Skip to main content
NCBI home page
EPA Author Manuscripts logoLink to EPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Aug 8.
Published in final edited form as: J Geophys Res Atmos. 2020 Aug 8;125(15):10.1029/2019JD032093. doi: 10.1029/2019JD032093

Impacts of tiled land cover characterization in the Model for Prediction Across Scales-Atmosphere (MPAS-A)

Patrick C Campbell 1,2,3, Jesse O Bash 4, Jerold A Herwehe 4, Robert C Gilliam 4, Dan Li 5
PMCID: PMC7788010  NIHMSID: NIHMS1629094  PMID: 33425636

Abstract

Parameterization of subgrid-scale variability of land cover characterization (LCC) is an active area of research, and can improve model performance compared to the dominant (i.e., most abundant tile) approach. The “Noah” land surface model implementation in the global Model for Predictions Across Scales-Atmosphere (MPAS-A), however, only uses the dominant LCC approach that leads to oversimplification in regions of highly heterogeneous LCC (e.g., urban/suburban settings). Thus, in this work we implement a subgrid tiled approach as an option in MPAS-A, version 6.0, and assess the impacts of tiled LCC on meteorological predictions for two gradually refining meshes (92–25 and 46–12 km) focused on the conterminous U.S for January and July 2016. Compared to the dominant approach, results show that using the tiled LCC leads to pronounced global changes in 2-m temperature (July global average change ~ −0.4 K), 2-m moisture, and 10-m wind speed for the 92–25 km mesh. The tiled LCC reduces mean biases in 2-m temperature (July U.S. average bias reduction ~ factor of 4) and specific humidity in the central and western U.S. for the 92–25 km mesh, improves the agreement of vertical profiles (e.g., temperature, humidity, and wind speed) with observed radiosondes; however, there is increased bias and error for incoming solar radiation at the surface. The inclusion of subgrid LCC has implications for reducing systematic temperature biases found in numerical weather prediction models, particularly those that employ a dominant LCC approach.

1. Introduction

The land cover characterization (LCC), i.e., the physical characteristics of Earth’s land surface (vegetated, wetlands, water, ice, or urban/impervious), is inherently heterogeneous, and is rapidly changing due to recent and projected future fluctuations in the LCC for both developed and developing countries. Changes in LCC due to human activities (e.g., deforestation, industrialization, agriculture, urban sprawl) produce physical changes in land surface albedo, latent (LH) and sensible heat (SH) fluxes, and atmospheric aerosol and greenhouse gas concentrations. Consequently, there is high confidence that changes in LCC and the feedbacks to surface properties, fluxes, and atmospheric composition accounts for as much as half of the human-caused global radiative forcing from 1850 to the present day (Hibbard et al., 2017).

Numerical weather prediction (NWPs) and climate models are used to predict the near- and long-term weather and climate changes, respectively, and are tightly bound to the land surface model (LSM) component that represents the lower physical boundary. The LSM component also controls the representation of LCC variability, is the memory of climatic changes, and apportions the physical responses in surface LH and SH fluxes. NWP models, however, consistently simulate warmer surface temperatures compared to ground observations, where most of the systematic 2-meter temperature biases appear by day-5 predictions, and the largest warm bias is found in the Central U.S. (Ma et al., 2014). A joint model-observation intercomparison project, the Clouds Above the United States and Errors at the Surface (CAUSES), evaluated the role of clouds, radiation, and precipitation processes in contribution to the surface temperature biases in the Central U.S (Morcrette et al., 2018). One of the important findings from the CAUSES project was that the large warm bias in NWP models are attributed to the simulation of deep convective clouds and the evaporative fraction (EF = LH/[LH+SH]) at the surface (Steiner et al., 2018; and references contained within). As the EF is strongly tied to the LCC, we extend this connection to assert that there also exists a relationship between the accuracy of the NWP models’ representation of the LCC and subgrid scale variability in its coupled LSM, and the predictive accuracy of EF and associated feedbacks with surface temperature and moisture, clouds, and precipitation.

The need for subgrid-scale LCC variability in LSMs has been known for some time, and has been an active area of research over the past three decades (Giorgi and Avissar, 1997). Avissar and Pielke (1989) first proposed a subgrid LCC parameterization that used a number of patches (or tiles), i.e., the “tiled” approach, and showed that it resulted in strong contrasts in total surface energy fluxes. In the tiled approach, the corresponding surface fluxes, energy, and water balances in the LSM are calculated for each explicit LCC with unique vegetation attributes in the model grid cell, and then are spatially averaged to produce the surface fluxes for each cell. Other subgrid LCC approaches are also found in multi-mode, offline/coupled land-ocean surface platforms, such as the “SURFEX” model (Masson et al., 2013). Other subgrid LCC methods include the “composite” approach, which is similar to the dominant approach, but the surface properties are either linearly or non-linearly aggregated based on the properties of all the tiles within the grid cell (Koster and Suarez, 1992; Verseghy et al., 1993). The statistical-dynamical approach assumes that the land surface parameters that are critical for calculating surface fluxes follow certain probability density functions (PDFs) (Avissar, 1991; Entekhabi and Eagleson, 1989; Famiglietti and Wood, 1991). The multivariate mosaic subgrid approach (i.e., “k-means clustering”) method is used to take an arbitrary number of input descriptors and objectively determine areas of similarity within a grid cell. This is in contrast to a “univariate approach” that only uses one spatially varying parameter to aggregate a catchment into a relatively few classes (Newman et al., 2014). The k-means clustering method may in fact be well suited to represent subgrid spatial complexity in LSM applications on the global to regional scales. Other global- to regional-scale LSMs have incorporated subgrid LCC, such as the Community Land Model (CLM) that has a nested subgrid hierarchy in which grid cells are composed of multiple land units (vegetated, lake, urban, glacier, and crop), snow/soil columns, and plant functional types (PFTs). A true tiled scheme called “newsnow” is also an option in the European High Resolution Limited Area Model (HIRLAM), and it includes seven individual subgrid tiles that are treated with unique values of vegetation, roughness length, and albedo (Samuelsson et al., 2006; Gollvik and Samuelsson, 2011).

