“chap03”—2004/1/20 — page 110 — #1 Chapter 3 High spatial resolution mapping of surface energy balance components with remotely sensed data Karen Humes, Ray Hardy,William P. Kustas, John Prueger and Patrick Starks 3.1 Introduction 3.1.1 Background In order to better understand the exchange of heat and moisture between the land surface and lower atmosphere, it is important to quantify the compo- nents of the surface energy balance in a distributed fashion at the landscape scale. Remotely sensed data can provide spatially distributed information on a number of key land surface characteristics and state variables that control the surface energy balance. When combined with near-surface meteorologi- cal measurements and a relatively simple model, satellite and aircraft-based remotely sensed data can be used to create “maps” of spatially distributed surface energy balance components over a watershed. Assuming no advec- tion of energy into an area, the simplest form of the surface energy balance is given by R net = G + H +LE (3.1) where R net refers to the net radiation balance, G refers to the soil heat flux (i.e. the energy used to warm the near-surface soil layers), H refers to the sensible heat flux (the energy used to transfer heat from the surface to the atmosphere), and LE refers to the latent heat flux (the energy used to transfer water vapor from the surface to the atmosphere). The influence of the land surface energy fluxes on regional and global atmospheric processes has become well recognized in the climate and meteo- rological modeling communities (e.g. Avissar and Pielke 1989; Chen and Avissar 1994; Betts et al. 1996). This has given rise to the development of quite a number of more sophisticated parameterizations for simulating land surface processes within mesoscale and global atmospheric models (Dickinson et al. 1986; Sellers et al. 1986; Entehkhabi and Eagleson 1989; Noilhan and Planton 1989; Avissar and Verstraete 1990; Xue et al. 1991). “chap03”—2004/1/20 — page 111 — #2 High spatial resolution mapping 111 The use of these schemes within atmospheric models has helped to improve the performance of both regional and mesoscale atmospheric models. However, most of these models, referred to as soil–vegetation–atmosphere transfer (SVAT) models, require a priori knowledge of a considerable num- ber of surface parameters and detailed information for initialization. They also require pertinent ground data and substantial human effort for model calibration. Additionally, when complex point-scale models are run within the context of mesoscale or global atmospheric models, the grid cell reso- lution is generally on the order of hundreds to thousands of meters in size. Many of the key parameters and variables in the complex physically based models would be expected to vary considerably within grid cells of that size. 3.1.2 Objectives of this study The primary objective of this study is to demonstrate the feasibility of using high spatial resolution remotely sensed data, combined with driving mete- orological data from a ground network and a relatively simple model, to compute spatially distributed values of surface energy balance components. The model employed here is a relatively simple “snapshot” model. That is, it does not simulate any of the processes as a function of time; rather, it uses satellite and ground data to estimate the fluxes at the time of the satellite overpass. Almost all the model parameters and variables used by the model (such as surface temperature, land cover type, and vegetation density) are estimated from remotely sensed data. The meteorological inputs required by the model were derived from a ground network. This approach has the advantage of being very “data driven” and the model does not need to be calibrated or “tuned” for a particular site. Thus, the fluxes estimated from this approach can be useful for validation or assimilation into more complex simulation models. The model was applied on a pixel-by-pixel basin across a watershed in a sub-humid climate zone. Although surface fluxes have been previously mapped using these types of approaches (Moran et al. 1990; Holwill and Stewart 1992; Humes et al. 1997), this study represents the application of a more complex (two-layer) model over more heterogeneous land cover types than these previous efforts. Additionally, the