• Betts, A. K., Chen F. , Mitchell K. E. , and Janjić Z. I. , 1997: Assessment of the land surface and boundary layer models in two operational versions of the NCEP Eta Model using FIFE data. Mon. Wea. Rev., 125 , 28962916.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Black, T. L., 1994: The new NMC mesoscale Eta model: Description and forecast experiments. Wea. Forecasting, 9 , 266278.

  • Black, T. L., ,and Coauthors. 1997: Changes to the Eta forecast systems. NWS Tech. Procedures Bull. 441, 31 pp. [Available from NCEP/EMC, 5200 Auth Rd., Camp Springs, MD 20746.].

    • Search Google Scholar
    • Export Citation
  • Champeaux, J-L., Arcos D. , Bazile E. , Giard D. , Goutorbe J-P. , Habets F. , Noilhan J. , and Roujean J-L. , 2000: AVHRR-derived vegetation mapping over western Europe for use in numerical weather forecast models. Int. J. Remote Sens., 21 , 11831199.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chang, J-T., and Wetzel P. J. , 1991: Effects of spatial variations of soil moisture and vegetation on the evolution of a prestorm environment: A numerical case study. Mon. Wea. Rev., 119 , 13681390.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chen, F., ,and Coauthors. 1996: Modeling of land-surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101 , 72517268.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Clark, C. A., and Arritt R. W. , 1995: Numerical simulations of the effect of soil moisture and vegetation cover on the development of deep convection. J. Appl. Meteor., 34 , 20292045.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Crawford, T. M., Stensrud D. J. , Mora F. , Merchant J. W. , and Wetzel P. J. , 2001: Value of incorporating satellite-derived land cover data in MM5/PLACE for simulating surface temperatures. J. Hydrometeor., 2 , 453468.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Crook, N. A., 1996: Sensitivity of moist convection forced by boundary layer processes to low-level thermodynamic fields. Mon. Wea. Rev., 124 , 17671785.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Douglas, M. W., Maddox R. A. , Howard K. , and Reyes S. , 1993: The Mexican monsoon. J. Climate, 6 , 16651677.

  • Eidenshink, J. E., 1992: The 1990 conterminous U.S. AVHRR data set. Photogramm. Eng. Remote Sens., 58 , 809813.

  • Gutman, G., and Ignatov A. , 1995: Global land monitoring from AVHRR: Potential and limitations. Int. J. Remote Sens., 16 , 23012309.

  • Gutman, G., and Ignatov A. , 1998: The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. Int. J. Remote Sens., 19 , 15331543.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Fritsch J. M. , 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.

    • Search Google Scholar
    • Export Citation
  • Lu, L., and Shuttleworth W. J. , 2002: Incorporating NDVI-derived LAI into the climate version of RAMS and its impact on regional climate. J. Hydrometeor., 3 , 347362.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mahrt, L., and Pan H-L. , 1984: A two-layer model of soil hydrology. Bound.-Layer Meteor., 29 , 120.

  • Markowski, P. M., and Stensrud D. J. , 1998: Mean monthly diurnal cycles observed with PRE-STORM surface data. J. Climate, 11 , 29953009.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mitchell, K., ,and Coauthors. 2000: Recent GCIP-sponsored advancements in coupled land-surface modeling and data assimilation in the NCEP Eta mesoscale model. Preprints, 15th Conf. on. Hydrology, Long Beach, CA, Amer. Meteor. Soc., 180–183.

    • Search Google Scholar
    • Export Citation
  • Mitchell, K., ,and Coauthors. 2002: Reducing near-surface cool/moist biases over snowpack and early spring wet soils in NCEP Eta Model forecasts via land surface model upgrades. Preprints, 16th Conf. on Hydrology, Orlando, FL, Amer. Meteor. Soc., J1–J6.

    • Search Google Scholar
    • Export Citation
  • NOAA, 2001: Climate Diagnostics Bulletin. 01/02–01/08, 595 pp. [Available from Climate Prediction Center, 5200 Auth Rd. Camp Springs, MD 20746; also available online at http://www.cpc.noaa.gov/products/analysis_monitoring/CDB_archive.html.].

    • Search Google Scholar
    • Export Citation
  • Ookuchi, Y. M., Segal M. , Kessler R. C. , and Pielke R. A. , 1984: Evaluation of soil moisture effects on the generation and modification of mesoscale circulations. Mon. Wea. Rev., 112 , 22812291.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pan, H-L., and Mahrt L. , 1987: Interaction between soil hydrology and boundary-layer development. Bound.-Layer Meteor., 38 , 185202.

  • Pielke R. A. Sr., , 2001: Influence of the spatial distribution of vegetation and soils on the prediction of cumulus convective rainfall. Rev. Geophys., 39 , 151177.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pielke R. A. Sr., , and Zeng X. , 1989: Influence on severe storm development of irrigated land. Natl. Wea. Dig., 14 , 1617.

  • Pielke R. A. Sr., , Dalu G. A. , Snook J. S. , Lee T. J. , and Kittel T. G. F. , 1991: Nonlinear influence of mesoscale land use on weather and climate. J. Climate, 4 , 10531069.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rabin, R. M., Stadler S. , Wetzel P. J. , Stensrud D. J. , and Gregory M. , 1990: Observed effects of landscape variability on convective clouds. Bull. Amer. Meteor. Soc., 71 , 272280.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, M. D., and Karl T. R. , 1990: Spring phenology: Nature's experiment to detect the effect of “green-up” on surface maximum temperatures. Mon. Wea. Rev., 118 , 883890.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Segal, M., Purdom J. F. W. , Song J. L. , Pielke R. A. , and Mahrer Y. , 1986: The predictability of sea breeze generated thunderstorms. Preprints, WMO Int. Workshop on Rain-Producing Systems in the Tropics and Extra-Tropics, San Jose, Costa Rica, WMO, 101–107.

    • Search Google Scholar
    • Export Citation
  • U.S. Department of Agriculture, 2001: Weekly Weather and Crop Report. Vol. 88, No. 15–32, 493 pp. [Available online at http://usda.mannlib.cornell.edu/reports/waobr/weather.].