On the contrary, the application of a “dominant” approach to LCC in LSMs, where each grid cell is assumed to be entirely composed of the most abundant LCC type, is a known oversimplification of the real-world LCC variability (realistically on the order of meters), even at relatively “high-resolution” mesoscale simulations (< 10 km) (Ament and Simmer, 2006). This becomes a greater issue for global-scale NWP or climate models, which are typically run at coarser horizontal resolutions (e.g., ~ 100 km or greater). Of course, in practice there must exist a balance between representing the myriad of processes that relatively coarse models cannot resolve, especially global climate models, and the available computational capacity and resources for the respective application of the model.

The unified National Center for Atmospheric Research (NCAR), Oregon State University, the U.S. Air Force, and National Centers for Environmental Prediction’s (NCEP’s) Office of Hydrology (“Noah”) LSM (Chen & Dudhia, 2001; Chen et al., 1996, 1997, 2007; Ek et al., 2003; Li et al., 2013; Mitchell et al., 2004; Niu et al., 2011; Pan & Mahrt, 1987; Yin et al.,2015) has been widely developed, applied, and evaluated in its parent atmospheric grid model, the Weather Research and Forecasting (WRF) model (Powers et al., 2017; Skamarock & Klemp, 2008). The Li et al. (2013) method of explicit tiling (referred to as the “mosaic” method) in WRF/Noah is intriguing as it maintains tile-specific surface energy flux calculations that are then weighted averages for the entire grid cell used in conservation checks. Furthermore, Li et al. demonstrated that the tiled LCC method demonstrates stark differences, better model performance, and less sensitivity to spatial grid resolution for surface energy fluxes, land surface temperature, near-surface states, boundary layer growth, as well as rainfall distribution compared against the dominant approach in Noah. However, the applications of tiled LCC in global LSMs are limited to simulated energy and carbon balances at select boreal, temperate and tropical locations across the world (Li and Arora, 2012), and do not truly investigate the global, coupled atmospheric feedbacks as a result of the tiling (Melton and Arora, 2014). Furthermore, studies that do investigate such atmospheric feedbacks to subgrid LCC are specific to regional-scale applications (Li and Arora, 2012; Li et al., 2013; Li et al., 2017; Mallard et al., 2018). We note that the above referenced studies show that certain surface parameters and energy fluxes may be sensitive to using a tiled LCC compared to a dominant or composite approach. Thus, there is a need to implement and carefully test the impacts of tiled LCC on the global scale to assess the impacts of more realistic LCC on surface energy fluxes and the feedbacks to the cloud and radiative model predictions. The effects of tiled LCC have implications for both the scientific and operational weather forecasting communities, especially in areas of highly contrasting LCC types (Manrique-Suñén et al., 2013).

The atmospheric component of the Model for Predictions Across Scales-Atmosphere (MPAS-A) uses an unstructured centroidal Voronoi, nominally hexagonal mesh (grid, or tessellation) and C-grid staggering of the state variables as the basis for the horizontal discretization in the dynamical solver (Skamarock et al., 2012 and references contained within). The MPAS-A is ideal for this work as it is a parent, global atmospheric model to the Noah LSM, and the unstructured variable resolution meshes can be generated having smoothly-varying mesh transitions. The Noah implementation in MPAS-A (hereafter referred to as MPAS/Noah), however, only uses the dominant LCC approach. This results in an oversimplification in regions of highly heterogeneous LCC (e.g., urban/suburban settings), which is also impacted by the gradually refining meshes in MPAS for global to mesoscale applications. Thus, in this work we implement the tiled LCC approach as an option in MPAS/Noah, version 6.0, and assess the global-to-mesoscale impacts of tiled LCC in MPAS/Noah on meteorological predictions for two gradually refining meshes (92–25 and 46–12 km) focused on the conterminous U.S for January and July 2016. The year 2016 was chosen as relatively fine scale initial conditions are available for that year (see Section 2.2), and the January/July months represent climatological cool/warm seasons for both the Northern and Southern Hemispheres.

2. Methods

2.1. A tiled approach to LCC in MPAS-A

In this work we implement a tiled LCC to MPAS/Noah, which is analogous to the tiled (i.e., “mosaic”) approach found in the WRF model described by Li et al. (2013) (hereafter referred to as the “L13-tiled”, or simply the “tiled” approach). Generally applying the L13 approach here, a certain user-defined number (N) of tiles, each representing a land cover category, is considered within a mesh cell. The atmospheric properties and soil properties are assumed to be homogenous over the mesh cell when surface fluxes and surface state variables are calculated for each tile, and all prognostic variables are maintained for each tile, some of which are aggregated to yield the mesh cell average variables (Li et al., 2013). In the L13-tiled approach the mesh cell average variables are weighted by the normalized area fraction accounting for the areas of each tile, where the tile with the largest normalized area fraction has a rank of 1. The smaller normalized area fractions for each land cover category are subsequently given lower rankings, and the total N tiles are assumed constant for all mesh cells. This is in contrast to the dominant LCC approach, which only considers the most dominant tile (i.e., tile rank = 1), and does not consider fractional impacts of subgrid, tiled heterogeneity in LCC (Figure 1). The reader is referred to Li et al. (2013) and the references contained within for further details regarding the L13-tiled approach.

Figure 1.

Figure 1.

Open in a new tab

Illustration of a hypothetical “real-world” land cover in a hexagonal mesh cell in MPAS-A, and the corresponding dominant vs. L13-tiled approach to LCC used in the MPAS/Noah LSM.

2.2. Model configuration and simulation design

Here we apply two MPAS-A version 6.0 global meshes that seamlessly refine from a relatively coarse to fine, 92–25 and 46–12 km, horizontal grid spacing over the conterminous U.S. (CONUS). The global domain and subset of the CONUS are shown for the 92–25 km mesh in Figure 2, which also include the corresponding average vegetation fractions for January and July 2016.

Figure 2.

Figure 2.

Open in a new tab

Global 92–25 km and seamlessly refined 25 km mesh over CONUS. The average vegetation fraction at 92–25 km is also shown for January and July 2016.

