+ All Categories
Home > Documents > t·'&'=> 'G ~ ~-nopr.niscair.res.in/bitstream/123456789/35983/1/IJRSP 22...important in many fields...

t·'&'=> 'G ~ ~-nopr.niscair.res.in/bitstream/123456789/35983/1/IJRSP 22...important in many fields...

Date post: 28-Apr-2019
Category:
Upload: dophuc
View: 213 times
Download: 0 times
Share this document with a friend
9
Indian Journal of Radio & Space Physics Vol. 22, June 1993, pp. 156-164 e i) r::..... r;pD :J ....' S- 3, '&'=> 'G ~ ~- oJ \ ' C:c" / ~'"'' Large-area soil moisture estimation using l"lN.SA l'~YHRl~;,infrared ~~t(!~ * P ffPathak, B MjRao &.M..!Vot<ki- '\ S (")- '!l ) I~. " (Remote Sensing Applications Group, Space Applications Centre, Ahmedabad 380 05~ ~O~t~~\(.o(\_~"')' ',.• .\ x-\J\ \~L ~~Kl9I1P lQ9llj ftUiiN ~~ary·{993;~h~p"l-~ \ It has been shown by-ea-Flier-workers that rate of change ofland surface temperature (An during mid- morning hours as determined from the thermal infrared sensor onboard geosynchronous satellite can pro- vide a good signature of area-averaged soil moisture. In order to examine the feasibility ofthis approach, a pilot study was undertaken using limiteddata ofINSA T-VHRR. Sinceno concurrent measurements of soil moisture and other meteorological data were available, the analysis was carried out making a number of simplifyingassumptions. The observed ATduring April 1991 showed a reasonably good association with arid and semi-arid regions of Rajasthan and Gujarat, respectively,while AT during September 1991 was found to be negatively correlated with the area-averaged total rainfall data over Gujarat. Although these results are encouraging, they can at best be considered as indicative of physical processes in view of the simplifying assumptions. More rigorous analysis is essential to establish the validity of the approach. t, (\J '11~r . 1 Introduction The moisture content of soil is the basic link bet- ween the energy budget ofland surface and the hydro- logical cycle. Estimation of soil moisture over large land areas using satellite remote sensing would be important in many fields such as climate modelling, regional drought monitoring, desertification studies, etc. The practical remote sensing methods of estimat- ing soil moisture are limited to two basic wavelength bands, namely, (i) the thermal infrared and (ii) the microwaves. The thermal infrared method depends on the modulation ofland surface temperature due to changes in thermal inertia which, in turn, are related to soil moisture. The microwave band depnds on the effect of soil moisture on the dielectric constant and thereby on the emissivity of the soil. While the thermal infrared method cannot be used during cloudy condi- tions, the microwave range has the special advantage that it can also be used during cloudy, but non-rain- ing conditions. However, it is important to note that suitable thermal sensors are available on both the pre- sently operational polar (NOAA) and geostationary (INSA T) satellites, whereas there are no operational satellites having appropriate microwave sensors for estimating soil moisture. The present work is based on the thermal infrared data from the INSAT satelli- te. *This paper was presented at the National Space Science Sympo- sium. held during 11-14·March 1992 at Physical Research Labo- ratory. Ahmedabad 3S0 009. Historically, the development of the estimation of soil moisture using thermal infrared wavelength has evolved out of the earlier work of geologists to de- tect surface-rock types through their thermal inertia effects on the diurnal range of land surface tempe- rature'. The thermal inertia refers to the resistance of a material to a change of temperature and is normally defined as p= (kpC)1/2 ... (1) where, P is the thermal inertia, k the thermal conduct- ivity, p the density and c tire specific heat of the materi- al. The addition of moisture to dry soil will significan- tly increase its thermal inertia as a result of increase in specific heat and thermal conductivity. As a result, the rise in the land surface temperature would be less for a given heat input. It was shown by Watson et al? that the thermal inertia of dry material is linearly cor- related with its density. Rocks and rock-forming mi- nerals have larger densities (2-4 g/cc) and, therefore, have higher thermal inertia, while dry soils have low densities ( < 2 g/cc) and thus have low values of ther- mal inertia. Hence, thermal inertia monitoring thro- ugh remote sensing has been found to be useful for detecting rock types. In addition to thermal inertia, albedo of the land surface is also an important factor which determines the land surface temperature. Hig- her the albedo, lower would be the land surface temp- erature and vice versa. Soil moisture determination by thermal infrared remote sensing essentially consists of two phases, viz. (i) measuring surface temperature Tand (ii) quantita-
Transcript