    • Search Google Scholar
    • Export Citation
  • Wetzel, P. J., and Boone A. , 1995: A parameterization for land–atmosphere–cloud exchange (PLACE): Documentation and testing of a detailed process model of the partly cloudy boundary layer over heterogeneous land. J. Climate, 8 , 18101837.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp.

  • Yucel, I., Shuttleworth W. J. , Washburne J. , and Chen F. , 1998: Evaluating NCEP Eta Model–derived data against observations. Mon. Wea. Rev., 126 , 19771991.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    The inner box represents the area covered by the version of the Eta Model used in this study. The outer box corresponds to the area covered by the NCEP operational Eta Model

  • View in gallery

    Vegetation fractions from 18 Apr 2001 from (a) the Eta Model climatological database, and (b) the NDVI-derived 1-km dataset as used in the VegEta. (c) The difference (VegEta − Eta) in the vegetation fractions. (d)–(f) Similarly, the same fields from 16 May 2001. Color bar on the upper right represents the vegetation fractions in % at intervals of 10%. The bottom-right color bar is for the vegetation fraction difference fields

  • View in gallery

    As in Fig. 2 but for (a)–(c) 17 Jun 2001 and (d)–(f) 7 Jul 2001

  • View in gallery

    Differences in sensible heat flux at (a) 9 and (b) 33 h, and latent heat flux at (c) 9 and (d) 33 h from the 13 May 2001 model forecasts. Differences are defined as VegEta minus Eta values. Keys indicate the values of the differences

  • View in gallery

    Differences in 2-m temperature at (a) 9 and (b) 33 h, and 2-m dewpoint temperatures at (c) 9 and (d) 33 h from the 13 May 2001 model forecasts. Differences are defined as VegEta minus Eta values. Isolines are every 0.2°C. Keys indicate the values of the differences.

  • View in gallery

    Mean error, or bias (forecast − observation), vs observed 2-m temperature of all 12 cases for forecast times valid at (a) 1200, (b) 1500, (c) 1800, (d) 2100, (e) 0000, and (f) 0600 UTC. Solid lines are for the Eta forecasts and dashed for the VegEta forecasts. Keys specify other details. Data are averaged over 5°C bins based upon the observed temperature. Model forecasts started at 1200 UTC

  • View in gallery

    (Continued)

  • View in gallery

    Mean error, or bias (forecast − observation), vs observed 2-m temperature and dewpoint temperature of all 12 cases for (a) 9-h (2100 UTC) and (b) 33-h (2100 UTC) forecast times. Solid lines are for the Eta forecasts and dashed lines are for the VegEta forecasts. The T and Td indicate the 2-m temperature and dewpoint curves, respectively. Data are averaged over 5°C bins based upon the observed temperature

  • View in gallery

    Mean error, or bias (forecast − observation), vs VegEta vegetation fraction (as derived from the NDVI data) for all 12 cases for (a) 9-h (2100 UTC) and (b) 33-h (2100 UTC) forecast times. Solid lines are for the Eta forecasts and dashed lines for the VegEta forecasts. The T and Td indicate the 2-m temperature and dewpoint curves, respectively. Data are averaged over 10% vegetation fraction bins based upon the VegEta vegetation fraction values

  • View in gallery

    As in Fig. 8 but from the 9-h forecast time only from (a) 18 Apr 2001 and (b) 7 Aug 2001. Note how the added vegetation information in the VegEta has a greater impact in Apr than in Aug

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 138 20 1
PDF Downloads 36 20 2

Assessment of Implementing Satellite-Derived Land Cover Data in the Eta Model

View More View Less
  • 1 NOAA/National Severe Storms Laboratory, Norman, Oklahoma
  • | 2 Cooperative Institute for Mesoscale Meteorological Studies, Norman, Oklahoma
Full access

Abstract

One of the challenges in land surface modeling involves specifying accurately the initial state of the land surface. Most efforts have focused upon using a multiyear climatology to specify the fractional coverage of vegetation. For example, the National Centers for Environmental Prediction (NCEP) Eta Model uses a 5-yr satellite climatology of monthly normalized difference vegetation index (NDVI) values to define the fractional vegetation coverage, or greenness, at 1/8° (approximately 14 km) resolution. These data are valid on the 15th of every month and are interpolated temporally for daily runs. Yet vegetation characteristics change from year to year and are influenced by short-lived events such as fires, crop harvesting, droughts, floods, and hailstorms that are missed using a climatological database. To explore the importance of the initial state vegetation characteristics on operational numerical weather forecasts, the response of the Eta Model to initializing fractional vegetation coverage directly from the National Oceanic and Atmospheric Administration's Advanced Very High Resolution Radiometer (AVHRR) data is investigated. Numerical forecasts of the Eta Model, using both climatological and near-real-time values of fractional vegetation coverage, are compared with observations to examine the potential importance of variations in vegetation to forecasts of 2-m temperatures and dewpoint temperatures from 0 to 48 h for selected days during the 2001 growing season. Results show that use of the near-real-time vegetation fraction data improves the forecasts of both the 2-m temperature and dewpoint temperature for much of the growing season, highlighting the need for this type of information to be included in operational forecast models.

Additional affiliation: NOAA/National Severe Storms Laboratory and the NOAA/NWS/Storm Prediction Center, Norman, Oklahoma

Corresponding author address: Dr. David J. Stensrud, NOAA/NSSL, 1313 Halley Circle, Norman, OK 73069. Email: David.Stensrud@nssl.noaa.gov

Abstract

One of the challenges in land surface modeling involves specifying accurately the initial state of the land surface. Most efforts have focused upon using a multiyear climatology to specify the fractional coverage of vegetation. For example, the National Centers for Environmental Prediction (NCEP) Eta Model uses a 5-yr satellite climatology of monthly normalized difference vegetation index (NDVI) values to define the fractional vegetation coverage, or greenness, at 1/8° (approximately 14 km) resolution. These data are valid on the 15th of every month and are interpolated temporally for daily runs. Yet vegetation characteristics change from year to year and are influenced by short-lived events such as fires, crop harvesting, droughts, floods, and hailstorms that are missed using a climatological database. To explore the importance of the initial state vegetation characteristics on operational numerical weather forecasts, the response of the Eta Model to initializing fractional vegetation coverage directly from the National Oceanic and Atmospheric Administration's Advanced Very High Resolution Radiometer (AVHRR) data is investigated. Numerical forecasts of the Eta Model, using both climatological and near-real-time values of fractional vegetation coverage, are compared with observations to examine the potential importance of variations in vegetation to forecasts of 2-m temperatures and dewpoint temperatures from 0 to 48 h for selected days during the 2001 growing season. Results show that use of the near-real-time vegetation fraction data improves the forecasts of both the 2-m temperature and dewpoint temperature for much of the growing season, highlighting the need for this type of information to be included in operational forecast models.

