to T, the ranges of projected departures in p are highly variable across the CONUS ( Figure 6). A consistent pattern for 2030, 2060, and 2090 is observed, with the largest range in p departures for western portions of the CONUS for JFM and OND, and the mid-Atlantic portion of the CONUS for all seasons to a varying spatial degree. The increasing range in p departures between 2030 and 2090 indicates increasing uncertainty in projected p into the future ( Figure 6). exception of the JFM (winter) season, which means that more than half of the climate simulations project decreases in R over most of the CONUS except in winter, which is in contrast to the projected increases in p for most of the CONUS.
The range in MWBM outputs produced using the 214 GCM climate simulations provides an indication of the uncertainty or variation in plausible future climatic projections. Figure 6 shows the range (95th percentile − 5th percentile) of departures in T, p, and R projected for 2030, 2060, and 2090 from 214 climate simulations mapped for the CONUS. The ranges of projected departures in R also are variable across the CONUS with generally smaller values than for p ( Figure 6). The largest range in R departures is in JFM in the Pacific northwestern and mid-Atlantic portions of the CONUS. Areas of the CONUS with the largest range in R departures do not necessarily correspond (either spatially or seasonally) to the areas with largest range in p departures. The increasing range in R departures between 2030 and 2090 is not as apparent as for T and p ( Figure 6) perhaps due to the large range in baseline R magnitudes ( Figure 1).

Ratio of the Range of Projected Changes (2030, 2060, and 2090) to Baseline
The absolute magnitude of departures in projected future p and R, and the magnitude of the range in these projected departures may not provide the best indication of potential aquatic and ecosystem vulnerability. This is due, in part, to the confounding influence of the strong correlations between baseline p and R and the respective magnitudes of projected ranges in projected p and R departures. An alternative indicator may be the ratio of the ranges in projected departures from baseline climatic conditions to those baseline conditions (calculated as [95th percentile − 5th percentile]/baseline). Larger ratios (more than 1) indicate that the range in projected departures for an area is larger than the baseline condition, demonstrating greater potential importance of those changes than for an area where this value is smaller (less than 1). For example, if the range in model projections for R were 20 mm and the baseline R were 20 mm, then that variation in model outputs could be important to consider; however, if the range were the same and the baseline R were 200 mm, then that variation in model outputs would be less important to consider.
The ratio of the ranges in project departures in p and R to baseline conditions across the CONUS for 2030, 2060, and 2090 are shown on Figure 7. By 2030, for most of the CONUS, the ratio for p is less than 0.5 which indicates that the uncertainty in projected changes in p are less than half the magnitude of baseline p. The ratio is greater than 1 for p only during the summer (JAS) months in 2060 and 2090 in parts of the arid western U.S. In contrast, the ratio is greater than 1 for R in some locations across the CONUS in all seasons starting in 2030 ( Figure 7). Hence, while in many areas the magnitudes of R departures are relatively small (Figure 6), the range (or uncertainty in model projections) is larger than the baseline conditions (Figure 7), indicating that it could have great potential importance to local hydro-climatic conditions.

Hydro-Climatic Change in the National Parks
Ranges in hydro-climatic changes experienced by parks in different parts of the CONUS have very different consequences depending on the initial (or baseline) climatic conditions. The following

Hydro-Climatic Change in the National Parks
Ranges in hydro-climatic changes experienced by parks in different parts of the CONUS have very different consequences depending on the initial (or baseline) climatic conditions. The following sections put these projected changes in perspective of the NPS parks by comparing them with local NPS park baseline conditions.

