A Performance Evaluation of Dynamical Downscaling of Precipitation Over Northern California

sustainability

Article A Performance Evaluation of Dynamical Downscaling of Precipitation over Northern California

Suhyung Jang 1,*, M. Levent Kavvas 2, Kei Ishida 3, Toan Trinh 2, Noriaki Ohara 4, Shuichi Kure 5, Z. Q. Chen 6, Michael L. Anderson 7, G. Matanga 8 and Kara J. Carr 2

1 Water Resources Research Center, K-Water Institute, Daejeon 34045, Korea 2 Department of Civil and Environmental Engineering, University of California, Davis, CA 95616, USA; [email protected] (M.L.K.); [email protected] (T.T.); [email protected] (K.J.C.) 3 Department of Civil and Environmental Engineering, Kumamoto University, Kumamoto 860-8555, Japan; [email protected] 4 Department of Civil and Architectural Engineering, University of Wyoming, Laramie, WY 82071, USA; [email protected] 5 Department of Environmental Engineering, Toyama Prefectural University, Toyama 939-0398, Japan; [email protected] 6 California Department of Water Resources, Sacramento, CA 95814, USA; [email protected] 7 California Department of Water Resources, Sacramento, CA 95821, USA; [email protected] 8 US Bureau of Reclamation, Sacramento, CA 95825, USA; [email protected] * Correspondence: [email protected]; Tel.: +82-42-870-7413

Received: 29 June 2017; Accepted: 9 August 2017; Published: 17 August 2017

Abstract: It is important to assess the reliability of high-resolution climate variables used as input to hydrologic models. High-resolution climate data is often obtained through the downscaling of Global Climate Models and/or historical reanalysis, depending on the application. In this study, the performance of dynamically downscaled precipitation from the National Centers for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) reanalysis data (NCEP/NCAR reanalysis I) was evaluated at point scale, watershed scale, and regional scale against corresponding in situ rain gauges and gridded observations, with a focus on Northern California. Also, the spatial characteristics of the simulated precipitation and wind fields, with respect to various grid sizes, were investigated in order to gain insight to the topographic effect on the atmospheric state variables. To this end, dynamical downscaling was performed using the mesoscale atmospheric model MM5, and the synoptic scale reanalysis data were downscaled to a 3 km grid spacing with hourly temporal resolution. The results of comparisons at point scale and watershed scale over a 50-year time period showed that the MM5-simulated precipitation generally produced the timing and magnitude of the observed precipitation in Northern California. The spatial distributions of MM5-simulated precipitation matched the corresponding observed precipitation reasonably well. Furthermore, the statistical goodness of fit tests of the MM5-simulated precipitation against the corresponding observed precipitation showed the reliability and capability of MM5 simulations for downscaling precipitation. A comparison of the spatial characteristics of the results with respect to various grid sizes indicated that precipitation and wind fields are significantly affected by the local topography. In particular, the banded structures and orographic effects on precipitation and wind fields can be well described by a mesoscale model at 3 km and 9 km grid resolutions while 27 km and 81 km grid model simulation may not be sufficient for watershed-scale or sub-watershed-scale studies.

Keywords: precipitation; regional climate model (RCM); watershed scale; reanalysis data; orographic effect; California

Sustainability 2017, 9, 1457; doi:10.3390/su9081457 www.mdpi.com/journal/sustainability Sustainability 2017, 9, 1457 2 of 17

1. Introduction GCMs (global climate models) are the primary tools for assessing the magnitude and impacts of climate change on water resources, floods, water quality, sea level rise, ecosystems, etc. Global historical reanalysis datasets such as the National Centers for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) reanalysis I [1] data can be used for the reconstruction of historical atmospheric data which can then be utilized as input for hydrologic or environmental modeling of the historical conditions in sparsely gauged or ungauged watersheds [2,3]. However, GCM and reanalysis datasets whose horizontal resolutions are more than 50 km in general are not suitable for the watershed scale because they cannot capture the mesoscale processes such as convection, orographic effects and sharp frontal gradients in order to account for the effect of local land conditions on the climate at these scales [4–8]. The fine spatial and temporal scales are also essential to the modeling of surface and subsurface hydrological processes (e.g., [9–11]) due to the heterogeneity in land characteristics within a watershed. As such, downscaling of the coarse-resolution global datasets becomes necessary to resolve the scale discrepancy between coarse-resolution GCMs (or reanalysis data) and the resolution required for an assessment at a regional scale or a watershed scale. Statistical downscaling and dynamical downscaling approaches are commonly used to resolve the scale discrepancy, but statistical downscaling is not suitable because it fundamentally assumes unchanged statistical relationships and unchanged spatial variability [8,12,13]. Meanwhile, dynamical downscaling by means of a regional climate model (RCM) is a technique that has several valuable features. RCMs are able to represent realistic regional climate features such as orographic precipitation, land surface characteristics [14–16], and extreme precipitation (e.g., [17,18]). RCMs can also account for the physical interactions of land-atmospheric processes with consideration of the heterogeneity in topography, soil, vegetation, and climate variables over a watershed by solving the full equation suite of mass, energy, and momentum conservation laws in the atmosphere. Moreover, the complete climate variables including precipitation, air temperature, wind speed, relative humidity, solar radiation, etc., can be produced by RCMs based upon the given GCM outputs or reanalysis data at very fine spatial and temporal resolutions. On the other hand, the major disadvantage of RCMs is the cost of their numerical computations that place restrictions not only on the simulation period, but also the computational grid size. For this reason, most of the RCM-related studies have been restricted to applications on the assessment of model performance (e.g., [19–24]), or on future climate change impact assessment with a relatively coarse grid resolution (normally more than 20 km) over specified time slices (e.g., [25–27]), with finer resolution over specified time slices (e.g., [28]), or with relatively coarse grid resolution over a long-term period [29,30]. The accuracy of downscaled climate variables depends on the biases inherited from the driving GCM [4,5] as well as the model characteristics, such as the model formulation, grid resolution, numerical scheme and other physical parameterizations [7,31,32]. GCM simulations of the present-day climate are forced by prescribed sea surface temperatures (SST), and are deterministically predictable in short and middle range [33,34]. Therefore, downscaled results from GCM simulations are comparable to observation data only in climatological features, as in the study by [29]. On the other hand, reanalysis data are the assimilated results from the analysis of various data sources (e.g., land surface, ship, rawinsonde, pibal, aircraft, and satellite), and are consistent with the observations [1]. Thus, downscaled results from reanalysis data have the potential to be used for more detailed comparison with observation data. Hence, reanalysis data are more suitable for the evaluation of downscaling performance. In this study, dynamically downscaled precipitation, which is already commonly used in climate change impact studies at watershed scales, is used to reconstruct historical climate variables at different scales. The dynamically downscaled precipitation is evaluated at point scale, watershed scale and regional scale against corresponding in situ rain gauges and gridded observations, with a focus on Northern California. A description of the study area and observational datasets is Sustainability 2017, 9, 1457 3 of 17

2017provided,, 9, 1457 the methodology is demonstrated, and dynamically downscaled precipitation is evaluated.3 of 17 Additionally, the spatial characteristics of the simulated precipitation and wind ﬁelds, with respect to tovarious various grid grid sizes, sizes, areinvestigated are investigated in order in toorder gain insightto gain to insight the topographic to the topographic effect on the effect atmospheric on the atmosphericstate variables. state variables.

2. Study Study Area Area and and Methodology Methodology

2.1. Study Study Area Area A region region of Northern Northern California (see Figure 1 1)) waswas selectedselected forfor thisthis studystudy asas itit isis thethe prime sourcesource of water for California. Two-thirds Two-thirds of of the the precipit precipitationation in in California falls falls on the Sacramento River watershed, and 80% of the precipitation occurs between November and March.

Figure 1. Study area and simulation domain (latter highlighted by the blue box). Shaded areas indicate Figure 1. Study area and simulation domain (latter highlighted by the blue box). Shaded areas indicate the eight selected watersheds and circles mark the locations of the eight observation stations used in the eight selected watersheds and circles mark the locations of the eight observation stations used in this study. Abbreviations to eight watersheds are provided in the main text and Table 1. this study. Abbreviations to eight watersheds are provided in the main text and Table1.

