搜档网
当前位置:搜档网 › Surface Boundary Conditions for Mesoscale Regional Climate Models

Surface Boundary Conditions for Mesoscale Regional Climate Models

Surface Boundary Conditions for Mesoscale Regional Climate Models
Surface Boundary Conditions for Mesoscale Regional Climate Models

Copyright?2005,Paper09-018;10,818words,10Figures,0Animations,5Tables.

https://www.sodocs.net/doc/4114177754.html,

Surface Boundary Conditions

for Mesoscale Regional

Climate Models

Xin-Zhong Liang,*Hyun I.Choi,and Kenneth E.Kunkel

Illinois State Water Survey,University of Illinois at Urbana–Champaign,Champaign, Illinois

Yongjiu Dai

Research Center for Remote Sensing and GIS,Beijing Normal University,Beijing, China

Everette Joseph

Department of Physics and Astronomy,Howard University,Washington,D.C. Julian X.L.Wang

National Oceanic and Atmospheric Administration/Air Resources Laboratory, Silver Spring,Maryland

Praveen Kumar

Department of Civil and Environmental Engineering,University of Illinois at Urbana–Champaign,Champaign,Illinois

Received8November2004;accepted25March2005

ABSTRACT:This paper utilizes the best available quality data from mul-

tiple sources to develop consistent surface boundary conditions(SBCs)for

mesoscale regional climate model(RCM)applications.The primary SBCs

include1)fields of soil characteristic(bedrock depth,and sand and clay

*Corresponding author address:Dr.Xin-Zhong Liang,Illinois State Water Survey,Univer-sity of Illinois at Urbana–Champaign,2204Griffith Dr.,Champaign,IL61820–7495.

E-mail address:xliang@https://www.sodocs.net/doc/4114177754.html,

fraction profiles),which for the first time have been consistently introduced to

define3D soil properties;2)fields of vegetation characteristic fields(land-

cover category,and static fractional vegetation cover and varying leaf-plus-

stem-area indices)to represent spatial and temporal variations of vegetation

with improved data coherence and physical realism;and3)daily sea surface

temperature variations based on the most appropriate data currently available or

other value-added alternatives.For each field,multiple data sources are com-

pared to quantify uncertainties for selecting the best one or merged to create a

consistent and complete spatial and temporal coverage.The SBCs so developed

can be readily incorporated into any RCM suitable for U.S.climate and hy-

drology modeling studies,while the data processing and validation procedures

can be more generally applied to construct SBCs for any specific domain over

the globe.

KEYWORDS:Surface boundary conditions;Regional climate model;Land

cover;Leaf area index;Fractional vegetation cover;Soil fraction profiles;

Bedrock depth;Sea surface temperature

1.Introduction

Mesoscale regional climate models(RCMs)are recognized as an increasingly important tool to address scientific issues associated with climate variability, changes,and impacts at local–regional scales(Giorgi and Mearns1999;Giorgi et al.2001;Leung et al.2003).Numerous RCMs have been developed,applied,and intercompared,demonstrating important downscaling skills,but also model defi-ciencies yet to be resolved(Takle et al.1999;Leung et al.1999;Roads et al.2003; and references therein).The most widely used RCM has been the second-generation regional climate modeling system(RegCM2),developed by Giorgi et al.(1993a;Giorgi et al.1993b)based on the fourth-generation Pennsylvania State University–National Center for Atmospheric Research(PSU–NCAR)Mesoscale Model(MM4;Anthes et al.1987).Over the years,the hydrostatic MM4was significantly improved and eventually replaced by the nonhydrostatic MM5(Grell et al.1994;Dudhia et al.2000).Several RCMs built upon the MM5then emerged to address a wide range of applications(Leung and Ghan1999;Liang et al.2001; Liang et al.2004a;Liang et al.2004b;Wei et al.2002).Meanwhile,the next-generation Weather Research and Forecasting(WRF)model has been developed (Klemp et al.2000;Michalakes2000;Chen and Dudhia2000;more information available online at https://www.sodocs.net/doc/4114177754.html,/index.php)to supersede the MM5. Accordingly,the Illinois State Water Survey then initiated an effort to develop a climate extension of the WRF(CWRF)by implementing numerous crucial improvements,including surface–atmosphere interaction,convection–cloud–radiation interaction,and system consistency throughout all process modules(see Liang et al.2005for an introductory overview).This extension inclusively incor-porates all WRF functionalities for numerical weather predictions while enhancing the capability for climate applications.

For all RCMs,one essential component is the representation of surface–atmosphere interactions,which generally requires specification of surface bound-ary conditions(SBCs)over both land and oceans.However,there is no universal, complete set of SBCs that satisfies all models.For example,the WRF release

version2included the six-layer Rapid Uptake Cycle(RUC;Smirnova et al.2000) and the four-layer Noah(Chen and Dudhia2001;Ek et al.2003)land surface models(LSMs),while the CWRF added the11-layer Common Land Model (CLM;Dai et al.2003;Dai et al.2004).Over oceans,observed daily sea surface temperature(SST)variations have been incorporated into the CWRF,integrated with all surface modules.Only the CLM predicts water temperature profiles for inland shallow and deep lakes(Bonan1995).The RUC and Noah require the soil texture category to define static soil properties uniformly distributed throughout all layers and assume that bedrock is below the bottom layer everywhere,both of which are unrealistic.By contrast,the CLM needs soil sand and clay fraction profiles to specify soil properties in individual layers and bedrock depth to set the soil bottom impermeable to water.Meanwhile,all modules use the land-cover category(LCC)to define static canopy(morphological,optical,physiological) properties,which are more comprehensive in the CLM.With regard to these aspects,the CLM approach,albeit more demanding,is physically more appropriate and,fortunately,viable with current data availability.