Additional affiliation: NOAA/National Severe Storms Laboratory and the NOAA/NWS/Storm Prediction Center, Norman, Oklahoma

Corresponding author address: Dr. David J. Stensrud, NOAA/NSSL, 1313 Halley Circle, Norman, OK 73069. Email: David.Stensrud@nssl.noaa.gov

1. Introduction

Vegetation is one of several parameters that play an important role in land–atmosphere interactions by helping to determine the partitioning of the surface sensible and latent heat flux. Ookouchi et al (1984) show that solenoidal circulations are able to develop between patches of moist and dry soil. Similarly, areas of very dense vegetation next to bare soil surfaces under favorable environmental conditions promote sea-breeze-type circulations (Segal et al. 1986). Observations indicate that harvesting winter wheat over Oklahoma alters the surface sensible heat flux, such that afternoon cumulus clouds develop over harvested fields before they form over adjacent areas with an active green canopy (Rabin et al. 1990). Markowski and Stensrud (1998) further show that the harvesting of winter wheat over Oklahoma and Kansas plays a major role in the spatial distribution of monthly mean diurnal cycles of conserved variables in the surface layer.

Schwartz and Karl (1990) find there are statistically and practically significant relationships between the timing of the onset of vegetation and surface daily maximum temperature. Their study demonstrates at least a 3.5°C reduction in surface daily maximum temperature at an agricultural inland area over any 2-week period subsequent to first leaf compared to a 2-week period prior to first leaf. Stations generally near major bodies of water show a smaller (1.5°C) reduction. Vegetation also has a strong influence on seasonal climate, as shown in a modeling study by Lu and Shuttleworth (2002).

Inhomogeneities in land surface interactions create differences in the boundary layer and mesoscale flow (Pielke and Zeng 1989; Pielke et al. 1991). Chang and Wetzel (1991) conclude that spatial variations of vegetation and soil moisture affect the evolution of surface baroclinic structures through differential heating. In addition, they find that without the effect of vegetation tapping the root zone soil moisture, the cooling due to surface evaporation alone is too weak to correctly predict the absolute magnitude of the surface temperature or its horizontal variability. Furthermore, soil moisture and vegetation cover variability affect the development of deep convection over land (Clark and Arritt 1995; Pielke 2001).

As a result of the important role that vegetation plays in land surface and land–atmosphere interactions, it needs to be represented adequately in numerical weather prediction models. Thus, it is essential that the vegetation conditions be gathered in an accurate and timely manner. Vegetation data originally came from ground-based surveys (Gutman and Ignatov 1995). However, there are several problems with this type of method. Foremost, the information comes from different organizations, which makes it difficult to harmonize. Maps of the survey data cannot be updated frequently due to the large amount of time required to obtain the ground-based samples. Last, the resolution of conventional global maps tends to be too coarse for the more advanced atmospheric models. Therefore, remote sensing devices such as the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) are implemented to gather vegetation data (Champeaux et al. 2000).

There are several advantages to utilizing the 1-km AVHRR data over that of the 30-m land remote sensing satellite system (LANDSAT) or 10-m Systeme Probatoire d'Observation de la Terre (SPOT). First, the revisit periods of LANDSAT and SPOT are near 2 weeks. Since cloud-free images are not always possible when the satellite passes over a given location (Crawford et al. 2001), a month or longer may pass before the land surface can be observed clearly. Second, AVHRR data have nominal cost, whereas high-resolution data from LANDSAT and SPOT are costly and cover only limited regions of the globe episodically (Gutman and Ignatov 1995). Other notable advantages include the availability of spectral information for vegetation studies, global coverage, and the long-term continuous observational period.

Two important vegetation parameters that are used within numerical weather models are the green vegetation fraction and the leaf area index (LAI). The green vegetation fraction, also referred to as the greenness fraction or fractional vegetation cover, is the model grid-cell fraction where midday downward solar radiation is intercepted by a photosynthetically active green canopy (Chen et al. 1996) and it acts as the weighing factor between bare soil and canopy transpiration. This, in turn, affects surface temperature forecasts through the alteration of surface fluxes. The LAI, a measure of the vegetation biomass, is defined as the sum of the one-sided area of green leaves above a specified area of ground surface and it plays a major role in determining the amount of transpiration from the vegetation canopy. Holding all other parameters constant, a larger LAI value produces greater canopy transpiration than a lower LAI value. Together, the vegetation fraction and LAI describe the state of the vegetation covering the land surface.

It is possible to compute both the vegetation fraction and LAI from the AVHRR through calculations based upon the normalized difference vegetation index (NDVI) (see Crawford et al. 2001). However, according to Gutman and Ignatov (1998), it is less difficult to compute the horizontal density, or vegetation fraction, than the LAI from the NDVI. Thus, for example, the LAI is assigned a constant value of 1 in the Eta Model and only the vegetation fraction is allowed to vary in time and space. Other land surface models allow both parameters to vary.

Currently, the NCEP Eta Model uses a 0.144° (approximately 14 km) resolution monthly database for vegetation fraction, based upon a 5-yr climatology (Black et al. 1997) developed at the National Environmental Satellite, Data, and Information Service (NESDIS). The monthly vegetation fraction values apply to the 15th of every month and these data are interpolated for daily values. These estimations are based upon observations of NDVI, which are obtained from NOAA's polar-orbiting satellites that carry the AVHRR.