Eight watersheds and eight individual ground observation stations were selected for the model performanceEight watersheds evaluation and considering eight individual the location, ground topography, observation vegetation, stations were and selectedsize of the for watershed, the model asperformance shown in evaluationTable 1 and considering Figure 1. theThe location, largest topography,watershed among vegetation, the andstudy size watersheds of the watershed, is the Sacramentoas shown in TableRiver1 andwatershed Figure1 .(SRW) The largest which watershed is a fundamental among the water study watershedssource, supplying is the Sacramento domestic, irrigation,River watershed industrial (SRW) and which environm is a fundamentalental water water in California. source, supplying It also contributes domestic, irrigation, contaminants industrial and nutrientsand environmental from agricultural water in land California. and mountainous It also contributes terrain contaminants to the Central and Valley nutrients and fromBay-Delta agricultural areas. Theland Central and mountainous Valley and terrain Bay-Delta to the areas Central are Valleycrucial and regions Bay-Delta for agriculture areas. The and Central ecosystems Valley and in California.Bay-Delta areasOroville are crucialDam, Folsom regions Dam, for agriculture Shasta Dam, and Englebright ecosystems Dam, in California. and Trinity Oroville Dam Dam, in the Folsom study regionDam, Shasta control Dam, the water Englebright resources Dam, ofand Northern Trinity California. Dam in the The study Cache region Creek control watershed the water (CCW) resources is a representativeof Northern California. hydrologic The unit Cache in Creekthe Coastal watershed Range, (CCW) and is athe representative Northern Central hydrologic Valley unit (NCV) in the watershedCoastal Range, comprises and the most Northern of the Central low-lying Valley and (NCV) agricultural watershed lands comprises in Central most California. of the low-lying The Sierra and Nevada,agricultural Coastal lands Range, in Central and California.their foothills The recharge Sierra Nevada, the groundwater Coastal Range, under and the theirCentral foothills Valley, recharge which isthe a groundwatervery important under source the of Central water Valley,during which the dry is aseason very importantin California. source Elevations, of water drainage during the areas dry andseason vegetation in California. covers Elevations, of the selected drainage watersheds areas and are vegetationdiverse (see covers Table of1 and the selectedFigure 2). watersheds For example, are mid-diverse and (see downstream Table1 and Figuresections2). of For the example, Upper mid-Feather and River downstream watershed sections (UFRW) of the and Upper Shasta Feather Dam watershed (SHA) are comprised of forest and steep canyon areas, but upstream sections of these watersheds are flat grasslands and croplands. The northwest coastal region which contains the Trinity dam watershed (TRI), and the Sierra Nevada which contains the American River watershed Sustainability 2017, 9, 1457 4 of 17

River watershed (UFRW) and Shasta Dam watershed (SHA) are comprised of forest and steep canyon areas, but upstream sections of these watersheds are flat grasslands and croplands. The northwest coastal2017, 9, 1457 region which contains the Trinity dam watershed (TRI), and the Sierra Nevada which contains4 of 17 the American River watershed (ARW), Yuba River watershed (YBW) and downstream of UFRW and SHA(ARW), in NorthernYuba River California, watershed receive (YBW) plenty and downstre of precipitationam of UFRW due to and a combination SHA in Northern of the orographicCalifornia, effectreceive and plenty moisture of precipitation influx from due the Pacific to a combination Ocean, which of the is sometimes orographic referred effect and to as moisture a pineapple influx express from orthe atmospheric Pacific Ocean, river. which For the is sometimes evaluations referred at point to scale, as a eight pineapple ground express stations or from atmospheric the California river. Data For Exchangethe evaluations Center at (CDEC, point[ 35scale,]) of theeight California ground Departmentstations from of Waterthe California Resources Data (CDWR) Exchange were selected Center based(CDEC, on [35]) their of locations the California and record Department length (seeof Wate Figurer Resources1 and Table (CDWR)1). were selected based on their locations and record length (see Figure 1 and Table 1). Table 1. Summary information on watersheds and ground observational stations selected for this study. TheTable eight 1. Summary ground observation information stations on watersheds are from theand California ground observational Data Exchange stations Center selected (CDEC, [for35 ])this of thestudy. California The eight Department ground observation of Water Resources stations are (CDWR). from the California Data Exchange Center (CDEC, [35]) of the California Department of Water Resources (CDWR). Watershed Ground Observationa WatershedDrainage Mean Ground Observationa ID Basin Name Drainage2 Area Mean ElevationID Station Name ElevationElevation (m) ID Basin Name Area (km ) Elevation (m) ID Station Name (km2) (m) (m) SRW Sacramento River 71,721 921 SHO Shasta dam 328 SRW Sacramento River 71,721 921 SHO Shasta dam 328 SHA Shasta dam 19,818 1419 WVR Weaverville RS 625 SHATRI TrinityShasta dam 170119,818 1417 1419 GNVWVR Greenville Weaverville RS RS 1088 625 TRINCV Northern Trinity Central dam Valley 27,351 1701 197 1417 BOW GNV Bowman Greenville RS 1641 1088 NCVUFRW Northern Upper Feather Central River Valley 9225 27,351 1532 197 CLF BOW Colfax Bowman 732 1641 UFRWARW Upper American Feather River River 5535 9225 1149 1532 ORL CLF Orland Colfax 77 732 ARWYBW American Yuba River River 35285535 1202 1149 UKH ORL UkiahOrland 193 77 YBWCCW CacheYuba River Creek 29703528 535 1202 DVSUKH Davis 2WSWUkiah 18 193 CCW Cache Creek 2970 535 DVS Davis 2WSW 18

FigureFigure 2.2. Digital elevation model (leftleft)) from U.S. Geological SurveySurvey (USGS)(USGS) 22 arc-secondarc-second gridgrid spacingspacing datadata [[36]36] andand LandLand use/landuse/land cover types ( right) from Global Land Cover CharacteristicsCharacteristics (GLCC) datadata versionversion 22 fromfrom USGSUSGS [[37]37] overover thethe studystudy area.area.

2.2.2.2. Methodology InIn thisthis study, study, the the Fifth-Generation Fifth-Generation NCAR/Penn NCAR/Pen Staten State Mesoscale Mesoscale Model, Model, MM5 MM5 [38], [38], was setwas up set over up theover Northern the Northern California California domain domain shown inshown Figure in1, downscalingFigure 1, downscaling the NCAR/NCEP the NCAR/NCEP reanalysis I reanalysis data [1,39]) I fromdata [1,39]) a 210 × from210 a km 210 horizontal × 210 km gridhorizontal spacing grid to aspacing ﬁne-scale to a 3 fine-scale× 3 km grid 3 × 3 resolution km grid resolution (see Figure (see3). FourFigure one-way 3). Four nested one-way grids nested are used grids for are this used study. for this Two-way study. nestingTwo-way techniques nesting techniques could make could the model make resultsthe model more results accurate; more however, accurate; two-wayhowever,nesting two-way accuracy nesting dependsaccuracy ondepends application on application regions and regions it is computationallyand it is computationally more expensive more expensive and time-consuming and time-consuming than the methodthan the applied method herein applied [40 herein]. [40]. Sustainability 2017, 9, 1457 5 of 17 2017, 9, 1457 5 of 17

FigureFigure 3. 3. RegionalRegional hydro-climate hydro-climate model model (MM5) (MM5) simulation simulation domains domains for for the the study study area. area.