Indian Journal of Radio & Space PhysicsVol. 22, June 1993, pp. 156-164

e i)r::..... r;pD :J

....•'

S- 3,t ·'&'=> 'G ~ ~- oJ \ ' C:c"

/ ~'"''

Large-area soil moisture estimation using l"lN.SAl'~YHRl~;,infrared ~~t(!~*P ffPathak, B MjRao &.M..!Vot<ki- '\ S (") - '!l

) I~. " • (Remote Sensing Applications Group, Space Applications Centre, Ahmedabad 380 05~~O~t~~\(.o(\_~"')' ',.• .\ x-\J\ \~L~~Kl9I1P lQ9llj ftUiiN ~~ary·{993;~h~p"l-~

\ It has been shown by-ea-Flier-workersthat rate of change ofland surface temperature (An during mid-morning hours as determined from the thermal infrared sensor onboard geosynchronous satellite can pro-videa good signature of area-averaged soil moisture. In order to examine the feasibilityofthis approach, apilot study wasundertaken using limiteddata ofINSA T-VHRR. Sinceno concurrent measurements of soilmoisture and other meteorological data were available, the analysis was carried out making a number ofsimplifyingassumptions. The observed ATduring April 1991 showed a reasonably good association witharid and semi-arid regions of Rajasthan and Gujarat, respectively,while AT during September 1991 wasfound to be negativelycorrelated with the area-averaged total rainfall data over Gujarat. Although theseresults are encouraging, they can at best be considered as indicative of physical processes in view of thesimplifying assumptions. More rigorous analysis is essential to establish the validity of the approach. t,

(\J '11~r.1 Introduction

The moisture content of soil is the basic link bet-ween the energy budget ofland surface and the hydro-logical cycle. Estimation of soil moisture over largeland areas using satellite remote sensing would beimportant in many fields such as climate modelling,regional drought monitoring, desertification studies,etc.

The practical remote sensing methods of estimat-ing soil moisture are limited to two basic wavelengthbands, namely, (i) the thermal infrared and (ii) themicrowaves. The thermal infrared method dependson the modulation ofland surface temperature due tochanges in thermal inertia which, in turn, are relatedto soil moisture. The microwave band depnds on theeffect of soil moisture on the dielectric constant andthereby on the emissivity of the soil. While the thermalinfrared method cannot be used during cloudy condi-tions, the microwave range has the special advantagethat it can also be used during cloudy, but non-rain-ing conditions. However, it is important to note thatsuitable thermal sensors are available on both the pre-sently operational polar (NOAA) and geostationary(INSA T) satellites, whereas there are no operationalsatellites having appropriate microwave sensors forestimating soil moisture. The present work is basedon the thermal infrared data from the INSAT satelli-te.

*This paper was presented at the National Space Science Sympo-sium. held during 11-14·March 1992 at Physical Research Labo-ratory. Ahmedabad 3S0 009.

Historically, the development of the estimation ofsoil moisture using thermal infrared wavelength hasevolved out of the earlier work of geologists to de-tect surface-rock types through their thermal inertiaeffects on the diurnal range of land surface tempe-rature'. The thermal inertia refers to the resistanceof a material to a change of temperature and isnormally defined as

p= (kpC)1/2 ... (1)

where, P is the thermal inertia, k the thermal conduct-ivity, p the density and c tire specific heat of the materi-al. The addition of moisture to dry soil will significan-tly increase its thermal inertia as a result of increase inspecific heat and thermal conductivity. As a result,the rise in the land surface temperature would be lessfor a given heat input. It was shown by Watson et al?that the thermal inertia of dry material is linearly cor-related with its density. Rocks and rock-forming mi-nerals have larger densities (2-4 g/cc) and, therefore,have higher thermal inertia, while dry soils have lowdensities ( < 2 g/cc) and thus have low values of ther-mal inertia. Hence, thermal inertia monitoring thro-ugh remote sensing has been found to be useful fordetecting rock types. In addition to thermal inertia,albedo of the land surface is also an important factorwhich determines the land surface temperature. Hig-her the albedo, lower would be the land surface temp-erature and vice versa.