A more troublesome issue is that the same SBC field may be inconsistently defined,used,or specified by different surface modules.For example,to distin-guish canopy versus bare soil contributions,all modules require the fraction of the vegetated area in an RCM grid box.The RUC and Noah prescribe this fraction by monthly climatological means(Gutman and Ignatov1998),but presently kept fixed at the initial condition.In accounting for the leaf density effect on canopy resistance,the Noah introduces the leaf area index(LAI),which is set to a uni-versal and fixed value of4over the globe.In contrast,the CLM uses the combi-nation of a static fractional vegetation cover(FVC)and time-varying LAI to describe dynamic canopy variations.A relevant question then is which parameter (FVC or LAI,or possibly both)should carry the information about the time variations of terrestrial vegetation phenology.Data for the global distribution at fine spatial and temporal scales can only be determined by means of remote sensing,such as the satellite product of normalized difference vegetation index (NDVI).Sellers et al.(Sellers et al.1996)incorporated all geographic and seasonal variations of NDVI into LAI distributions.Gutman and Ignatov(Gutman and Ignatov1998)revealed that the limited information contained in NDVI precludes construction of seasonal variations for both FVC and LAI,and thus derived time-varying FVC while prescribing a constant LAI.Recently,Zeng et al.(Zeng et al. 2000)argued that the assumption of a static FVC and varying LAI is more realistic from a modeling perspective and the model implementation of this assumption is made feasible by current data availability.As such,FVC is determined by distinct vegetation categories and long-term edaphic and climatic controls,whereas LAI includes all dynamic canopy variability.This study concurs with Zeng et al.and parameterizes,as in the CWRF CLM,the3D canopy effects by the combination of the static FVC for the fractional area of vegetation covering a model grid (horizontal extent)and varying LAI for the abundance of green leaves of the vegetated area(vertical density).

It is advantageous that data sources of unprecedented scope are currently avail-able to specify the SBC fields discussed above.The data quality problem,how-ever,is often overlooked.No single data source can provide a long-term continu-ous record of a specific field,nor can multiple sources ensure consistency between

fields.The goal of this study is to develop a coherent,realistic set of SBCs that are most suitable for mesoscale climate and hydrology modeling.This study focuses on those fields that have multiple data sources but contain significant uncertainty, inconsistency,and incompleteness.This requires both objective procedures and manually intensive efforts to process data.Although the data processing is specific to the CWRF,the resulting SBCs by design can be incorporated into all RCMs for climate studies in North America and the procedures are more generally applicable over the globe.

2.General consideration

The SBCs data quality and value representation largely depend on the RCM computational domain and grid resolution.For our simulations of U.S.climate,the domain is centered at(37.5°N,95.5°W)using the Lambert Conformal Conic map projection and30-km horizontal grid spacing,with total points of196(west–east)×139(south–north),covering most of North America.The domain has been demonstrated to facilitate skillful simulations of temporal and spatial variations of precipitation over North America(Liang et al.2001;Liang et al.2004a;Liang et al.2004b).In this study,all SBCs are constructed and displayed on this RCM domain,suitable for U.S.applications.For convenience,the geographic location of a point is hereafter referred to as a“pixel”for raw data and a“grid”for the RCM result.A given value at a pixel or grid represents the area surrounding the point as defined by its respective horizontal spacing.

A critical requirement in constructing the SBCs is that each field must be globally defined with no missing value and physical consistency must be main-tained across all relevant parameters.Missing data,if any,must be appropriately filled.For mesoscale weather and climate modeling,the raw data should be avail-able at the finest possible resolution.This will facilitate a more realistic represen-tation of surface heterogeneity effects.When the data resolution is sufficiently finer than the RCM grid,the subgrid effects can be further incorporated using composite,mosaic,or statistical–dynamical approaches(Avissar and Pielke1989; Koster and Suarez1992;Dickinson et al.1993;Giorgi1997;Leung and Ghan 1998).Although many raw datasets collected in this study have adequate resolu-tions(as fine as1km)to account for the subgrid effects,this study presents the SBCs only on a given RCM grid where a single dominant surface type is assumed. Note that Masson et al.(Masson et al.2003)developed a global database at1-km resolution for several land surface parameters,1but their data are insufficient for the CWRF applications,especially in terms of those fields discussed in this study. The existing observational databases have various resolutions,finer or coarser than the RCM grid,a wide range of map projections and data formats,and often contain missing values or inconsistencies between variables.This presents signifi-cant challenges and requires labor-intensive efforts to process the data onto the 1Among the variables listed in Table1,Masson et al.only included FVC and LAI.Their FVC was parameterized as1?e?0.6LAI for crops and assigned with a constant for eight other vegetation types,while LAI was given as only a climatological mean.This study derives both FVC and LAI from satellite measurements and also includes interannual variations of LAI.