The simulated precipitation by the MM5 has been successfully validated and used for The simulated precipitation by the MM5 has been successfully validated and used for precipitation precipitation analysis over California by several research groups [41–44]. Furthermore, Ishida et al. analysis over California by several research groups [41–44]. Furthermore, Ishida et al. [45] showed that [45] showed that the MM5 can occasionally show better precipitation results than a next-generation the MM5 can occasionally show better precipitation results than a next-generation mesoscale numerical mesoscale numerical weather prediction system, the Weather Research and Forecasting Model weather prediction system, the Weather Research and Forecasting Model (WRF, [46]), and the MM5 (WRF, [46]), and the MM5 requires less computational costs. requires less computational costs. According to the one-third ratio rule of the nesting domain, recommended by [38], the first According to the one-third ratio rule of the nesting domain, recommended by [38], the first through through fourth nesting domains (D1–D4) were configured with the spatial grid resolutions of 81 km, fourth nesting domains (D1–D4) were configured with the spatial grid resolutions of 81 km, 27 km, 9 km, 27 km, 9 km, and 3 km, respectively. Thus, the domains have 24 × 22, 37 × 31, 73 × 49, and 175 × 109 and 3 km, respectively. Thus, the domains have 24 × 22, 37 × 31, 73 × 49, and 175 × 109 horizontal horizontal grid points, respectively, with 25 vertical levels. MM5 is a nonhydrostatic and fully grid points, respectively, with 25 vertical levels. MM5 is a nonhydrostatic and fully compressible compressible three-dimensional model which can be downscaled even up to a 1 km spatial resolution. three-dimensional model which can be downscaled even up to a 1 km spatial resolution. It is hence It is hence able to capture the orographic effects and land surface/land use conditions of watersheds. able to capture the orographic effects and land surface/land use conditions of watersheds. The NCAR/NCEP reanalysis I data used for the initial and boundary conditions of the MM5 The NCAR/NCEP reanalysis I data used for the initial and boundary conditions of the MM5 simulations are a combination of a numerical weather model that runs reforecasts over historical simulations are a combination of a numerical weather model that runs reforecasts over historical periods, assimilating observations from in situ sources such as rawinsonde, marine surface, land periods, assimilating observations from in situ sources such as rawinsonde, marine surface, land surface, aircraft, satellite, and so on [1]. Therefore, the MM5-simulated precipitation from the surface, aircraft, satellite, and so on [1]. Therefore, the MM5-simulated precipitation from the NCAR/NCEP reanalysis I data may be suitable for evaluating the model performance by comparing NCAR/NCEP reanalysis I data may be suitable for evaluating the model performance by comparing it against observations. For the 50-year climate simulation from 1950 to 1999, the model simulation it against observations. For the 50-year climate simulation from 1950 to 1999, the model simulation was reinitiated at every water year (from 1 October to 30 September) of the simulation period with was reinitiated at every water year (from 1 October to 30 September) of the simulation period with an additional two days at the beginning for spin-up. MM5-simulated precipitation over the study an additional two days at the beginning for spin-up. MM5-simulated precipitation over the study area is dominated by boundary conditions. Several trials before the yearly simulations found that area is dominated by boundary conditions. Several trials before the yearly simulations found that initial conditions affect precipitation over the study area only for a few days. In addition, the study initial conditions affect precipitation over the study area only for a few days. In addition, the study area has little precipitation at the beginning of a water year, in early October. Thus, the re-initiation area has little precipitation at the beginning of a water year, in early October. Thus, the re-initiation at at every water-year during the simulation period does not affect the results of the yearly simulations. every water-year during the simulation period does not affect the results of the yearly simulations. No nudging technique was utilized in this study for this reason. On the other hand, all the variables No nudging technique was utilized in this study for this reason. On the other hand, all the variables used as boundary conditions, including sea surface temperature, were from the NCAR/NCEP used as boundary conditions, including sea surface temperature, were from the NCAR/NCEP reanalysis data at 6 h temporal resolution. reanalysis data at 6 h temporal resolution. For the performance evaluation of dynamically downscaled precipitation by MM5, the historical For the performance evaluation of dynamically downscaled precipitation by MM5, the historical precipitation data were collected from various public sources. This includes eight CDEC ground precipitation data were collected from various public sources. This includes eight CDEC ground observation stations from CDWR (see Table 1), PRISM data from the PRISM Climate Group [47], and observation stations from CDWR (see Table1), PRISM data from the PRISM Climate Group [ 47], NCEP/EMC U.S. Gridded Precipitation (gauge-only, radar, and Stage IV) from NCAR/EOL under the sponsorship of the National Science Foundation [48]. The ground-observed precipitation data were used for time-series comparisons at point locations, while PRISM data were used for evaluations Sustainability 2017, 9, 1457 6 of 17 and NCEP/EMC U.S. Gridded Precipitation (gauge-only, radar, and Stage IV) from NCAR/EOL under the sponsorship of the National Science Foundation [48]. The ground-observed precipitation data were used for time-series comparisons at point locations, while PRISM data were used for evaluations of both basin-average precipitation at eight watersheds and for spatial precipitation ﬁelds over Northern California. NCEP/EMC U.S. Gridded Precipitation data were used in order to evaluate the spatial variability2017 at, 9, hourly,1457 24 h and monthly time scales within the simulation domain. 6 of 17

It shouldof both be basin-average noted that precipitation model bias at correction, eight watersheds which and is a for common spatial precipitation procedure forfields improving over the quality ofNorthern data reconstruction, California. NCEP/EMC has not U.S. been Gridded used Precipit in thisation study. data were All used evaluation in order to results evaluate are the based on simulatedspatial precipitation variability at without hourly, any24 h and bias monthly correction time inscales order within to testthe simulation the model domain. performance objectively. It should be noted that model bias correction, which is a common procedure for improving the 3. Resultsquality and of Discussion data reconstruction, has not been used in this study. All evaluation results are based on simulated precipitation without any bias correction in order to test the model performance objectively. 3.1. Evaluation at Point Scale 3. Results and Discussion MM5-simulated precipitation was evaluated by the corresponding ground observations at 8 CDEC stations3.1. (see Evaluation Table1 andat Point Figure Scale 1). The elevations of the selected eight ground stations range from 18 m to 1641MM5-simulated m. precipitation was evaluated by the corresponding ground observations at 8 CDEC Scatterstations graphs (see Table of the 1 and simulated Figure 1). The and elevations observed of the monthly selected precipitationeight ground stations at eight range ground from 18 stations m to 1641 m. are shown in Figure4. Although the model slightly underestimates the precipitation at the GNV Scatter graphs of the simulated and observed monthly precipitation at eight ground stations are (Greenvilleshown RS) in andFigure UKH 4. Although (Ukiah) the stations model slightly and overestimates underestimates at the the precipitation WVR (Weaverville at the GNV RS) and BOW (Bowman)(Greenville stations,RS) and UKH both (Ukiah) the observedstations and andoverestimates the simulated at the WVR values (Weaverville are generally RS) and BOW comparable. The coefﬁcient(Bowman) of determination,stations, both the Robserved2, values and are the higher simulated than values 0.75. are However, generally thecomparable. model performance The 2 could becoefficient different of depending determination, on R the, values locations are higher of the than gauging 0.75. However, stations the withinmodel performance the watersheds. could Hence, be different depending on the locations of the gauging stations within the watersheds. Hence, it may it may notnot bebe appropriate to evaluate to evaluate the downscaling the downscaling performance performance by comparisons by comparisonsat only a few gauging at only a few gaugingstation station point point locations. locations.

Figure 4. Scatter plots of the ground observed precipitation against the MM5-simulated (NCEP) Figure 4. Scatter plots of the ground observed precipitation against the MM5-simulated (NCEP) precipitation at eight ground stations during 1950–1999. precipitation at eight ground stations during 1950–1999. 3.2. Evaluation at Watershed Scale 3.2. EvaluationFor at evaluation Watershed at Scalethe scale of watersheds, MM5-simuated precipitation values were compared Foragainst evaluation the corresponding at the scale PRISM of watersheds, (parameter-elevation MM5-simuated regressions on precipitation independent slopes values model, were [49]) compared precipitation data at eight watersheds within the simulation domain. PRISM provides interpolated against theground corresponding precipitation PRISMdata over (parameter-elevation the United States at 4 km regressions horizontal ongrid independent spacing from slopes1895 to model,the [49]) precipitationpresent data [47]. at eight watersheds within the simulation domain. PRISM provides interpolated ground precipitation data over the United States at 4 km horizontal grid spacing from 1895 to the present [47]. 2017, 9, 1457 7 of 17

As stated previously in the section on the study area, the selected eight watersheds have various characteristics with respect to topography, geomorphology, vegetation, drainage area, and climate. Figure 5 shows the scatter plots of basin-average precipitation between PRISM and MM5-simulated precipitation for eight Northern California watersheds during 1950–1999. Among the eight watersheds, the MM5-simulated precipitation for SHA, TRI and ARW is slightly higher than the PRISM precipitation at higher values. These three watersheds are located in high-elevation mountainous terrain. MM5-simulated and PRISM precipitation values were also evaluated by statistical comparisons of the means, standard deviations (STDEV), coefficients of determination (R2), root- mean-square errors, and Nash-Sutcliffe efficiency (ENS) during 1950 to 1999 for the eight watersheds (Table 2). The means and STDEVs are similar to the PRISM values. High values of R2 and ENS were found at the eight watersheds. Sustainability 2017, 9, 1457 7 of 17 Table 2. Statistical comparisons of MM5-simulated (NCEP) versus PRISM precipitation during 1950–1999.