Here we employ the default MPAS-A v6.0 default physics suite (based on the Advanced Research WRF model), except for the new implementation of the L13-tiled approach as an option to dominant LCC (default) in MPAS/Noah (Table 1). The physics options used here are a very common configuration in WRF, and thus are well documented on the WRF User’s Page and references contained within (http://www2.mmm.ucar.edu/wrf/users/).

Table 1.

MPAS-A v6.0 Model Configuration

Model Mesh/Process Configuration
Mesh Resolutions with Seamless Refinements 92–25 km and 46–12 km
Time Steps 100 s (92–25 km); 40 s (46–12 km)
Land Surface Model Dominant and L13-tiled Noah (N=15 tiles per cell)
LCC Data Combined 40-Category NLCD (conterminous U.S.) and IGBP-MODIS (Global)
Surface Layer Monin-Obukhov (MO)
Planetary Boundary Layer Yonsei University (YSU)
Grid Microphysics/Subgrid Convection WRF Single Moment 6-class (WSM6)/Kain-Fritsch (KF)
Radiation Rapid Radiative Transfer Model for GCMs (RRTMG)
Open in a new tab

The meteorological initial conditions are based on NCEP operational Global Forecast System analysis and forecast grids on a 0.25° x 0.25° global latitude longitude grid (https://rda.ucar.edu/datasets/ds083.3/). A combined 40-category dataset is used to represent the LCC, where the National Land Cover Database (NLCD) is used within the CONUS, and elsewhere the International Geosphere-Biosphere Programme (IGBP)-Modified Moderate Resolution Imaging Spectroradiometer (MODIS) satellite database. Independent tests of a similar WRF model configuration/domain over the U.S. also indicates that setting the number of tiles (N) = 8 results in about 97% of all model grid cells having 99% of their LCC categories represented (Campbell et al., 2019). To ensure all MPAS/Noah mesh cells have ≥ 99% of their LCC categories represented, here we employ a very conservative value of N = 15.

The simulation design consists of 1-month simulations using dominant and tiled LCC for January and July 2016, both at 92–25 and 46–12 km variable mesh grid spacing (total of 8 simulations). Each simulation applies a continuous 10-day spin-up (not used in analysis) and a collection of 5-day reinitializations (Table 2), with no overlaps, which both reduces the error ingested from the initial conditions and helps avoid model divergence typical of longer simulation periods (e.g., multiple weeks or months).

Table 2.

MPAS-A v6.0 Model Simulation Design

Run MPAS/Noah LCC Period Mesh Resolution
#1 Dominant January 2016 92–25 km
#2 Dominant July 2016 92–25 km
#3 L13-tiled January 2016 92–25 km
#4 L13-tiled July 2016 92–25 km
#5 Dominant January 2016 46–12 km
#6 Dominant July 2016 46–12 km
#7 L13-tiled January 2016 46–12 km
#8 L13-tiled July 2016 46–12 km
Open in a new tab

The simulation design in Table 2 allows for the analysis of the impacts of L13-tiled compared to dominant LCC during both a winter and summer month, while also providing insight into the impact of the L13-tiled approach on reduction of the sensitivity of the MPAS/Noah model to different mesh resolutions.

2.3. Observations and Evaluation Protocol

Observations from both surface and upper-air platforms are used for the evaluation of MPAS-A dominant versus the tiled method and the sensitivity to the refining mesh resolution. The near-surface observations of 2-meter temperature (T2), 2-meter specific humidity (Q2), and 10-meter wind speed (WSPD10) are based on the Surface Weather Observations and Reports for Aviation Routine Weather Reports (METAR) which are collected by NCEP’s Meteorological Assimilation Data Ingest System (MADIS) (https://madis.ncep.noaa.gov/madis_metar.shtml). The shortwave radiation at the ground (SWDOWN) observations are obtained from the World Radiation Monitoring Center’s (WRMC) Baseline Solar Radiation Network (BSRN) (https://bsrn.awi.de/; Driemel et al., 2018;). Vertical profile observations of temperature, relative humidity, and wind speed are obtained from the National Oceanic and Atmospheric Administration (NOAA), Earth System Research Laboratory’s (ESRL) Radiosonde Database (RAOB) (https://ruc.noaa.gov/raobs/).

Typical meteorological statistical metrics are used to evaluate the performance of MPAS-A dominant versus the tiled approach, which include the mean bias (MB), root mean square error (RMSE), Pearson’s correlation coefficient (R), and index of agreement (IOA). Such statistical metrics have been well defined in the available literature (e.g., Table 3 in Emery et al., 2016).

3. Results

3.1. Impacts of the tiled approach for the MPAS-A 92–25 km mesh

Globally, the tiled method’s top ranked tiles (i.e., ranking of LCC by dominance) show a high heterogeneity in LCC categories and associated fractions compared to the dominant category (Figure 3a–f). In the western U.S., the tiled method allows for forest fractions in cells dominated by grasses/shrubs, while in the eastern U.S., there are urban and grass/shrub fractions in cells dominated by forest (Figure 3g–l). Ultimately, the amount of tiled LCC heterogeneity depends on the combination of the specific input LCC dataset and model cell resolution, which in this case varies from the global (IGBP-MODIS and 92 km) to the U.S. scale (NLCD and 25 km).

Figure 3.

Figure 3.

Open in a new tab

Spatial plots of the top three ranked (#1 = dominant LCC, #2 and #3), tiled LCC categories and their associated fractions for the combined 40-category NLCD (CONUS) and (IGBP)-Modified MODIS satellite database (elsewhere global). The IGPB-MODIS (1–17) and NLCD (21 – 40) LCC categories are (combined for simplification): Forests: 1–5 & 28–30; Grasses, Shrubs, or Savannahs: 6–10 & 31–34 & 37; Lichens/Moss: 35–36; Wetlands: 11 & 39–40; Croplands: 12, 14, & 38; Urban/Developed: 13 & 23–26; Snow and Ice: 15 & 22; Barren/Sparsely Vegetated: 16 & 27; Water: 17 & 21; Regions of LCC fraction <0.01 or Unclassified: 0 (shaded grey for land) & 18– 20 (shaded pure white for water). Areas with zero associated LCC fractions are also shaded in grey for land and pure white for water.