Table1.The list of the primary SBCs incorporated into the CWRF.

Name Description Units Level Time DBED Bedrock,lakebed,or seafloor depth m1Static SAND Soil sand fraction profile11Static CLAY Soil clay fraction profile11Static LCC Land-cover category1Static FVC Fractional vegetation cover1Static LAI Leaf area index m2m?21Daily SAI Stem area index m2m?21Daily SST Sea surface temperature K1Daily RCM-specific grid mesh.Our objective procedures employ the Geographic Infor-mation System(GIS)software application tools2to do horizontal data remapping (Liang et al.2005).In particular,the GIS tools are used to first determine the geographic conversion information from a specific map projection of each raw data to the identical RCM grid system.The information includes location indices, geometric distances,or fractional areas of all input cells contributing to each RCM grid.It can then be applied to remap all variables of the same projection.The remapping is completed by a bilinear interpolation method in terms of the geo-metric distances if the raw data resolution is low or otherwise a mass conservative approach as weighted by the fractional areas.Some raw input data available only at coarse resolutions(e.g.,1°),especially those for land or ocean only,contain gaps along coastal regions and over islands that are resolved by the RCM.These gaps are filled by extrapolating from available adjacent values using the bilinear ap-proximation.Even the relatively finer resolution(1and8km)input data such as soil fraction and LAI have missing value pixels.They are filled by the average over the nearby data pixels having the same LCC within a certain radius around a missing point.Here the number of pixels and the range of radius used for filling depend on the resolution of the raw input data(see below).

3.Surface boundary conditions

Table1lists the key SBCs that are the focus of this study.The soil characteristic fields(DBED,SAND,CLAY)have for the first time ever been consistently introduced into climate or hydrology models.The vegetation characteristic fields (LCC,FVC,LAI,SAI)have been improved with data coherence and physical realism.The incorporation of daily SST variations is the minimal requirement enabling the CWRF for climate applications.Liang et al.(Liang et al.2005) presented the details about the raw data sources and processing procedures for a 2The GIS tools are Arc/Info and Arc/Map from the Environmental Systems Research Insti-tute,Inc.In particular,IMAGEGRID and GRIDPOLY convert input data from the image to the ArcGIS raster grid and to the polygon coverage formats,respectively;PROJECT remaps the raw input data onto the CWRF grid projection;UNION and CLIP geometrically intersect polygon features of input data with the CWRF grid mesh and extract the fractional area of each pixel contributing to the grid;GRID DOCELL and IF statements conditionally merge,replace,or adjust different input datasets for an improved product.

comprehensive set of SBCs that have been developed for CWRF use.This study describes the scientific rationale,insights,and decisions that were made to develop the key SBCs in a manner that is reasonable and consistent with our understanding of the surface.

3.1.Soil characteristics

The CWRF CLM uses the bedrock depth to determine thermal and hydraulic properties in terms of the sand and clay fraction profiles for the soil layers above the bedrock or otherwise of rocks.The CLM also predicts water temperature profiles separately for shallow and deep lakes,distinguished by the lakebed depth. The CWRF dynamic ocean module needs specification of the seafloor depth to define the lower boundary of the water circulation,which can be integrated with the terrestrial hydrology module and a comprehensive routing scheme to predict the water level of major inland water bodies,including the Mississippi River and the Great Lakes.These features are important for regional climate simulations, although other RCMs have yet to incorporate them.This study thus develops static soil characteristics(DBED,SAND,CLAY)for general application in regional climate and hydrology models.

3.1.1.Bedrock,lakebed,or seafloor depth

The DBED consists of the bedrock,lakebed,and seafloor depth,and when com-bined,defines the bottom of all surface modules impermeable to water3over the entire RCM domain.The lakebed depth is calculated by subtracting the lake topographic data from mean water surface elevations.Currently,only the Great Lakes topographic data are available at2.56-km spacing from the National Oce-anic and Atmospheric Administration(NOAA)Great Lakes Environmental Re-search Laboratory,whereas all others within the domain have no digital data available and are set to be10m deep.The seafloor depth is based on the global 2-min bathymetry data from National Geophysical Data Center(Smith and Sand-well1997;Jakobsson et al.2001).

The bedrock depth is defined as the depth of soil and/or unconsolidated material that lies between the land surface and the geologic substratum(Miller and White 1998).The data are a combination of the Continental United States Multi-Layer Soil Characteristics Dataset(CONUS–SOIL)and,outside of the United States,the Food and Agriculture Organization of the United Educational,Scientific,and Cultural Organization(FAO–UNESCO)Soil Map of the World.The FAO–UNESCO includes the global5-min distribution of mapping units(FAO1996), each containing1–8soil units among the106categories.For each mapping unit, all soil units are first assigned with their respective depth upper bounds(i.e.,10, 3Strictly speaking,water can penetrate through the bedrock between gaps of consolidated material.This penetration represents surface and groundwater interactions,an aspect that can be explored where detailed bedrock information is available.