Watershed Model Mean (mm) STDEV (mm) RMSE (mm) R² Nash-Sutcliffe Efficiency PRISM 79.63 91.57 As statedSRW previously in the section on the study area, the selected eight watersheds have MM5-simulated 88.20 104.70 32.72 0.92 0.87 various characteristicsPRISM with respect71.56 to topography,72.84 geomorphology, vegetation, drainage area, SHA and climate. FigureMM5-simulated5 shows the scatter97.60 plots of100.22 basin-average 48.94 precipitation0.87 between 0.55 PRISM and PRISM 128.91 153.14 TRI MM5-simulated precipitationMM5-simulated for eight155.57 Northern 193.10 California watersheds79.07 0.87 during 1950–1999. 0.73 Among the PRISM 50.92 63.78 eight watersheds,NCV the MM5-simulated precipitation for SHA, TRI and ARW is slightly higher than MM5-simulated 53.66 74.11 25.15 0.89 0.83 the PRISM precipitationPRISM at higher100.52 values. These120.76 three watersheds are located in high-elevation UFRW mountainous terrain.MM5-simulated MM5-simulated 106.17 and PRISM 125.79 precipitation 39.29 0.90 values were 0.89 also evaluated by PRISM 101.59 121.49 ARW statistical comparisonsMM5-simulated of the means,114.85 standard 152.47 deviations 57.54 (STDEV 0.89), coefﬁcients 0.78 of determination 2 PRISM 136.31 165.91 (R ), root-mean-squareYBW errors, and Nash-Sutcliffe efﬁciency (E ) during 1950 to 1999 for the eight MM5-simulated 140.50 180.41 56.27 NS 0.90 0.88 2 watersheds (Table2). ThePRISM means and72.16STDEVs are97.28 similar to the PRISM values. High values of R and CCW MM5-simulated 65.77 95.75 31.18 0.90 0.90 ENS were found at the eight watersheds.

Figure 5.FigureScatter 5. plots Scatter of basin-averageplots of basin-average precipitation precip betweenitation between PRISM PRISM and MM5-simulated and MM5-simulated precipitation for eightprecipitation Northern California for eight Northern watersheds California during watersheds 1950–1999. during 1950–1999.

Table 2. Statistical comparisons of MM5-simulated (NCEP) versus PRISM precipitation during 1950–1999.

Watershed Model Mean (mm) STDEV (mm) RMSE (mm) R2 Nash-Sutcliffe Efﬁciency PRISM 79.63 91.57 SRW MM5-simulated 88.20 104.70 32.72 0.92 0.87 PRISM 71.56 72.84 SHA MM5-simulated 97.60 100.22 48.94 0.87 0.55 PRISM 128.91 153.14 TRI MM5-simulated 155.57 193.10 79.07 0.87 0.73 PRISM 50.92 63.78 NCV MM5-simulated 53.66 74.11 25.15 0.89 0.83 PRISM 100.52 120.76 UFRW MM5-simulated 106.17 125.79 39.29 0.90 0.89 PRISM 101.59 121.49 ARW MM5-simulated 114.85 152.47 57.54 0.89 0.78 PRISM 136.31 165.91 YBW MM5-simulated 140.50 180.41 56.27 0.90 0.88 PRISM 72.16 97.28 CCW MM5-simulated 65.77 95.75 31.18 0.90 0.90 2017Sustainability, 9, 1457 2017, 9, 1457 88 of of 17 17

3.3. Evaluation of Spatial Characteristics over the Whole Modeling Domain at Regional Scale 3.3. Evaluation of Spatial Characteristics over the Whole Modeling Domain at Regional Scale As a key input for models of hydrological processes over a watershed, the spatial characteristics As a key input for models of hydrological processes over a watershed, the spatial characteristics of precipitation fields are important because hydrological processes are locally diverse due to the of precipitation fields are important because hydrological processes are locally diverse due to the heterogeneity of the land surface and soil hydraulic properties within a watershed. In this study, the heterogeneity of the land surface and soil hydraulic properties within a watershed. In this study, simulated precipitation fields were first compared against PRISM precipitation fields in order to the simulated precipitation fields were first compared against PRISM precipitation fields in order to evaluate the spatial distribution of the mean monthly precipitation during 1950–1999 over the whole evaluate the spatial distribution of the mean monthly precipitation during 1950–1999 over the whole modeling domain of Northern California. modeling domain of Northern California. Figure 6 shows some example comparisons of mean monthly precipitation in December, Figure6 shows some example comparisons of mean monthly precipitation in December, January, January, and February between MM5-simulated precipitation and PRISM data over Northern and February between MM5-simulated precipitation and PRISM data over Northern California during California during 1950–1999. Spatial distributions of mean monthly precipitation by the dynamic 1950–1999. Spatial distributions of mean monthly precipitation by the dynamic downscaling (MM5) downscaling (MM5) from NCAR/NCEP reanalysis data match the PRISM data reasonably well over from NCAR/NCEP reanalysis data match the PRISM data reasonably well over the simulation domain, the simulation domain, especially at low and median elevation ranges. The magnitudes of precipitation especially at low and median elevation ranges. The magnitudes of precipitation along the mountain along the mountain ridges in the Northwest Coastal region and the Sierra Nevada region in Northern ridges in the Northwest Coastal region and the Sierra Nevada region in Northern California are slightly California are slightly higher than the PRISM precipitation. Figure 7 shows basin-average higher than the PRISM precipitation. Figure7 shows basin-average comparisons of mean monthly comparisons of mean monthly precipitation between PRISM and MM5-simulated precipitation for precipitation between PRISM and MM5-simulated precipitation for eight watersheds in Northern eight watersheds in Northern California during 1950–1999. In general, the results in Figures 6 and 7 California during 1950–1999. In general, the results in Figures6 and7 are similar with respect to are similar with respect to the mean monthly comparisons of basin-average precipitation for the eight the mean monthly comparisons of basin-average precipitation for the eight selected watersheds in selected watersheds in Northern California. The basin-average precipitation values are slightly Northern California. The basin-average precipitation values are slightly higher at high-elevation higher at high-elevation watersheds (e.g., TRI, ARW and SHA) but are comparable at low-elevation watersheds (e.g., TRI, ARW and SHA) but are comparable at low-elevation watersheds (e.g., NCV watersheds (e.g., NCV and CCW). and CCW).

FigureFigure 6. 6. ExampleExample spatial spatial comparisons comparisons for for the the mean mean mo monthlynthly precipitation precipitation (December, (December, January, January, and and February)February) between between PRISM PRISM and and MM5-sim MM5-simulatedulated precipitation during 1950–1999. Sustainability 2017, 9, 1457 9 of 17 2017, 9, 1457 9 of 17

Figure 7. 7. Basin-averageBasin-average comparisons comparisons of mean of mean mont monthlyhly precipitation precipitation between between PRISM PRISM and MM5- and MM5-simulatedsimulated precipitation precipitation for eight for watershe eight watershedsds in Northern in Northern California California during during 1950–1999. 1950–1999.