Including additional LCC categories in the tiled method results in global differences for the top three ranked tiled (i.e., tile #2 - #1 and #3 - #1) surface/skin temperature (TSKtile), surface specific humidity (QSFCtile), sensible heat flux (SH_flxtile), latent heat flux (LH_flxtile), ground heat flux (Gtile), and aerodynamic roughness length (Ztile) in January (Figure 4) and July 2016 (Figure 5) for the 92–25 km mesh (Supporting Figure S1 also shows the albedo, ALBtile, and emissivity, EMIStile differences). Clearly the regions of appreciable LCC fractions for tile rankings #2 and #3 (Figure 3) spatially agree well with the areas of largest changes in surface variables and fluxes, where the magnitude of TSK change is generally larger in July (e.g., Figure 5a; global avg. ΔTSKtile#2-#1 = −0.13 K) compared to January (e.g., Figure 4a; global avg. ΔTSKtile#2-#1 = −0.04 K), especially at the refined mesh scale (~25 km) in the CONUS region that uses the NLCD.

Figure 4.

Figure 4.

Open in a new tab

January 2016 average differences in the top ranked tiled LCC categories (#2-#1 and #3-#1) for a)-b) TSKtile, c)-d) SHtile, e)-f) Gtile, g)-h) QSFCtile, i)-j) LHtile, and k)-l) Ztile on the 92–25 km resolution mesh. Areas shown in grey for land and white for water have zero change for each tiled variable and are not significant (see Figure 3 for LCC ranking and fraction).

Figure 5.

Figure 5.

Open in a new tab

Same as in Figure 4, but for July 2016.

On average, comparing the #2 and #3 ranked tiles to the dominant (#1 rank) leads to decreased global TSKtile and SH_flxtile, and slightly increased G (more heat flux into the ground) in both January and July (Figures 4a–f and 5a–f). There are exceptions, however, where the #2 and #3 ranked LCC tiles demonstrates increased TSKtile, SH_flxtile, and Gtile, especially in the eastern CONUS for Tile #2–1 and #3–1 in July. These increases are due to the effects that urban and crop/grasslands in the #2 and #3 ranked tiles have on the surface energy balance compared to the dominant deciduous and evergreen forest in mesh cells found in this region (Figure 3). In the western U.S. in July, the #2 and #3 ranked tiles have appreciable fractions of forests compared to the dominant grasses, shrubs, or savannahs in this region that leads to a strong cooling effect with widespread decreases in TSKtile, SH_flxtile, and Gtile.

The tiled method also impacts the aerodynamic roughness length (Ztile), which have the same changes for January and July 2016 because Ztile is solely a function of the tabulated LCC category in the Noah LSM. On average, there is a slight decrease in Ztile globally; however, there are locally larger increases and decreases dependent on the level of contrast in roughness lengths for different categories. For example, there are relatively large decreases in Ztile in the eastern U.S. due to smaller average roughness lengths for croplands (~0.1; #2 and #3 ranked tiles) compared to the dominant forests (~0.5). Changes in roughness lengths, Ztile, will have impacts on the diagnosed average wind speeds above the surface, as discussed further below.

There are prominent changes in the average January and July 2016 difference plots (tiled – dominant) for MPAS-A diagnostic variables T2, Q2, and WSPD10 at the 92–25 km resolution mesh (Figure 6). In fact, mostly all of the differences are deemed statistically significant based on a calculated p-value test statistic that is ≤ 0.05 for each cell. The p-value (i.e., probability) is used to determine the likelihood of a false null hypothesis, where the smaller the p-value, the higher the significance differences (tiled – dominant). There are only localized areas of statistically insignificant changes near the regions of smallest differences, and are mostly found in January over the oceans and Antarctica (Figure 6; p-values > 0.05 masked out as grey for land and white for water).

Figure 6.

Figure 6.

Open in a new tab

Average January (left) and July 2016 (right) dominant and difference plots (tiled – dominant) for the diagnostic variables a)-d) T2, e)-h) Q2, and i)-l) WSPD10 on the 92–25 km resolution mesh. Areas shaded as grey for land and pure white for water represent a p-value > 0.05, and are statistically insignificant differences (tiled-dominant).

The globally widespread cooling in both January (~ −0.1 K) and July 2016 (−0.4 K) in Figure 6 are due to global increases in LH_flx, widespread decreases in SH_flx, and consequently widespread increases in the EF (Evaporative Fraction; See Section 1 for explanation), especially in July (Figure 7). These results suggest that using a tiled LCC may act to mitigate the systematic warm bias found in global NWP model applications that currently use a dominant LCC approach.

Figure 7.

Figure 7.

Open in a new tab

Average January (left) and July 2016 (right) difference plots (tiled – dominant) for a)-b) LH_flx, c)-d) SH_flx, and e)-f) evaporative fraction (EF) on the 92–25 km mesh. Areas shaded in grey for land and pure white for water represent a p-value > 0.05 and are statistically insignificant differences (tiled-dominant)

There is also a qualitative agreement for areas that are typical of higher (lower) temperatures (Figures 6a–b) with areas of decreased (increased) temperature due to tiled LCC (Figures 6c–d). The regions of lower (higher) humidity (Figures 6e–f) also agree with regions of increased (decreased) humidity (Figures 6g–h) due to tiling. The opposing directions of change in T2 and Q2, particularly for the U.S. in July (see dipoles of change in the west and east U.S.), further indicates that the impacts of the tiled method are a result of changes in the surface energy balance and a shift in the partitioning of the SH_flx and LH_flx, which is due to the incorporation of appreciable fractions of various LCCs in the mesh cells. There are spatially variable impacts on WSPD10 with both increases and decreases (Figure 6i–l), but the impact of the tiled LCC is largest in July, and also leads to secondary (e.g., effects of modulated north-south temperature gradient) interactions in the northern high latitudes (~ > 60°N).

