Detecting Change in Precipitation Indices Using Detecting Change in Precipitation Indices Using Observed (1977-2016) and Modeled Future Climate Observed (1977-2016) and Modeled Future Climate Data in Portland, Oregon, USA Data in Portland, Oregon, USA

This study addresses how regional changes to precipitation may be identi ﬁ ed by exploring the effect of temporal resolution on trend detection. Climate indices that summarize precipitation characteristics are used with Mann – Kendall monotonic testing to investigate precipitation trends in Portland, Oregon (OR) from 1977 to 2016. Observational records from rain gages are compared with downscaled global climate models to determine trends for the historic (1977 – 2005) and future (2006 – 2100) periods. Standard indices created by the Expert Team on Climate Change Detection and Indices (ETCCDI) are deployed. ETCCDI indices that summarize conditions at the annual level are generated alongside a limited number of ETCCDI indices summarized at the monthly level. For the future climate, the indices summarized at the annual level demonstrate trends indicative of an intensifying hydrologic cycle. The historical record depicted by annual indices does not show trends. The historical record is viewed differently by changing the indices to monthly summaries, which causes trend detection to increase and hallmark indicators of an intensifying hydrologic cycle to become apparent.

• We compared observed precipitation records (1977-2005) with future (2006-2100) projections from five downscaled global climate models for the Portland area.
• While four of five climate models projected an intensifying hydrologic cycle from 2006 to 2100, trends were not detected in the 1977-2016 observational record.
• When data were disaggregated from annual to monthly, many of the hallmarks of an intensified hydrologic cycle were observed in the 1977-2016 Portland record in spring, winter, and fall months.
• Trend detection of increasing precipitation intensity was detected more at a finer temporal scale (i.e., hourly data), indicating a finer temporal analysis is critical for urban flood risk management.

INTRODUCTION
As a consequence of rising levels of greenhouse gases, the global hydrologic cycle is likely to intensify (Huntington ).Changes to patterns of precipitation are confidently expected in part because a warmer atmosphere can hold additional water vapor, as described by the Clausius-Clapeyron model of gas behavior under conditions of temperature increase.The effects of temperature rise on precipitation can vary but include the possibilities for altered routes of water vapor transport in atmospheric circulation and different seasonal precipitation patterns (Held & Soden ; Wentz et al. ; Trenberth ).
Global shifts in precipitation regime are expected to already be occurring because global temperature rise is highly certain (IPCC ).An increase in atmospheric water vapor has been observed globally, and some studies suggest changes in tropical and sub-tropical rainfall cannot be explained without a global greenhouse gas (GHG) signal (Hense et al. ; Zhang et al. ).Many other precipitation trends have been detected around the world, but with varying periods of study and measurement techniques, these studies may be site-specific rather than connected by the underlying warming trend (Groisman et

).
It is important to consider regional and seasonal  (IPCC ).This is evidence of the need for regional studies and the inadequacy of using a single metric for the entire globe.
Many studies of regional precipitation exist but do not lend themselves to a clear narrative about climate change because of widely varied research methods and results reporting (Moberg et al. ).The development of common metrics for measuring change to precipitation and other weather variables can improve communication and cooperation.In response to such needs, scientific communities created a number of common indices.These indices include the Expert Team on Climate Change Detection and Indices (ETCCDI) and also the Expert Team on Sector-specific Climate Indices (ET-SCI), both of which emerged from the World Meteorological Organization (WMO) (Zhang et al. ).The European Climate Assessment & Dataset (ECA&D) created further indices (Klok & Klein Tank ).Many of the indices sets are appropriate for different contexts.Sector-specific indices were the basis for ET-SCI and may be applied to the industries of agriculture, health, and water resources.Not all indices may be appropriate for any region.Some drought indices may outperform others depending on the region, and evaluating fitness to the region is often appropriate (Shamshirband et al. ).
The ETCCDI indices are aimed at global ubiquity and have been adopted in studies of many regions.Reviews of regional studies employing ETCCDI indices shown in            5)) Annual total precipitation in wet days: Let RR ij be the daily precipitation amount on day i in period j.If i represents the number of days in j, then Runoff ratio Rx1day (Equation ( 6)) Monthly maximum 1-day precipitation: Let RR ij be the daily precipitation amount on day i in period j.The maximum 1-day value for period j are: Surface runoff rate Rx5day (Equation ( 7)) Monthly maximum consecutive 5-day precipitation: Let RR kj be the precipitation amount for the 5-day interval ending k, period j.Then maximum 5-day values for period j are: Surface runoff rate, recharge rate, soil moisture availability SDII (Hourly) (Equation ( 8)) Hourly simple precipitation intensity index: Let RR wj be the daily precipitation amount on wet days, w (RR !1 mm) in period j.If H represents number of wet hours in j, Surface runoff rate, recharge rate TXx (Equation ( 9)) Monthly maximum value of daily maximum temperature: Let TX x be the daily maximum temperatures in month k, period j.The maximum daily maximum temperature each month is then: Evapotranspiration rate, soil moisture availability TNn (Equation ( 10)) Monthly minimum value of daily minimum temperature: Let TN n be the daily minimum temperatures in month k, period j.The minimum daily minimum temperature each month is then: Evapotranspiration rate, soil moisture availability