watershed studied here has a special instrumentation network that makes possible more detailed spatial analysis of the factors influencing the surface fluxes. The motiva- tion for applying this model at relatively high spatial resolution over the watershed is twofold: (a) at higher spatial resolution the approach is more easily validated using ground-based point measurements and (b) mapping the fluxes at high spatial resolution allows an evaluation of the relative impor- tance of various surface and atmospheric variables in determining the surface fluxes. “chap03”—2004/1/20 — page 112 — #3 112 Humes et al. 3.2 Study area The USDA /Agricultural Research Service (ARS) Little Washita Watershed (LWW), operated by the ARS Grazinglands Research Station, is located in central Oklahoma. The land cover types present in the watershed include a mixture of cultivated areas (primarily winter wheat, soybeans, alfalfa, and corn), pastures with native grasslands and non-native species, varied man- agement practices, and (depending considerably on climatological variables that vary considerably from east to west) wooded areas. The LWW has also been the site of several special experimental campaigns involving the simultaneous acquisition of ground and remotely sensed data. The water- shed was a US Supersite for the SIR-C (Shuttle Imaging Radar) Experiments in 1992 and 1994. The SIR-C experiments became the focal point for one field campaign in 1992 and three field campaigns in 1994 which included many different ground measurements, as well the acquisition of many types of remotely sensed data from ground, aircraft, and satellite-based sensors (Jackson and Scheibe 1993; Starks and Humes 1996). Remotely sensed data sets included passive microwave, active microwave, and optical sensors. Among the many special ground observations acquired during these cam- paigns were the measurement of surface energy fluxes by Bowen ratio and eddy correlation techniques (Prueger 1996; Kustas et al. 1999). These ground-based measurements were used for validation of the surface energy fluxes produced by this modeling effort. Observations also included ground- based radiometric measurements of surface reflectance and temperature. These were acquired with a backpack-type apparatus that facilitated the acquisition of ground data over a large, relatively uniform target area at the time of the Landsat satellite overpass. These data were used to vali- date the atmospherically corrected radiometric surface temperatures derived from satellite data. Additionally, the ARS operates the Micronet network in the LWW, which consists of 42 monitoring stations on a 5-km grid. These stations record meteorological variables such as incoming solar radi- ation and near-surface (1.8 m) air temperature and relative humidity. These measurements were used for meteorological input to the model. The data sets used in this analysis were from the August 1994 field cam- paign on the CWW. A false color composite image from Landsat 5 Thematic In this image, the data from the TM band 4 (near-infrared) are displayed as red, data from the TM band 3 (red) are displayed as green, and data from the TM band 2 (green) are displayed as blue. In August, the winter wheat fields are typically bare and thus appear bluish green on the false color com- posite image. It can be observed from the image that these areas are most extensive in the western portion of the watershed. The bright red areas of the image correspond to riparian vegetation along drainage areas, the relatively small watershed area corresponding to cultivated crops that are green at this Mapper (TM) data acquired on August 18, 1994, is shown in Figure 3.1. “chap03”—2004/1/20 — page 113 — #4 High spatial resolution mapping 113 Figure 3.1 False color composite image from the Landsat TM sensor for the LWW from time of year (such as corn and alfalfa), and, to a lesser degree, the spatially extensive pastures of various density and species composition. In the early morning hours of August 18, a relatively intense thunder- storm moved through the watershed. The cumulus clouds that can be seen in Figure 3.1, and the cirrus cloud contamination over a portion of the water- shed evident in the thermal band, were remnants from that storm. The system moved out of the watershed region approximately 1 h before the image was acquired. 