Baseline Climatic Conditions (1981-1999)
Average baseline (1981-1999) climatic conditions for T, p, and R for each season for each of the 374 NPS parks were calculated by area weighting the appropriate HRU values. The results are summarized in Table 1 and provided for each NPS park in Tables S2-S5. Baseline T ranged from a minimum of −11.5 • C (coldest park in coldest season) to a maximum of 40.5 • C (warmest park in warmest season). Baseline p ranged from a minimum of 15.3 mm (driest park in driest season) to a maximum of 414 mm (wettest park in wettest season). Baseline R ranged from a minimum of 0.30 mm (least runoff in any park and season) to a maximum of 313 mm (most runoff in any park and season) ( Table 1). As indicated by the CONUS results, uncertainty in the projections of T, p, and R increases from 2030 to 2090. The 2030 (i.e., 2021-2039) period is of immediate concern to the NPS, given planning processes generally focus on the next 5-20 years [38]. Therefore, the following sections present projected departures in T, p, and R for parks specifically for the 2030 period. Departures in seasonal average monthly T, p, and R by percentiles (5th, 25th, 50th, 75th, and 95th) projected for 2030 from 214 climate simulations for each of the 374 NPS parks were calculated by area weighting the appropriate HRU values. These departures from GCM baseline conditions are summarized by season in Table 2 and provided for each NPS park in Tables S2-S5. The dominant patterns of projected 2030 changes in T for NPS parks were increases for all seasons for almost all percentiles (Tables S2-S5). Medians of the 50th percentile T departures in 2030 for the 374 parks were all increases ranging from 1.67 • C in JFM to 2.16 • C in JAS ( Table 2). The 50th percentile climate simulations used in this study projected increases in temperature of 1.67 • C or more in 52% of parks in JFM, 68% of parks in AMJ, 80% of parks in JAS, and 72% of parks in OND (Tables S2-S5). More extreme projections of change from the 5th and 95th percentile models ranged from a minimum median T departure of 0.38 • C to a maximum median T departure of 3.42 • C ( Table 2).
The patterns of projected 2030 changes in p for NPS parks for each season indicated a mix of increases and decreases (Tables S2-S5). Medians of the 50th percentile p departures in 2030 for the 374 parks were all increases ranging from 2.92 mm in JAS to 5.18 mm in JFM ( Table 2). The 50th percentile climate simulations projected increases in precipitation in 93% of parks in JFM, 74% of parks in AMJ, 80% of parks in JAS, and 94% of parks in OND (Tables S2-S5). More extreme projections of change from the 5th and 95th percentile models ranged from a minimum median p departure of −13.2 mm to a maximum median p departure of 25.0 mm ( Table 2).
The patterns of projected 2030 changes in R for NPS parks for each season indicate a mix of increases and decreases that were generally smaller in magnitude than projected changes in p (Tables S2-S5). Medians of the 50th percentile departures in R in 2030 for the 374 parks ranged from −0.75 mm in AMJ to 1.80 mm in JFM ( Table 2). The 50th percentile climate simulations projected decreases in R in 22% of parks in JFM, 82% of parks in AMJ, 63% of parks in JAS, and 52% of parks in OND (Tables S2-S5). More extreme projections of change from the 5th and 95th percentile models ranged from a minimum median R departure of −12.3 mm to a maximum median p departure of 16.1 mm ( Table 2). (2030) The projected ranges (95th percentile − 5th percentile) of seasonal departures in T, p, and R in 2030 for the 374 parks are summarized in Table 3 and provided for each NPS park in Tables S2-S5. The projected range of T, p, and R departures ( Figure 6) and the ratio to baseline conditions ( Figure 7) allow park managers to assess the potential changes a park may experience, with a wider range or larger ratio indicating greater uncertainty in future conditions and greater potential importance of departures to local hydro-climatic conditions. For example, these ranges and ratios could be used to construct divergent scenarios of plausible future hydro-climatic conditions a park may experience to evaluate vulnerability for a given set of resources with known climate sensitivities [39] and against which they can assess and test adaptation planning strategies. Medians of the 2030 ranges in seasonal T departures for the 374 parks were between 2.23 • C in AMJ and 2.64 • C in JFM (Table 3) and do not vary much by season. The minimum 2030 range in seasonal T departures was 1.05 • C and the maximum range was 4.44 • C. The small variation in ranges (maximum minus minimum, Table 3) of seasonal T departures indicate relative certainty in projections of future T.

Range of Projected Changes
Medians of the 2030 ranges in seasonal p departures for the 374 parks were between 33.1 mm in JFM and 38.7 mm in AMJ (Table 3) and do not vary much by season. The minimum 2030 range in seasonal p departures was 7.11 mm and the maximum range was 103 mm. The large variation in ranges of seasonal p departures indicates large uncertainty in projections of future p.
Medians of the 2030 ranges in seasonal R departures for the 374 parks were between 9.90 mm in JAS and 27.1 mm in JFM (Table 3) and they vary substantially by season. The minimum 2030 range in seasonal R departures was 0.33 mm and the maximum range was 121 mm. Projections of R incorporate the uncertainty in projections of both T and p. The larger variation in ranges (maximum minus minimum, Table 3) of seasonal R departures indicates greater uncertainty in projections of future R than for future p.