50,100,150,or300cm,provided by Dr.C.Reynolds of the U.S.Department of Agriculture Foreign Agricultural Service),and then integrated with their occur-rence rates to estimate the mean bedrock depth.The CONUS–SOIL,developed from the State Soil Geographic Database(STATSGO),has a finer resolution of 1-km spacing(Miller and White1998).About one-third of the data pixels,how-ever,were coded as152cm,which generally indicates the maximum depth of soil data normally examined where bedrock was not actually encountered.A compari-son showed that most regions with bedrock deeper than152cm in the CONUS–SOIL data are overlaid with certain FAO–UNESCO mapping units having soil depths of150cm.Given large uncertainties involved in these estimates,a uniform bedrock depth of600cm(deeper than the bottom of the last CLM soil layer)is subjectively reassigned to all CONUS–SOIL pixels with values of152cm and the corresponding FAO–UNESCO mapping units.

Figure1depicts the geographic distribution of DBED over the RCM domain. Sizeable areas of deep soils(bedrock below300cm)are found in the central United States extending into south-central Canada and along the Gulf and South Atlantic coasts.Many of the mountainous areas of the western United States and along the Appalachians as well as much of Canada and Mexico are characterized by shallow soils.The bedrock acts as a bottom lid that effectively prevents down-ward water flux.It raises the water table and limits moisture storage available in the soil column above the lid.Consequently,the bedrock controls the subsurface moisture dynamics,which in turn has significant impact on surface energy and water flux dynamics(Chen and Kumar2001).Exposed and shallow bedrock keeps water close to the topographic surface and available for evaporation during wet periods while having a disproportionately large sensible heat but little to no evapo-

Figure1.The geographic distribution of DBED.

rative flux under dry conditions(Spence and Rouse2002).Local bedrock topog-raphy may be highly significant for runoff generation(Freer et al.2002)and hillslope–riparian linkage(Katsuyama et al.2005).Thus the bedrock geographic distribution can cause the terrestrial hydrology to exhibit considerable spatial variability,with greater soil moisture memory in deeper zones.General neglect of this variability in most LSMs likely results in unrealistic representation of the regional water recycling process.The deep nature of the Great Lakes is in clear contrast to other lakes as well as the large extent of shallow water along the Atlantic and Gulf Coast.Such contrast,when incorporated,will enable physically realistic simulations of processes related to the thermal inertia and vertical mixing of water.

3.1.2.Sand and clay fraction profiles

Increasing evidence shows the necessity to incorporate both horizontal and vertical soil heterogeneity effects for realistic hydrology modeling(Choi et al.2005,un-published manuscript).The key element in such modeling is the accurate speci-fication of soil thermal and hydraulic properties,including specific heat capacity of dry soil,thermal conductivity of dry soil,porosity,saturated negative potential, saturated hydraulic conductivity,and the exponent B defined in Clapp and Horn-berger(Clapp and Hornberger1978).The CLM requires SAND and CLAY to parameterize these properties(Bonan1996;Dai et al.2003)following Cosby et al. (Cosby et al.1984).Consistent with the bedrock depth,these profiles are deter-mined by the combination of the CONUS–SOIL in the United States and FAO–UNESCO for the rest,and presented in terms of the CLM soil layer structure (Table2).Similar profiles can be readily constructed for given layers of any LSM. The FAO–UNESCO global5-min distributions of sand and clay fractions for the two layers(0–30and30–100cm)was produced by Reynolds et al.(Reynolds et al.2000).The top layer data are uniformly assigned for the five CLM layers above28.91cm,while the bottom layer values are assigned for the remaining. Over the United States,they are replaced by the CONUS–SOIL1-km distributions of sand and clay fractions in11standard layers,divided at5,10,20,30,40,60, 80,100,150,200,and250cm(Miller and White1998).Since the original values below152cm were likely not direct measurements,the raw data for10th and11th Table2.The CLM soil layer thickness and depth(cm).

Layer Layer thickness Layer node depth Layer bottom depth

1 1.750.71 1.75

2 2.76 2.79 4.51

3 4.55 6.239.06

47.5011.8916.55

512.3621.2228.91 620.3836.6149.29 733.6061.9882.89 855.39103.80138.28 991.33172.76229.61 10150.58286.46380.19 11187.45473.92567.64

standard layers(150–250cm)are discarded.The data in the top nine standard layers are interpolated with thickness weighting to the eight CLM layers above 138.28cm,while those of the ninth standard layer are extended uniformly down to the remaining layers.When bedrock is located within a CLM layer,the aver-aging applies an additional thickness weight for the portion of the layer above bedrock.Some regions have soil texture classified as“organic material”without sand and clay data.They are mainly distributed in Florida,Minnesota,and several western states.These regions are assigned with a negative unit of sand or clay as an indicator to use the organic material properties in the CLM.The CONUS–SOIL also contains points with soil texture classified as“others,”giving no sand and clay data.Each missing point is filled by averaging over all nearby data pixels having the same land-cover category(see below)within a certain radius starting from10 km(440pixels)around the point and increasing until a minimum of50data pixels are obtained.

Figure2shows the geographic distributions of SAND and CLAY for the first and eighth CLM layers on the RCM domain.The central United States is char-acterized by low sand and rather moderate clay fractions,a combination that promotes high water-holding capacity and easy root penetration by plants.High sand fractions are found in southeast Canada and the southeast United States, reducing the water-holding capacity and increasing the likelihood of soil moisture

Figure2.The geographic distributions of SAND and CLAY(%)for the(a),(b)the first (0–1.75cm)and(c),(d)the eighth(82.89–138.28cm)CLM layers.

stress.Although the variability is smaller in the vertical than horizontal,both spatial contrasts have been demonstrated to be important for modeling moisture movement in soil(Choi et al.2005,unpublished manuscript).