3.3 Model description and implementation 3.3.1 Model description The model utilized here is described in detail in Norman et al. (1995). It is a two-source model, meaning that separate energy balance computations are done for the soil and vegetation layers of the surface. It was run on a pixel- by-pixel basis to compute spatially distributed energy fluxes over the LWW during the time of the Landsat TM overpass during the August 1994 field The conceptual model formulation is summarized here. The four components of net radiation are quantified as follows: (a) incoming solar radiation is a model input typically provided by ground mea- surements; (b) outgoing solar radiation is computing using incoming solar campaign. A diagram of model inputs and outputs is shown in Figure 3.2. August 18, 1994 (see Colour Plate XII). “chap03”—2004/1/20 — page 114 — #5 114 Humes et al. Model inputs/outputs Landsat TM derived surface characteristicsPoint-based meteorological data from Micronet stations Air temperature Relative humidity Solar radiation 120 m Atmospherically corrected radiometric surface temperatures 30 m RED and NIR reflectance from TM data 30 m Land cover classification from TM data Aggregate to 120 m pixel Grids Kriging of each data parameter to 120 m pixel grids Norman two-source model Model output 120 m Latent heat flux 120 m Ground heat flux 120 m Sensible heat flux 120 m Net radiation Figure 3.2 Conceptual diagram of the input and output quantities used for the application of the Norman et al. (1995) model to data from the August 18, 1994, Landsat scene over the LWW. radiation and assumed values of surface albedo for different land cover types; (c) incoming longwave radiation is estimated using ground-based measure- ments of air temperature and relative humidity and an empirical expression for clear sky conditions (Idso 1981); (d) outgoing longwave radiation is computed using the surface temperature from the satellite data and an assumed emissivity of 0.98. It should be noted that for some “snapshot”- type models for estimating fluxes, surface albedo is calculated using empirical functions that relate surface hemispherical albedo to reflectance in the finite wavebands of the Landsat TM sensor. This approach was not utilized in this application because of uncertainty in the atmospheric correction of the satellite data to absolute surface reflectance. The net radiation at the surface is partitioned between the soil and vege- tation layers using a typical “Beers law” formulation. The exponent in this relationship is controlled by an estimate of the fractional vegetation cover (which is estimated from remotely sensed data in the manner described in more detail below), and an assumption of spherical leaf inclination angle distribution. Soil heat flux is assumed to be a constant fraction (0.35) of the net radiation reaching the soil. The total sensible and latent heat fluxes are simply taken to be the sum of the vegetation and soil contributions. Those contributions are determined by doing a separate surface energy balance on the soil and vegetation lay- ers and assuming that the flux of heat from the soil and vegetation layers “chap03”—2004/1/20 — page 115 — #6 High spatial resolution mapping 115 act in parallel. This gives rise to a simpler resistance formulation than some multi-layer models (e.g. Shuttleworth and Wallace 1985; Xue et al. 1991), but several studies have shown that for low to moderate vegetation cover, the various levels of complexity in resistance networks are indistin- guishable because air temperature gradients are small in the upper canopy (Norman and Campbell 1983). The key to estimating both contributions to the sensible heat flux is in the decomposition of the radiometric surface tem- perature (T rad ), derived from satellite observations, into soil and vegetation components using T rad (θ) =[f (θ)T n c + (1 − f (θ))T n s ] 1/n (3.2) where T c is the vegetation canopy temperature, T s is the soil surface temper- ature, n is the power of the temperature and approximates the blackbody function when considered over the entire wavelength of the sensor, θ is the view angle of the sensor, f (θ) is the fraction of the field of view of the radiometer occupied by canopy and is given by f (θ) = 1 − exp  −0.5F cos θ  (3.3) where F is the leaf area index. At the value of θ = 0: f c = 1 − exp(−0.5F) (3.4) where f c is the fractional vegetation cover. The component surface temperatures and the turbulent flux components for the soil and vegetation layers are derived using a series of steps that some- times require iteration. In the following equations, the symbols R net,c , H c , and LE c refer to the canopy portion of the net radiation, sensible, and latent heat fluxes, respectively, and the symbols R net,s , H s , and LE s refer to the soil contribution to the net radiation, sensible, and latent heat fluxes. First, an approximation for the canopy portion of the latent heat flux is estimated using a Priestly and Taylor (1972) type formulation with the canopy portion of the net radiation: LE c = 1.26f g  s s + γ  R net,c (3.5) where f g is the fraction of the vegetation cover which is green, s is the slope of the saturation vapor pressure versus temperature curve, γ is the psychrometric constant (0.66 kPa C −1 ). “chap03”—2004/1/20 — page 116 — #7 116 Humes et al. The sensible heat flux of the canopy layer is then computed as the residual in the energy balance for the canopy layer: H c = R net,c − LE c (3.6) The canopy temperature is then estimated by inverting the equation for a simple resistance formulation for sensible heat flux from the canopy: H c = (T c − T air )/r ah (3.7) where T c is the surface temperature of the canopy, T air is the near-surface air temperature and r ah is the aerodynamic resistance. The formulation for the aerodynamic resistance is derived from the diabatically corrected log temperature profile equations (Brutsaert 1982). The roughness lengths used in the calculation of aerodynamic resistance were set according to the land cover type as shown in Table 3.1. Using this approximation for T c and the satellite measurement of T rad , equation (3.2) is used to solve for T s , the soil temperature. This value of T s is used to calculate the soil contribution to sensible heat flux using a bulk resistance formulation for the soil layer, given by H s = ρC p (T s − T air )/(r ah + r s ) (3.8) where r s is the soil-surface resistance as derived in Norman et al. (1995). The soil component of latent heat flux is then computed as the residual in the soil energy balance: LE s = R net,s − G − H s (3.9) If the soil evaporation which results from this calculation is less than zero, then LE s is set equal to zero and H s is recomputed using equation (3.9), T s Table 3.1 Roughness length (Z 0 ), canopy height (h), and albedo for each land cover type Land cover Roughness length (m) Canopy height (m) Albedo Water 0.00035 — 0.10 Urban 0.25 — 0.25 Woodland 0.625 5 0.15 Mixed crops 0.0125 0.1 0.20 Pasture – dense 0.0625 0.50 0.15 Pasture – moderate density 0.0625 0.5 0.15 Pasture – less dense 0.0625 0.5 0.175 Sparse or senescent 0.0125 0.1 0.20 Bare soil and wheat stubble 0.01 0.1 0.15 “chap03”—2004/1/20 — page 117 — #8 High spatial resolution mapping 117 is recomputed using equation (3.8), new values of T c and H c are computed using equations (3.2) and (3.7), respectively, and a new value of LE c is computed using equation (3.6). One advantage of this model formulation over other resistance-based for- mulations and more complicated SVAT schemes is that it does not require an estimate of the canopy surface resistance to evaporation. Since this quantity is highly spatially variable, very dynamic in time, and not readily obtained from remotely sensed data, a formulation that can reliably estimate surface fluxes without the use of this parameter can be more readily applied to new areas and larger spatial scales. 3.3.2 Inputs derived from ground data As discussed above, the meteorological inputs required for the data include: incoming solar radiation, near-surface air temperature, relative humidity, and windspeed. Spatially distributed values for the near-surface air temper- ature (1.8 m above the surface) and incoming solar radiation are shown in from the USDA/ARS Micronet network of 42 stations located across the watershed, shown on the maps. The point data correspond to the data from the 5-min averaging interval closest in time to the satellite overpass time of Air temperature Little Washita Watershed Micronet station Air temperature (°C) 30–30.4 30.5–30.9 31–31.4 31.5–31.9 32–32.4 32.5–32.9 33–33.4 33.5–33.9 34–34.4 34.5–34.9 35–35.4 35.5–35.9 August 18, 1994 1640 UTC 5,000 