Ratio of the Range of Projected Changes (2030) to Baseline Conditions
The ratios of the projected ranges in future p and R departures to their respective baselines by season for 2030 are summarized in Table 4 and provided for each NPS park in Tables S2-S5. The median seasonal ratios of the projected range in p departures to baseline p were between 0.28 and 0.34 and do not vary much by season ( Table 4). The minimum seasonal ratio was 0.18 and the maximum seasonal ratio was 0.66; hence, no parks had p ratios that exceeded 1, and the range in ratio values was relatively small ( Table 4). The median seasonal ratios of baseline R to the projected range in R departures were between 0.50 and 0.80 and vary some by season ( Table 4). The minimum seasonal ratio was 0.22, and the maximum seasonal ratio was 4.03. Forty-five parks (12%) had R ratios greater than 1 in JFM, 7 parks (1.9%) had R ratios greater than 1 in AMJ, 30 parks (8%) had R ratios greater than 1 in JAS, and 112 parks (30%) had ratios greater than 1 in OND (Tables S2-S5).

Comparison of Projected Changes in Precipitation and Runoff
A comparison of projected changes in p to changes in R provides an indication of where future changes in hydro-climatic conditions may affect park resources. Figure 8 summarizes the output from the 214 climate simulations and indicates how many parks fall into one of four categories: (1) more than half of climate simulations projected increasing p and R; (2) more than half of climate simulations projected increasing p and decreasing R; (3) more than half of climate simulations projected decreasing p and R; and (4) more than half of climate simulations projected decreasing p and increasing R.
In JFM, the 50th percentile climate simulations projected increasing p in 346 NPS parks (92.5%) and decreasing p in 28 parks (7.5%) (Table S2). In parks with the 50th percentile climate simulations projecting increases in JFM p, the 50th percentile climate simulations projected increasing R in 286 parks and decreasing R in 60 parks (Figure 8). Parks where JFM 2030 p and R are both projected to increase tend to occur in the northern CONUS, whereas parks where JFM 2030 p is projected to increase and R is projected to decrease occur in the southern CONUS. In parks with the 50th percentile climate simulations projecting decreases in JFM p, the 50th percentile climate simulations projected decreasing R in 21 parks and increasing R in 7 parks. All parks with the 50th percentile climate simulations projecting decreases in JFM p are in the southern or southwestern CONUS (Figure 8).
In AMJ, the 50th percentile climate simulations projected increasing p in 276 NPS parks (60.2%) and decreasing p in 98 parks (39.8%) (Table S3). In parks with the 50th percentile climate simulations projecting increases in AMJ p, the 50th percentile climate simulations projected increasing R in 51 parks and decreasing R in 225 parks (Figure 8). Parks where AMJ 2030 p and R are both projected to increase tend to occur along the eastern seacoast and in the north-central CONUS, whereas parks where AMJ 2030 p is projected to increase and R is projected to decrease occur broadly across the CONUS. In parks with the 50th percentile climate simulations projecting decreases in AMJ p, the 50th percentile climate simulations projected decreasing R in 83 parks and increasing R in 15 parks. All parks with the 50th percentile climate simulations projecting decreases in AMJ p are in the western or southern CONUS (Figure 8). In JFM, the 50th percentile climate simulations projected increasing p in 346 NPS parks (92.5%) and decreasing p in 28 parks (7.5%) (Table S2). In parks with the 50th percentile climate simulations projecting increases in JFM p, the 50th percentile climate simulations projected increasing R in 286 parks and decreasing R in 60 parks (Figure 8). Parks where JFM 2030 p and R are both projected to increase tend to occur in the northern CONUS, whereas parks where JFM 2030 p is projected to increase and R is projected to decrease occur in the southern CONUS. In parks with the 50th percentile climate simulations projecting decreases in JFM p, the 50th percentile climate simulations projected decreasing R in 21 parks and increasing R in 7 parks. All parks with the 50th percentile climate simulations projecting decreases in JFM p are in the southern or southwestern CONUS (Figure 8).