It is noted that many physiographic and climatic effects were incorporated in the creation of the PRISM precipitationprecipitation data.data. AsAs such, such, this this dataset dataset is is one one of of the the most most reliable reliable and and comparable comparable datasets datasets for modelfor model calibration calibration or validation. or validation. However, However, the basic the assumption basic assumption in PRISM isin thatPRISM given is a that homogenous given a hillslope,homogenous precipitation hillslope, canprecipitation be predicted can throughbe predic ated moving-window through a moving-window regression function regression with function a linear relationshipwith a linear with relationship the elevation. with The the precipitation elevation. The generated precipitation by PRISM generated depends by on groundPRISM observationsdepends on whichground are observations located in similar which physiographicare located in similar and climatic physiographic areas. However, and climatic ground areas. observations However, may ground not beobservations available or may may not be be limited. available As such,or may precipitation be limited. comparisonsAs such, precipitation to PRISM comparisons data at high elevationsto PRISM maydata notat high be appropriate. elevations may not be appropriate. For example,example, TRI TRI is is located located in ain high-elevation a high-elevation mountainous mountainous terrain, terrain, and it isand affected it is byaffected moisture by influxmoisture from influx the Pacific,from the e.g., Pacific, atmospheric e.g., atmospheric rivers. In the rivers. TRI watershed,In the TRI watershed, 11 ground stations 11 ground were stations found fromwere found the CDEC from CDWR the CDEC dataset, CDWR and dataset, most observationsand most observations since the latesince 1980s the late are 1980s available are available with the exceptionwith the exception of two stations of two (CLE stations and CFF)(CLE which and CFF) are located which inare valleys. located There in valleys. are no observationalThere are no stationsobservational along stations the mountain along the ridges mountain from northridges to from east. north Within to east. this Within framework this framework the MM5-simulated the MM5- resultssimulated show results the effect show of the orography effect of onorography precipitation on precipitation while PRISM while data PRISM show adata smooth show appearance a smooth ofappearance mean monthly of mean precipitation monthly fields precipitation in December fiel duringds in December 1950–1999 (Figureduring8 ).1950–1999 Historically, (Figure extreme 8). precipitationHistorically, extreme events beginprecipitation in late December.events begin Therefore, in late December. December Therefore, has been selectedDecember in has order been to investigateselected in order the spatial to investigate structures the of spatial extreme struct precipitationures of extreme in complex precipitation terrain. in complex terrain. MM5 simulation results werewere alsoalso comparedcompared withwith NCEP/EMCNCEP/EMC U.S. Gridded PrecipitationPrecipitation products (gauge-only, radar,radar, andand StageStage IV)IV) inin orderorder toto evaluateevaluate thethe MM5MM5 atmosphericatmospheric model–based downscaling performance on reproducing the spatial characteristics of of hourly, 24 h and monthly accumulated precipitation fieldsfields during December 2005.2005. The precipitation event in December of 2005 is oneone of of the the most most extreme extreme events events which haswhich occurred has occurred since the NCEP/EMCsince the NCEP/EMC U.S. Gridded U.S. Precipitation Gridded Precipitation products, including Stage IV datasets, were made available in 2002 [48]. These U.S. Gridded Precipitation products provide sub-monthly precipitation at fine spatial grid scales (about 4 Sustainability 2017, 9, 1457 10 of 17 products, including Stage IV datasets, were made available in 2002 [48]. These U.S. Gridded Precipitation2017, 9, 1457 products provide sub-monthly precipitation at fine spatial grid scales (about10 of 4 17 km). The gauge-only data is the NCEP Stage II gauge-based precipitation, which is derived from automated precipitationkm). The gauge-only gauge observations, data is the wherein NCEP Stage erroneous II gauge-based gauge data precipitation, are removed which according is derived to a reject-list from automated precipitation gauge observations, wherein erroneous gauge data are removed according of apparently faulty gauges. The radar-only product with quality control is created by the WSR-88D to a reject-list of apparently faulty gauges. The radar-only product with quality control is created by Radar Product Generator at each radar site. Each individual radar estimate is merged with other radar the WSR-88D Radar Product Generator at each radar site. Each individual radar estimate is merged estimates on the national Hydrologic Rainfall Analysis Project (HRAP) grid. Radar estimates that with other radar estimates on the national Hydrologic Rainfall Analysis Project (HRAP) grid. Radar fall into the same bin are averaged together within that bin by a simple inverse-distance weighted estimates that fall into the same bin are averaged together within that bin by a simple inverse-distance method. The Stage IV multi-sensor product (Stage IV) combines the precipitation products from weighted method. The Stage IV multi-sensor product (Stage IV) combines the precipitation products thefrom regional the regional hourly/6 hourly/6 h multi-sensor h multi-sensor (radar (radar and and rain rain gauge) gauge) precipitation precipitation analyses, analyses, produced produced by by the 12the river 12 forecastriver forecast centers, centers, with qualitywith quality control–performed control–performed manual manual NWS NWS forecasts. forecasts. Therefore, Therefore, the the Stage IVStage product IV product may be consideredmay be considered to be a reliableto be a referencereliable reference for precipitation for precipitation observations. observations. The detailed The informationdetailed information on NCEP/EMC on NCEP/EMC U.S. Gridded U.S. Gridded Precipitation Precipitation products products is available is available at [48]. at [48].

FigureFigure 8. 8.Example Example spatial spatial comparisons comparisons ofof mean monthl monthlyy precipitation precipitation in in December December between between PRISM PRISM andand MM5-simulated MM5-simulated precipitation precipitation for for TRI TRI during during 1950–1999 1950–1999 withwith thethe locationlocation ofof 1111 groundground stations in thein CDEC the CDEC DWR DWR dataset dataset (DEM (DEM = Digital = Digital Elevation Elevation Model). Model).

Sustainability 2017, 9, 1457 11 of 17 2017, 9, 1457 11 of 17

Figure 99 showsshows thethe spatial comparisons of of gauge-only, gauge-only, radar radar with with bias bias removal removal and and Stage Stage IV observationsIV observations together together with with MM5-simulated MM5-simulated prec precipitationipitation fields fields for for the the hourly hourly accumulatedaccumulated precipitation fieldsfields (6 a.m. GMT on 31 December 2005).2005). Figure 99a,ba,b areare thethe samesame precipitationprecipitation fieldsfields from each product, but the color schemes for the magnitudes of precipitation precipitation are different in order to distinguish the spatial patterns among variousvarious products.products. The radar product shows precipitation magnitudes that that are are too too small small for for spatial spatial comparison, comparison, but but the the spatial spatial patterns patterns of precipitation of precipitation are capturedare captured well wellby the by radar the radar product. product. As can As be can seen be in seen Figure in Figure 9a, the9a, spatial the spatial distribution distribution of MM5- of simulatedMM5-simulated precipitation precipitation is comparable is comparable to the to theStage Stage IV IVproducts, products, but but the the radar-only radar-only product product shows precipitation magnitudes that are too small when compared to the other products.products. The gauge-only product has a smooth appearance (similar to the distribution of the PRISM precipitation field field in Figure8 8).). OnOn thethe otherother hand,hand, thethe otherother productsproducts showshow aa bandedbanded structurestructure forfor thethe precipitationprecipitation fieldsfields due to wind and orographic effects (Figure9 9).). TheThe clearerclearer bandedbanded structurestructure inin thethe MM5-simulatedMM5-simulated results may may be be due due to to the the grid grid resolution resolution of of the the simulation. simulation. Although Although the theMM5-simulated MM5-simulated results results are theare theclosest closest to the to theStage Stage IV IVproduct product in interms terms of of both both the the magnitudes magnitudes and and the the spatial spatial pattern of precipitation, thethe StageStage IV IV product product has has a spatiala spatial coverage coverage problem problem around around high-elevation high-elevation regions regions such suchas the as Sierra the Sierra Nevada Nevada and the and northern the northern coastal coastal regions regions in California. in California.

(a) Same color scheme (b) Different color scheme

Figure 9.9. SpatialSpatial comparisons comparisons of hourlyof hourly accumulated accumulated precipitation precipitation ﬁelds atfields 6 a.m. at GMT6 a.m. on 31GMT December on 31 December2005 from gauge-only2005 from gauge-only (a-1,b-1), radar (a-1, withb-1), biasradar removal with bias (a-2 removal,b-2), and (a-2 Stage,b-2 IV), and (a-3 ,Stageb-3) observations, IV (a-3,b-3) observations,together with together NCEP/NCAR-based with NCEP/NCAR-based MM5 simulations MM5 ( a-4simulations,b-4). (a-4,b-4).