Soil moisture determination by thermal infraredremote sensing essentially consists of two phases, viz.(i) measuring surface temperature Tand (ii) quantita-

PATHAK et aL: SOIL MOISTURE ESTIMATION USING INSAT-VHRR INFRARED DATA 157

tively relating Tto the amount of water present in thesoil. The temperature T is measured by remote sen-sors operating in the 8-14 .urn atmospheric "window"(or its portion, such as 9.5-11 urn, 10.5-12.5 urn etc.),where the atmospheric effects on the passing radia-tion are relatively small. Nevertheless, the radiationreceived by the sensor includes parameters due to fac-tors related to atmosphere and land surface emissivi-ty, thereby complicating the interpretation.

Wetzel et al.' have examined the feasibility of esti-mating soil moisture from land surface temperatureinferred through the IR data of geosynchronous sate-llite. Using a one-dimensional boundary layer-sur-face-soil model, the diurnal variation of surface tem-perature was simulated and sensitivity tests were per-formed by varying various parameters such as surfacealbedo, biomass, roughness height, soil moisture,wind speed etc. As a result of these sensitivity tests, anumber of potential signatures sensitive to soil mois-ture were found out. These are:(i) Rate of change ofland surface temperature W.r.t.the absorbed solar radiation (AT/AS) during mid-morning and in late afternoon,(ii) Diurnal range of land surface temperature,(iii) Lag-time of the peak temperature W.r.t. the localnoon time, and(iv) Integral area under the land surface diurnal tem-perature curve.

Out of the above parameters, AT/AS during mid-morning hours (0800-1000 hrs) was found toshow the strongest signature due to soil moisture andleast effects due to other variables like albedo, landsurface emissivity etc. However, wind speed, rough-ness height and biomass were found to have notice-able effects on AT/AS which should be taken intoaccount. Based on the above approach, Wetzel andWoodward", in a case study over Kansas and Nebras-ka, obtained a multiple linear regression between !!J.T,normalized difference vegetation index (NDVI) fromNOAA, representing biomass, wind speed and soilmoisture. It was found that it is possible to distinguishat least four classes of soil moisture using this appro-ach. The method is applicable only for arid, semi-aridand marginal agricultural areas.

The purpose of this study is to present the prelimi-nary results of a pilot study using limited data ofINSA T and NOAA satellites. Since no ground truthdata on soil moisture were available either throughactual measurement or estimated using rainfall data,an approximate analysis was carried out by correlat-ing the observed A Twith the total rainfall data. This isa crude assumption and, therefore, the results can atbest be considered as only suggestive or indicative ofthe actual physical processes. Since we have not cons-

ide red the effects of meteorological parameters like,wind speed etc., and in view of the limited data onNDVI, the results of the present study are only preli-minary in nature. The present work forms a part of theGeosphere- Biosphere studies.

2 Data and methodologyThe present study has made use of data from

INSAT-VHRR (Very High Resolution Radiomet-er) and NOAA-A VHRR (Advanced VHRR) sens-ors. The INSA T-VHRR has two bands-cone in visi-ble (VIS) (0.55-0.75 urn) and other in IR (10.5-12.5urn) range at resolutions of2.75 and 11km, respectiv-ely. The INSAT data are available at the interval ofevery 30 min and are most suitable for monitoring themid-morning rise in surface temperatures. The NO-AA-A VHRR has 5 bands, out of which the first twoare in VIS (0.58-0.68 urn) and near-IR (0.725-1.10urn) with a resolution of 1.1 km. These two bands areuseful in determining the quantity called NDVI,which is indicative of the greenness or biomass".

The INSA T-1D VHRR data were recorded at Ah-medabad Earth Station (AES) on 18 April 1991 at0830 and 1130hrs 1ST. Another set of data was recor-ded on 24 Sep. 1991at 0900 and ll00hrs 1ST. Duringboth these days the data were totally cloud free. Itshould be noted here that whereas, the April dataduring pre-monsoon period represented generallydry soil conditions, the September data in the post-monsoon period represented relatively wet conditi-ons due to the soil moisture input from the south-westmonsoon rainfall during June-September 1991.Thus, both these data sets were expected to show extr-eme conditions of soil moisture depending on the rain-fall. The NOAA-AVHRR data for 18 Apr. 1991were available for computing NDVI. The NOAAdata of September 1991 were not available for thepresent study. In this study, we have selected test siteover the states of Gujarat and Rajasthan. Figure 1shows the test site in reference to India. Some import-ant rivers in this area are also shown. The land areas inthese states are classified as arid, semi-arid and sub-humid types and, therefore, they offer contrasting soilmoisture conditions in general. The climatologicalaverage rainfall" during June-September is lowest(~ 10ern) at the western edge of Rajasthan and gradu-ally increases towards east reaching a value of 75 cm.