The impacts of the tiled method on EF also has feedbacks to the topmost soil temperature (TSLB) and moisture (SMOIS), as well as the planetary boundary layer height (PBLH) (Figure S2). In January, the TSLB increases in the high Northern latitudes and decreases in the low Northern latitudes and Southern Hemisphere. The increases (decreases) in TSLB qualitatively agree with decreases (increases) in SMOIS due to the effects of EF changes. In July, the effects are similar but exacerbated, where there are widespread decreases in TSLB and increases in SMOIS, respectively, which is most prominent in the Northern Hemisphere mid- to high latitude regions, and may also impacted by increases in the low cloud fraction (not shown). As expected, the increases (decreases) in PBLH are spatially well correlated with the regions of increases (decreases) in T2.

The spatial differences in tiled and diagnostic variables in January and July 2016 (Figures 4–6) are further elucidated when comparing the diurnal patterns of tiled G, LH, SH, QSFC, and TSK for the dominant and 2nd ranked LCC category, most notably in July (Figure 8).

Figure 8.

Figure 8.

Open in a new tab

Diurnal analysis of the differences in dominant – 2nd ranked tiles forTSK, QSFC, G, SH, and LH in a)-e) January and f)-j) July 2016. Analysis has been averaged over all CONUS grid cells. Approximate daytime hours for CONUS are shaded in light yellow/grey.

In much of the central and western U.S. the widespread shrubs/grasses dominate the landscape (Figure 1), where including appreciable fractions of evergreen and decidous forest in the tiled approach leads to a net increase in heat flux into the ground (ΔG > 0) at the expense of sensible heat flux (ΔSH < 0), which results in cooler surface temperatures (ΔTSK < 0) at night in July (black line; Figure 8f–j). During the morning transition period, the ΔG, ΔSH, and change in latent heat flux (ΔLH) approach zero, and there is a minimum in ΔTSK due to the tiled effects. Later in the daytime hours in July, however, the presence of more evergreen and deciduous forest result in a net loss of ground heat flux (ΔG < 0), which mostly enhances latent heat flux to the atmosphere (ΔLH>0), where energy partitioning also requires that ΔSH<0 and consequently cooler surface temperatures, ΔTSK < 0. There is also a net increase in specific humidity (ΔQSFC>0) during both night and day in July due to the presence of more forest canopy and enhanced evapotranspiration (black line; Figure 8f–j). The opposite is true when including tiled fractions of shrubs/grasses in either the dominant evergreen or deciduous forest regions, which are found mainly in the eastern U.S, and leads to predominantly drier (ΔQSFC<0) and warmer surface conditions (ΔTSK > 0) (red and blue lines; Figure 8f–j). A similar drier and warmer pattern is also true when including tiled fractions of evergreens in cells that are dominated by decidous forest, as evergreens typically have less daytime transpiration (ΔLH<0) compared to deciduous trees in July (green lines; Figure 8f–j). In January, the diurnal patterns of G, LH, SH, QSFC, and TSK in January are similar to July, but have smaller amplitudes due to smaller net radiation energy for the U.S. winter, and consequently smaller magnitudes for the ΔG, ΔLH, and ΔSH partitioning (Figures 8a–8e).

3.2. Model evaluations of the MPAS-A dominant and tiled LCC approach

The tiled approach results in widespread reductions in MB for T2, Q2, and WSPD10 across the U.S. for a 92–25 km mesh during January and July 2016 (Figure 9). The largest, and most prolific reductions in MB are found in the western U.S. in July, where there are large decreases in T2 and increases in Q2 (Figures 9c–d and 9g–h). There are some smaller areas of increased MB for T2 and Q2, most notably in the southeast U.S. for July where increased temperatures exacerbate the simulated warm bias for T2, and in parts of the Central U.S. where decreases in predicted mixing ratio exacerbates the model dry bias for Q2. While more variable in nature, there are predominantly decreased MB for WSPD10 across the U.S. There are also widespread decreases in the RMSE for the tiled approach for T2, Q2, and WSPD10 (Supporting Figure S3). Similar to the global changes (Figure 6), mostly all the major changes in T2, Q2, and WSPD10 are statistically significant in CONUS (p-value ≤ 0.05), with only very localized areas of insignificant changes (i.e., p > 0.05).

Figure 9.

Figure 9.

Open in a new tab

Average January (left) and July 2016 (right) absolute and MB difference (|tiled MB| – |dominant LCC MB|) compared against available MADIS-METAR stations for a)-d) T2, e)-h) Q2, and i)-l) WSPD10 on the 92–25 km resolution mesh. Areas shaded in grey for land and pure white for water represent a p-value > 0.05, and are deemed statistically insignificant differences (tiled-dominant).

The results from the 92–25 and 46–12 km meshes show a reduction in diurnal MB and RMSE for T2, Q2, and WSPD10 in the western U.S. during July, and there is preliminary indication of decreasing model sensitivity to the mesh resolution for T2 when using the tiled approach (i.e., MB red lines closer than blue lines) (Figures 10–11); however, testing of more resolutions are necessary for a full investigation of grid sensitivity. The impacts of the tiled approach are less for January in the eastern U.S., with some model degradation for similar reasons as discussed previously. The average CONUS and global statistical summaries (i.e., R, MB, RMSE, and IOA) are found in Supporting Tables S1 and S2. Overall, the largest model performance change in CONUS is for T2, where the average MB is reduced by a factor of ~ 4 due to the tiled approach. There is also lower MB for the WSPD10 in the western U.S. for July; however, there are increases in MB and RMSE for Q2 and WSPD10 in the eastern U.S. This dipole in model performance change apparent across CONUS is consistent with the strong east-west vegetation and moisture gradient and its interaction with the tiled compared to dominant LCC approaches.

Figure 10.

Figure 10.

Open in a new tab

Diurnal MB comparisons for a)-d) T2, e)-h) Q2, and i)-l) WSPD10 against MADIS-METAR for 92–25 and 46–12 km over eastern and western CONUS.

Figure 11.

Figure 11.

Open in a new tab

Same as in Figure 10, but for RMSE.

The effects of the tiled LCC approach on the partitioning of SH_flx and LH_flx also has implications for the total incoming SWDOWN (Figure 12).