3.2.Vegetation characteristics

Several vegetation characteristics are widely used in LSMs,mainly by prescription from observational proxies.There exist many data sources,which contain sub-stantial biases in each and inconsistencies between.These deficiencies certainly affect model simulations,causing result uncertainty and incomparability.This study thus develops a consistent set of vegetation characteristics(LCC,FVC,LAI, SAI)for general application.In particular,the study elaborates on how to achieve the consistency between different data sources so that a coherent long-term record can be established to realistically represent the vegetation variability.

https://www.sodocs.net/doc/4114177754.html,nd-cover category

This study adopts the U.S.Geological Survey(USGS)24-category land-cover classification system,developed using the global1-km resolution Advanced Very High Resolution Radiometer(AVHRR)satellite-derived NDVI composites from April1992to March1993.Within each RCM grid,the contributing area for each of the24land-cover categories is summed over all pixels of the same category. The majority category that contributes the largest area is chosen as the LCC for the grid.When the fractional area of water bodies(shallow or deep lakes,sea ice,or ocean)is less than0.5but dominates in a grid,the second major category is chosen as the LCC for the grid.Instead of the USGS system,some LSMs have used the International Geosphere Biosphere Program(IGBP)17-category land-cover clas-sification system(Belward1996;Loveland et al.2000).Obviously,the correspon-dence between the USGS and IGBP categories4is not1to1,but contains cross references.For a consistent conversion when needed(see below),the two raw land-cover distribution maps are intersected using the GIS tools to determine the fractional areas of all contributing categories within each RCM grid and thus the USGS–IGBP correspondences.Table3summarizes the total percentage coverage of each land-cover category in the RCM domain and over the globe for both the USGS and IGBP classification systems as well as their correspondences. Figure3illustrates the USGS-based LCC geographic distribution.There is a general transition from deciduous and evergreen forests in the east and southeast United States to dryland cropland in the central United States to grassland and shrubland in the northern and western United States to forests in the northwest and mixed forests in southeast Canada.Model differences in energy budget partition-ing and friction drag can be expected across the boundaries between regions because of differences in vegetative-cover fraction,roughness length,seasonal cycle,and root penetration.Note that the raw data do not contain categories4and 20over the globe,and additionally12,17,and23within the RCM domain.

4This study prefers the USGS to IGBP classification system mainly because the former has been used as the basic land-cover identification in the MM5and WRF and also contains more categories(24versus17)than the latter.

Table3.The comparison of the total percentage coverage and the NDVI value for a complete coverage(N c,?)of each USGS land-cover category in the RCM domain and over the globe as well as the correspondences of all contributing IGBP clas-sification categories.

USGS land-use/land-cover legend IGBP land-cover legend

N

c,?Distribution

ratio

(%)

Contributing ratio

for corresponding

USGS legend(%)

Type Description RCM Global RCM Global Type Description N

c,?

RCM Global 1Urban and built-up land0.620.620.340.0313Urban and built-up0.62100100 2Dryland cropland and

pasture

0.610.61 5.49 1.9612Croplands0.61100100

3Irrigated cropland and

pasture 0.610.610.480.4512Croplands0.6110094.41

14Cropland/natural

vegetation mosaic

0.650 5.59

4Mixed dryland/irrigated

cropland and pasture*

5Cropland/grassland

mosaic 0.650.65 4.420.7012Croplands0.610 1.33

14Cropland/natural

vegetation mosaic

0.6510098.67

6Cropland/woodland

mosaic 0.650.65 2.26 1.0514Cropland/natural

vegetation mosaic

0.65100100

7Grassland0.490.49 6.10 1.9210Grasslands0.49100100

8Shrubland0.600.608.23 2.386Closed shrublands0.60 6.6714.81

7Open shrublands0.6079.5281.00

8Woody savannas0.6213.81 4.19

9Mixed shrubland/

grassland 0.600.590.110.336Closed shrublands0.6010015.28

7Open shrublands0.60076.53

10Grasslands0.4908.19

10Savanna0.610.600.99 2.078Woody savannas0.6286.8850.48

9Savanna0.5813.1249.52

11Deciduous broadleaf

forest 0.690.70 5.46 1.072Evergreen broadleaf

forest

0.69020.41

4Deciduous broadleaf

forest

0.7068.7868.26

5Mixed forest0.6831.2211.33

12Deciduous needleleaf

forest**0.630.000.494Deciduous needleleaf

forest

0.63100

13Evergreen broadleaf

forest 0.690.690.08 1.562Evergreen broadleaf

forest

0.69100100

14Evergreen needleleaf

forest 0.630.6310.32 1.051Evergreen needleleaf

forest

0.63100100

15Mixed forest0.680.687.59 1.615Mixed forest0.68100100 16Water bodies43.0967.5717Water bodies100100 17Herbaceous wetland**0.560.000.0111Permanent wetlands0.56100 18Wooded wetland0.560.560.840.2211Permanent wetlands0.56100100

19Barren or sparsely

vegetated 0.600.600.46 2.4716Barren or sparsely

vegetated

0.60100100

20Herbaceous tundra*

21Wooded tundra0.600.61 3.35 1.867Open shrublands0.6010068.11

8Woody savannas0.62031.89 22Mixed tundra0.600.600.370.7116Barren or sparsely

vegetated

0.60100100 23Bare ground tundra**0.600.000.0216Barren or sparsely

vegetated

0.60100 24Snow or ice0.0310.4715Snow and ice100100

*Land-cover type does not exist in the global dataset.