In the present work, in place of AS, where, S is theabsorbed solar radiation, we have used the climatolo-gically observed global solar radiation 7 for S. Thiscould be slightly different on account of the surfacealbedo. We have not made any quantitative estimate ofthis effect. However, it is felt that, if the change inalbedo during mid-morning is not appreciable, thiseffect would be negligible.

158 INDIAN J RADIO & SPACE PHYS, JUNE 1993

20~--------~~~--------~

Fig. I-Test site over Rajasthan and Gujarat states with referenceto India (Important rivers in the test area are also shown.)

In the absence of actual ground truth soil moisturedata, we have used total south-west monsoon rainfallfor June-September over Gujarat and Rajasthan.This crude assumption implies that both evaporationand run-off are totally neglected. Figure 2 shows thetest site along with the total number of raingaugesavailable in each lOx 10 lat./long. grid box.

2.1 Processing of the INSAT-VHRR dataThe IR data were separated from the VIS, and IR

data stream ofINSAT- VHRR and the data over thetest area were extracted for further processing. Thetwo INSA T images recorded two or three hours apartwere exactly registered within one pixel by takingclear ground control points (GCPs) into considerati-on. Next, the two images were digitally subtracted toget the difference image in terms of gray count diffe-rence L1Cfor each pixel. Based on the laboratory/cali-bration of the INSAT- YHRR IR channel, the appar-ent brightness temperatures Tappof a given pixel attime (I and 12 in terms of the corresponding gray countC. are given as

30...,

! 1:..,i'/ \/

'"(~V._../2 3 4 5 6 {,K(-) (-) (1) (2 ) (- ) (1)I'"

t....•. 7 8 9 10 11 12 ;;;} (1) ( -) ( -) (1) (1) (2 )

1\ __ .13 14 1~ 16 17 18 (H(. (1) (1) (1) (1) (2) '319 20 21

• 2~~

~~

,~ ....•.. .'\'")(3) 'r'N" , (11 (1)1 ~

~23 24 25 '. 26 27 CJ'(4) ( 5) (15) (11')", ,(,1)(--'728 29 JO 31 ;~ (10) (10) ~111 (17)-, 32 33 ~( 3~~

(16) (15) (14) ~<.'-..-- ))

28

z° .26l1Jo::I•....•....<..J

24

22

68 70 72 74

LONGITUDE,oE76 78

Fig. 2- Test site along with total number of rain-gauge ,(each gridbox is assigned a number which is given on the right-hand top corner.The total number of raingauge stations available in each box is

shown in brackets.)

Tapp (tl) = A - 0.125 C (II) (2)Tapp (t2) = A - 0.125 C (12) (3)

By digitally subtracting the two images we get thechange in the apparent temperature as

L1Tapp=0.125xL1C , .. (4)

However, the retrieval ofland surface temperature Tfrom Tappis modified due to the absorption/emissionby atmosphere. This introduces a contrast (Tatm) bet-ween the actual land surface temperature and the ap-parent temperature observed by the satellite sensor.Secondly, the land surface emissivity which generallyvaries between 0.95-0.99 also has to be taken intoaccount. In the present work, the surface emissivity isassumed to be 1. According to Prices, for moderaterange of land surface temperature (300 ± 10 K), theapparent temperature is approximately linearly rela-ted to the surface temperature T.The apparent tem-peratures at times II and 12 can thus be expressed as

Tapp «(1)= T (11)+ Tatm (II) (5)Tapp (12) = T (12) + Tatm (12) (6)

Assuming that the temperature and water vapourstructure do not change within the time interval(12 - II), Tatmwill remain same at both the times. The

. change in the apparent temperature is, therefore, gi-

PATHAK et at.: SOIL MOISTURE ESTIMATION USING INSAT-VHRR INFRARED DATA 159

ven by

L1Tapp= T (t2) - T (td = L1T

or,L1T=0.125 x L1C ... (7)

Thus, approximately the difference image ofINSAT-VHRR IR channel can be considered to bethe same as the L1T image, giving mid-morning cha-nge in the land surface temperature.