Figure 10 shows the spatial comparisons of gauge-only, radar with bias removal, and Stage IV Figure 10 shows the spatial comparisons of gauge-only, radar with bias removal, and Stage observations together with the corresponding MM5-simulated precipitation fields for the 24 h (Figure 10a) IV observations together with the corresponding MM5-simulated precipitation fields for the 24 h and monthly (Figure 10b) accumulated precipitation fields. From this figure it may be seen that the (Figure 10a) and monthly (Figure 10b) accumulated precipitation fields. From this figure it may be seen MM5-simulated precipitation fields are quite similar to the corresponding Stage IV products with that the MM5-simulated precipitation fields are quite similar to the corresponding Stage IV products respect to the magnitudes and the spatial patterns of precipitation. The gauge-only products show a with respect to the magnitudes and the spatial patterns of precipitation. The gauge-only products smoother appearance and less precipitation when compared to MM5-simulated precipitation and the show a smoother appearance and less precipitation when compared to MM5-simulated precipitation Stage IV estimates, especially for monthly accumulation. However, the banded structure of precipitation and the Stage IV estimates, especially for monthly accumulation. However, the banded structure fields is no longer seen in the Stage IV products in Figure 10 (24 h and monthly accumulation) because of precipitation fields is no longer seen in the Stage IV products in Figure 10 (24 h and monthly the Stage IV 24 h and monthly products are accumulated from 6 h Stage IV products (not hourly). accumulation) because the Stage IV 24 h and monthly products are accumulated from 6 h Stage IV Meanwhile, the radar product cannot be used in the Stage IV 6 h product estimates due to the spatial products (not hourly). Meanwhile, the radar product cannot be used in the Stage IV 6 h product coverage problem of the radar product. estimates due to the spatial coverage problem of the radar product. Sustainability 2017, 9, 1457 12 of 17 2017, 9, 1457 12 of 17

(a) 24-hour accumulated precipitation (b) 1-month accumulated precipitation

Figure 10. Spatial comparisons of (a) 24 h (1 p.m. GMT on 30 December 2005—12 p.m. GMT on 31 Figure 10. Spatial comparisons of (a) 24 h (1 p.m. GMT on 30 December 2005—12 p.m. GMT on December 2005) and (b) one-month (1 p.m. GMT on 1 December 2005—12 p.m. GMT on 1 January 31 December 2005) and (b) one-month (1 p.m. GMT on 1 December 2005—12 p.m. GMT on 1 January 2006) 2006) accumulated precipitation fields from gauge-only (a-1,b-1), radar with bias removal (a-2,b-2), accumulated precipitation ﬁelds from gauge-only (a-1,b-1), radar with bias removal (a-2,b-2), and Stage andIV (a-3Stage,b-3 IV) observations (a-3,b-3) observations together with together NCEP/NCAR-based with NCEP/NCAR-based MM5 simulations MM5 simulations (a-4,b-4). (a-4,b-4).

The simulated precipitation and wind fields by the MM5 model over the Northern California The simulated precipitation and wind fields by the MM5 model over the Northern California simulation domain together with the digital elevation model of the domain at (a) 3, (b) 9, (c) 27, and simulation domain together with the digital elevation model of the domain at (a) 3, (b) 9, (c) 27, (d) 81 km grid resolutions on (6 a.m. GMT on 31 December 2005) are shown in Figure 11 in order to investigateand (d) 81 kmthe grideffect resolutions of banded on structures (6 a.m. GMT and on or 31ographic December effects 2005) on are precipitation shown in Figure with 11respect in order to variousto investigate grid sizes. the effectThe time of bandedof Figure structures 11 (6 a.m. and GMT orographic on 31 December effects on 2005) precipitation is the same with as that respect of Figureto various 9. It gridis clearly sizes. Theseen timein Figure of Figure 11 that11 (6 the a.m. spatial GMT details on 31 December of the topography 2005) is the are same smoothed as that significantlyof Figure9. Itwith is clearly the increase seen inin Figurethe grid 11 resolution that the spatials from detailsthe 3 km of grid the topographysize to the 81 are km smoothed grid size withinsignificantly the simulation with the domain. increase At in thethe grid3 km resolutions grid resolution from (Figure the 3 km 11a) grid the size spatial to the details 81 km of gridthe topographysize within are the described simulation well domain. from low-lying At the 3 kmflat gridareas resolution to high-elevation (Figure 11mountaina) the spatial regions. details Those of spatialthe topography details are aredescribed described relatively well from well up low-lying to the 9 flatkm areasgrid resolution to high-elevation (Figure 11b). mountain However, regions. the spatialThose spatialdetails detailsof the aretopography, described seen relatively at the well 3 km up and to the 9 km 9 km grid grid resolutions, resolution (Figureare no longer11b). However, seen at thethe 27 spatial km grid details resolution of the topography, (Figure 11c) seenand 81 at thekm 3grid km resolutions and 9 km grid (Figure resolutions, 11d). From are these no longer figures seen it mayat the also 27 kmbe gridinferred resolution that the (Figure precipitation 11c) and and 81 kmwind grid fields resolutions are significantly (Figure 11 d).affected From by these the figures local topographyit may also in be terms inferred of their that thespatial precipitation structures and and wind magnitudes. fields are For significantly example, the affected effects byof thebanded local structurestopography and in orography terms of theiron precipitation spatial structures and wind and fields magnitudes. are described For example,well at the the 3 km effects grid resolution of banded (Figurestructures 11a) and and orography at 9 km ongrid precipitation resolution and(Figure wind 11b) fields while are describedthe spatial well details at the of 3 kmprecipitation grid resolution and wind(Figure fields 11a) fade and away at 9 km at gridthe 27 resolution km grid (Figure(Figure 11 11c)b) whileand 81 the km spatial grid (Figure details of11d) precipitation resolutions. and wind fieldsSince fade precipitation away at the 27provides km grid the (Figure main 11 sourcec) and of 81 water km grid for (Figurehydrological 11d) resolutions. processes that occur in time andSince space precipitation within a watershed, provides the the main spatial source variation of water of precipitation for hydrological has processesa fundamental that occurimpact in ontime the and outcome space within of hydrological a watershed, processes the spatial within variation a watershed. of precipitation Similarly, has a the fundamental wind field impact over ona watershedthe outcome is ofan hydrologicalimportant factor processes in evapotranspira within a watershed.tion processes Similarly, and the in wind the spatial field over distribution a watershed of snowis an importantwithin a watershed. factor in evapotranspiration From the above analysis processes one and may in infer the spatial that the distribution dynamical ofdownscaling snow within of a precipitationwatershed. From and wind the above to 3 km analysis and 9 onekm grid may resolu infer thattions the can dynamical properly account downscaling for the of effect precipitation of local topographyand wind to and 3 km land and cover 9 km gridon the resolutions spatial distribu can properlytions of account these forclimate the effectvariables. of local In topographyturn, such realisticand land modeling cover on theof the spatial precipitation distributions and of wind these climatefields would variables. positively In turn, impact such realistic the hydrologic modeling modelingof the precipitation of streamflow, and wind snow, fields and wouldevapotranspira positivelytion impact processes the hydrologic within a watershed. modeling of Meanwhile, streamflow, thesnow, spatial and evapotranspirationdistributions of precipitation processes withinand wind a watershed. are not well-resolved Meanwhile, theat 27 spatial km and distributions 81 km grid of resolutions.precipitation As and such, wind arethe notresults well-resolved of hydrologic at 27 km modeling and 81 km that grid utilizes resolutions. such Ascoarse-resolution such, the results precipitation and wind as inputs may not produce reliable hydrologic process simulation results. Sustainability 2017, 9, 1457 13 of 17 of hydrologic modeling that utilizes such coarse-resolution precipitation and wind as inputs may not produce2017, 9 reliable, 1457 hydrologic process simulation results. 13 of 17

FigureFigure 11. The11. The simulated simulated precipitation precipitation and and wind wind fields fields by MM5 by MM5 over theover Northern the Northern California California simulation domainsimulation together domain with together the digital with elevation the digital model elevation of the model domain of the at domain (a) 3, (b )at 9, (a ()c 3,) 27, (b) and9, (c () d27,) 81 and km grid resolution(d) 81 km at 6grid a.m. resolution GMT on at 31 6 a.m. December GMT on 2005. 31 De Thecember contour 2005. map The contour shows precipitationmap shows precipitation fields; the vector mapfields; shows the wind vector fields; map shows the 3D wind color fields; map the shows 3D color the map digital shows elevation the digital model elevation foreach model grid for each resolution (3, 9,grid 27 andresolution 81 km). (3, 9, 27 and 81 km). Sustainability 2017, 9, 1457 14 of 17