m5,000 0 Figure 3.3 Gridded field of air temperature 2 m above the surface interpolated from Figures 3.3 and 3.4, respectively. These maps were derived using point data measurements at Micronet stations (see Colour Plate XIII). “chap03”—2004/1/20 — page 118 — #9 118 Humes et al. Solar radiation Little Washita Watershed Micronet station Solar radiation (W m –2 ) 550–575 575–600 600–625 625–650 650–675 675–700 700–725 725–750 750–775 775–800 800–825 825–850 850–875 875–900 August 18, 1994 1640 UTC 5,000 m5,000 0 Figure 3.4 Gridded field of incoming solar radiation measurements interpolated from approximately 1640 UTC. A universal kriging algorithm was used for spatial interpolation between the point measurements. The Micronet does not include measurements of windspeed, which are required for model calculations of aerodynamic resistance. To obtain a reasonable watershed-wide average value of windspeed, data from four Oklahoma Mesonet stations were used. The Oklahoma Mesonetwork is a state-wide monitoring network consisting of 112 stations that provide measurements of meteorological and surface variables at 5-min intervals. Four of the Mesonet stations are located just outside the boundaries of the watershed. Values of the windspeed (at 9 m above the surface) and relative humidity from these four stations were averaged to compute a watershed- wide average for these variables for the 5-min interval closest to the satellite overpass time. 3.3.3 Inputs derived from remotely sensed data Radiometric surface temperature One of the key inputs to the model is the radiometric surface temperature, in this case derived from TM Band 6 (bandpass 10.9–12.3 µm). Data from measurements at Micronet stations (see Colour Plate XIV). “chap03”—2004/1/20 — page 119 — #10 High spatial resolution mapping 119 the Landsat thermal band were corrected for atmospheric effects by running the radiative transfer code Lowtran 7 (Kniezys et al. 1988). Atmospheric temperature and water vapor profiles from on-site radiosonde data acquired by a team from the National Severe Storms Lab at the time of the satellite overpass was used as input to the radiative transfer model. The resulting corrections were applied on a pixel-by-pixel basis across the scene. The radiometric temperature of a large ground target area was measured at the time of the satellite overpass with instruments mounted on two back- pack type apparatuses. The satellite pixels that most closely corresponded to this large target area were extracted from the scene and compared with the ground-based temperature measurement. The TM-derived temperature was slightly higher than the ground-based temperature (approximately 1.5 ◦ C). The ground radiometric measurements and radiosonde measurements were made just adjacent to one another at a site near the center of the watershed. The map of surface temperature for the watershed is shown in Figure 3.5. The cool areas in the east-central portion of the image correspond to contam- ination by cirrus clouds, and the cool spots in the far southern and western portions of the watershed correspond to cumulus clouds and shadows. Corrected radiometric surface temperature Little Washita Watershed Temperature (°C) 30–30.5 30.6–31 31.1–31.5 31.6–32 32.1–32.5 32.6–33 33.1–33.5 33.6–34 34.1–34.5 34.6–35 35.1–35.5 35.6–36 36.1–36.5 36.6–37 37.1–37.5 37.6–38 38.1–38.5 5,000 m5,000 0 Figure 3.5 Atmospherically corrected radiometric surface temperature derived from a Landsat 5 TM scene acquired over the Little Washita Watershed on August 18, 1994. The dark areas in the east-central portion of the image corresponds to contamination by cirrus clouds, and the dark spots in the far southern and west- ern edges of the watershed correspond to contamination by cumulus clouds (see Colour Plate XV). [...]... G (mod) H (obs) H (mod) LE (obs) LE (mod) 1 2 3 4 RMS 582 528 527 6 03 566 541 550 591 46 82 77 74 80 93 91 148 132 100 90 14 83 49 44 34 404 34 6 36 0 515 402 39 9 415 409 16.57 41.68 43. 