2.2 NOAA-A VHRR dataThe NOAA-A VHRR data were used to derive the

vegetation information for the study area. The vege-tation information is obtained in the form of a norma-lized difference vegetation index (NOVI) defined as

NOVI =(Ch2- Chl)/(Ch2+Chl) ... (8)

where, Ch I and Ch2 are reflectances in the first twochannels in visible and near infrared, respectively, ofA VHRR. The data were first geometrically correctedand the NOVI was then computed over the test site.Since the ground resolution of NO VI data is 1.1 km,averaging over 10 pixels was carried out to make theNDVI data comparable with the INSA T -VHRR/IRresolution of 11 km. All the satellite data (INSAT andNOAA) were processed on the image processing co-mputer-ISROVISION, using SACIMAGEsoftware routines for generating colour coded imagesof sr and NDVI.

3 Analysis of the data and resultsFigure 3 is the colour-coded imagery of

INSAT-fiTfor 18 Apr. 1991. Note the few squaressurrounded by rosette-like structure. These denotenoise in the data which have very large gray countsand occur randomly in the image data. It can be seenthat, based on fi T, 3-4 categories ofland regions canbe separated with fiT showing a gradual decreasefrom west to east. The highest fiT values of 6-7 K(green colour) are limited to a small region in westernRajasthan, which is itself embedded in a much largerregion (pink colour) with fiT values of 5-6 K. Thisregion covers a large part of Rajasthan and some por-tion of north Gujarat. The next fiT range of 4-5 K(yellow colour) occupies almost the whole of Gujaratand eastern Rajasthan. The lowest L1Trange of2-4 K(brown colour) is limited to very small regions alongthe coastal areas in Saurashtra (and south Gujarat)and in the Rann of Kachchh as well as in some por-tions along the three rivers in south Gujarat and theIndus river.These lowest L1Tvalues could be mostlikely due to wet/marshy coastal areas and due tosome vegetation along the rivers. Another reason for

the lower temperatures could be the effect of sea/landbreeze in the coastal areas. It is interesting to note thatthe climatological demarcation of the arid, semi-aridand sub-humid regions in Rajasthan and Gujarat?(Fig. 4) is very much similar to the INSA T-fi Timage,such that high fiTvalues could be roughly associatedwith arid regions, and medium and low fiT valueswith semi-arid and sub-humid regions, respectively.This observation, although circumstantial, lends in-direct support to the basic idea of the present workthat fi Tis directly related to aridity. For the presentstudy, however, it was not possible to obtain any gro-und truth for aridity or soil moisture and, therefore,the comparison has been made only with the climato-logical features.

The observed fiTvalues by INSAT also compareweJl with the climatological data" of soil temperatu-res during April. These data show that the change inthe soil temperature at 5 em depth from 0700 to 1400hrs 1ST over the test site is about 12-13 K. Based onthis, the 3-hourly change works out to be about 5 K.At the surface, the corresponding change would besomewhat more than 5 K, which is consistent with thepresent observations.

The NOAA-NOVI image, averaged over 10 pix-els, is shown in Fig. 5. This shows that 3-4 regions canbe clearly separated based on their NOVI values. Itmay be noted that the pre-monsoon dry condition ofApril gives, in general, low values of NOVI almosteverywhere over the test site. The highest value ofNOVI is 0.3. This could be partly due to the averagingover 10 pixels in order to bringthe resolution in con-formity with INSA T data.

The different features of NOAA-NOVI data art,compared with the lNSAT-fiTin Table I. In general,there seems to be a negative correlation betweenNOAA-NOVI and INSAT-fiT, with low NDVI ass-ociated with high fi Tand vice versa. This is consistentwith the expected behaviour, since the canopy tempe-rature over vegetation is lower than the bare arid soildue to evapotranspiration by vegetation Io. A scatterplot between the 5 x 5 pixel, i.e., values, averaged over55 km x 55 krn, of the various representative featuresofNDVI and fiT(Fig. 6), does, in fact, show negativecorrelation as expected.

Figure 7 gives the INSA T -fi T image for 24 Sep.1991. As in the INSAT-fiTimage for April, the Sep-tember fi Timage also contains some noise in the data.It can be seen that there are three distinct legionshaving different L1Tranges. The fourth region (greencolour, at the bottom right-hand corner) which isactually outside the test site. has resulted due to clo-udy condition in one of the data sets. This is not taken

160 INDIAN J RADIO & SPACE PHYS, JUNE 1993

0-16

0-2 2-4

33-40

4-5

41-48

5-6 6-70.11-0.15

GRAY COUNT D1FF.

c. T (K)

NDVI

(APPROX)

Fig. 5-NOAA-NDVI image for 18 Apr. 1991.