**Land-cover type does not exist in the CWRF domain.

Moreover,category 24is not chosen as the majority type for LCC.Therefore,the final LCC includes only 18land-cover categories over this RCM domain.

3.2.2.Fractional vegetation cover

The FVC is the one ecological parameter that determines the contribution parti-tioning between bare soil and vegetation for surface evapotranspiration,photosyn-thesis,albedo,and other fluxes crucial to land –atmosphere interactions.It is

as-

Figure 3.The geographic distribution of LCC,with only 18categories occurring over

the RCM domain.Outlined are five key regions of interest,each with a predominant category:Texas (grassland),the Southwest (shrubland),the Midwest (dryland cropland and pasture),the Southeast (evergreen needleleaf forest),and the Northeast (deciduous broadleaf forest)United States.

sumed to be time invariant or static,and derived following Zeng et al.(Zeng et al.2000;Zeng et al.2002),from the same global 1-km AVHRR satellite product as for LCC.The 10-day composites from April 1992to March 1993were used to determine the annual maximum NDVI (N p ,max )for each land-cover category,minimizing the effect of cloud contamination on data quality.For each pixel,the vegetation cover is computed by

C ?=N p ,max ?N s N c ,??N s ,(1)

where N c,?is the NDVI value for a complete coverage of a specific USGS land-cover category over the pixel and N s for bare soil.Zeng et al.(Zeng et al.2000),using a commercial imagery database,determined N c,?by examining percentiles of the N p ,max histogram for each IGBP land-cover category.To avoid redundant data processing,the N c,?value for each USGS land-cover category is calculated from those of all contributing IGBP categories as weighted by their corresponding fractional areas (Table 3).After Zeng et al.(Zeng et al.2000),a uniform value of 0.05is assigned to N s for all USGS land-use categories.

There exist significant differences between the NDVI from the AVHRR and the most recent Moderate Resolution Imaging Spectroradiometer (MODIS)sensors.Gallo et al.(Gallo et al.2004)compared the concurrent 16-day composite data during 2001and showed that a linear relationship exists between the two.The regression intercept and slope values change with land-cover categories,but all are significantly different from 0and 1,respectively.The MODIS has generally larger values than the AVHRR,causing Equation (1)to produce greater C ?values.On the other hand,Zeng et al.(Zeng et al.2002;Zeng et al.2003)demonstrated that,using the same method with Equation (1),the C ?derived from 8-km AVHRR NDVI during 1982–2000(James and Kalluri 1994)is consistent with that from the 1-km data for April 1992–March 1993.Given the good agreement with field surveys and observational studies and the small interannual variability over areas with minimal anthropogenic impact,the FVC derived from the AVHRR NDVI was believed to be robust.

The MODIS is providing quality-controlled data for numerous variables that are necessary for terrestrial modeling,such as developing a new land surface albedo parameterization (Liang et al.2004c).It is thus desirable to have a consistent FVC based on the MODIS data.One appealing approach is to scale C ?from the MODIS toward AVHRR.For each USGS land-cover category,a scaling factor f p,?is first defined to remove the systematic difference of MODIS from AVHRR in N p ,max averaged over all pixels.Assuming the same N s and multiplying N p ,max by f p,?in Equation (1),the corresponding N c,?is then estimated to minimize the C ?differ-ence between MODIS and AVHRR.Table 4lists the resulting f p,?and N c,?values as well as the correlation coefficients and root-mean-square (rms)differences between the C ?based on the AVHRR and MODIS after scaling.The f p,?ranges from 0.50to 0.81,while the N c,?remains close to the respective AVHRR value except for category 19.The correlations are generally excellent,mostly above 0.5,around 0.4for categories 18and 22,but quite low (~0.3)for categories 6and 19.Nonetheless,the rms differences are small for all categories.It is noteworthy that the f p,?correction has almost no impact on the correlations but reduces the rms

differences.The low correlations reflect the poor correspondences between the raw MODIS and AVHRR N p ,max data at 1-km resolution,for which no clear explana-tion can be given.

The final FVC is obtained by the area-weighted averaging of C ?values for all pixels within each RCM grid.Figure 4compares the FVC geographic distributions derived from the AVHRR and scaled MODIS data over the RCM domain.The two distributions are very similar,both in pattern and magnitude.This similarity is a direct result of the above described correction procedure;without this correction,the MODIS values are excessive almost everywhere.There are certain minor differences between the two.The values are somewhat lower in the MODIS compared to AVHRR for some western portion of the domain,most apparent in West Texas,New Mexico,and northern Mexico,whereas the opposite situation occurs for some eastern areas,particularly around Hudson ’s Bay and Florida.These slight differences are unlikely to have a major impact on model simulations.

