The Earth Observer, September/October, 1995

Workshop on Thermal Remote Sensing of the Energy and Water Balance Over Vegetation in Conjuntion with Other Sensors

[Report of NASA-EOS, CETP, CEMAGREF, CNES, CNRS, USDA-ARS, ESA-sponsored Workshop held September 20-24, 1993 at la Londe les Maures, France.]

--Robert R. Gillies (gillies@essc.psu.edu), Earth System Science Center, The Pennsylvania State University, PA, 16802-5094 USA
--Alain Vidal, CEMAGREF-ENGREF, Remote Sensing Laboratory, France
--Odile Taconet, CEPT, 78140 Velizy, France
--Toby N. Carlson, Department of Meteorology, The Pennsylvania State University, PA 16802-5094, USA

The Workshop on Thermal Remote Sensing of the Energy and Water Balance Over Vegetation in Conjunction with Other Sensors was created by a small community of international scientists who felt that the potential of thermal remote sensing of the surface energy balance for studying land surface climate had not been fully realized by the larger scientific community. Various international field programs (FIFE, 1987, 1989; HAPEX-MOBILHY, 1986; MAC-HYDRO '90, MONSOON '90, 1990; HAPEX-SAHEL, 1992) (a list of acronyms appears at the end of this report) have been conducted since 1986 expressly to study the issue of scaling up from point- to regional-scale estimates of the surface energy fluxes over heterogeneous terrain. Still, there have been conflicting results, e.g., Hall et al. (1992), on the utility of thermal infrared to satisfactorily resolve important land surface parameters. In light of this, the principal goal of the workshop was to appraise the current state of the science with consideration not just given to thermal infrared measurement but also to consider and incorporate other spectral, e.g., surface reflectivity, and sensor, e.g., radar, data as a means for addressing some of the challenges posed in the area of climate change. Further complexities of terrain, which might consist of a mixture of vegetation (including forests), sloping surfaces, water bodies, bare soil, and urban landscapes were also incorporated into the workshop theme.

This report condenses the contents of the workshop proceedings*. It broadly identifies a set of critical problems and/or issues in the field of thermal infrared remote sensing, i.e., electromagnetic measurements within the 8-14 ľm window region. It also addresses the use of additional measurements coupled with thermal remote measurements--the so-called synergistic or multispectral approach. We present such topics within the context of the future programs of Mission to Planet Earth/EOS, CNES (French Space Agency), and the European Space Agency (ESA). These agencies intend to employ many new sensors aboard their next-generation satellites.

The accuracy of the measurements recorded at the satellite sensor (radiometer) is determined by several factors. These are a function of instrument design and the physics which relates the measured quantity (radiance) at the sensor to that at the surface. In general, accurate estimates of surface radiant temperature and reflectance involve calibrating the data into physical units and then correcting the data to account for a host of other effects which are due to the physical properties of the surface and the atmosphere. In addition, other geometric properties of the system, e.g., scan angle of the satellite, have to be considered. Some of the more important of these factors are listed here along with the corresponding workshop recommendations as appropriate.

The fundamental observable in the field of remote sensing is radiance; so sensor calibration is crucial. All attendees at the workshop were in agreement that some type of onboard sensor calibration was ideal. Validation work with in situ measurements is being carried out, but these validations are always dependent on atmospheric corrections (see below) since sensor calibration and atmospheric correction cannot be deconvolved in comparisons between ground-based measurements and satellite measurements. Such inter-dependencies mean that it is particularly difficult to make a definitive statement about the overall accuracy of the measurements.

Surface emissivity can vary due to many factors. Moreover, the emissivity in the 8-14 ľm window region may differ significantly from the emissivity within smaller spectral bands. In general, the emissivity of land surfaces in the 10-12.5 ľm band is slowly varying with wavelength and lies between 0.94-0.98. A reasonable estimate of emissivity for vegetation-soil systems with a leaf area index greater than 0.5 is 0.97 ą0.03; thus, associated uncertainties in temperature measurements are usually less than 1-2 degree C.

Recommendation: Estimates of emissivity are not a priority when considering vegetation; however, care must be taken when considering partial vegetation cover and variations in emissivity with radiation wavelength.