Figure 12.

Figure 12.

Open in a new tab

Average diurnal time series of SWDOWN and bias comparisons against BSRN for 10 CONUS sites for both 92–25 and 46–12 km meshes.

For an average of 10 BSRN sites across CONUS (Supporting Figure S4 contains a map of the U.S. sites), the tiled approach leads to an overall reduction in total SWDOWN during the local peak time, which leads to an overall increase in MB (and RMSE; see Supporting Figure S5) compared to the dominant LCC approach. This effect is more prominent during summer in July due to appreciable forest fractions included in the dominant shrublands/grasslands across the western U.S. (Figure 3h), and the resulting increase in EF and Q2 here (Figures 6h and 7d). A spatial evaluation of SWDOWN against the global BSRN observation sites also shows increases in MB and RSME in the early to late afternoon hours due to the tiled approach, particularly in July 2016 for BSRN sites in the Northern Hemisphere (Supporting Figures S6–S7).

Incorporating more detailed and realistic LCC in MPAS-A leads to widespread model performance improvements (decreased MB and RMSE) for T2 and Q2, particularly in the western U.S., but a degradation (increased MB and RMSE) in the evaluation of SWDOWN possibly driven by competing cloud-radiative feedback effects, both in the U.S. and globally. This result may be indicative of a relatively “tuned” model performance towards more accurate predictions of near-surface temperature and moisture in MPAS-A (and other NWP models), at the expense of degrading performance (and more unknown) for cloud-radiative feedback processes that affect the surface radiative balance (Ma et al., 2014).

The impacts of the tiled approach extend above the surface as well, and there are increases in the IOA for the temperature, relative humidity, and wind speed profiles compared to RAOB sites across CONUS (Figure 13). The tiled approach shows increases in IOA for temperature up to about 500–600 hPa model heights for all the RAOB sites shown except in the northwest (Boise, Idaho; KBOI) and southwest U.S. (Salt Lake City, UT; KSLC). The relative humidity also shows increased IOA across an increased depth of the atmosphere (up to 200 hPa) for the central (Lincoln, IL; KILX) and northeast U.S. (Pittsburgh, PA; KPIT) compared to the dominant approach. There are slight decreases in IOA for the tiled approach in the lower atmosphere (> 800 hPa) in the upper midwest (Detroit, MI; KDTX) and western U.S. (Oakland, CA; KOAK), but overall there are larger increases in IOA compared to the decreases across the RAOB sites (i.e., generally improved model column performance). There are also larger increases in IOA for wind speed compared to the decreases for the tiled approach, where in some cases these increases are across a significant depth of the model column, e.g., in the south (Amarillo, TX; KAMA) and northeast U.S. (KPIT).

Figure 13.

Figure 13.

Open in a new tab

Vertical profiles of IOA compared against select RAOB sites across CONUS for temperature (top), relative humidity (middle), and wind speed (bottom) for the dominant (black) and tiled (red) approach for the 92–25 km mesh.

4. Summary and Implications

In this work a tiled approach to land cover characterization (LCC) in the Noah land surface model, following Li et al. (2013), is implemented in the Model for Prediction Across Scales – Atmosphere (MPAS-A), version 6.0, and was tested for January and July 2016 on both the 92–25 and 46–12 km refining meshes (focused on the conterminous U.S.; CONUS). Implementation of the tiled LCC in MPAS shows significant impacts on global soil conditions, surface fluxes, and near-surface atmospheric properties compared to the largely simplified dominant LCC approach. Specifically, the tiled LCC leads to both moderate warming and cooling in the Northern and Southern Hemisphere in January, respectively, with more dramatic, globally widespread cooling in July. For CONUS, there is a strong dichotomy of cooler and moister conditions in the west, and warmer and drier conditions in the east due to the tiled LCC. Such temperature and moisture changes are a result of shifts in tiled evergreen and deciduous forests, grasslands/shrublands, and urban LCC in the eastern and western U.S. compared to the dominant approach, which alter the overall cloud-radiative balance, available energy, and diurnal partitioning between the ground, sensible, and latent heat fluxes. These changes in turn effect the development of near-surface wind flow, boundary layer heights, cloud formation processes, and resulting cloud-radiative feedbacks.

The tiled LCC has a strong impact on model performance, where there are significant reductions in both mean bias and root mean square error in CONUS for 2-m temperature, 2-m specific humidity, and 10-m wind speed; however, there are increases in model bias and error for incoming solar radiation. There is a preliminary indication that the tiled LCC also reduces the sensitivity of predicted 2-m temperature to the finer 46–12 km mesh resolution in the eastern U.S.; however, more studies of the mesh sensitivity are necessary. Depending on the specific variable (temperature, relative humidity, or wind speed) and U.S. region, there is also a moderate indication that the tiled LCC impact may extend upwards the atmospheric column, while demonstrating improved agreement compared to many radiosonde locations.

An important implication of this work is the effect of the tiled LCC on the evaporative fraction, cloud-radiative feedbacks, and the overall reduction in global temperatures in July (Northern Hemisphere summer). As demonstrated by the improved model performance for 2-meter temperature in CONUS, use of a tiled LCC could mitigate the systematic, global summertime warm biases that are apparent in many climate and numerical weather prediction (NWP) models, particularly those that apply a dominant LCC approach. The improved near-surface meteorology, but degraded performance in incoming solar radiation due to the more detailed tiled LCC further demonstrates that NWP models such as MPAS-A have experienced prolonged deficiencies in the LCC representation and processes, while being preferentially “tuned” to improve the above ground meteorological predictions despite unresolved cloud-feedbacks. The need for more iterative model developments with respect to LCC methodologies in LSMs and the impacts on soil/surface, meteorological, and cloud-feedbacks in NWP models cannot be overstated. While further testing is needed (e.g., a full calendar year and detailed cloud-radiative analysis/evaluation), it is further recommended that computationally efficient subgrid LCC schemes be continually developed and integrated in the LSMs coupled to global weather forecast models.

Supplementary Material

Supplement1

Key Points.

  • The use of tiled land cover characterization (LCC) has significant impacts on global meteorological predictions in MPAS-A.