Vegetation characteristics change from year to year, however, and are influenced by circumstances such as fires, irrigation, deforestation, desertification, crop harvesting, drought-affected vegetation, hailstorms, and the early onset of spring vegetation (Crawford et al. 2001). Using near-real-time satellite-derived values of vegetation fraction and LAI improves the 2-m temperature simulations of the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) when using the Parameterization for Land–Atmosphere–Cloud Exchange (PLACE; Wetzel and Boone 1995) for the land surface (Crawford et al. 2001). Thus, it is hypothesized that an improved initialization of vegetation fraction in the Eta Model can lead to smaller surface temperature forecast errors, which likely result in part from the use of climatological land use information (Mitchell et al. 2000). Comparisons of numerical forecasts of the Eta Model are made, using both climatological and satellite-derived values of fractional vegetation coverage from the NOAA AVHRR data, in order to examine the potential importance of real-time vegetation fraction information to forecasts of 2-m temperatures and dewpoint temperatures from 0 to 48 h. The cases selected span the growing season of 2001.

The Eta Model is discussed briefly in section 2, along with a description of the NOAA AVHRR data and the calculation of vegetation fraction. Results of the Eta Model forecasts are presented in section 3, followed by a final discussion in section 4.

2. Methods and procedures

a. Eta Model description

The workstation version of the Eta Model (Black 1994) is used in this study. This model uses a domain that is approximately 20% of the operational Eta domain (Fig. 1), although it has the same horizontal grid spacing of 22 km and the same 50 vertical layers that were used operationally in the Eta Model during most of 2001. The initial and boundary conditions are obtained directly from a subset of the operational Eta Model fields. However, there are two major differences between the operational Eta Model and the version run for this study. This workstation version uses the Kain–Fritsch convective parameterization scheme (Kain and Fritsch 1993) and uses fourth-order horizontal diffusion instead of second-order horizontal diffusion. Neither of these changes is expected to influence significantly the results of the present study regarding the importance of vegetation fraction on the model forecasts.

The land surface model (LSM) used within the Eta Model is a multilayer soil–vegetation–snowpack model. It was originally developed at Oregon State University (Mahrt and Pan 1984; Pan and Mahrt 1987) and was then modified at NCEP for use in the Eta Model (Chen et al. 1996; Black et al. 1997). This LSM has one canopy layer and eight prognostic variables: soil moisture and temperature in the three soil layers, water stored on the canopy, and snow stored on the ground (Chen et al. 1996). Vegetation fraction influences the surface latent heat flux calculations directly by weighting the relative contributions of bare soil and canopy transpiration. Thus, a larger value of the vegetation fraction indicates that the vegetated surface contributes a greater portion of the total latent heat flux than a smaller value of the vegetation fraction. This also has a large influence on the model predictions of 2-m dewpoint temperature. In addition, since the vegetation fraction influences the latent heat flux, it also influences the sensible heat flux through the surface energy balance. Thus, vegetation fraction has a potentially large influence on both the 2-m temperature and dewpoint temperature and, thus, affects the development of the planetary boundary layer in the model forecasts.

b. Calculation of vegetation fraction

Since 1989, the U.S. Geological Survey's Earth Resources Observing Systems (EROS) Data Center (EDC) has used the 1-km AVHRR daily observations to produce biweekly maximum NDVI composites over the conterminous United States (Eidenshink 1992). Since clouds are associated with low values of NDVI, biweekly maximum NDVI composites tend to have minimal cloud contamination. Biweekly composites of NDVI from USGS/EROS for 2001 are available to customers approximately 1 day after the composite intervals ends. These data are downloaded and navigated using a geographic information system. The biweekly period that contains the forecast day of interest is selected for use in order to mimic the data that could be available if daily updates of this information were provided routinely.

The values for vegetation fraction (fveg) are derived from the near-real-time NDVI using the following equation (Chang and Wetzel 1991):
i1520-0434-18-3-404-e1
where the values of vegetation fraction are constrained to lie between 0 and 1. Since the Eta Model has a spatial resolution of 22 km, while the NDVI data resolution is 1 km, an objective analysis is necessary to determine a representative value for the vegetation fraction on the Eta grid. Data points for the 1-km vegetation fraction that lie within a 10-km radius of each Eta grid point are averaged and then applied to that Eta grid point. While more sophisticated analyses are possible, this approach is reasonable for an initial study into the importance of vegetation fraction in the Eta Model. Locations outside of the region of the biweekly composite data use the Eta Model climatology for the vegetation fraction, which can produce large gradients in vegetation fraction in southern Canada.

c. Running the Eta Model

Two different runs of the Eta Model are performed: one with the climatological value for vegetation fraction (Eta; Gutman and Ignatov 1998) and the other with the real-time NDVI-derived vegetation fraction (VegEta). The model grid, physical process schemes, and initial and boundary conditions are identical in these two runs except for the vegetation fraction defined in the surface parameter input file. The model provides output every 3 h, beginning at the 1200 UTC initialization time and ending at 48 h.

A total of 12 case studies are examined for April–August of 2001 (Table 1). Two forecasts are completed for each of these months, except for April, which has four forecasts. Each day is chosen because a majority of the United States for that day is free of deep convective clouds, thereby minimizing the influence of the convective parameterizations. This case selection maximizes the potential for the effects of vegetation to be seen and diagnosed in the model forecasts, as the influences of many of the other model physical process schemes are minimized.

Once the forecasts and postprocessing are complete, the model 2-m temperature and dewpoint temperature are bilinearly interpolated to National Weather Service (NWS) surface observing sites. Since the model domain covers the contiguous 48 states, there are over 1600 NWS surface stations that provide hourly reports and can be used to compare the performance of the two forecasts at each output time. The number of NWS stations actually used in the analyses is somewhat less, varying from approximately 1350 to 1600. While the hourly surface observations are examined for obvious problems, errors may exist. However, any errors in the observational data influence the results of both of the forecasts in the same manner and should not influence the subsequent comparisons. The bias, root-mean-square error (rmse), and Pearson (ordinary) correlation [see Wilks (1995) for documentation on all three of these measures] are computed both for the operational forecasts (Eta) and the real-time-NDVI-derived VegEta, and the results are compared. We now turn to the forecast comparisons.