In AMJ, the 50th percentile climate simulations projected increasing p in 276 NPS parks (60.2%) and decreasing p in 98 parks (39.8%) (Table S3). In parks with the 50th percentile climate simulations In JAS, the 50th percentile climate simulations projected increasing p in 299 NPS parks (79.9%) and decreasing p in 75 parks (20.1%) (Table S4). In parks with the 50th percentile of climate simulations projecting increases in JAS p, the 50th percentile climate simulations projected increasing R in 137 parks and decreasing R in 162 parks (Figure 8). Parks where JAS 2030 p and R are both projected to increase tend to occur in the eastern and southwestern CONUS, whereas parks where JAS 2030 p is projected to increase, and R is projected to decrease occur broadly across the CONUS. In parks with the 50th percentile climate simulations projecting decreases in JAS p, the 50th percentile climate simulations projected decreasing R in 72 parks and increasing R in 3 parks. The parks with the 50th percentile climate simulations projecting decreases in JAS p are in the central and western CONUS (Figure 8).
In OND, the 50th percentile climate simulations projected increasing p in 351 NPS parks (93.9%) and decreasing p in 23 parks (6.1%) (Table S5). In parks with the 50th percentile climate simulations projecting increases in OND p, the 50th percentile climate simulations projected increasing R in 163 parks and decreasing R in 188 parks (Figure 8). Parks where OND 2030 p and R are both projected to increase occur broadly across the CONUS, as do parks where OND 2030 p is projected to increase and R is projected to decrease. In parks with the 50th percentile climate simulations projecting decreases in OND p, the 50th percentile climate simulations projected decreasing R in 9 parks and increasing R in 14 parks. The parks with the 50th percentile climate simulations projecting decreases in OND p are in the southwestern CONUS (Figure 8).
In selected NPS parks, the seasonal interaction between p and T and processes, such as T driven evapotranspiration or changes in albedo effect [40], can result in unexpected relations between changes in p and changes in the magnitude and timing of R. Processes included in the MWBM, including snow accumulation and melt, soil-moisture storage, and potential evapotranspiration, can result in a variety of relations between p and R. For example, more than half of the climate simulations projected 2030 p to increase but R is projected to decrease in 16% of parks in JFM, 60% of parks in AMJ, 43% of parks in JAS, and 50% of parks in OND. When considering all parks and seasons together, the 50th percentile climate simulations project increasing p and R in 637 park-seasons (42.6%) (374 parks × 4 seasons = 1496 park-seasons), increases in p and decreasing R in 635 park-seasons (42.4%), decreasing p and R in 185 park-seasons (12.4%), and decreasing p and increasing R in 39 park-seasons (2.6%). Hence, in many NPS parks, R is projected to decrease even when p is projected to increase in part because of increasing T in all parks [25].

Comparison of Projected Hydro-Climatic Changes in the CONUS to Changes in National Parks
Analysis by Gonzalez et al. [11] indicated that NPS lands in CONUS experienced statistically significant warming, while p did not change significantly, over the historical period 1895-2010.
Their results indicate that future T is projected to increase in almost all areas, rising by between 1.7 and 5.0 • C per century in CONUS and between 1.6 and 4.9 • C per century in CONUS NPS parks for the period 2000-2100. Their analysis of projections of future p indicates a mixture of increases and decreases with the total land area of the CONUS showing increases between 5% and 7% per century, and the land area of parks in CONUS showing increases between 6% and 7% per century for the period 2000-2100 [11]. Those results are largely in agreement with results from this study but do not include estimates of historical or future R. Figure 9 uses boxplots to compare the projected ranges (95th percentile − 5th percentile) of departures in T, p, and R for 2030 from the 214 climate simulations for all HRUs in the CONUS, with only the HRUs in the 374 selected NPS parks. The projected ranges in the 2030 departures for the HRUs associated with the 374 parks are not visibly different from the projected ranges of departures for all HRUs in the CONUS (Figure 9), indicating the range of uncertainty in seasonal departures for T, p, and R are similar for CONUS NPS parks and the CONUS as a whole.
(95th percentile-5th percentile) of departures in T, p, and R for 2030 from the 214 climate simulations for all HRUs in the CONUS, with only the HRUs in the 374 selected NPS parks. The projected ranges in the 2030 departures for the HRUs associated with the 374 parks are not visibly different from the projected ranges of departures for all HRUs in the CONUS (Figure 9), indicating the range of uncertainty in seasonal departures for T, p, and R are similar for CONUS NPS parks and the CONUS as a whole. The line across the box shows the median projected range. The whiskers extend to the largest value that is less than the 75th percentile value plus 1.5 times the interquartile range (75th percentile-25th percentile) and to the smallest value that is greater than the 25th percentile value minus 1.5 times the interquartile range. Outliers are shown as circles. The line across the box shows the median projected range. The whiskers extend to the largest value that is less than the 75th percentile value plus 1.5 times the interquartile range (75th percentile-25th percentile) and to the smallest value that is greater than the 25th percentile value minus 1.5 times the interquartile range. Outliers are shown as circles.