Model results in conjunction with measurements indicate that angular effects on the radiant temperature and the reflectivity measurements are certainly significant at the local scale and depend on such factors as the water stress level of the vegetation and the canopy structure. Angular effects seem to be insignificant over dense well-watered vegetation and over many plants such as alfalfa. Over bare soil, e.g., plowed fields, the row effects are noticeable in small-scale measurements. It is not clear at what point the effects related to sensor view angle are manifested in the data or if they are important on the scale of the AVHRR, which has viewing angles in excess of 25 degrees from nadir. It is still not known if, or how, one can use local results to correct satellite data or to what extent the correction is related to other parameters such as emissivity. It may be useful to obtain measurements at different angles (as proposed for next-generation sensors such as AATSR, IRSUTE, and the middle-spatial-resolution optical scanner on PRIRODA). Despite these questions, it was generally felt that multi-angle measurements would improve present methods. Acquisition of measurements at two or three angles would allow some sort of extrapolation to nadir, which could then represent an a priori standard value.

Much of the work reported at the workshop showed generally positive results in comparing satellite-based and ground-based radiant temperatures. Schemes for computing and testing atmospheric corrections depend to some extent on the sensor being used. For sensors with multiple bands, various split-window approaches are generally practicable. Applied initially for sea-surface temperature correction where the split-window coefficients are widely applicable, there is evidence to suggest that local calibration of these coefficients is necessary over land surfaces if desirable precision is to be obtained. For sensors with only one thermal band, one-channel procedures--those which use radiosonde data and a radiative transfer model--are required. In this case, a 1 degree C accuracy is possible with appropriate local measurements of atmospheric structure. Another approach, proposed for Landsat TM thermal infrared data, performs an atmospheric correction based upon the surface energy balance, and avoids the requirement for atmospheric or ground-based measurements. For proposed and new sensors with multiple view angle capabilities, e.g., AATSR, techniques involve a combination of bands and view angles to obtain the atmospheric correction. All methods of atmospheric correction for thermal infrared data, however, require knowledge of the surface emissivity and are also a function of viewing angle and the radiometer.

In general, the measurement error for the surface radiant temperature, considering only those errors associated with instrument calibration, emissivity, angular effects, and atmospheric correction, tends to lie between 2 degree and 3 degree C, at best. This raised the question as to whether it would be productive to make corrections of this magnitude if there are multiple sources of error with the same magnitude. For example, the magnitudes of errors in conventional meteorological radiosonde data are also between 1 degree and 2 degree C, and so we should not expect to determine radiant temperatures at the surface with an accuracy of much better than 1 degree C. Corrections which account for errors less than this magnitude are probably not worth considering. Errors of this magnitude are likely to be less serious over dry, bare soil than over unstressed vegetation, where the differences between surface and air temperature are of this magnitude. In addition, limitations in the accuracy of measurements due to the digital nature of instruments employed should be considered.

One of the challenges in thermal infrared remote sensing is to apply such satellite observations to yield land surface parameters. Such data are useful in broader scientific studies associated with the surface energy balance and are important in climate, meteorological, hydrological, ecological, and environmental studies as required input to such modeling, as verification studies, or simply as an improved description of the Earth's surface. Central to this though are some critical issues regarding the capability of thermal remote sensing to infer the surface energy balance to acceptable accuracies.

There has been substantial progress over the last 15 years in the use of thermal infrared measurements to infer surface energy fluxes (sensible and latent). Many problems with current techniques are related to an inconsistency in the science, i.e., the fact that the aerodynamic temperature (which is used in the formula for computing sensible heat flux) is defined differently from the radiant temperature, which is the measurement from the satellite. Various papers indicated that the inclusion of an additional term (cited in the literature as kB-1) greatly improved the accuracy of the surface energy flux estimates. The kB-1 factor is a useful parameter which not only accounts for the difference between aerodynamic and radiant temperatures, but also a combination of other effects--the vertical distribution of thermal radiation fluxes within the vegetation, the angular effect on radiative temperature measurements, energy exchanges between different surfaces (bare soil and vegetation in partial canopies), and perhaps even the effect of surface roughness on the transition layer at the top of the plant canopy. All these factors change with canopy structure and in particular with the leaf area index or the fractional vegetation cover and also with air-surface temperature differences or as a function of scale or viewing angle.

