Thermal Remote Sensing in Land Surface Processes - Chapter 11 (end) doc

11.2 The MUST mission and related applications11.2.1 Applications The application of thermal infrared measurements from space are based on the relation existing between surface temperatu

MUST – a medium scale surface temperature mission dedicated

to environment and agriculture

Alain Vidal, Philippe Duthil, Catherine Ottlé, Vicente Caselles, Antonio Yagüe and

John Murtagh

11.1 Introduction

The Medium Scale Surface Temperature (MUST) study was carried out inthe framework of the European Commission (DG XII) fourth “Researchand Development Work Programme.” The objective of this study was thedefinition and demonstration of interest of a large swath, medium resolutionthermal infrared imager mission, named MUST More precisely speaking thespecific objectives were:

• to demonstrate the relevance and efficiency of the products of the MUSTmission in the relevant application fields and to assess the economicalbenefits of the mission;

• to further develop methodologies for retrieving thermal- and related surface parameters from the sensor data;

water-• to design a medium-resolution, large-swath thermal imager, that is,compact and affordable;

• to analyze the operational implementation of the ground segment

The study was co-ordinated by Matra Marconi Space (MMS) and theirpartners Cemagref (France), CNRS/CETP (France), the Universitat deValencia (Spain), INFOCARTO (Spain), and the NRSCL (UK) It includedthe whole Mission and System definition process, starting with the defini-tion of the user requirements, including the space and ground segments, thecost estimates, and ending with the evaluation of the MUST mission benefitsversus costs and the final recommendations on the potential continuation ofthe programme A development and implementation of the MUST sensorwas then proposed in the framework of the European Space Agency CoastalZone Earth Watch mission

11.2 The MUST mission and related applications

11.2.1 Applications

The application of thermal infrared measurements from space are based

on the relation existing between surface temperature and the soil and etation hydric state as introduced later They can be classified into threemain classes: (a) the assessment of the vegetation hydric state, importantfor applications such as agriculture (crop yield forecasts, potential stress due

veg-to drought, illness, or other pests), irrigation management, and forest firesrisks assessment; (b) the assessment of surface (soil and vegetation) evapo-transpiration, and thereby the evaluation of water consumption, useful forirrigation management and the evaluation of soil moisture that is helpful

in hydrology applications; (c) the assessment of surface temperature itself

or the air temperature as a by-product of surface temperature The relatedapplications are mapping frosts on agricultural surfaces or heat islands onurban surfaces In addition, the MUST thermal infrared data are expected

to be useful for the global monitoring of the biosphere and as a contribution

to the Global Circulation Models providing data on the water fluxes at theglobal scale The different fields of operational applications for the thermalinfrared data are listed in Table 11.1

11.2.2 The MUST information products

The MUST information products can be classified into three types, based onequation (11.1):

where Ts is the surface temperature measured by MUST and Ta the air

temperature This simple equation explains the double dependence of Tson:

(a) the climatic conditions, expressed through Ta; (b) the energy balance of

Table 11.1 Main land applications identified for a thermal imager

Domain Parameter of interest

and irrigation managementAreas of frost risks

Scientific biosphere global

monitoring, Global

Circulation Models

Complement to VEGETATION data: water fluxes,hydric state of vegetation and soils

the considered surface, where equilibrium is the difference between surfaceand air temperatures(Ts− Ta).

Product type 1: vegetation stress index product

Measured through Ts − Ta, this product mainly concerns crop yield mation in agriculture, irrigation monitoring, and risk assessment of forestfires The evaluation of vegetation stress is derived from the analysis of thesurface energy balance terms The energy balance is usually expressed withthe following equation:

evapotranspi-reaches its maximal (or potential) value LE = LEp(Moran et al 1994; Vidal

et al 1997) LEp depends on the atmospheric conditions (air temperatureand moisture) and on the plant characteristics (resistance to heat exchangewith air and resistance to evapotranspiration)

The ratio of actual LE to LEp(LE/LEp) provides a precise assessment of

the vegetation stress, which is minimal when LE /LEp = 1, and maximal

when LE /LEp = 0 Several indices have been developed to estimate this

ratio, LE /LEp, using remote sensing measurements The more classical onesare based on the CWSI (Crop Water stress Index) approach where (Jackson

been derived by many authors (Jackson et al 1977; Seguin and Itier 1983;

Vidal and Perrier 1988) from the surface energy balance for estimatingthe daily evapotranspiration from an instantaneous midday remote sensing

measurement of Ts− Ta:

where LEdand Rndare the daily evapotranspiration and net radiation, A and

B are constants depending on the canopy, and Ts− Tais the instantaneousdifference between surface and air temperatures measured near midday