4. Summary and Conclusions In this study, coarse resolution NCEP/NCAR Reanalysis I precipitation for Northern California (approximately 170,000 km2) was dynamically downscaled to an hourly, 3 km grid resolution using the MM5 atmospheric model over the period 1950–1999. The dynamically downscaled precipitation was then evaluated at point scale, watershed scale and regional scale against corresponding in situ rain gauges and gridded observations. The spatial structures of MM5-simulated hourly, daily and monthly accumulated precipitation fields in December 2005 were also evaluated in order to investigate the effects of banded structures and orography on precipitation fields. This analysis was based on various observations from NCEP/EMC U.S. Gridded Precipitation products (gauge-only, radar, and Stage IV). Additionally, the effect of orography on precipitation and wind fields with respect to horizontal grid sizes (3 km, 9 km, 27 km, and 81 km grids) from four nested grids used for the MM5 simulations was also investigated. From this study the following conclusions may be reached, as detailed below. From the results of point-scale evaluation based on the precipitation data from eight ground gauges during 1950–1999, the MM5-simulated precipitation values matched the corresponding ground-observed values generally well with respect to the timing as well as the magnitude of monthly precipitation. On scatter plots, the simulated precipitation values were distributed well along the 1:1 lines against the corresponding observed values. For the evaluation at watershed scale, the MM5-simulated precipitation was compared to the PRISM precipitation for eight watersheds (SRW, SHA, TRI, NCV, UFRW, YBW and CCW) in Northern California during 1950–1999. The MM5-simulated basin-average monthly precipitation values provided an excellent match with PRISM-based basin-average precipitation values. The results of the scatter plots showed that the model-simulated and PRISM values are distributed closely along the 1:1 line for all eight watersheds, especially SRW, UFRW, NCV, YBW, and CCW. The mean and the standard deviation of MM5-simulated basin-average precipitation were found to be close to the corresponding mean and standard deviation (STDEV) values from the PRISM data. Also, high values for the coefficient of determination (R2) and for Nash-Sutcliffe efficiency (ENS) statistic were found when model-simulated precipitation was compared against PRISM data for the eight selected watersheds in Northern California. Although, in general, the spatial distributions of the MM5-simulated mean monthly precipitation fields matched well against the corresponding PRISM precipitation fields during 1950–1999, the MM5-simulated values were slightly higher at the high-elevation mountain areas than the corresponding PRISM data. This may be due to PRISM precipitation values being based on the ground observations which are generally located at low-elevation areas near roads or around residential areas in order to be operationally accessible. As such, the spatial structures of the MM5-simulated precipitation fields were evaluated further based on various grid-based observation fields from NCEP/EMC U.S. Gridded Precipitation products (gauge-only, radar, and Stage IV). In this further investigation, the radar products showed much smaller precipitation magnitudes than other products (gauge-only, Stage IV), but the radar products described the banded precipitation structures and the orographic effects on precipitation well. The gauge-based precipitation fields did not capture the banded precipitation structures and orographic effects on precipitation, and showed a spatially smooth precipitation appearance (similar to the PRISM precipitation fields) over Northern California. The MM5-simulated precipitation fields showed spatial structures that are similar to those from radar and Stage IV products, and showed precipitation magnitudes similar to those from gauge-only and Stage IV products over Northern California. Therefore, one may conclude that MM5 simulations can capture not only the banded precipitation structures and orographic effects on precipitation but also the magnitudes of precipitation quite well over Northern California. From the results of the spatial comparisons of topography, precipitation and wind fields with respect to various grid sizes, it was found that the spatial distributions of the modeled precipitation and wind fields over Northern California are significantly impacted by the variation in the model grid resolutions. The grid resolutions of 3 km and 9 km resolve the local topography well, yielding realistic spatial patterns for precipitation and wind. However, the 27 km and 81 km grid sizes could not resolve Sustainability 2017, 9, 1457 15 of 17 the local topography, and, hence, could not resolve the spatial variation of precipitation and wind fields over Northern California. In conclusion, in this study the performance of dynamically downscaled precipitation from NCEP/NCAR reanalysis data was evaluated at point scale, watershed scale and spatial scale against corresponding ground observations, PRISM data, and NCEP/EMC U.S. Gridded Precipitation products. Also, the spatial characteristics of the simulated precipitation and wind fields with respect to various grid sizes were investigated in order to gain insight to the topographic effect on the atmospheric state variables. The results of this study indicate that the MM5 regional atmospheric model can dynamically downscale the climate variables from the synoptic-scale products such as reanalysis data to a finer grid scale satisfactorily, while also quantifying the effect of the topography and land cover on atmospheric information.

Acknowledgments: This work is partially supported by the Korea Agency for Infrastructure Technology Advancement (KAIA) grant funded by the Ministry of Land, Infrastructure and Transport (Grant 17AWMP-B0830660-04). Author Contributions: Z. Q. Chen set up the regional climate model; Noriaki Ohara, Suhyung Jang and Shuichi Kure reconstructed and simulated the atmospheric and hydrological data; Toan Trinh collected the observations; Suhyung Jang and Kei Ishida analyzed and wrote the paper with support from M. Levent Kavvas, Michael L. Anderson, G. Matanga, and Kara J. Carr. Conﬂicts of Interest: The authors declare no conﬂict of interest.

References

1. Kalnay, E.; Kanamitsu, M.; Kistler, R.; Collins, W.; Deaven, D.; Gandin, L.; Iredell, M.; Saha, S.; White, G.; Woollen, J.; et al. The NCEP/NCAR 40-Year Reanalysis Project. Bull. Am. Meteorol. Soc. 1996, 77, 437–471. [CrossRef] 2. Kavvas, M.L.; Kure, S.; Chen, Z.Q.; Ohara, N.; Jang, S. WEHY-HCM for Modeling Interactive Atmospheric-Hydrologic Processes at Watershed Scale. I: Model Description. J. Hydrol. Eng. 2013, 18, 1262–1271. [CrossRef] 3. Kure, S.; Jang, S.; Ohara, N.; Kavvas, M.L.; Chen, Z.Q. WEHY-HCM for Modeling Interactive Atmospheric-Hydrologic Processes at Watershed Scale. II: Model Application to Ungauged and Sparsely Gauged Watersheds. J. Hydrol. Eng. 2013, 18, 1272–1281. [CrossRef] 4. Grotch, S.L.; MacCracken, M.C. The Use of General Circulation Models to Predict Regional Climatic Change. J. Clim. 1991, 4, 286–303. [CrossRef] 5. Fowler, H.J.; Blenkinsop, S.; Tebaldi, C. Linking climate change modelling to impacts studies: Recent advances in downscaling techniques for hydrological modelling. Int. J. Climatol. 2007, 27, 1547–1578. [CrossRef] 6. Christensen, J.H.; Christensen, O.B. Climate modelling: Severe summertime ﬂooding in Europe. Nature 2003, 421, 805–806. [CrossRef][PubMed] 7. Maraun, D.; Wetterhall, F.; Ireson, A.M.; Chandler, R.E.; Kendon, E.J.; Widmann, M.; Brienen, S.; Rust, H.W.; Sauter, T.; Themeßl, M.; et al. Precipitation downscaling under climate change: Recent developments to bridge the gap between dynamical models and the end user. Rev. Geophys. 2010, 48, RG3003. [CrossRef] 8. Jang, S.; Kavvas, M.L. Downscaling global climate simulations to regional scales: Statistical downscaling versus dynamical downscaling. J. Hydrol. Eng. 2015, 20.[CrossRef] 9. Kavvas, M.L.; Chen, Z.Q.; Tan, L.; Soong, S.-T.; Terakawa, A.; Yoshitani, J.; Fukami, K. A regional-scale land surface parameterization based on areally-averaged hydrological conservation equations. Hydrol. Sci. J. 1998, 43, 611–631. [CrossRef] 10. Shrestha, R.; Tachikawa, Y.; Takara, K. Input data resolution analysis for distributed hydrological modeling. J. Hydrol. 2006, 319, 36–50. [CrossRef] 11. Haltas, I.; Kavvas, M.L. Scale Invariance and Self-Similarity in Hydrologic Processes in Space and Time. J. Hydrol. Eng. 2011, 16, 51–63. [CrossRef] 12. Diaz-Nieto, J.; Wilby, R.L. A comparison of statistical downscaling and climate change factor methods: Impacts on low ﬂows in the River Thames, United Kingdom. Clim. Chang. 2005, 69, 245–268. [CrossRef] Sustainability 2017, 9, 1457 16 of 17