3.2.3.Leaf and stem area index

LAI and SAI are defined as the total one-sided area of all green canopy elements and stems plus dead leaves,respectively,over the vegetated ground area.They are constructed from the global monthly mean distributions of green vegetation leaf area index,based on the AVHRR NDVI data,during July 1981–December 1999at 8-km spacing (Zhou et al.2001;Buermann et al.2002).There exist missing data

Table 4.The estimated f p,?and N c,?for C ?based on MODIS NDVI (2000–03).Type

USGS land-cover legend f p,?N c,?Correlation Rms error 1

Urban and built-up land 0.770.630.810.142

Dryland cropland and pasture 0.780.610.630.123

Irrigated cropland and pasture 0.810.620.530.164

Mixed dryland/irrigated cropland and pasture*5

Cropland/grassland mosaic 0.780.650.800.116

Cropland/woodland mosaic 0.780.650.260.087

Grassland 0.770.510.690.188

Shrubland 0.750.640.820.139

Mixed shrubland/grassland 0.810.620.640.1410

Savanna 0.770.620.590.1211

Deciduous broadleaf forest 0.810.690.660.0812

Deciduous needleleaf forest**13

Evergreen broadleaf forest 0.770.690.780.0614

Evergreen needleleaf forest 0.750.640.640.1115

Mixed forest 0.740.680.580.1216

Water bodies 17

Herbaceous wetland**18

Wooded wetland 0.590.560.420.1319

Barren or sparsely vegetated 0.530.960.280.1020

Herbaceous tundra*21

Wooded tundra 0.630.620.610.1622

Mixed tundra 0.500.660.360.19

23

Bare ground tundra**24Snow or ice *Land-cover type does not exist in the global dataset.

**Land-cover type does not exist in the CWRF domain.

zones in some regions with land cover of urban and built-up,permanent wetlands,marshes,tundra,barren,desert,or very sparsely vegetated area.These missing zones are filled by the average over nearby data pixels having the same land-cover category within a certain radius starting from 16km (24pixels)around a missing point and increasing until a minimum of three pixels are obtained.The product is denoted as L raw .

Since L raw is defined with respect to unit ground area,it is divided by local vegetation cover C ?to define L gv representing the green leaf area index with respect to vegetated area only (Zeng et al.2002).Due to inconsistency between C ?and L raw data at individual pixels,some L gv values are abnormally large,up to several hundreds.The inconsistency arises mainly because C ?was derived based on the 24USGS land-cover categories at 1-km spacing while L raw was derived in terms of six alternative biomes with distinct vegetation structures at an 8-km interval.Zeng et al.(Zeng et al.2002)determined C ?at every point,while defining LAI for each IGBP land-cover category by a mean seasonal variation within a 10°latitude zone.Here the 1-km C ?data is first integrated onto the 8-km L raw map to compute initial estimates of L gv and then a smoothing filter is applied to remove abnormal values.The filter is designed,through trial and error,by examining the frequency distribution of abnormal L gv values and considering the canopy dis-placement height in the CLM for each USGS land-cover category.The point value that exceeds the filter threshold listed in Table 5is filled by the average over nearby data pixels having the same land-cover category within a certain radius starting from 16km (24pixels)around the point and increasing until a minimum of three pixels are obtained.In addition,L gv data contain large uncertainties in winter due to cloud contamination,especially for the USGS categories 13and 14(evergreen broadleaf and needleleaf forests).Following Zeng et al.(Zeng et al.2002),L gv values in winter months for these two categories are adjusted by

L gv =max ?L gv ,c L gv,max ?,(2)

where correction coefficient c is 0.8(0.7)for category 13(14),and L gv,max is the maximum L gv .For the climatology,the maximum can be determined over

all

Figure 4.The geographic distributions of FVC derived using the NDVI data from (a)

the AVHRR (Apr 1992–Mar 1993)and (b)the scaled MODIS (Jan 2000–Dec 2003).

monthly values during the entire period,while for interannual variations it is taken in three consecutive years.

For each USGS land-cover category,SAI is then approximated as in Zeng et al.(Zeng et al.2002)by

SAI m =max{SAI min ,??SAI m ?1+max ?LAI m ?1?LAI m ,0??},(3)

where m denotes m th month,SAI min the prescribed minimum SAI,and (1??)the monthly removal rate of dead leaves.Both ?and SAI min are listed in Table 5.The resulting SAI for most land-cover categories reach the minimum in winter and the maximum in fall (October or November).This seasonal trend may not be appro-priate for certain categories,especially those with croplands where nothing may remain on the field after crops are harvested in fall.

A serious concern is the systematic difference in the L raw products based on the AVHRR (Zhou et al.2001)and the recent MODIS (Knyazikhin et al.1998;Myneni et al.2002)data.The MODIS L raw ,available from February 2000onward,has a finer resolution at 1-km spacing.Following the same procedure described above,the corresponding LAI and SAI can be constructed from the MODIS L raw .Figure 5depicts the April and July mean LAI distributions of the AVHRR and

Table 5.The ecological parameters in deriving LAI and SAI for each USGS land cover.