3. Model results

a. Description of vegetation fraction differences

Comparisons between the vegetation fraction from the Eta and VegEta model initial conditions (Figs. 2 and 3) indicate that the VegEta vegetation fraction contains more horizontal structure than the values from the Eta (owing to both the selected radius of influence in the objective analysis and the high-resolution data used). Thus, the real-time satellite data are capturing the temporal and horizontal variations in vegetation growth that likely are masked by the use of a 5-yr monthly climatology. Early in the growing season the largest differences in vegetation fraction are over the southeast United States (Fig. 2c), with the values from the VegEta being larger than those from the Eta, but these differences decrease as the growing season progresses. By mid-June (Fig. 3c) the largest differences in vegetation fraction are over the plains states, stretching from Oklahoma to North Dakota, and in southeastern Canada. The region of largest differences shifts farther northward into the northern plains during July (Fig. 3f) and a more localized difference is also seen in western Mexico associated with vegetation green-up during the North American monsoon (Douglas et al. 1993). Average absolute differences in vegetation fraction calculated across the entire model grid from the Eta and VegEta are slightly more than 0.10 for the 12 days examined.

Since the method used by Gutman and Ignatov (1998) to calculate vegetation fraction for the Eta Model produces slightly larger values of vegetation fraction for a given NDVI value than the method used in this study [Eq. (1)], it is somewhat surprising that the VegEta often has larger vegetation fractions than the Eta. However, vegetation fractions can increase by 0.30 or more over a 1-month period, suggesting that one explanation for the larger vegetation fractions in the VegEta is the improved temporal and spatial resolution of the data. Crop moisture estimations during 2001 indicate that much of the eastern United States had moist to abnormally moist soil conditions (U.S. Department of Agriculture 2001), and surface temperatures during April and May over the eastern United States exceeded normal values by 1°–3°C (NOAA 2001), with the largest temperature anomalies in April centered over Illinois, suggesting that an early start to the growing season is likely and would explain many of these differences in vegetation fraction. This assessment reinforces the conclusions of Crawford et al. (2001) on the need for real-time vegetation information.

b. Comparisons between Eta and VegEta

Strong relationships exist between the spatial distribution of the vegetation fraction differences between Eta and VegEta and the corresponding differences in surface fluxes, 2-m temperature, and 2-m dewpoint temperature. For example, recall that for mid-May 2001 the VegEta contains broad regions with vegetation fractions larger than the Eta, with differences approaching 40% in the Southeast and stretching northward along the East Coast into Canada (see Fig. 2f). In addition, areas in the far northwest also contain larger values of vegetation fraction in the VegEta. Differences in the instantaneous flux values between the two models show large, coherent regions in which both the VegEta latent and sensible heat fluxes are different from those of the Eta at 9 and 33 h (Fig. 4), near the time of daytime maximum heating during forecast days 1 and 2. Many of these coherent areas of flux difference correspond well with areas of vegetation fraction difference between the two forecasts (cf. Figs. 2f and 4).

While the fluxes typically differ by slightly more than 25 W m–2, these differences lead to 2-m temperature differences exceeding 1°C and dewpoint temperature differences exceeding 2°C (Fig. 5). In the southeastern United States, the VegEta temperatures are warmer than those from the Eta with differences often exceeding 1°C (Figs. 5a,b). On the other hand, Eta forecast temperatures are larger than VegEta values in those regions where Eta contains larger vegetation fractions, such as portions of the northern Great Plains. A similar pattern is found in the dewpoint difference fields (Figs. 5c,d) except as expected the sign of the differences is reversed. These differences are due in large part to the accumulated effects of differences in the predicted surface fluxes over the length of the forecast period. Thus, even though the instantaneous flux differences may only approach 25 W m–2, the cumulative effect is discernable and important to various forecast scenarios. For example, Crook (1996) shows that boundary layer temperature differences of ±1°C and ±1 g kg–1 in mixing ratio (∼2°C difference in dewpoint temperature) make operationally significant differences in simulated thunderstorms using a cloud-scale model. The rather unexpected result, that the forecasts with the larger vegetation fraction produce warmer temperatures and lower dewpoint temperatures, is discussed in detail following the general verification of all the forecasts that follows.

To further explore the differences in the Eta and VegEta forecasts, the forecast temperature errors are binned with respect to the observed temperature and the mean errors (biases, defined here as forecast − observation) plotted. Results indicate that both model forecasts display a general cold bias in forecasting 2-m temperatures throughout the growing season (Fig. 6). At the model initial time, all 2-m temperatures have a cold bias (Fig. 6a) that generally increases with the observed temperature. The differences between the Eta and VegEta are most apparent during the daytime at the 6–12-h forecast times (1800–2400 UTC; Figs. 6c–e) and again at the 30–36-h forecast times (1800–2400 UTC day 2). The cold bias also is less pronounced at these times. For most observed temperatures, the VegEta forecasts have smaller biases than the Eta forecasts. In contrast, the forecast biases are very similar to each other and larger during the nighttime. Betts et al. (1997) indicate that the Eta Model nighttime temperature minima are often too low because of a model underestimate of downwelling radiation, which may explain the nighttime cold biases seen here in both the initial condition and the forecasts.

Since the differences between the two forecasts are most evident at 9 and 33 h, the dewpoint temperature biases also are examined at these two times. Results show that the dewpoint biases are large for observed dewpoints near 0°C, but decrease dramatically as the observed dewpoint temperature increases (Fig. 7). Also note that the VegEta forecasts consistently have smaller biases than the Eta forecasts for both times examined.

Upon examining the bias and rmse averaged for each individual case, it appears that both model forecasts become more accurate in predicting 2-m temperatures as the season changes from spring to summer (Table 2). Similar results are found for the 2-m dewpoint temperatures (now shown). Note that in April, the temperature bias for each case study is approximately −2°C for both the Eta and VegEta, with the VegEta biases being slightly smaller than the Eta values. However, by mid-May, the bias magnitude decreases to near −1°C. As the months progress into June and July, the bias magnitudes decrease even more. These results also reinforce the cold bias as mentioned previously. Similarly, the rmse tends to decrease slightly from April to July. The correlation between the observed and forecast temperatures peaks in May, although the values do not show a very large variation and the differences likely are not statistically significant.