35 “chap 03 — 2004/1/20 — page 122 — # 13 65 .33 High spatial resolution mapping 1 23 this site) is computed as the residual in the energy balance for the soil, all the errors in the other three components affect the model... watershed, combined with minimal radiation loading that occurred that morning before the satellite data were acquired, had the effect of minimizing the spatial variability in some of the surface state variables that control surface fluxes, such as surface temperature It also gave rise to the rather unusual situation in which there was more variation in near -surface air temperature than in surface temperature... values are in units of W m−2 Land cover Pixel Median STD Median STD Median STD Median STD count Rnet Rnet G G H H LE LE Water Woodland Crops Pasture – dense Pasture – moderate Pasture – less dense Sparse or senescent Bare soil 571 8,590 6,814 3, 705 2,507 3, 8 93 1,497 8,648 585 579 555 579 576 5 73 5 73 579 28.6 96 29.9 75 27.2 84 25.7 84 23. 6 90 28.1 90 25.5 90 30 .9 144 95.7 44.9 42 .3 38 .3 38.9 39 .0 41.4... maps To further assist in the interpretation of the flux maps, the median flux values observed within each land cover type are shown in Figure 3. 14 The numerical data corresponding to these plotted medians, together with the standard deviation among all the pixels belonging to a particluar land cover type, are summarized in Table 3. 3 Both the flux maps and the data shown in Figure 3. 14 indicate that, overall,... (1990) Mapping surface energy balance components by combining Landsat Thematic mapper and ground-based meteorological data Remote Sens Environ 30 : 77–87 Noilhan, J and S Planton (1989) A simple parameterization of land surface processes for meteorological models Mon Weather Rev 117: 536 –49 Norman, J.M and G.S Campbell (19 83) Application of a plant–environment model to problems in irrigation In D.J Hillel... Colorado Entekhabi, D and P.S Eagleson (1989) Land surface hydrology parameterization for atmospheric general circulation models including sub-grid scale variability J Appl Meteorol 2: 817 31 Gillies, R.R and T.N Carlson (1995) Thermal remote sensing of surface soil water content with partial vegetation cover for incorporation into climate models J Appl Meteorol 34 : 745–56 Holwill, C.J and J.B Stewart (1992)... mapping 129 Table 3. 4 Correlation coefficients for each input model grid and output flux map Norman Model Input Grids (18 August 1994) Corrected radiometric surface temperature Land cover NDVI Solar radiation Air temperature Storm total precipitation Surface Energy Flux Map Rnet G H LE −0.106 −0.006 −0.066 −0.020 0.004 −0.065 0.577 0. 034 0.1 83 0.2 63 −0 .39 4 0.1 13 −0. 037 0.102 −0.157 0.222 −0.078 −0 .31 6... the watershed The rainfall totals were highest in the eastern third and extreme western edge of the watershed The texture classes for the surface soils over the watershed, as derived from the STATSGO database are shown in Figure 3. 13 The precipitation and soils data shown in Figures 3. 12 and 3. 13 were not used in the calculation of the fluxes; they are provided here to aid in the interpretation of the... 81% of these points were accurately classified The land cover map was derived at the original 30 -m pixel resolution for the TM reflective bands, then aggregated to 120-m resolution to match the resolution of the thermal band data The aggregation procedure assigned the land cover type that occupied the majority of the area of the 120-m pixels The resulting map is shown in Figure 3. 6 This land cover map... models 3. 5 Summary A relatively simple, “snapshot”-type model for computing components of the surface energy balance data was run on a pixel-by-pixel basis for the LWW in central Oklahoma The model uses ground and remotely sensed “chap 03 — 2004/1/20 — page 129 — #20 130 Humes et al data to calculate a separate energy balance for the vegetation and soil layers The remotely sensed data, in this case Landsat . radiometric surface temperature Little Washita Watershed Temperature (°C) 30 30 .5 30 .6 31 31 .1 31 .5 31 .6 32 32 .1 32 .5 32 .6 33 33 .1 33 .5 33 .6 34 34 .1 34 .5 34 .6 35 35 .1 35 .5 35 .6 36 36 .1 36 .5 36 .6 37 37 .1 37 .5 37 .6 38 38 .1 38 .5 5,000. (°C) 30 30 .4 30 .5 30 .9 31 31 .4 31 .5 31 .9 32 32 .4 32 .5 32 .9 33 33 .4 33 .5 33 .9 34 34 .4 34 .5 34 .9 35 35 .4 35 .5 35 .9 August 18, 1994 1640 UTC 5,000 m5,000 0 Figure 3. 3 Gridded field of air temperature 2 m above the surface interpolated. 5-min averaging interval closest in time to the satellite overpass time of Air temperature Little Washita Watershed Micronet station Air temperature (°C) 30 30 .4 30 .5 30 .9 31 31 .4 31 .5 31 .9 32 32 .4 32 .5 32 .9 33 33 .4 33 .5 33 .9 34 34 .4 34 .5 34 .9 35 35 .4 35 .5 35 .9 August