The parameter kB-1 is well defined for smooth surfaces. It is, however, more difficult to define for more-complex surfaces such as vegetation canopies and partially vegetated areas. Nevertheless, it is possible to obtain useful bulk estimates of kB-1 by comparison with surface field measurements or with a detailed energy balance model; however, such estimates tend to be restricted to the type of surface conditions for which they were obtained. In general, a value of kB-1 set at "seven" seems to suit a wide variety of applications over partial vegetation cover, whereas others suggest values closer to "three" for full cover.

Recommendation: The present empirical approach must be reassessed. A more-rigorous analytical approach must be developed for complex surfaces with partial vegetation cover. It was recommended that the FIFE results be re-examined in light of new ideas on the use of kB-1.

Accuracy is generally good for high evapotranspiration since the sensible heat flux can be found within 20% accuracy with a model and an infrared surface temperature measurement. Note that this estimate of error pertains to ground level, without including the inherent errors associated with making atmospheric corrections, etc.

The perceived relative accuracy in measuring the correct surface radiant temperature at the regional and local scales is a maximum of 2-3 degree C when all uncertainties are factored in (1 degree C gives 10-20% accuracy in the fluxes). Errors in sensible heat flux are likely to be larger in dry conditions but are more important in wet conditions; the reverse is true with regard to evapotranspiration. Anticipated errors of 20% in surface flux estimates are more optimistic than those published by Hall et al. (1992). The accuracy of the fluxes depends not only upon the accuracy of the measurement of the surface radiant temperature but also upon other parameters and there are also issues of magnitude to be considered. A target of better than 30% accuracy in the fluxes is certainly realistic and would be considered acceptable given the accuracy of flux measurements is generally 20%. Thus, errors in estimates of kB-1 up to 100% may be acceptable to maintain the aforementioned level of accuracy in the surface fluxes.

References were made to the need to validate the estimated surface fluxes with field data. This should be done by using existing data sets such as those from FIFE, 1987, 1989; HAPEX-MOBILHY, 1986 ; MAC-HYDRO '90, MONSOON '90, 1990; and Washita '92.

Another primary topic that was raised continually was the spatial and temporal scaling of surface radiant temperature measurements and resultant fluxes. This, however, was never satisfactorily resolved, and it was suggested that larger problems lie in understanding the changes in parameters, variables, and functional relationships at the intermediate scale (10 m up to 1 km). This is a particularly relevant question pertaining to the treatment of current and future satellite data.

The complexity that scale introduces lies in the difficulty in defining prerequisite surface parameters, e.g., roughness, air temperature, emissivity, etc., other than the surface radiant temperature, with precision or uniqueness with increasing scale. As a result, it seems that methods applicable to deriving fluxes at the local scale may not be wholly appropriate at the larger scale. Some results presented at the workshop did, however, suggest that for some applications the microscale variability seems to be integrated into the macroscale.

Recommendation: In general, more work is required to understand and develop procedures for scaling up from the canopy scale to regional scales. There is a need to utilize data from large-scale experiments to evaluate new large-scale parameters and variables as well as relationships between them.

It seems feasible to use mid-morning temperatures (corresponding to the overpass of Landsat-TM) rather than the later NOAA-AVHRR overpass (~1400 LST) temperature, without losing much information. However, it may be important to consider and investigate the influence of the onset time for water stress in vegetation. This tends to be limited to the period from roughly 1000 to 1400 hrs.

New sensors will have significantly greater spatial resolution. IRSUTE will have a surface resolution of 50 m; ASTER-90 m ; AATSR/ERS-1 (launched in 1991)-1 km; MODIS-250 m, 500 m, and 1 km; MSG (Meteosat Second Generation)-3 km; and ESA-PRISM-30-50 m. Some of these sensors will also have look-back capability, i.e., viewing at 2 angles, which is an essential addition if improvements in existing techniques are to be made.

It also seems possible to use the relationship between a vegetation index and surface (canopy) radiant temperature to deduce a regional air temperature. This will tend to be very close to the surface radiant temperature in densely vegetated regions, provided that there is no contamination of the pixel by clouds or standing water.

Research has shown the existence of a strong mid-day depression in surface evapotranspiration over stressed vegetation, suggesting that the differences between vegetation temperature and air temperature may be large for a period of time around mid-day during periods of plant water stress. It appears that differences between air and surface temperature can be up to 5 degree C, which corresponds to theoretical values. Studies suggest, however, that such large differences are not observed in remotely sensed aircraft or satellite measurements.

A water deficit index was presented which extends the crop water stress index to partial canopies.