Uncertainty in Projections of Temperature, Precipitation, and Runoff
As noted above, the same projected departure in T, p, or R can have substantially different effects on local hydro-climatic conditions depending on the baseline climatic conditions. Figure 10 compares baseline conditions to projected ranges (95th percentile − 5th percentile) of departures in seasonal average monthly T, p, and R for 2030 from the 214 climate simulations for all HRUs in the CONUS (black dots) and NPS parks in the CONUS (green dots). Additional colored dots indicate individual NPS parks that will be discussed later (ROMO and CONG). The green dots do not appear to fall as outliers in any of the plots indicating that the NPS parks in this study have ranges of uncertainty in T, p, and R that are similar to those seen for the CONUS as a whole.
A comparison of the projected ranges in T, p, and R with baseline climatic conditions demonstrate a relation of increasing uncertainty with greater baseline values for p and R, but less so for T. Examining the NPS parks alone (green dots in Figure 10), there was limited correlation (Pearson's correlation) between the baseline T in NPS parks and the magnitude of the projected range in 2030 T departures by season (−0.42 in JFM, 0.16 in AMJ, 0.39 in JAS, and −0.10 in OND), and only correlations for JFM and JAS were significant at the p < 0.01 level. These weak correlations indicate a slightly greater range in model projections of future T for NPS parks with colder baseline temperatures in JFM and slightly smaller range in model projections of future T in parks with colder baseline temperatures in JAS (Figure 10). The larger the range in the GCM projections in JFM and OND indicates there is increased uncertainty associated with projections in snowmelt-dominated watersheds due to larger ranges of projected cold regime temperatures.
Water 2020, 12, x FOR PEER REVIEW 19 of 28 As noted above, the same projected departure in T, p, or R can have substantially different effects on local hydro-climatic conditions depending on the baseline climatic conditions. Figure 10 compares baseline conditions to projected ranges (95th percentile -5th percentile) of departures in seasonal average monthly T, p, and R for 2030 from the 214 climate simulations for all HRUs in the CONUS (black dots) and NPS parks in the CONUS (green dots). Additional colored dots indicate individual NPS parks that will be discussed later (ROMO and CONG). The green dots do not appear to fall as outliers in any of the plots indicating that the NPS parks in this study have ranges of uncertainty in T, p, and R that are similar to those seen for the CONUS as a whole.  Instances where the 2030 range of projected departures approaches the magnitude of baseline conditions may indicate locations of relatively large uncertainty in future conditions. The dotted red lines on the p and R plots in Figure 10 show where the 2030 range of projected departures would equal the baseline value. All HRU and park values for p fall below this line indicating that the range in p departure projections are always less than the actual baseline p, but the uncertainty (range) in p projections increases for all seasons with increasing baseline p. Examining the NPS parks alone (green dots in Figure 10), there were significant positive correlations between baseline p and the magnitude of projected ranges in 2030 p departures by season (0.91 in JFM, 0.90 in AMJ, 0.84 in JAS, and 0.91 in OND), and all correlations were significant at the p < 0.01 level. These strong correlations indicate a larger range (or uncertainty) in projected future p in NPS parks with larger baseline p ( Figure 10). In contrast to p, some of the HRU and park R values plot above the red line ( Figure 10). For these locations, the range in projections of R departures are greater than the baseline values indicating a greater potential importance of the R changes in those parks. In NPS parks there was strong correlation between baseline R and the magnitude of the projected range in 2030 R departures by season (0.90 in JFM, 0.91 in AMJ, 0.94 in JAS, and 0.91 in OND) and all correlations were significant at the p < 0.01 level. These strong correlations indicate a larger range (or uncertainty) in projected future R departures for NPS parks with larger baseline R (Figure 10).