Methods for Question 3: comparing daily versus and hourly data
For the last research question, the SDII index for rainfall intensity was examined using hourly data rather than daily data to determine the effect of high temporal resolution on trend detection.The ETCCDI formula for SDII uses daily rainfall totals, as given in Equation ( 4).The adapted formula to calculate SDII using hourly rainfall total is shown in Equation ( 8).
Two calculations of SDII were therefore performed.All trends were analyzed with the Mann-Kendall and bootstrapping approach described in Question 1 methods.The null hypothesis of trend tests is that no monotonic trends exist.

RESULTS
Comparisons with downscaled global climate models dictions that rainfall will be more intense in the Pacific Northwest.Figure 4 shows index trends from all models.
Overall CDD appears to increase with CWD models showing increase and decrease.The expected increase in CDD was significant in the CNRM-CM5 model (tau ¼ 0.19, p-value ¼ 0.01) and the HADGEM-ES model (tau ¼ 0.12, p-value ¼ 0.09).A significant increase in CWD was detected by the CanESM2 model (tau ¼ 0.15, p-value ¼ 0.04).In contrast, there was a decrease in CWD detected by the HADGEM2 (tau ¼ À0.17, p-value ¼ 0.02).
Historic simulations and observations using annual indices showed non-significant mixed direction trends.
However, in future projections, multiple indices showed an intensified hydrologic cycle.As shown in Table 5, trends from these projections were mostly positive but not uniformly significant.Only the GFDL-ESM2M model projections showed no significant trends.Results from autocorrelation testing showed that autocorrelation decreased as the period of study is decreased.