  • Tiled LCC reduces bias and error for near-surface temperature, moisture, and wind speed over the U.S.

  • The tiled LCC approach is important and can help mitigate systematic temperature biases in weather and climate models.

Acknowledgments

This research was performed while Patrick Campbell held a National Research Council Research Associateship Award at the U.S. Environmental Protection Agency. We thank Kiran Alapaty and Russell Bullock Jr. (U.S. EPA) and Fantine Ngan (NOAA/ARL) for providing valuable comments on a draft of this manuscript. We further acknowledge the MADIS-METAR, WRMC-BSRN, and ESRL-Observational networks for their data records. A pre-release version 1.0.0 of the modified MPAS-A version 6.1 source code, which includes the tiled Noah LCC option, is available at https://doi.org/10.5281/zenodo.3734982. The raw and processed MPAS simulation data are very large (total size > 10 TB) and are freely available in two different ways: 1) On the local U.S. EPA repositories that can be directly transferred via SFTP or SCP in pieces, or manually copied in their entirety and provided via external hard drives. 2) On a publicly available repository/transfer in pieces or via another high-speed file transfer service such as Globus (https://www.globus.org/data-transfer).

Footnotes

Publisher's Disclaimer: Disclaimer

Publisher's Disclaimer: The scientific results and conclusions, as well as any views or opinions expressed herein, are those of the author(s) and do not necessarily reflect the views of U.S. EPA, NOAA, or the Department of Commerce.