The Wilcoxon signed-rank test, a nonparametric test for the significance of the difference between the distributions of two nonindependent samples involving matched pairs (Wilks 1995), is performed on the 9- and 33-h forecast 2-m temperatures and dewpoint temperatures for each case day. Results show that the differences in the temperatures between the VegEta and Eta forecasts are significant at the 99% level for the forecasts during April and May. However, the significance level decreases below 99% for the remaining half of the forecast days, with the exception of the 9-h forecasts for 7 July and 7 August and the 33-h forecasts for 16 June and 5 July. Differences in the dewpoint temperatures between the VegEta and Eta forecasts are significant at the 99% level for all the cases from April to July, but are not significant for the August days. This analysis supports the conclusion that the real-time vegetation information is most important early in the growing season and has decreased importance later during the summer when the vegetation growth largely is complete.

To explore how the vegetation fraction information influences the forecasts, both the 2-m temperature and dewpoint temperature errors at 9 and 33 h are binned with respect to the value of the VegEta vegetation fraction. Results indicate that both the Eta and VegEta provide the worst 2-m temperature forecasts for bare soil conditions (Fig. 8). Both runs, however, have better 2-m temperature forecasts as the vegetation fraction increases, while both forecasts also have worsening dewpoint temperature forecasts as the vegetation fraction increases from 0% to 60%.

The Eta and VegEta mean bias curves are nearly identical for vegetation fractions below 60% and separate for vegetation fractions above this value, with the VegEta biases smaller than those from the Eta. For vegetation fractions between 90% and 100%, the VegEta temperature forecast bias is smaller by 0.5°C when compared with the bias in the Eta forecasts and, similarly, the VegEta dewpoint temperature bias is reduced by 1.0°C compared to bias in the Eta forecasts. As the growing season progresses, the influence of the real-time vegetation information decreases as illustrated by analyses from 18 April and 7 August (Fig. 9). In April, the VegEta bias in 2-m temperature is 1°C smaller than those from the Eta, and for 2-m dewpoint temperature is 3°C smaller than those from the Eta, for vegetation fractions above 90%. However, by August these differences are only a fraction of a degree. This decrease in the forecast difference is expected; the average absolute differences in vegetation fraction between the VegEta and Eta also decrease during the growing season, and large regions showing only small differences in the vegetation fraction between climatology and the real-time data become apparent by July (Fig. 3f). However, for most vegetation fractions the VegEta biases continue to be less than those from the Eta.

The reduction in the 2-m temperature cold bias when using the real-time vegetation information is somewhat counterintuitive. The vegetation fraction typically appears to be larger from the real-time data than from climatology (Figs. 2 and 3) leading one to expect that canopy transpiration would increase and that the surface sensible heat flux would decrease. Yet it appears that the sensible heat flux increases as vegetation fraction increases. While it is difficult to determine cause and effect within complex modeling systems, one possible explanation is an overestimate of evaporation from the bare soil physics formulation. Thus, as vegetation fraction is increased, more of the surface fluxes are determined by the canopy physics and less are determined by the bare soil physics. Another possible explanation is that the net radiation at the surface increases as vegetation increases, owing to a decrease in albedo over vegetation as compared with some light-colored bare soils, leading to larger sensible heat flux over more vegetated surfaces.

Recent results by Mitchell et al. (2002) support the hypothesis that the bare soil parameterization may be the cause of the model cold bias. They indicate that for moist soil conditions, as observed during 2001, the bare soil scheme produces too much evaporation from the ground, thereby reducing the model near-surface temperatures. Yucel et al. (1998) also report a cold bias in Eta Model forecasts over Arizona, a region in which the bare soil scheme plays a dominant role. This would exacerbate the problem in the bare soil parameterization.

Curiously, the mean forecast temperature biases are cold throughout the growing season (Table 2). This contrasts with the monthly average results from Yucel et al. (1998) who indicate an Eta Model warm bias over Oklahoma and from Mitchell et al. (2000, 2002) in which the Eta Model has a cold bias in March and a warm bias in July over the north-central United States. A regional verification of the 12 cases indicates that a warm bias exists for several of the regional areas from June through August, generally in the NE, SE, and SW1 during daytime hours (not shown). No warm bias is found for the NW quadrant of the contiguous United States.

4. Discussion

The main goal of this study is to investigate the impact of using real-time vegetation fraction information in the Eta Model. Since methods could be developed to provide these data to operational centers in near–real time, it is important to evaluate what would happen if we could detect drought-affected vegetation, the early onset of spring green-up, regions of burned vegetation from forest fires, and crop harvesting, which otherwise are missed by the climatological vegetation data used in the operational Eta. Thus, values of the biweekly NDVI-derived vegetation fraction as determined using satellite data are inserted into the Eta Model for 12 cases spanning most of the growing season during 2001 and forecasts using these near-real-time data as compared with those from both the operational version of the model and observations.

Calculations of bias for the 2-m temperature and dewpoint temperature indicate that the Eta Model during 2001 has a cold temperature bias and a wet (warm) dewpoint temperature bias throughout the growing season. The use of the real-time vegetation fraction reduces both the cold temperature and wet dewpoint temperature biases, particularly for the regions with the largest vegetation fractions. This counterintuitive result is attributed to the Eta Model LSM having a significant cold bias in the bare soil physical parameterization scheme. Thus, when more of the surface energy budget is controlled by the canopy transpiration calculations, the model biases are reduced. This agrees with the analysis of Mitchell et al. (2002), who also indicate that this bare soil physics problem was improved upon in late 2001 with changes to the operational Eta Model LSM.

Results also show that the use of the real-time satellite-derived vegetation fraction improves the forecasts of both the 2-m temperature and dewpoint temperature at the 99% significance level for much of the growing season. This result clearly illustrates the value of real-time satellite-derived vegetation information and the need for a system to get this type of information into the operational Eta Model. However, until this happens the Eta Model will continue to use the monthly climatology to specify vegetation information. This leaves the forecaster in the uncomfortable position of knowing that possible biases exist, but with no clear way to evaluate their operational significance from day to day.

Thankfully, there are sources of real-time NDVI data that can provide temporary assistance until real-time vegetation data becomes standard in the Eta Model initialization process. One could envision the following procedure occurring every few weeks, especially during the first half of the growing season.

  1. Print out the monthly vegetation fractions used by the Eta Model valid on the 15th of each month and interpolated for daily values. These are available online (http://www.emc.ncep.noaa.gov/mmb/gcp/sfcimg/gfrac/index.html) for each month of the year.