Results for
At the annual scale, autocorrelation in annual PCRPTOT and SDII values was detected.Since only PRCPTOT was significant, bootstrapping was used to check the range of      Improving the resolution is likely to improve the change that important regional and seasonal changes are detected.
In seeking tools that use standard methodologies and are easy to understand for scientists and practitioners alike, the changes in precipitation in order to identify emerging change to the hydrologic cycle.Changes in the seasonal distribution of precipitation could affect drinking water resources in regions where reservoirs rely on snowmelt filling (Barnett et al. ).Decreased rain in the arid regions of the Middle East and Northern Africa can lead to failed harvests and disruption of the agricultural sector (Dai ; Homsi et al. ).Although mean precipitation increased in mid-latitude land areas since 1951, trends at other latitudes are less confident trends were found in precipitation indices like rainfall intensity (SDII) and precipitation total (PRCPTOT), trends are not consistently positive or negative.There is a distinct lack of clarity on the mechanisms creating spatially incoherent trends.Increases to intensity (SDII) and maximum single-day rainfall (Rx1day), which can potentially affect the magnitude of runoff and the intensity of the hydrologic cycle, are present in many studies, but these trends are rarely consistent across the entire study area (Rahimzadeh et al. ; Dumitrescu et al. ).Temporal scale is another important feature in the study of precipitation, and current ETCCDI methodology does not dictate multiple scales of analysis.The RClimDex software widely employed by researchers to calculate ETCCDI indices is designed so that the period of study is annual, although monthly calculations are available for some indices (Zhang & Yang ).Many researchers have opted to conduct research on an annual or seasonal basis, and a few have used monthly (Table 1).Further, ETCCDI indices are based on the use of daily precipitation and temperature data.Considering that modern precipitation research can now look at sub-daily rain rates through innovations in remote sensing and computation, the use of the daily scale may be limiting (Munoz et al. ; Sanò et al. ).Further, the use of daily records obscures the fact that precipitation is a phenomenon occurring at a sub-daily scale (Trenberth ).Daily data have been shown to mask trends in precipitation intensity when compared to sub-daily data (Cooley & Chang ).It is possible that the use of annual time scales and daily data may contribute to the incoherent precipitation trends observed.One reason for the use of daily data in the ETCCDI indices is that the indices are designed to emphasize global collaboration and high-resolution records are not available in many regions (Peterson & Manton ).Further, ETCCDI indices are not only used for historical analysis but also for future projections made by global climate models.Although the resolution of climate models steadily increases, the ability of these models to deliver precipitation projections that resemble realistic precipitation at the hourly scale remain novel (Xu et al. ; Seneviratne et al. ; Prein et al. ).The atmospheric dynamics that lead to precipitation in the Pacific Northwest are expected to be altered by climate change (Chou et al. ; Rupp et al. a, b).The results of this include an intensification of convergence zones like the North Pacific storm track that transport water vapor to the Pacific Northwest (Salathé ).Atmospheric rivers that are responsible for many of the flood events in the region are anticipated to increase in duration and frequency (Dettinger ).A regional climate model indicates an increased magnitude of single-day rainfall events in the twenty-first century (Salathé et al. ).Elevation increase of snow lines may contribute to increased precipitation if a greater portion of moisture falls as rain rather than snow (Tohver et al. ).Annual precipitation has increased in the Pacific Northwest over the twentieth century, although whether this is attributable to climate change is debatable (Mote ).The seasonal distribution of the increased precipitation from climate change is expected to occur in winter, but most observed significant increasing trends are in spring (Abatzoglou et al. ).Global climate models predict annual mean precipitation changes of À10% to 20% by the 2080s for the Pacific Northwest (Mote ; Salathé ; Mote & Salathé ; Rupp et al. a, b).The seasonal distribution of precipitation is likely to skew towards winter months with summers becoming drier.Although many studies have investigated the long-term climate change in the Pacific Northwest, hydrologic studies of observational records have tended to focus on streamflow, snowpack, and modes of climate variability (Chang et al. ).El Nino and the Pacific Decadal Oscillation have been shown to have an important influence over temperature but only moderate influences on precipitation (Redmond & Koch ; Praskievicz & Chang ).A number of recent studies of precipitation have been conducted in British Columbia.Predictions of increased winter rainfall and decreased summer rainfall have not yet been born out in studies of observational records (Burn & Taleghani ).Instead, increased frequency of heavy precipitation events in summer has been observed while winter has mixed signals.Some indication of increased intensity in spring has also been observed ( Jakob et al. ).The nature and direction of trends are different across sub-regions, suggesting that different convective processes are creating diverse precipitation response.Examining the spatial extent of extreme precipitation in the Pacific Northwest, heterogeneous topography was found to be a key limiting factor on the spatial extent of extreme rainfall (Parker & Abatzoglou ).For this reason, closely related areas with different topography may have different responses to climate change, suggesting the need for a spatially explicit analysis.
Figure 1 | Map of the study area.
GHG forcing it uses relative to other CMIP5 scenarios provides a greater likelihood of detecting a GHG-forced signal.This scenario represents a future with a global population of 12 billion in 2100, where minimal gains from technological advances in energy efficiencies occur (Riahi et al. ).Selecting a study with weaker GHG forcing is less likely to show hydrologic change and less likely to represent actual conditions.RCP 8.5 is considered the most probable future considering the current rate of GHG emissions (Jackson et al. ).Historical temperature records were 100% complete and were acquired from National Climatic Data Center (NCDC) Station 356751 located at Portland International Airport (PDX).Daily maximum and minimum temperature were obtained for 1977-2016.Methods Methods for Question 1: comparisons with downscaled climate models Changes to the hydrologic cycle from climate change were hypothesized to show in aspects of the GHG-forced climate record.Precipitation indices that measure characteristics of rainfall were selected from ETCCDI.Giorgi et al. () predicted that consecutive dry days (CDD) would increase and that rainfall events would be shorter, more intense, and wetter (Giorgi et al. , ).Figure 2 shows the selected indices and hypotheses about the future state of these variables.Annual indices include simple daily rainfall intensity (SDII), maximum CDD, maximum consecutive wet days (CWD), and total days of precipitation above the 95th percentile (R95p).SDII, R95pTOT, CDD increase while CWD decreases.Table 3 presents the equations for these as defined by ETCCDI.Indices were selected based on a focus on regional precipitation and storm duration and intensity.These parameters are relevant to wastewater and stormwater engineering because of their use in design storm standards.This study presents less focus on indices, such as temperature that could be used to study the physical mechanisms driving hydrologic change.The RClimDex software created by ETCCDI was used to generate the selected indices (Zhang & Yang ).The RClimDex software is designed to run parallel to the statistical software R. Input climate variables to the RClimDex software include daily precipitation, maximum temperature, and minimum temperature.The software uses these inputs to generate a unique index value for every year of the study.Alternatives to the RClimDex include ClimPact2 that also calculates several of the same indices.ClimPact2 is a more recently built software package and has the advantage of a new user interface and documentation.However, ClimPact2 is less well established in the literature.The
Figure 3 | Trends in annual indices from observations (a) and climate models (b) for 1977-2005.
Question 2: comparing annual and monthly trends Analysis of the observational data for 1977-2016 using annual indices PRCPTOT, Rx1day, Rx5day, and SDII indicate consistent increasing trends observed in only PRCPTOT, but no other precipitation indices (Figure 5).Temperature minimum index TNn is increasing and barely nonsignificant (tau ¼ 0.18, p-value ¼ 0.12).Temperature maximum index TXx is nonsignificantly decreasing (tau ¼ À0.12, p-value ¼ 0.28).When examining precipitation indices on the monthly scale, a different picture emerges.Rx1day is increasing and significant in January and March at the 5% significance level (Figure 6(a)).Rx5day is increasing and significant in January, March, and June at the 10% significance level (Figure 6(b)).PRCPTOT is significant in March, October, and November at 5% significance (Figure 6(c)).