References

  1. Ament F, and Simmer C (2006). Improved representation of land-surface heterogeneity in a non-hydrostatic numerical weather prediction model. Boundary-layer meteorology, 121(1), 153–174. 10.1007/s10546-006-9066-4 [DOI] [Google Scholar]
  2. Avissar R (1991). A statistical-dynamic approach to parameterize subgrid scale land-surface heterogeneity in climate models. Surv. Geophys, 12(1–3), 155–178. 10.1007/Bf01903417 [DOI] [Google Scholar]
  3. Avissar R, and Pielke RA (1989). A parameterization of heterogeneous land surfaces for atmospheric numerical-models and its impact on regional meteorology. Mon. Weather Rev, 117(10), 2113–2136. [DOI] [Google Scholar]
  4. Campbell PC, Bash JO, & Spero TL (2019). Updates to the Noah land surface model in WRF‐CMAQ to improve simulated meteorology, air quality, and deposition. Journal of Advances in Modeling Earth Systems, 11 10.1002/2018MS001422 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Driemel A, Augustine J, Behrens K, Colle S, Cox C, Cuevas-Agulló, et al. (2018). Baseline Surface Radiation Network (BSRN): structure and data description (1992–2017). Earth Syst. Sci. Data, 10, 1491–1501. 10.5194/essd-10-1491-2018 [DOI] [Google Scholar]
  6. Chen F, & Dudhia J (2001). Coupling an advanced land surface-hydrology model with the Penn State-NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev, 129, 569–585. [DOI] [Google Scholar]
  7. Chen F, Mitchell K, Schaake J, Xue Y, Pan H, Koren V, et al. (1996). Modeling of land-surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res, 101, 7251–7268. 10.1029/95JD02165 [DOI] [Google Scholar]
  8. Chen F, Janjic Z, & Mitchell KE (1997). Impact of atmospheric surface-layer parameterizations in the new land-surface scheme of the NCEP mesoscale Eta model. Bound. Layer Meteorol, 85, 391–421. 10.1023/A:1000531001463 [DOI] [Google Scholar]
  9. Chen F, Manning KW, LeMone MA, Trier SB, Alfieri JG, Roberts R, et al. (2007). Description and evaluation of the characteristics of the NCAR high-resolution land data assimilation system. J. Appl. Meteorol. Climatol, 46, 694–713. 10.1175/JAM2463.1 [DOI] [Google Scholar]
  10. Ek MB, Mitchell KE, Lin Y, Rogers E, Grunmann P, Koren V, et al. (2003). Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys. Res, 108, D22, 10.1029/2002JD003296 [DOI] [Google Scholar]
  11. Emery C, Liu Z, Russell AG, Odman MT, Yarwood G, & Kumar N (2017). Recommendations on statistics and benchmarks to assess photochemical model performance. Journal of the Air & Waste Management Association, 67:5, 582–598. 10.1080/10962247.2016.1265027 [DOI] [PubMed] [Google Scholar]
  12. Entekhabi D, and Eagleson PS (1989). Land surface hydrology parameterization for atmospheric general circulation models including subgrid scale spatial variability. J. Clim, 2(8), 816–831, [DOI] [Google Scholar]
  13. Famiglietti JS, and Wood EF (1991), Evapotranspiration and runoff from large land areas: Land surface hydrology for atmospheric general circulation models, Surv. Geophys, 12(1–3), 179–204, 10.1007/Bf01903418 [DOI] [Google Scholar]
  14. Giorgi F, and Avissar R (1997). Representation of heterogeneity effects in earth system modeling: Experience from land surface modeling. Rev. Geophys, 35(4), 413–437, 10.1029/97RG01754 [DOI] [Google Scholar]
  15. Gollvik S, and Samuelsson S, (2011). A tiled land-surface scheme for HIRLAM Swedish Meteorological and Hydrological Institute; SE-601 76 Norrkoping, Sweden. [Google Scholar]
  16. Hibbard KA, Hoffman FM, Huntzinger D, & West TO (2017). Changes in land cover and terrestrial biogeochemistry. In: Climate Science Special Report: Fourth National Climate Assessment, Volume I [Wuebbles DJ, Fahey DW, Hibbard KA, Dokken DJ, Stewart BC, and Maycock TK(eds.)]. U.S. Global Change Research Program, Washington, DC, USA, pp. 277–302. 10.7930/J0416V6X [DOI] [Google Scholar]
  17. Koster RD, and Suarez MJ (1992b). A comparative analysis of two land surface heterogeneity representations. J. Clim, 5(12), 1379–1390. [DOI] [Google Scholar]
  18. Li R, and Arora VK (2012). Effect of mosaic representation of vegetation in land surface schemes on simulated energy and carbon balances. Biogeosciences, 9, 593–605. 10.5194/bg-9-593-2012. [DOI] [Google Scholar]
  19. Li D, Bou-Zeid E, Barlage M, Chen F, & Smith JA (2013). Development and evaluation of a mosaic approach in the WRF-Noah framework. J. Geophys. Res. Atmos, 118, 11,918–11,935. 10.1002/2013JD020657. [DOI] [Google Scholar]
  20. Li H, Wolter M, Wang X, & Sodoudi S (2017). Impact of land cover data on the simulation of urban heat island for Berlin using WRF coupled with bulk approach of Noah-LSM. Theoretical and Applied Climatology, 134(1–2), 67–81. 10.1007/s00704-017-2253-z [DOI] [Google Scholar]
  21. Ma H, Xie S, Klein SA, Williams KD, Boyle JS, Bony S, et al. (2014). On the Correspondence between Mean Forecast Errors and Climate Errors in CMIP5 Models. J. Climate, 27, 1781–1798. 10.1175/JCLI-D-13-00474.1 [DOI] [Google Scholar]
  22. Mallard MS, and Spero TL (2018). Effects of Mosaic Land Use on Dynamically Downscaled WRF Simulations of the Contiguous United States. J. Geophys. Res. Atmos, 124, 16, 9117–9140. 10.1029/2018JD029755 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Manrique-Suñén A, Nordbo A, Balsamo G, Beljaars A, & Mammarella I (2013). Representing Land Surface Heterogeneity: Offline Analysis of the Tiling Method. J. Hydrometeor, 14, 850–867. 10.1175/JHM-D-12-0108.1 [DOI] [Google Scholar]
  24. Masson V, Le Moigne P, Martin E, Faroux S, Alias A, Alkama, et al. (2013). The SURFEXv7. 2 land and ocean surface platform for coupled or offline simulation of earth surface variables and fluxes. Geosci. Model Dev, 6, 929–960. 10.5194/gmd-6-929-2013 [DOI] [Google Scholar]
  25. Melton JR, & Arora VK (2014). Subgrid scale representation of vegetation in global land surface schemes: implications for estimation of the terrestrial carbon sink. Biogeosciences, 11, 1021–1036. 10.5194/bg-11-1021-2014. [DOI] [Google Scholar]
  26. Mitchell KE, Lohmann D, Houser PR, Wood EF, Schaake JC, Robock A, et al. (2004). The multi-institution North American Land Data Assimilation System (NLDAS): utilizing multiple GCIP products and partners in a continental distributed hydrological modeling system. J. Geophys. Res, 109, D07S90 10.1029/2003JD003823 [DOI] [Google Scholar]
  27. Morcrette CJ, Van Weverberg K,Ma H-Y, Ahlgrimm M, Bazile E,Berg LK, et al. (2018). Introduction to CAUSES: Description of weather and climate models and their near-surface temperature errors in 5day hindcasts near the Southern Great Plains. Journal of Geophysical Research:Atmospheres,123, 2655–2683. 10.1002/2017JD027199 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Newman AJ, Clark MP, Winstral A, Marks D, & Seyfried M (2014). The Use of Similarity Concepts to Represent Subgrid Variability in Land Surface Models: Case Study in a Snowmelt-Dominated Watershed. J. Hydrometeor, 15, 1717–1738, 10.1175/JHM-D-13-038.1 [DOI] [Google Scholar]
  29. Niu G-Y, Yang Z-L, Mitchell KE, Chen F, Ek MB, Barlage M, et al. (2011). The community Noah land surface model with multiparameterization options (Noah-MP): 1. Model description and evaluation with local-scale measurements. J. Geophys. Res 116, D12109, 10.1029/2010JD015139 [DOI] [Google Scholar]
  30. Pan H-L, and Mahrt L (1987). Interaction between soil hydrology and boundary layer development. Boundary Layer Meteorol, 38, 185–202. 10.1007/BF00121563 [DOI] [Google Scholar]
  31. Powers JG, Klemp JB, Skamarock WC, Davis CA, Dudhia J, Gill DO, et al. (2017). The weather research and forecasting model: Overview, system efforts, and future directions. Bulletin of the American Meteorological Society, 98(8), 1717–1737. 10.1175/BAMS-D-15-00308.1 [DOI] [Google Scholar]
  32. Samuelsson S, Gollvik S, & Ullerstig A (2006). The land surface scheme of the Rossby Centre Regional Atmospheric Climate Model (RCA3). Meteorology 122, SMHI, SE-601 76 Norrköping, Sweden. [Google Scholar]
  33. Skamarock WC, and Klemp JB (2008). A time‐split nonhydrostatic atmospheric model for weather research and forecasting applications. Journal of Computational Physics, 227(7), 3465–3485. 10.1016/j.jcp.2007.01.037 [DOI] [Google Scholar]
  34. Skamarock WC, Klemp JB, Duda MG, Fowler LD, Park S, & Ringler TD (2012). A Multiscale Nonhydrostatic Atmospheric Model Using Centroidal Voronoi Tesselations and C-Grid Staggering. Mon. Wea. Rev, 140, 3090–3105. 10.1175/MWR-D-11-00215.1 [DOI] [Google Scholar]
  35. Steiner A (2018). Diagnosing the warm bias in the central United States, Eos, 99 10.1029/2018EO095669. Published on 23 April 2018. [DOI] [Google Scholar]
  36. Verseghy DL, McFarlane NA, & Lazare M (1993). CLASS – a Canadian land surface scheme for GCMs. 2. Vegetation model and coupled runs. Int. J. Climatol, 13, 347–370. 10.1002/joc.3370130402 [DOI] [Google Scholar]
  37. Yin J, Zhan X, Zheng Y, Hain C, Ek M, Wen J, et al. (2015). Improving Noah land surface model performance using near real time surface albedo and green vegetation fraction. Agricultural and Forest Meteorology, 218–219. 10.1016/j.agrformet.2015.12.001 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement1
NIHMS1629094-supplement-Supplement1.docx (4.6MB, docx)

RESOURCES