A new idea brought up independently by several participants in one form or another was that by placing constraints on remotely sensed measurements it is possible to reduce the number of unknown factors, thereby improving the accuracy of estimated parameters. It was suggested that by analyzing patterns of multispectral plots of vegetation indices versus surface radiant temperature, limits can be placed on the data with the aid of physical reasoning and/or numerical models. An example of this constraining technique was demonstrated in the so-called "triangle" (or trapezoid) methods based on the observed (theoretical) plots of vegetation indices versus surface radiant temperature.

Synergy between microwave and optical data would be different for high-resolution, watershed-scale, process-based studies, typically using data with a spatial resolution on the order of 100 meters or better, than the approaches for synergy of these data types using sensors with coarse spatial resolution, e.g., SSM/I and AVHRR.

In this context (taking advantage of different sensors), it is possible to augment measurements in the thermal infrared and gain alternative or supplementary information. For example, Landsat-TM thermal data coupled with ERS-1 radar data can be used to assess, respectively, water stress and small branch water content. Taken together, these factors constitute a reasonable index for the so-called forest "flammability"--the implied susceptibility of forests to fire.

Recommendations: New indices--one should be aware of new indices related to surface conditions. Multi-source approaches--new approaches should involve a combination of different information sources, i.e., multi-spectral, multi-domain and multi-viewing angle sources. The synergy of microwave and optical data should be further explored using available measurements, e.g., FIFE, MONSOON '90, HAPEX and global-scale satellite measurements. It was also recommended that the scientific community study and address the possible problems arising from combining data from different sensors.

A brief attempt was made to classify uses of thermal remote sensing over vegetation, listing the requirements for each application and outlining a hierarchy of models (Carlson et al., 1995) used to obtain the results, i.e., land surface parameters or indices.

In general, all the models can be used to simulate fluxes (hourly or daily) and other variables concerning vegetation, ecosystem, or atmospheric behavior. In addition, some models can simulate thermal infrared temperature and other remote sensing variables--optical reflectance, microwave brightness temperature, and radar backscatter reflectance.

Uses and applications of these models to analyze surface radiant temperature (surface energy fluxes) were presented with regard to equating surface thermal infrared temperatures to surface aerodynamic temperatures (determination of kB-1), simulation of thermal infrared and microwave brightness temperatures, or surface soil water content. Very little emphasis was placed on combining passive microwave measurements with thermal infrared temperatures and vegetation indices. In fact, it was pointed out that any comparison or integration between the two does not seem to have been attempted.

The simulation models can also be used for inverse and assimilation procedures and this topic was discussed extensively. These models may be used to obtain leaf area index and evapotranspiration using measurements of vegetation indices and surface radiant temperatures, but they can also be used as components in crop simulation or hydrological models. Assimilation methods, in which derived parameters are introduced into atmospheric prediction, crop, or hydrological models, have the potential to contribute significantly to these longer-term models.

Recommendations: Satellite data should be assimilated in regional-scale hydrological, meteorological, ecological, environmental, crop, weather forecast models, and climate models as this is the ultimate goal of obtaining the satellite-derived quantities. Such quantities may not be obtainable on an operational basis, but they can be used to update such models as the opportunity arises. Perspectives should be identified precisely for different types of applications, such as land surface climatology, agriculture, and landscape ecology.

By the end of the workshop, virtually all of the ideas on current and future scientific methodologies that permit conversion of remotely sensed measurements to useful land surface parameters, as well as existing problems and achievements, had been touched upon and discussed in some depth. In general, a strategy that includes thermal infrared temperature measurements in a multispectral/synergistic approach seems to be desirable, if not essential, in future remote sensing endeavors over the land surface. Such approaches furnish additional degrees of freedom which not only improve the accuracy of useful geophysical parameters but also improve our understanding of the underlying processes which control the partitioning of the energy balance at the Earth's surface.

Hall, F.G., K.F. Huemmrich, S.J. Goetz, P.J. Sellers, and J.E. Nickerson, 1992: Satellite remote sensing of the surface energy balance: Success, failures, and unresolved issues in FIFE. J. Geophys Res., 97 D17; 19,061-19089.

Carlson, T.N., O. Taconet, A. Vidal, R.R. Gillies, A. Olioso, and K. Humes, 1995: An overview of the workshop on thermal remote sensing held at la Londe les Maures, France, September 20-24, 1993. Remote Sensing Reviews., 12, 147-158.