13. Mearns, L.O.; Giorgi, F.; Whetton, P.; Pabon, D.; Hulme, M.; Lal, M. Guideline for Use of Climate Scenarios Developed from Regional Climate Model Experiments; Data Distribution Centre of the Intergovernmental Panel on Climate Change Task Group on Data and Scenario Support for Impact and Climate Analysis: Geneva, Switzerland, 2003. 14. Jones, R.G.; Murphy, J.M.; Noguer, M. Simulation of climate change over europe using a nested regional-climate model. I: Assessment of control climate, including sensitivity to location of lateral boundaries. Q. J. R. Meteorol. Soc. 1995, 121, 1413–1449. [CrossRef] 15. Frei, C.; Christensen, J.H.; Déqué, M.; Jacob, D.; Jones, R.G.; Vidale, P.L. Daily precipitation statistics in regional climate models: Evaluation and intercomparison for the European Alps. J. Geophys. Res. 2003, 108. [CrossRef] 16. Christensen, J.H.; Christensen, O.B. A summary of the PRUDENCE model projections of changes in European climate by the end of this century. Clim. Chang. 2007, 81, 7–30. [CrossRef] 17. Fowler, H.J.; Kilsby, C.G.; O’Connell, P.E.; Burton, A. A weather-type conditioned multi-site stochastic rainfall model for the generation of scenarios of climatic variability and change. J. Hydrol. 2005, 308, 50–66. [CrossRef] 18. Frei, C.; Schöll, R.; Fukutome, S.; Schmidli, J.; Vidale, P.L. Future change of precipitation extremes in Europe: Intercomparison of scenarios from regional climate models. J. Geophys. Res. 2006, 111, D06105. [CrossRef] 19. Murphy, J. An Evaluation of Statistical and Dynamical Techniques for Downscaling Local Climate. J. Clim. 1999, 12, 2256–2284. [CrossRef] 20. Leung, L.R.; Qian, Y.; Bian, X. Hydroclimate of the Western United States Based on Observations and Regional Climate Simulation of 1981–2000. Part I: Seasonal Statistics. J. Clim. 2003, 16, 1892–1911. [CrossRef] 21. Leung, L.R.; Qian, Y.; Bian, X.; Hunt, A. Hydroclimate of the Western United States Based on Observations and Regional Climate Simulation of 1981–2000. Part II: Mesoscale ENSO Anomalies. J. Clim. 2003, 16, 1912–1928. [CrossRef] 22. Leung, L.R.; Qian, Y. Atmospheric rivers induced heavy precipitation and ﬂooding in the western U.S. simulated by the WRF regional climate model. Geophys. Res. Lett. 2009, 36.[CrossRef] 23. Rauscher, S.A.; Coppola, E.; Piani, C.; Giorgi, F. Resolution effects on regional climate model simulations of seasonal precipitation over Europe. Clim. Dyn. 2010, 35, 685–711. [CrossRef] 24. Liu, J.; Bray, M.; Han, D. Sensitivity of the Weather Research and Forecasting (WRF) model to downscaling ratios and storm types in rainfall simulation. Hydrol. Process. 2012, 26, 3012–3031. [CrossRef] 25. Leung, L.R.; Qian, Y.; Bian, X.; Washington, W.M.; Han, J.; Roads, J.O. Mid-Century Ensemble Regional Climate Change Scenarios for the Western United States. Clim. Chang. 2004, 62, 75–113. [CrossRef] 26. Salathé, E.P.; Steed, R.; Mass, C.F.; Zahn, P.H. A High-Resolution Climate Model for the U.S. Paciﬁc Northwest: Mesoscale Feedbacks and Local Responses to Climate Change. J. Clim. 2008, 21, 5708–5726. [CrossRef] 27. Salathé, E.P.; Leung, L.R.; Qian, Y.; Zhang, Y. Regional climate model projections for the State of Washington. Clim. Chang. 2010, 102, 51–75. [CrossRef] 28. Gao, Y.; Leung, L.R.; Lu, J.; Liu, Y.; Huang, M.; Qian, Y. Robust spring drying in the southwestern U.S. and seasonal migration of wet/dry patterns in a warmer climate. Geophys. Res. Lett. 2014, 41, 1745–1751. [CrossRef] 29. Caldwell, P.; Chin, H.-N.S.; Bader, D.C.; Bala, G. Evaluation of a WRF dynamical downscaling simulation over California. Clim. Chang. 2009, 95, 499–521. [CrossRef] 30. Gao, Y.; Fu, J.S.; Drake, J.B.; Liu, Y.; Lamarque, J.-F. Projected changes of extreme weather events in the eastern United States based on a high resolution climate modeling system. Environ. Res. Lett. 2012, 7, 044025. [CrossRef] 31. Bachner, S.; Kapala, A.; Simmer, C. Evaluation of daily precipitation characteristics in the CLM and their sensitivity to parameterizations. Meteorol. Z. 2008, 17, 407–419. [CrossRef] 32. Fowler, H.J.; Ekström, M. Multi-model ensemble estimates of climate change impacts on UK seasonal precipitation extremes. Int. J. Climatol. 2009, 29, 385–416. [CrossRef] 33. Tatsumi, K.; Yamashiki, Y.; Takara, K. Effect of uncertainty in temperature and precipitation inputs and spatial resolution on the crop model. Hydrol. Res. Lett. 2011, 5, 52–57. [CrossRef] 34. Nakaegawa, T.; Nakakita, E. Comment on “Effect of uncertainty in temperature and precipitation inputs and spatial resolution on the crop model” by Kenichi Tatsumi, Yosuke Yamashiki and Kaoru Takara. Hydrol. Res. Lett. 2012, 6, 13–14. [CrossRef] Sustainability 2017, 9, 1457 17 of 17

35. California Data Exchange Center (CDEC). Available online: http://cdec.water.ca.gov/ (accessed on 10 August 2017). 36. National Elevation Dataset (NED). Available online: https://nationalmap.gov/elevation.html (accessed on 10 August 2017). 37. Global Land Cover Characteristics (GLCC) Data Base Version 2.0. Available online: https://lta.cr.usgs.gov/ glcc/globdoc2_0 (accessed on 10 August 2017). 38. Grell, G.A.; Dudhia, J.; Stauffer, D. A Description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5); Mesoscale and Microscale Meteorology Division, National Center for Atmospheric Research: Boulder, CO, USA, 1994; p. 121. 39. NCAR/NCEP reanalysis I data. Available online: http://rda.ucar.edu/datasets/ds090.1/ (accessed on 10 August 2017). 40. Giorgi, F.; Gutowski, W.J. Regional downscaling and the CORDEX initiative. Ann. Rev. Environ. Resour. 2015, 40, 467–490. [CrossRef] 41. Grubiši´c,V.; Vellore, R.K.; Huggins, A.W. Quantitative Precipitation Forecasting of Wintertime Storms in the Sierra Nevada: Sensitivity to the Microphysical Parameterization and Horizontal Resolution. Mon. Weather Rev. 2005, 133, 2834–2859. [CrossRef] 42. Reeves, H.D.; Lin, Y.-L.; Rotunno, R. Dynamic Forcing and Mesoscale Variability of Heavy Precipitation Events over the Sierra Nevada Mountains. Mon. Weather Rev. 2008, 136, 62–77. [CrossRef] 43. Ohara, N.; Kavvas, M.L.; Kure, S.; Chen, Z.Q.; Jang, S.; Tan, E. Physically Based Estimation of Maximum Precipitation over American River Watershed, California. J. Hydrol. Eng. 2011, 16, 351–361. [CrossRef] 44. Ishida, K.; Kavvas, M.L.; Jang, S.; Chen, Z.Q.; Ohara, N.; Anderson, M.L. Physically Based Estimation of Maximum Precipitation over Three Watersheds in Northern California: Atmospheric Boundary Condition Shifting. J. Hydrol. Eng. 2015, 20, 04014052. [CrossRef] 45. Ishida, K.; Kavvas, M.L.; Jang, S. Comparison of Performance on Watershed-Scale Precipitation between WRF and MM5. Am. Soc. Civ. Eng. 2015, 989–993. [CrossRef] 46. Skamarock, W.C.; Klemp, J.B.; Dudhia, J.; Gill, D.O.; Barker, D.M.; Wang, W.; Powers, J.G. A Description of the Advanced Research WRF Version 2; Mesoscale and Microscale Meteorology Division, National Center for Atmospheric Research: Boulder, CO, USA, 2005; p. 88. 47. PRISM Preciptation Data. Available online: http://prism.oregonstate.edu (accessed on 10 August 2017). 48. NCEP/EMC U.S. Gridded Precipitation. Available online: http://data.eol.ucar.edu/ (accessed on 10 August 2017). 49. Daly, C.; Halbleib, M.; Smith, J.I.; Gibson, W.P.; Doggett, M.K.; Taylor, G.H.; Curtis, J.; Pasteris, P.P. Physically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol. 2008, 28, 2031–2064. [CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).