A simple interpretation of the surface temperature/vegetation index space
for asssment of surface moisture status
Inge Sandholt a,*,Kjeld Rasmusn a ,Jens Andern b
a
Institute of Geography,University of Copenhagen,Ostervoldgade 10,1350Copenhagen,Denmark
b
Department of Hydrodynamics and Water Resources,Technical University of Denmark,2800Lyngby,Denmark
Received 9August 1999;received in revid form 6November 2000;accepted 19February 2001
Abstract
A simplified land surface dryness index (Temperature–Vegetation Dryness Index,TVDI)bad on an empirical parameterisation of the relationship between surface temperature (T s )and vegetation index
(NDVI)is suggested.The index is related to soil moisture and,in comparison to existing interpretations of the T s /NDVI space,the index is conceptually and computationally straightforward.It is bad on satellite derived information only,and the potential for operational application of the index is therefore large.The spatial pattern and temporal evolution in TVDI has been analyd using 37NOAA-A VHRR images from 1990covering part of the Ferlo region of northern,miarid Senegal in West Africa.The spatial pattern in TVDI has been compared with simulations of soil moisture from a distributed hydrological model bad on the MIKE SHE code.The spatial variation in TVDI reflects the variation in moisture on a finer scale than can be derived from the hydrological model in this ca.D 2002Elvier Science Inc.All rights rerved.
1.Introduction
Hydrological modelling of large watersheds requires input of data for a substantial number of variables at short time intervals.Proper description of large and heterogen-eous watersheds requires that models are ‘‘distributed,’’i.e.,that the model takes into account the spatial distribution of each of the variables.Provision of data for such distributed models is particularly problematic in developing countries with little infrastructure and few resources for continuous monitoring of environmental variables.Where watersheds transcend national boundaries,the problems become even more pron
ounced.It is obvious that u of Earth Obr-vation (EO)data is potentially of great interest in such contexts.Widely ud distributed hydrological models,such as MIKE SHE (Refsgaard &Storm,1995),have not been designed to u EO data as inputs,and methods for extracting hydrological information from EO data have not been developed with this particular u in mind.Thus,there is a need to develop or modify methods and algorithms for determining key parameters in distributed hydrological
models from EO data,respecting the requirements on spatial and temporal resolution.
One of the EO data sources of greatest relevance at the prent time is NOAA-A VHRR.It produces daily coverage with a spatial resolution corresponding well to what is required by a distributed hydrological model of a large watershed,such as the Senegal River basin studied here.Its five bands in the visible,near,shortwave,and thermal infrared parts of the electromagnetic spectrum carry information on at least three key variables in hydrological models:vegetation cover and leaf area index (LAI),albedo and evapotranspiration,or soil humidity,the later of which are both related to the obrvable radiative surface temper-ature.Other prently available satellite nsor systems,including GOES/Meteosat and SSM/I,are uful as ,for estimating rainfall (e.g.,Zeng,1999).New systems,such as SPOT Vegetation,MODIS and MERIS,and —not the least —
Meteosat Second Generation,add further to the potential of EO for distributed hydrological mapping.
The prent paper aims at demonstrating how NOAA-A VHRR and other similar data may be ud to estimate temporal and spatial patterns of soil moisture,a key variable in distributed hydrological models.The basic approach is to interpret the so-called T s /NDVI space in terms of surface soil moisture status.The results are compared with soil
0034-4257/01/$–e front matter D 2002Elvier Science Inc.All rights rerved.PII:S 0034-4257(01)00274-7
*Corresponding author.Fax:+45-3-532-2501.E-mail address :is@geogr.ku.dk (I.Sandholt).
/locate/r
Remote Sensing of Environment 79(2002)213–
224
moisture derived from a hydrological model.The study area is the Senegale part of the miarid Sahel.dsds
2.Background
kuwThe potential of obtaining information about the energy and water status of a surface or for classification of land cover through the relation between remotely nd surface temperature(T s)and vegetation ,the normalid difference index,NDVI)has been investigated by veral ,Carlson,Gillies,&Perry,1994;Clarke,1997; Gillies,Carlson,Gui,Kustas,&Humes,1997;Goetz,1997; Lambin&Ehrlich,1996;Moran,Clarke,Inoue,&Vidal, 1994;Nemani,Pierce,Running,&Goward,1993;Nemani &Running,1997,1989;Smith&Choudhury,1991).The complementary information in the thermal and the visible/ near infrared wavelengths has proven to be well suited to monitoring vegetation status and stress,specifically in relation to water stress.NDVI is a rather conrvative indicator of water stress,becau vegetation remains green after initial water stress.In contrast,the surface temperature can ri rapidly with water ,Goetz,1997).The amount of vegetation prent is an important factor,how-ever,in surface temperature estimation.