  2. Find the most recent NDVI or vegetation fraction information on the Internet and see where the Eta Model vegetation fraction is significantly different from that suggested by real-time satellite observations. One source of real-time information is provided by the U.S. Department of Agriculture and can be accessed online (http://www.nass.usda.gov/research/avhrr/avhrrmnu.htm). Another source is the Kansas Applied Remote Sensing Program through their Green Report (online at http://www.kars.ukans.edu/). Other good Internet sites can be found through a Web search.

  3. By watching how the Eta Model forecasts compare with observations with respect to these vegetation fraction differences, it often is possible to detect and then to correct for the model biases in subsequent forecasts. Note that comparing maps with different color schemes sometimes can be challenging, suggesting that real-time difference fields between the Eta Model climatological values and present observations of vegetation fraction would be very helpful.

This study focuses solely on the influence of the implementation of near-real-time NDVI-derived vegetation fraction in numerical models. However, other land surface parameters that influence the forecasts are not altered within the model (see Crawford et al. 2001). Soil moisture specification, LAI, surface roughness, and surface albedo also are important in partitioning the surface energy budget and methods for improving their initialization, such as the Land Data Assimilation System (Mitchell et al. 2000), should be vigorously pursued. However, until these data are incorporated more completely into the operational models, forecasters may benefit from comparing the model land surface parameters, such as vegetation fraction, with those derived from other sources in real time to discern where the model flux predictions may be in error.

Acknowledgments

Thanks are due to Drs. Kelvin Droegemeier and Alan Shapiro of the University of Oklahoma for being members of the M.S. committee of the lead author (NPK), whose research is summarized in this manuscript. The NDVI composites were made available by the EROS Data Center, and we appreciate the assistance provided by Dr. Thomas Loveland. Constructive and helpful reviews provided by Kevin Gallo and two anonymous reviewers led to an improved presentation and are greatly appreciated. We express our thanks to the ITS group at NSSL who provided needed computer assistance. The lead author (NPK) was funded by the NOAA Graduate Scientist Program, and this support is gratefully acknowledged.

REFERENCES

  • Betts, A. K., Chen F. , Mitchell K. E. , and Janjić Z. I. , 1997: Assessment of the land surface and boundary layer models in two operational versions of the NCEP Eta Model using FIFE data. Mon. Wea. Rev., 125 , 28962916.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Black, T. L., 1994: The new NMC mesoscale Eta model: Description and forecast experiments. Wea. Forecasting, 9 , 266278.

  • Black, T. L., ,and Coauthors. 1997: Changes to the Eta forecast systems. NWS Tech. Procedures Bull. 441, 31 pp. [Available from NCEP/EMC, 5200 Auth Rd., Camp Springs, MD 20746.].

    • Search Google Scholar
    • Export Citation
  • Champeaux, J-L., Arcos D. , Bazile E. , Giard D. , Goutorbe J-P. , Habets F. , Noilhan J. , and Roujean J-L. , 2000: AVHRR-derived vegetation mapping over western Europe for use in numerical weather forecast models. Int. J. Remote Sens., 21 , 11831199.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chang, J-T., and Wetzel P. J. , 1991: Effects of spatial variations of soil moisture and vegetation on the evolution of a prestorm environment: A numerical case study. Mon. Wea. Rev., 119 , 13681390.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chen, F., ,and Coauthors. 1996: Modeling of land-surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101 , 72517268.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Clark, C. A., and Arritt R. W. , 1995: Numerical simulations of the effect of soil moisture and vegetation cover on the development of deep convection. J. Appl. Meteor., 34 , 20292045.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Crawford, T. M., Stensrud D. J. , Mora F. , Merchant J. W. , and Wetzel P. J. , 2001: Value of incorporating satellite-derived land cover data in MM5/PLACE for simulating surface temperatures. J. Hydrometeor., 2 , 453468.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Crook, N. A., 1996: Sensitivity of moist convection forced by boundary layer processes to low-level thermodynamic fields. Mon. Wea. Rev., 124 , 17671785.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Douglas, M. W., Maddox R. A. , Howard K. , and Reyes S. , 1993: The Mexican monsoon. J. Climate, 6 , 16651677.

  • Eidenshink, J. E., 1992: The 1990 conterminous U.S. AVHRR data set. Photogramm. Eng. Remote Sens., 58 , 809813.

  • Gutman, G., and Ignatov A. , 1995: Global land monitoring from AVHRR: Potential and limitations. Int. J. Remote Sens., 16 , 23012309.

  • Gutman, G., and Ignatov A. , 1998: The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. Int. J. Remote Sens., 19 , 15331543.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Fritsch J. M. , 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.

    • Search Google Scholar
    • Export Citation
  • Lu, L., and Shuttleworth W. J. , 2002: Incorporating NDVI-derived LAI into the climate version of RAMS and its impact on regional climate. J. Hydrometeor., 3 , 347362.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mahrt, L., and Pan H-L. , 1984: A two-layer model of soil hydrology. Bound.-Layer Meteor., 29 , 120.

  • Markowski, P. M., and Stensrud D. J. , 1998: Mean monthly diurnal cycles observed with PRE-STORM surface data. J. Climate, 11 , 29953009.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mitchell, K., ,and Coauthors. 2000: Recent GCIP-sponsored advancements in coupled land-surface modeling and data assimilation in the NCEP Eta mesoscale model. Preprints, 15th Conf. on. Hydrology, Long Beach, CA, Amer. Meteor. Soc., 180–183.

    • Search Google Scholar
    • Export Citation
  • Mitchell, K., ,and Coauthors. 2002: Reducing near-surface cool/moist biases over snowpack and early spring wet soils in NCEP Eta Model forecasts via land surface model upgrades. Preprints, 16th Conf. on Hydrology, Orlando, FL, Amer. Meteor. Soc., J1–J6.

    • Search Google Scholar
    • Export Citation
  • NOAA, 2001: Climate Diagnostics Bulletin. 01/02–01/08, 595 pp. [Available from Climate Prediction Center, 5200 Auth Rd. Camp Springs, MD 20746; also available online at http://www.cpc.noaa.gov/products/analysis_monitoring/CDB_archive.html.].