Figure 4 |
Figure 4 | Trends in annually calculated indices from climate models for 2006-2100.

Figure 5 |
Figure 5 | Annual index trends for all rain gages and one temperature gage for 1977- 2016.
when precipitation indices are adapted to a higher temporal resolution and tested for monthly periods, a greater extent of hydrologic change can be observed.Changes in the frequency, timing, and intensity of heavy precipitation events will have large consequences on the water cycle.However, the role of natural climate variability in these changes is unknown.Natural variability is thought to be responsible for an increase in mean precipitation observed in the Pacific Northwest over the twentieth century and may play a role in rainfall intensification(Mote   ).Expanding this work to include more measures to evaluate observations against models could make this study more conclusive.CONCLUSIONSThis study addresses the relationship between climate change and precipitation by examining how the lack of identifiable trends is affected by temporal resolution.ETCCDI indices that summarize precipitation characteristics at the monthly and annual levels are generated and tested for trends over the observed record and long-term future.The annual level indices demonstrate an intensifying hydrologic cycle from 2006 to 2100.These annually calculated indices do not show trends towards intensification in the observed record (1977-2005).However, when the indices are run to summarize data monthly instead of annually, trends showing intensification of the hydrologic cycle are apparent.Trend detection of increasing precipitation intensity became even more robust after the records themselves are moved from daily observations to hourly observations.The results demonstrate that the study of the effects of climate change on precipitation needs to reflect the ephemeral nature of rainfall.Signal change in precipitation patterns is likely to be obscured where annual characteristics such as annual mean rainfall are examined.
use of common indices like ETCCDI is of benefit.A reliance on annual summaries is a current limitation of the ETCCDI indices.Only a few parameters are available at the monthly scale, which makes the detection of seasonal changes difficult and therefore limits the detection of local and regional effects of climate change on precipitation.This limited availability represents a tradeoff between usability and accuracy.Several precipitation characteristics cannot be addressed at the annual scale.Still, the indices are based on daily observations that do allow places with simple observational networks to participate.Exploring additional indices that strike a balance to emphasize monthly and seasonal indices from daily observations would be an area for future study.Further, refining the performance measures by which indices are evaluated would allow for a more rigorous exploration of the success of these common metrics.

Table 1
demonstrate consistent temperature increase and disparate precipitation trends around the globe.Although

Table 2
was selected for both historic projectionsand future projections, except for theHadGEM2-  ES model (2006-2099).The Representative Concentration Pathway (RCP) 8.5 high emissions scenario was selected because the large

Table 2 |
Global climate models used and the creator of each model

Table 3 |
Definition of ETCCDI indices used for Research Question 1(Karl et al. 1999; Peterson et al. 2001)Let RR wj be the daily precipitation amount on a wet day w (RR !1 mm) in period j and let RR wn 95 be the 95th percentile of precipitation on wet days in the base period.If W represents the number of wet days in the period, then: w¼1RR wj where RR wj > RR wn 95 Surface runoff rate, recharge rate SDII (Equation (4)) Simple precipitation intensity index: Let RR wj be the daily precipitation amount on wet days, w (RR !1 mm) in period j.If W represents number of wet days in j, then:

Table 4 |
Definition of ETCCDI indices used for research questions 2 and 3(Karl et al. 1999; Peterson et al. 2001) ; Aragon et al. ).Simulation modeling shows that Jakob et al. ; Burn & Taleghani ).For this reason, gage spatial density and temporal resolution that is appropriate for temperature may not be appropriate for precipitation.The high spatial and temporal resolution of the data available in Portland's HYDRA network thus is an asset for the local area, although these trends cannot necessarily be extrapolated to the larger region.Analysis of Portland's gage network from 1977 to 2016 indicates that Compared to temperature, precipitation occurs at smaller spatial and temporal scales that are more difficult to measure(Boer et al. ).