abr
T s and NDVI in combination can provide information on vegetation and moisture conditions at the surface. Several studies focus on the slope of the T s/NDVI curve for this ,Friedl&Davis,1994;Nemani& Running,1989;Smith&Choudhury,1991).The T s/NDVI slope is related to the evapotranspiration rate of the surface,and has been ud to estimate air temperature (Boegh,Soegaard,Hanan,Kabat,&Lesch,1998;Pri-hodko&Goward,1997).Analysis of the T s/NDVI slope has also been ud to asss information related to areal averaged soil moisture conditions(Goetz,1997;Goward, Xue,&Czajkowski,2001),Nemani and Running(1989) related the slope of the T s/NDVI relationship to the stomatal resistance and the evapotranspiration of a decidu-ous forest.Their approach was later extended to u the information in the T s/NDVI ,Moran,Clarke, Inoue,et al.,1994).
udownThe first part of the ntence should read:‘‘A scatterplot of remotely nd surface temperature and a vegetation index often results in a triangular shape(Price,1990, Carlson et al.,1994),or a trapezoid shape(Moran,Clarke, Inoue,et al.,1994)if a full range of fractional vegetation cover and soil moisture contents is reprented in the data. Since different surface types may have different T s/NDVI slope and intercept for equal atmospheric and surface moisture conditions,the choice of scale may influence the shape of the relationship.
The location of a pixel in the T s/NDVI space is influ-enced by many factors,and a number of studies have been done to provide interpretations.Some of the may have had a theoretical ,Moran et al.,1994),some relied on simulations using Soil–Vegetation–Atmosphere Transfer(SVAT)models,(Gillies et al.,1997;Moran, Clarke,Kustas,Weltz,&Amer,1994),some have been bad on in situ measurements(Friedl&Davis,1994), while others are largely bad on analys of remotely nd data(Clarke,1997).A range of vegetation types and crops have been studied under a variety of climatic con-ditions,and the scales studied range from meters to global. It may not be surprising,therefore,that the interpretations differ widely.The approach described by Moran,Clarke, Inoue,et al.(1994)accounts for partially vegetated surfa-ces.They suggest a Water Deficit Index(WDI),related to the actual and potential evapotranspiration rates of a surface.Once the trapezoid defining the temperature/ vegetation index was estimated using a SVAT model, WDI was estimated from remotely nd and meteoro-logical data,using the relationship between surface tem-perature minus air temperature and a vegetation index. Carlson,Gillies,&Schmugge(1995)showed,using a SV AT model,how surface soil moisture availability and fractional vegetation cover can be derived from analysis of the temperature/vegetation index space.Assuming that the vegetation index is linearly related to fractional vegetation cover,and plotting the surface temperature–air temperature gradient as a function of the vegetation index,Mora
n, Clarke,Kustas,et al.(1994)derived the shape of T s/NDVI space from model simulations,and gave a theoretical justification of the concept.They suggested that the energy balance at the surface is a controlling factor,which justifies the u of the surface air temperature gradient via its relation to nsible heat flux.
Interpretation of surface temperature for spar canopies is not straightforward becau the measured temperature integrates the temperature of the soil surface as well as the temperature of the vegetation,and the components may not ‘‘mix’’linearly.Other studies have shown that,at least for well-watered surfaces,the relationship between surface temperature and NDVI is more directly related to surface soil moisture,through the increa of the thermal inertia of the soil with surface soil moisture rather than as a limiting control on latent heat(Friedl&Davis,1994).
We explored an empirical simplification of the WDI.The simplification takes into account variations in air temper-ature,radiation balance,and atmospheric forcing by empir-ical estimation of the T s/NDVI space.The method is conceptually and computationally straightforward,and only satellite-derived information is ud.theme是什么意思
2.1.Interpretation of the T s/NDVI space
The following mechanisms are suggested as tho deter-mining the location of a pixel in the T s/NDVI space.The interpretation is bad on findings in the , Carlson et al.,1994;Gillies et al.,1997;Goetz,1997; Lambin&Ehrlich,1996;Moran,Clarke,Inoue,et al., 1994;Nemani et al.,1993;Nemani&Running,1997,1989).
I.Sandholt et al./Remote Sensing of Environment79(2002)213–224 214
高考答案2.1.1.Fractional vegetation cover
The fractional vegetation cover can be related to spectral vegetation indices,through a simple,yet not necessarily linear transformation.Likewi,fractional vegetation cover influences the amount of bare soil and vegetation visible to a nsor and differences in radiative temperature between the soil and the vegetation canopy will affect the spatially integrated T s.