Example by-Park Interpretations
Each national park is unique, with unique susceptibilities to potential changes in climate. Providing information on potential hydro-climatic changes by NPS parks provides managers with information that can help to prioritize locations for resource conservation, an important step in the hydro-climatic change adaptation process [41]. The following sections present some interpretation of hydro-climatic changes in two example NPS parks: Rocky Mountain National Park and Congaree National Park. These two parks were selected because they represent different hydrologic conditions (headwaters versus flow-through) and hydro-climatic conditions (snowmelt dominated versus rainfall dominated). The ranges of projected changes in seasonal average T, p, and R for 2030 for HRUs in ROMO by percentiles from 214 climate simulations are shown as departures from baseline conditions in Figure 11. The spread of these boxplots shows the range in projected departures for individual HRUs in ROMO. All 214 climate simulations project increases in T in all seasons, and there is little variability in projected departures in T by HRU. The 50th percentile climate simulations project T increases of approximately 2 • C or more in all four seasons with the largest increases projected for JAS ( Figure 11).
The 50th percentile climate simulation projects increases in p in 2030 in all four seasons (hence more than half of climate simulations project increases in p in all HRUs). In the more extreme models (5th and 95th percentiles), there can be as much as a 10 mm range in the projected p departures for different HRUs within ROMO, which is likely a result of the large elevation driven range in baseline p amounts ( Figure 11). The 2030 ratios of the projected range in p departures to baseline p in ROMO were less than 0.35 in all four seasons (Tables S2-S5). These ratios indicate that the uncertainty in projections of 2030 p are small relative to baseline p values for all seasons. mm (JFM), 84.9 mm (AMJ), 34.6 mm (JAS), and 13.5 mm (OND).
The ranges of projected changes in seasonal average T, p, and R for 2030 for HRUs in ROMO by percentiles from 214 climate simulations are shown as departures from baseline conditions in Figure  11. The spread of these boxplots shows the range in projected departures for individual HRUs in ROMO. All 214 climate simulations project increases in T in all seasons, and there is little variability in projected departures in T by HRU. The 50th percentile climate simulations project T increases of approximately 2 °C or more in all four seasons with the largest increases projected for JAS ( Figure  11).  Projections of R for 2030 show large seasonal and within park variability. The 25th percentile climate simulations projects increases in R during JFM, whereas the 75th percentile climate simulation projects decreases in R during JAS (hence more than three quarters of climate simulations project decreases in all HRUs), and approximately 50% project increases in R during AMJ and OND ( Figure 11). The 2030 ratios of the projected range in R departures to baseline R in ROMO were about 1.04 in JFM, 0.40 in AMJ, 0.59 in JAS, and 1.05 in OND (Tables S2-S5). These ratios indicate that the uncertainty in projections of 2030 R departures are as large as baseline R values for the JFM and OND seasons.
These projections indicate that increases in T and increases in fall and winter p in ROMO may result in a temporal shift in R with more R in winter and less R in summer and with a large range in the model projections of R relative to baseline conditions in OND (fall) and JFM (winter). The wide range in projections of p and R in HRUs within ROMO (Figure 11) indicate that a more detailed analysis of projected changes that accounts for elevation bands within the park is merited.

Congaree National Park
Congaree National Park (CONG) consists of old-growth forested floodplains and wetlands of international importance [44] and approximately 106.3 km 2 of area in central South Carolina, USA.
Much of CONG is at low elevation, which ranges from 24 to 43 m above the NGVD 29, and the Congaree River forms the southern boundary of the park. CONG had approximately 146,000 visitors in 2018 [45]. All or parts of 21 HRUs fall within the boundary of CONG.
Spatially averaged baseline T in CONG varies substantially by season with values of about 7.5 • C in JFM, 20.1 • C in AMJ, 24.8 • C in JAS, and 10.8 • C in OND (Tables S2-S5). Spatially averaged baseline p in CONG was also variable by season with average monthly values of about 104 mm in JFM, 97.3 mm in AMJ, 113 mm in JAS, and 70.2 mm in OND. Spatially averaged baseline R in CONG was less variable by season than p, with average monthly values of about 38.8 mm in JFM, 19.5 mm in AMJ, 13.9 mm in JAS, and 14.0 mm in OND.