Type

USGS land-cover legend FVC (1km)Displacement height c (m)L gv filter threshold ?SAI min RCM Global 1

Urban and built-up land 0.7670.7350.66770.000.12

Dryland cropland and pasture 0.9410.8750.66770.000.13

Irrigated cropland and pasture 0.8820.8040.66770.000.14

Mixed dryland/irrigated cropland and pasture a 0.6675

Cropland/grassland mosaic 0.8230.7290.66770.250.56

Cropland/woodland mosaic 0.9580.8690.66770.250.57

Grassland 0.8050.7110.66760.50 1.08

Shrubland 0.4170.3810.33350.50 1.09

Mixed shrubland/grassland 0.7220.3910.33350.50 1.010

Savanna 0.8990.8480.66770.50 1.011

Deciduous broadleaf forest 0.9470.87113.33380.50 1.012

Deciduous needleleaf forest b 0.92013.33380.50 1.013

Evergreen broadleaf forest 0.9550.95323.33380.50 1.014

Evergreen needleleaf forest 0.8980.89513.33380.50 1.015

Mixed forest 0.8480.87513.33380.50 1.016

Water bodies 0.66717

Herbaceous wetland b 0.94713.33360.50 1.018

Wooded wetland 0.7290.8350.66780.50 1.019

Barren or sparsely vegetated 0.0610.0730.33340.50 1.020

Herbaceous tundra a 0.66721

Wooded tundra 0.7040.7140.66760.50 1.022

Mixed tundra 0.3960.3230.33360.50 1.023

Bare ground tundra b 0.0180.33360.50 1.024

Snow or ice 0.667a

Land-cover type does not exist in the global dataset.b

Land-cover type does not exist in the CWRF domain.c This is taken from the CLM.

MODIS climatologies over the RCM domain,while Figure 6presents seasonal variations of the five key regions outlined in Figure 3.The MODIS values are clearly smaller,which is not a result of long-term trends.Figure 7compares AVHRR and MODIS monthly mean LAI variations averaged over five key re-gions for the respective predominant LCC types.These include Texas (grassland),the Southwest (shrubland),the Midwest (dryland cropland and pasture),the South-east (evergreen needleleaf forest),and the Northeast (deciduous broadleaf forest)United States.Apparent discontinuities exist between the two datasets,where the MODIS values are systematically smaller,especially for the Midwest cropland.Analyses indicate that certain relationships exist,but vary greatly with regions.No physically sound and statistically robust adjustment can be made for consistency.A first-order correction is to obtain a same climatology (i.e.,identical monthly means averaged over all years)while retaining the interannual variability at each grid.Such corrected regional mean time series using the AVHRR or MODIS climatology are also shown in Figure 7.

The question is which climatology,AVHRR or MODIS,is more realistic.While over 1000published estimates during 1932–2000at nearly 400field sites over the globe have been compiled (Scurlock et al.2001)and validation is under way,no direct intercomparison between these field measurements and satellite products is currently available.Figure 8compares monthly mean LAI variations for the Mid-west cropland based on the AVHRR 8-km (January –December 1999)and 16-km (January 1999–May 2001)and MODIS 1-km (February 2000–May 2001)data with filed measurements at a central Illinois soybean/corn site (June –September

in

Figure 5.The geographic distributions of Apr and Jul mean LAI based on the original

data of (a),(b)the AVHRR (1981–99)and (c),(d)the MODIS (2000–03).

Figure6.The annual cycle of the LAI climatologies for the predominant LCC types over the five key regions outlined in Figure3,as derived from the original

AVHRR(1981–99;thin solid)and MODIS(2000–03;thick solid).

Figure7.Interannual variations of LAI averaged over the five key regions outlined in Figure3for the respective predominant LCC types7(Texas),8(South-

west),2(Midwest),14(Southeast),and11(Northeast)as derived from the

original AVHRR(1981–99;thin solid)and MODIS(2000–03;thick solid)and

their bias-corrected correspondences(thin,thick dashed).

1999–2000;by courtesy of Dr.Steven Hollinger of the Illinois State Water Sur-vey).The two AVHRR-based LAI estimates are in good agreement and well capture the peak values of the field observations.Clearly the MODIS-based esti-mates substantially underestimate the LAI of the growing season for the cropland.Since the MODIS measurement is continuing and providing finer resolution and quality-controlled data with improved atmospheric correction and cloud screening (Justice et al.1998)compared to the AVHRR (Goward et al.1991),its LAI product is preferred.Until a comprehensive evaluation or validation is completed with improved products made available,it is suggested that the AVHRR LAI is corrected to have the same monthly mean climatology as the MODIS except for the cropland-related LCC categories (2–6),where the opposite correction is ap-plied due to the obvious MODIS underestimation.The result is a long-term LAI dataset with continuation and consistency.This is particularly important when applied in an integrated model because,for example,the surface albedo param-eterization developed from the MODIS data depends on LAI (Liang et al.2004c).

3.3.Sea surface temperature

Given the lack of fine-resolution data,most mesoscale models have been using the weekly optimum interpolation SST (OISST)analysis at 1°spacing,a blend of multichannel AVHRR infrared measurements with in situ ship and buoy obser-vations (Reynolds et al.2002).Following Liang et al.(Liang et al.2004b),

daily

Figure 8.The comparison of LAI (Midwest dryland cropland and pasture)based on

AVHRR 8-km (dot)and 16-km (solid)data,and MODIS 1-km (dash)data with Illinois soybean/corn field measurements (spot)during Jan 1999–May 2001.

相关主题