2.1.2.Evapotranspiration
Evapotranspiration can largely control the surface tem-perature through the energy balance of the surface.The less the evapotranspiration,the more energy available for n-sible heating of the surface.Stomatal resistance to transpi-ration is a key factor,which is partly controlled by soil moisture availability.
2.1.
3.Thermal properties of the surface
Heat capacity and conductivity—and thus the thermal inertia—influence T s in the ca of partly vegetated surfa-ces.The thermal properties are a function of soil type,and change with surface soil moisture.
peru2.1.4.Net radiation
The available energy incident at the surface affects T s.The radiative control of surface temperature implies that areas with a lower net shortwave radiation ,due to a high albedo)will have lower temperatures,all el equal. Albedo is controlled by soil type,surface soil moisture,and vegetation cover.Incident radiation also affects the stomatal resistance to transpiration,which factors into the partitioning of net radiation into nsible and latent heat.
2.1.5.Atmospheric forcing and surface roughness
The ability to conduct heat away from the surface into the atmosphere is an important component in the control of surface temperature.This,in part,explains how vegetated surfaces with higher roughne
ss have lower surface temper-atures(all el equal)compared to bare soil and influences the shape of the T s/NDVI space.Similarly,homogeneous surfaces with unlimited water supply(for instance,irrigated surfaces)may have higher surface temperature than expected,if heat conductivity into the atmosphere is reduced by poor mixing(Nemani&Running,1997).
2.1.6.Interacting factors
The factors have been summarid in Fig.1.No direct relation between surface temperature and surface soil mois-ture is evident,but soil moisture is clearly a critical factor in the mechanisms involved.For bare soils at constant irradi-ance,surface temperatures are primarily determined by soil moisture content,via evaporative control and thermal prop-erties of the surface.Fig.2shows the conceptual T s/NDVI space,with T s plotted as a function of NDVI.The left edge reprents bare soil from the range dry to wet(top–down). As the green vegetation amount increas along the x-axis (with NDVI),the maximum surface temperature decreas. For dry conditions,the negative relation is defined by the upper edge,which is the upper limit to surface temperature for a given surface type and climatic forcing.
Many of the suggested controlling parameters are strongly interlinked,and the emingly very differe
nt results obtained by rearchers using different approaches are not necessarily mutually exclusive.
3.Methods
3.1.Temperature–Vegetation Dryness Index,TVDI
Following the concept in Figs.2and3,isolines can be drawn in the triangle defining the T s/NDVI space.As a first iteration to obtain information on the surface soil moisture content,a dryness index(TVDI)having the values of1at the‘‘dry edge’’(limited water availability)and0at the‘‘wet edge’’(maximum evapotranspiration and thereby unlimited water access)can be defined:
TVDI¼
T sÀT s
min
aþb NDVIÀT s
min
ð1Þ
where T s
min
is the minimum surface temperature in the triangle,defining the wet edge,T s is the obrved surface temperature at the given pixel,NDVI is the obrved normalid difference vegetation index,and a and b are parameters defining the dry edge modelled as a linear fit to data(T s
max
=a+b NDVI),where T s
max
is the maximum surface temperature obrvation for a given NDVI.The parameters
a Fig.1.Illustration of the factors determining surface brightness temper-ature.The circled variables can be estimated using satellite data. Sn=shortwave net radiation balance;Rn=net radiation balance;GLAI= green leaf area index;Fc=fractional vegetation cover;ET=evapotranspi-ration;rs=stomatal resistance;M1=soil moisture content(root zone); M0=top soil moisture content.
I.Sandholt et al./Remote Sensing of Environment79(2002)213–224215
and b are estimated on the basis of pixels from an area large enough to reprent the entire range of surface moisture contents,from wet to dry,and from bare soil to fully vegetated surfaces.The uncertainty of TVDI is larger for high NDVI values,where the TVDI isolines are cloly t. The simplification of reprenting the T s/NDVI space with a triangle in contrast to a ,Moran et al.,1994) adds uncertainty to TVDI at high NDVI values.Likewi, the‘‘wet edge’’is modelled as a horizontal line as oppod to a sloping wet edge in the trapezoid approach,which may lead to an over estimation of TVDI at low NDVI values.