    • Search Google Scholar
    • Export Citation
  • Ookuchi, Y. M., Segal M. , Kessler R. C. , and Pielke R. A. , 1984: Evaluation of soil moisture effects on the generation and modification of mesoscale circulations. Mon. Wea. Rev., 112 , 22812291.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pan, H-L., and Mahrt L. , 1987: Interaction between soil hydrology and boundary-layer development. Bound.-Layer Meteor., 38 , 185202.

  • Pielke R. A. Sr., , 2001: Influence of the spatial distribution of vegetation and soils on the prediction of cumulus convective rainfall. Rev. Geophys., 39 , 151177.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Pielke R. A. Sr., , and Zeng X. , 1989: Influence on severe storm development of irrigated land. Natl. Wea. Dig., 14 , 1617.

  • Pielke R. A. Sr., , Dalu G. A. , Snook J. S. , Lee T. J. , and Kittel T. G. F. , 1991: Nonlinear influence of mesoscale land use on weather and climate. J. Climate, 4 , 10531069.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rabin, R. M., Stadler S. , Wetzel P. J. , Stensrud D. J. , and Gregory M. , 1990: Observed effects of landscape variability on convective clouds. Bull. Amer. Meteor. Soc., 71 , 272280.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schwartz, M. D., and Karl T. R. , 1990: Spring phenology: Nature's experiment to detect the effect of “green-up” on surface maximum temperatures. Mon. Wea. Rev., 118 , 883890.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Segal, M., Purdom J. F. W. , Song J. L. , Pielke R. A. , and Mahrer Y. , 1986: The predictability of sea breeze generated thunderstorms. Preprints, WMO Int. Workshop on Rain-Producing Systems in the Tropics and Extra-Tropics, San Jose, Costa Rica, WMO, 101–107.

    • Search Google Scholar
    • Export Citation
  • U.S. Department of Agriculture, 2001: Weekly Weather and Crop Report. Vol. 88, No. 15–32, 493 pp. [Available online at http://usda.mannlib.cornell.edu/reports/waobr/weather.].

    • Search Google Scholar
    • Export Citation
  • Wetzel, P. J., and Boone A. , 1995: A parameterization for land–atmosphere–cloud exchange (PLACE): Documentation and testing of a detailed process model of the partly cloudy boundary layer over heterogeneous land. J. Climate, 8 , 18101837.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp.

  • Yucel, I., Shuttleworth W. J. , Washburne J. , and Chen F. , 1998: Evaluating NCEP Eta Model–derived data against observations. Mon. Wea. Rev., 126 , 19771991.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The inner box represents the area covered by the version of the Eta Model used in this study. The outer box corresponds to the area covered by the NCEP operational Eta Model

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 2.
Fig. 2.

Vegetation fractions from 18 Apr 2001 from (a) the Eta Model climatological database, and (b) the NDVI-derived 1-km dataset as used in the VegEta. (c) The difference (VegEta − Eta) in the vegetation fractions. (d)–(f) Similarly, the same fields from 16 May 2001. Color bar on the upper right represents the vegetation fractions in % at intervals of 10%. The bottom-right color bar is for the vegetation fraction difference fields

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 3.
Fig. 3.

As in Fig. 2 but for (a)–(c) 17 Jun 2001 and (d)–(f) 7 Jul 2001

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 4.
Fig. 4.

Differences in sensible heat flux at (a) 9 and (b) 33 h, and latent heat flux at (c) 9 and (d) 33 h from the 13 May 2001 model forecasts. Differences are defined as VegEta minus Eta values. Keys indicate the values of the differences

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 5.
Fig. 5.

Differences in 2-m temperature at (a) 9 and (b) 33 h, and 2-m dewpoint temperatures at (c) 9 and (d) 33 h from the 13 May 2001 model forecasts. Differences are defined as VegEta minus Eta values. Isolines are every 0.2°C. Keys indicate the values of the differences.

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 6.
Fig. 6.

Mean error, or bias (forecast − observation), vs observed 2-m temperature of all 12 cases for forecast times valid at (a) 1200, (b) 1500, (c) 1800, (d) 2100, (e) 0000, and (f) 0600 UTC. Solid lines are for the Eta forecasts and dashed for the VegEta forecasts. Keys specify other details. Data are averaged over 5°C bins based upon the observed temperature. Model forecasts started at 1200 UTC

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 6.
Fig. 7.
Fig. 7.

Mean error, or bias (forecast − observation), vs observed 2-m temperature and dewpoint temperature of all 12 cases for (a) 9-h (2100 UTC) and (b) 33-h (2100 UTC) forecast times. Solid lines are for the Eta forecasts and dashed lines are for the VegEta forecasts. The T and Td indicate the 2-m temperature and dewpoint curves, respectively. Data are averaged over 5°C bins based upon the observed temperature

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 8.
Fig. 8.

Mean error, or bias (forecast − observation), vs VegEta vegetation fraction (as derived from the NDVI data) for all 12 cases for (a) 9-h (2100 UTC) and (b) 33-h (2100 UTC) forecast times. Solid lines are for the Eta forecasts and dashed lines for the VegEta forecasts. The T and Td indicate the 2-m temperature and dewpoint curves, respectively. Data are averaged over 10% vegetation fraction bins based upon the VegEta vegetation fraction values

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Fig. 9.
Fig. 9.

As in Fig. 8 but from the 9-h forecast time only from (a) 18 Apr 2001 and (b) 7 Aug 2001. Note how the added vegetation information in the VegEta has a greater impact in Apr than in Aug

Citation: Weather and Forecasting 18, 3; 10.1175/1520-0434(2003)18<404:AOISDL>2.0.CO;2

Table 1.

List of all days used in this study

Table 1.
Table 2.

The mean bias (°C), correlation, and rmse (°C), calculated from the forecast and observed 2-m temperatures, averaged over all forecast times (0–48 h) for each case day. Each output time has over 1300 NWS surface station observations available for comparison with the model forecasts

Table 2.

1

Here, NE is defined as the area north of 40°N, south of 52°N, and east of 100°W; SE is north of 25°N, south of 40°N, and east of 100°W; SW is north of 25°N, south of 40°N, and west of 100°W; and NW is north of 40°N, south of 52°N, and west of 100°W.

Save