Product type 3: interpolated air temperature,Ta

This is derived by correlating surface and air temperature, assuming airtemperature to be known at some meteorological station point Some of theprimary applications include frosts prediction and detection of urban heatislands A strong correlation is found between surface and air temperatures,when low air temperatures occur, which are the usual conditions when frostsmaps or urban heat island maps, are required

11.2.3 Methodology followed for assessing the user

requirements and benefits

The User Requirements phase has been a major step in the definition of theMUST Mission and System, as no structured user community exists Thescientific community has not necessarily evaluated all the issues related toend-user requirements for information products using land surface temper-ature The user requirements and benefit assessments have therefore beenestablished with three National user groups in United Kingdom, Spain, andThe user groups were involved in two main steps of the process First, theyexpressed their requirements in terms of products and services Second, afterthe products had been simulated, they indicated more precisely their interestfor the products This provided an assessment of the benefits derived fromMUST products by the user community

11.2.4 The information products’ requirements

and simulations

The main applications in agriculture, water resources, and forest fires will bepresented henceforth In all the cases, MUST surface temperatures were sim-ulated from Landsat TM thermal IR data (120-m resolution) Since 250-mresolution was envisaged for MUST, Landsat thermal data were resampled

at 250-m resolution using bicubic convolution The maps presented in thischapter derived from such resampled thermal IR data

France (Table 11.2)

Table 11.2 Composition of user groups in the three partner countries of the MUST

project

Institute (remote sensing department), CerealsTrader, Sugar Beet Technical Institute

Provider in Forest Fire mitigation

Producers Association

research, Farming online,Value added company

MUST information products for agriculture

INPUTS TO YIELD PREDICTION MODELS

Users described that yield prediction models do not sufficiently take intoaccount the actual vegetation stress In this field, remote sensing is alreadyused (e.g by the EU MARS project), but it primarily involves the estimation

of biomass using reflected solar wavelengths Following the present cies in the use of EO data for yield prediction, it was suggested to use MUSTdata as a direct input in “efficiency” models, for example, the Monteithmodel (Monteith 1972), or the 3M “Modified Monteith Model” recently

tenden-developed by the MARS project with Cemagref (Laguette et al 1995, 1997).

In these models, the dry matter (DM) is estimated as a cumulative product ofefficiencies and global radiation(R g ), then transformed into crop yield using

harvest indexes (HI) In this case, a MUST-derived water requirement faction index SI can be used in the expression of the conversion efficiency,which is usually considered as a constant:

nor-of CWSI, εs is the climatic efficiency, εi0 is the interception efficiency formaximal NDVI, and εc0 is the conversion efficiency for maximal SI The

product of Rgwith efficiencies is integrated from the beginning of the ping season to the date of the cycle where yield is estimated/predicted.The aforementioned authors have shown that, when the “3M model” isused with a continuous series of NOAA-AVHRR images, the final yield

crop-of wheat can be retrieved with a precision crop-of 1.2 tons ha−1 instead of2.4 tons ha−1 obtained when not accounting for water stress effects onyield

SIMULATED PRODUCTS

The 3M model was applied on maize fields in the Orthez region (South West

of France) Yield prediction figures obtained with remote sensing data have

been compared to actual yield figures derived from in situ measurements

in sample plots The ideal process would have been to acquire remotelysensed data along the whole crop season with a sampling interval of typically

10 days and integrate them Unfortunately, this was not possible becauseLandsat TM images were available in cloud-free conditions on a single date(20 July, 1996) Consequently, it was decided to compare this single dateremote sensing result (which is actually the DM accumulation derivative)

with the in-situ DM variation measurement averaged on the period around

the available date

The results, sketched in Figure 11.1, are not conclusive on the capability

of IR-derived water stress information to improve the crop DM and yieldprediction Since this result is not coherent with the aforementioned MARSproject research results, it is believed that it is a consequence of the single-dateavailable acquisition

Figure 11.1 Comparison of the daily dry matter (DM) production estimated from

one-date MUST-simulated thermal IR data with the ground measured final DMproduction on maize (Orthez – France)

MUST information products for irrigation and water resources

The users involved in irrigation, from both agricultural and water agement points-of-view, identified three information products In order ofpriority, these are: the spatial distribution of water consumption (derivedfrom the evapotranspiration LE), maps of irrigated surfaces, and maps ofcrop water stress for monitoring water application and irrigation scheduling.The users involved in water quality management (the domestic water dis-tribution companies) were interested in soil moisture maps at the scale ofsmall to medium watershed area This information provides the means foridentifying and assessing the importance of water contributing areas, as inputfor water quality models They were also considering the crop water con-

man-sumption (LE estimation) to derive infiltration/runoff as input for water

quality models

SIMULATED PRODUCTS

The objective of the simulations was mostly to show the users tially distributed evapotranspiration information at 250-m resolution todemonstrate its advantage in comparison to sampled information and to1-km resolution information The simulated products are therefore dailyevapotranspiration maps on the sites of Orthez (France) (Figure 11.2), the

spa-LE < 3 mm day–1

Maize area Rivers

Figure 11.2 Daily evapotranspiration map obtained from MUST-simulated thermal

Orgeval river basin (France, part of the Seine river basin), and of Barrax(Albacete–La Mancha–Spain), using the approach in equation (11.4).