The isolines of TVDI correspond to the temperature–vegetation index(TVX)propod by Prihodko and Goward (1997),who estimated TVX as a bulk slope in the T s/NDVI space for a homogeneous area(13Â13NOAA-AVHRR pixels),with little or no variation in surface moisture conditions.The T s/N
DVI space emerges when the study area is incread and variability in surface moisture con-ditions introduced,in other words,the TVDI isolines can be regarded as veral superimpod TVX lines.Studies of T s/NDVI slopes report steeper slopes for dryer conditions (e.g.,Goetz,1997;Nemani et al.,1993),which is in accordance with TVDI(e Fig.3).The approach taken by veral studies of the slope of T s vs.NDVI,are thus in agreement with the concept behind TVDI,however,since TVDI may be estimated for each pixel,the full
spatial Fig.3.Definition of the TVDI.TVDI for a given pixel(NDVI/T s)is estimated as the proportion between lines A and B(e Eq.
亚特兰大奥运会主题曲(1)).
Fig.2.Simplified T s/NDVI(after Lambin&Ehrlich,1996).
本初子午线在哪I.Sandholt et al./Remote Sensing of Environment79(2002)213–224
216
resolution of data is maintained.In contrast TVX can only be computed for an area large enough to allow determina-tion of the slope in T s/NDVI ,contextual).
The advantage of the TVX and TVDI methods outlined here,or an approach estimating the vertices of the T s/NDVI space ,Clarke,1997),are their complete independence of ancillary data.Other , Moran et al.,1994),require detailed information on the meteorological conditions,including vapour pressure defi-cit,wind speed,and aerodynamic resistance,to define the limits of the T s/NDVI space.A simpler empirical estimation of the T s/NDVI space implicitly takes the parameters into account.The approach,however,requires a large number of remotely nd obrvations to ensure that the boundaries of the space are established.
3.2.Assumptions and sources of error
The empirical estimation of TVDI is bad on assump-tions that:(i)soil moisture is the main source of variation for T s and(ii)TVDI is related to surface soil moisture due to changes in thermal inertia and evaporative control(evap-oration and transpiration)on net radiation partitioning (energy balance).
For the operational estimation of TVDI from satellite data,a number of error sources exist:(i)no account of view angle effects on T s and NDVI,which affects the fraction of bare soil and vegetation visible to the nsor;(ii)the ‘‘triangle’’may not be determined correctly from the EO data,if the area of interest does not include a full range of variability in land surface ,dry bare soil, saturated bare soil,water stresd vegetation and well-watered vegetation);(iii)no account of errors in estimation of T s(unknown and varying land surface emissivity and atmospheric effects);(iv)no account of clouds,shadows, and associated variation in net radiation;(v)decoupling of the top surface soil layer from lower layers(Capehart& Carlson,1997);(vi)dependence of T s and NDVI on surface type due to differences in aerodynamic resistance(Friedl& Davis,1994;Lambin&Ehrlich,1995).
3.3.Model simulations of soil moisture
Surface soil moisture content is not well defined over large areas,which makes validation of TVDI using surface obrvations difficult.In order to compare the performance of TVDI to soil moisture,a h
ydrological model(MIKE SHE)was ud to simulate soil moisture(Andern, Refsgaard,&Jenn,in press;Refsgaard&Storm,1995). MIKE SHE is a development of the European Hydrological System(Systeme Hydrologique Europeen,SHE,Abbott, Bathurst,Cunge,&J.,1986).The model version ud in this study is oriented towards surface water studies,so that detailed description of groundwater conditions is not required.The model is fully distributed and physically bad except for the groundwater module that is only midis-tributed.The model solves for interception,evapotranspira-tion,overland flow,channel flow,unsaturated flow,and routing of saturated subsurface flow.In this way,the major flow process of the entire land pha of the hydrological cycle are described.The model in its current tup has been tested and calibrated for the whole Senegal River basin using daily rainfall data from112stations and discharge from11gauging stations covering a10-year period.Input to the model included maps of vegetation type,topography, river networks,soil type and depths,and time ries of precipitation,potential evapotranspiration,river discharge, LAI,and root depths.Despite spatial discretisation is 4Â4km2,the actual spatial resolution in model output is more coar due to low spatial resolution of the input data,particularly precipitation.The application of the model in the Senegal River basin has been described in more detail in Andern,Sandholt,Jenn,and Refsgaard (in press)and Sandholt et al.(1999).
4.Test area and data
The140Â140km2study area in the northern part of Senegal(Fig.4)includes part of the Senegal river valley as well as dry savanna on sandy soil.The area is generally flat, with very modest relief.Mean annual precipitation is less than200mm with a large north–south gradient.Local variability in rainfall is high.In general,the rainy ason is short,from June to September.The land cover is mainly dry grasslands with scattered trees and bushes and smaller cultivated areas in the river valley,some of them
irrigated, Fig.4.The Podor subt in the northern part of Senegal.The subt is 140Â140km2.Locations of rain gauge stations are indicated.
I.Sandholt et al./Remote Sensing of Environment79(2002)213–224217
>bapa