The ranges of projected changes in seasonal average monthly T, p, and R for 2030 for HRUs in CONG by percentiles from 214 climate simulations are shown as departures from historical conditions in Figure 12. All models project increases in T in all seasons, and there is little variability in projected departures in temperature by HRU. The 50th percentile climate simulations project temperature increases of approximately 1 • C or more in all four seasons with the smallest increases projected for JFM ( Figure 12). The spread of these boxplots shows the range in projected changes for individual HRUs in CONG. Water 2020, 12, x FOR PEER REVIEW 23 of 28 The 50th percentile climate simulations project increases in p in all seasons (hence, more than half of the climate simulations project increases in p in all seasons and all HRUs). There is not much difference in the projected departures for different HRUs within CONG which is likely a result of the lack of an elevation driven range in baseline p (see interquartile range in Figure 12). The 2030 ratios of the projected range in p departures to baseline p in CONG were less than 0.45 in all four seasons (Tables S2-S5)  The 50th percentile climate simulations project increases in p in all seasons (hence, more than half of the climate simulations project increases in p in all seasons and all HRUs). There is not much difference in the projected departures for different HRUs within CONG which is likely a result of the lack of an elevation driven range in baseline p (see interquartile range in Figure 12). The 2030 ratios of the projected range in p departures to baseline p in CONG were less than 0.45 in all four seasons (Tables S2-S5) These 2030 projections indicate that increases in T and increases in JAS p in CONG may not result in a large change in R; however, there is a large range in the model projections of R departures relative to baseline conditions in OND (fall). The narrow range in projections of p and R in HRUs within CONG ( Figure 12) indicate that a more detailed analysis of projected changes for individual HRUs within the park may not be merited.

Conclusions
Current and future CONUS wide projections of T, p, and R from an MWBM application of the NHM were analyzed to show how projections of hydro-climatic changes produced from 214 climate simulations can be interpreted for 374 NPS parks. The methodology provides decision makers with a range of plausible hydro-climatic futures to develop local adaptation strategies.
In the CONUS, T increased for all models in all seasons for each of the three time periods (2030, 2060, and 2090). Across most of the CONUS, the 50th percentile T departures increased to approximately 2-3 • C in 2030 and to more than 4 • C by 2090. In contrast to the consistently positive T departures, the patterns in p departures indicated large uncertainty and a wide range in future p projections. In 2030, 2060, and 2090, the 50th percentile p departures were positive in all four seasons, which means that more than half of the climate simulations projected increases in p in all seasons over most of the CONUS. Similar to patterns of departures in p, the R departure patterns indicated large uncertainty in future R projections. In 2030, 2060, and 2090 the 50th percentile R departures were negative in three of four seasons for most of the CONUS, with the exception of the JFM (winter) season, which means that more than half of the climate simulations projected decreases in R over most of the CONUS except in winter. The near future 2021-2039 period (centered on 2030) is within many NPS planning horizons and thus of priority interest. The 2030 projections indicate T increased for all parks with the 50th percentile climate simulations projecting increases of 1.67 • C or more in more than half of the parks in all seasons. The 2030 projections of p indicated a mix of mostly increases and some decreases in parks, with the 50th percentile climate simulations projecting increases in more than 70% of parks in all seasons. In contrast to p, departures in 2030 R indicated mostly decreases with some increases, and the 50th percentile climate simulations projected decreases in more than half of the parks in all seasons except winter (January-March). Hydro-climatic processes resulting from the interaction of T and p (e.g., snow accumulation and melt, soil-moisture storage, and potential evapotranspiration) can influence the relation between p and R, demonstrating that R trends cannot be predicted from p trends alone. When considering all parks and seasons together (374 parks × 4 seasons = 1496 park-seasons), the 50th percentile climate simulations projected increasing p and R in 42.6% of park-seasons, increasing in p and decreasing R in 42.4% of park-seasons, decreasing p and R in 12.4% of park-seasons, and decreasing p and increasing R in 2.6% of park-seasons.
The results indicate that similar magnitude hydro-climatic changes and ranges of changes can occur in several parks, but these changes can have very different consequences locally, depending on baseline climatic conditions. There was strong positive correlation between baseline p and R and the range in projections of changes in 2030 p and R indicating greater uncertainty in model projections in parks with larger baseline values of p and R.
Interpretation of results by park depends on the relative size of the HRUs associated with a park and park size. In larger parks with more topographic diversity, interpreting these results in terms of vulnerabilities for specific resources would require a finer, sub-park, analysis. In smaller parks or parks without distinct elevation gradients, whole park analyses may be appropriate. The work presented here provides each national park a range of plausible hydro-climatic futures that can be used to help assess the possible effects of future climatic conditions on park ecosystems and water dynamics related to park operations.