Forest fires

Fire-fighting authorities have been using short-term fire risk indexes for along time These indexes are usually based on actual and predicted mete-orological parameters, such as wind speed, air moisture, and temperature.Vegetation stress is usually represented by a simple budget between rainfalland potential evapotranspiration, which is difficult to transpose to forestareas, mainly due to spatial variations in the terms of this budget, and onhow this budget is exploited by soil and tree root zones It has recently beenshown that using surface temperature measurements to derive the vegetationstress improved the fire risk prediction on both a short-term (daily forecast)

and mid-term (weekly–monthly) range (Vidal et al 1994; Vidal and

Devaux-Ros 1995) Based on this rationale and on the operational way to fight fires

in Corsica, two types of requirements were expressed by the fire fightingusers:

• a real-time, daily-risk index integrating climatic and vegetation stress, atthe scale of large forested areas (typically larger that 50,000 ha) usefulfor a better positioning of the fire fighting teams put in alert duringsummer months;

• a weekly risk index at a more local scale, usually for areas rangingfrom 5,000 to 20,000 ha, needed in order to support decisions onconcentrating or moving means (staff and material) of fire watch patrols

In addition, the forests officials were interested in two types of products:

• long-term risk maps on usually stressed areas to be used for theestablishment of risk prevention plans at a 1/50,000 scale;

• fire damage maps: the thermal infrared data to be used in combination

with visible, near-infrared (NIR), and short wave infrared (SWIR) dataare expected to significantly enhance the accuracy of the damage mapsestablished with visible, NIR, SWIR data only

SIMULATED PRODUCTS

every day or 2–3 days In the case of Corsica, an extension of CWSI (seeequation 11.3) to sparse vegetation, called Water Deficit Index (WDI), has

been used This index, introduced by Moran et al (1994) and applied to

forests by Vidal and Devaux-Ros (1995), is based on the representationThe different types of products have been simulated for Corsica (Figure 11.4)and Spain (Figure 11.5), assuming that MUST would enable an observation

4: Dry bare soil 3: Saturated bare soil

2: Water-stressed vegetation 1: Well-watered vegetation

1 0.8 0.6 0.4 0.2 0

Ts – Ta (°C)

Figure 11.3 The theoretical trapezoidal shape showing the different biomass versus water

stress conditions of the canopy–soil continuum (from Moran et al 1994) The

WDI of point C is given by AC/AB as shown in equation (11.6)

0.2–0.4 0.4–0.5

Haute-Corse, June 1993

250-m resolution WDI map

derived from Landsat TM

WDI

< 0.2

0.5–0.6 0.6–0.7

Vegetation units

Median

Weekly fire danger map

Re-location of mitigation means

0 5 10 15 20 km

Figure 11.4 Daily and weekly fire risk index on the right part are the results of sub-sampling

a full scale risk index obtained from MUST-simulated thermal data (on the

of the soil-canopy continuum conditions in a fractional vegetation coverversus the difference between surface and air temperature(Ts−Ta) diagram.

Actually, its position is theoretically comprised within a trapezoidal pattern:Figure 11.3 presents such a pattern and the definition of its limits

TM 3 45 color composite TM 6 45 color composite

Burnt area

Figure 11.5 Classification of fire damaged areas using different bands of a LandsatTM image.

Respectively, red, NIR, SWIR (on the left), and thermal infrared, NIR and SWIR

These authors have proposed both a theoretical and a graphic simpleestimation of the soil–canopy evaporation for a given fractional vegetation

cover, knowing its potential evaporation LEp:

LE

LEp =(T (Ts− Ta) − (Ts− Ta)dry

s− Ta)wet− (Ts− Ta)dry =BC

where Tsis the composite surface temperature of the soil–canopy continuum

as estimated from thermal infrared measurements, BC and AB are the

dis-to the left and right limits of the trapezoid

The main interest of this approach is the possibility of estimating both

Ts− Taand fractional vegetation cover from remote sensing measurements

In the WDI approach, both NDVI and Soil Adjusted Vegetation Index (SAVI)have been used to estimate fractional vegetation cover:

NDVI= ρ ρNIR− ρR

SAVI= (ρNIR− ρR)/(ρNIR+ ρR+ L)(1 + L) (11.8)whereρNIRandρRare the reflectances in the sensor’s near-infrared and red

wavebands, and L is a unitless constant assumed to be 0.5 for a wide variety

of leaf area index values (Huete 1988)

11.2.5 System requirements derived from user requirements

From the above step of identification of MUST applications and informationduring each user meeting and after national interviews

tances represented inFigure 11.3, and the wet and dry indices correspond

products, a synthetic table (Table 11.3) was prepared and validated by users

Table 11.3 Synthesis of user requirements

Agriculture Irrigation Frosts heat

Air temperature maps

Water stress maps

Soil moisture evapo- transpiration map Frequency Daily to

Delivery 1–2 days Week Month Real-time Real-time

MUST frequency and overpass

A 1-day revisit was requested by most but actually only some applicationsrequire a real daily revisit (frosts, heat islands, forest fires) and this can beachieved quite easily since these applications concern atmospheric conditions

or areas where the cloud coverage is low to null The overpass time might

be a critical issue Most applications request data within 2 h before or afterthe time of maximal surface temperature (between 12.30 and 13.00 localsolar time) This would be satisfied by a late morning overpass However,frost monitoring requires night-time acquisition, which should preferablycorrespond to a late night overpass

MUST spatial resolution

This critical issue was finally solved with 250-m resolution, thought to besufficient for most applications However, 100-m resolution was recom-mended only for irrigation For this specific area, the expected informationproducts are compatible with 250-m resolution, but might be perturbed by

a succession of irrigated and non-irrigated fields

MUST temperature precision

This issue was the most discussed A simple look at bibliography and sical figures issued from other thermal IR missions (Landsat TM, AVHRR,ASTER, PRISM) actually showed that users required a 1 K precision whereasground calibration cannot achieve a precision better than 1 K Therefore,

clas-to be as user-oriented as possible, our assessment of user requirements

on MUST temperature precision was based on the analysis of information

products precision, as it is easy to figure that the final temperature precisionstrongly depends on the expected information product.

APPLICATIONS BASED ON VEGETATION STRESS ESTIMATION

For all these applications, the users require that data from instrument becapable of discerning five stress classes for stress indexes ranging from 0 to 1.This means that a 10–20% precision on the stress index is fully acceptable,which, as shown in equation (11.3), corresponds to a 10–20% precision on

the Ts min− Ts maxrange The expected precision on the surface temperaturemeasurement depends on meteorological conditions and was simulated for

wheat (Figure 11.6) and forest (Figure 11.7) as follows The Ts min− Ts max

range mostly depends on canopy aerodynamic resistance ra, air

tempera-ture Ta, global radiation Rg, and relative air moisture H% To simplify the simulation conditions, Tawas supposed to be related with Rg, and(Ta, Rg)

couples were combined with different values of H% to derive values of Ts

precisions corresponding to 20% of Ts min–Ts max

These simulations show that for crop stress assessment a precision of1–2 K is always acceptable, even in cold and humid conditions typical innorthern Europe Furthermore, a precision of 2–3 K is still acceptable forconditions with higher temperature and radiation values (e.g for almost allirrigated regions) For forests on the opposite, simulations show that a pre-cision of 0.5–1 K may be required for certain conditions (lower temperatureand radiation values), which are rarely met during forest fire periods.APPLICATIONS BASED ON SURFACE FLUX ESTIMATION

For these applications, two precision levels were requested “Irrigation”users requested a precision of better than 10% on the estimation of thedaily evapotranspiration, essentially to improve existing estimations “Waterquality” users required a precision of about 50%, but insisted on having

a good description of the spatial distribution of such a typical three-classinformation to improve their hydrological models If we consider the mostconstraining requirement and equation (11.4) with a maximal value of

parameter B of 0.6 mm K−1 (Brasa et al 1996), the required precision on

the surface temperature can be estimated through:δLE(mm) ≈ 0.6δTs, with

an expectedδLE ranging from 0.3 (temperate regions) to 0.6 mm

(Mediter-ranean and tropical regions), which yields aδTsof 0.5–1.0 K for the worstcase, that is, a final precision of 0.5–1.5 K

APPLICATIONS BASED ON AIR TEMPERATURE INTERPOLATION

For these applications, interpolation of air temperature restricts the

require-ment to a relative error to be added to the precision obtained on Taground-measurements, usually considered to be about 0.5 K The best pre-cision was required for frost monitoring on orchards, as a difference of0.5–1.0 K may be very important for certain fruit varieties

11.3 The MUST system derived main

characteristics

11.3.1 Main mission characteristics

The MUST instrument is to be accommodated on a low earth orbit spaceplatform The most likely candidate is an Earth Watch satellite of the Euro-pean Space Agency to be launched around 2004 (Coastal Zone Earth Watch)