Fig. 3-INSAT-c.T image for 18 Apr. 1991 (recorded during0830-1130 hrs LT) Table I-Comparison between INSAT-c.Tand NOAA-NDVI

images for April 1991

m ARID~ SEMI-ARID~ SUB-HUMID

Fig. 4-Arid, semi-arid and sub-humid regions in the test site9.

c. T range and correspondingland features (Fig. 3)

6-7 K (Highest values of c.T);Arid desert rocky land ofRajasthan (Green)

5-6 K; Regions surroundingthe Rajasthan desert and in theparts of Rann of Kachchh(Pink)

4-5 K; Large parts of Gujarat,Kachchh, Saurashtra andIndus valley (Yellow)

2-4 K; Coastal regions inGujarat, Saurashtra, Kachchhand isolated patches in theIndus valley region (Brown)

NDVI range and correspond­ing land features (Fig. 5)

(0.05-0.10); Very low vegeta­tion mainly in the desert regionand in the Rann of Kachchh

(Brown)

(0.11-0. 15); Low vegetation;Region surrounding the Raja­sthan desert and parts of Indusvalley, Saurashtra and Gujarat(Yellow)

(0.16-0.20); Moderate vegeta­tion; P!rts of Gujarat, coastalregions in Saurashtra and Indus'valley region (Pink)

(0.21-0.30); High vegeta-tion; Parts of coastal region inSouth Gujarat and Kheda dis­trict (Green)

into consideration for obvious reasons. A compari­son with April INSA T-LlTimage (Fig. 3) shows thatthe overall LlT values during September are some­what lower than those in April. During April, a regionofhighes.t LlTwas embedded in another region which

occupied a large part of Rajasthan with a small port­ion over north Gujarat. However, in September,there is only one region over Rajasthan with highestLlT values of 3·4 K (pink colour). Rather, the tworegions appear to have merged into one. The shape of

PATHAK et af.: SOIL MOISTURE ESTIMATION USING INSAT-VHRR INFRARED DATA 161

18 APRIL 1991

Fig. 6-Scatter plot ofNDVI versus INSA T -!!.T data for 18Apr.1991 [Both data are averaged over (5 x 5) pixels.]

!!. T (K)

Fig. 7-INSA T-~T image for 24 Sep. 1991 (recorded during0900-1100 hrs LT)

The general decrease of L\ T in September could bepossibly due to the following three factors.(i) The September data were takeh for a time intervalof2 h (0900-11 00 hrs 1ST), while the April data werefor 3 h (0830-1130 hrs 1ST).(ii) Seasonal decrease in solar radiation input.(iii) The third and most important physical factorcould be the input of soil moisture through the totalrainfall during the south-west monsoon period (Jun­e-September 1991)and the consequent increase in thevegetation cover, both of which could decrease L\ T.

The factors (i) and (ii) can be taken into account byusing L\S values appropriate for the required time in­tervals (2 or 3 h) during the two months in question.As mentioned before, we have used climatologicalvalues7 of L\S from the published data.

It was not possible to obtain the NOAA data forSeptember to compute NDVI and get an estimate ofthe extent of vegetation growth over the test site. Ba­sed on the Drought Bulletin No.9 for the states ofGujarat and Rajasthan!! for the period 24 Sep.-07Oct. 1991, it was found'out that there was a substan­tial increase in the green vegetation almost everywh­ere over the test site. However, it was not P9ssible toquantitatively derive the average NDVI value foreach lOx i0 grid box using these data which were in theform of imageries.

Next, the L\Tvalues fOl(April and September wereaveraged over each lOx 10 lat./long. grid box over the.test site. While averaging, care was taken to excludenoisy data which have large gray counts. Also, forsome of the grid boxes which were near to sea coast,very low L\Tvalues due to coastal wet land were notconsidered for averaging. The standard deviation va­lues of L\Tf'Or most of the boxes were generally foundto be ± 0.25 K. This may be compared with the noiseequivalent difference temperature (NE.L\1) ofVHRR-IR channel, which is also 0.2 K for 300 Ktarget temperature ..

The average L\T values for April and Septemberwere normalized by using appropriate values of L\S,and L\t/L\S was computed for each lOx 10 lat./long.grid box for April and September 1991. In order tostudy the decrease of L\Tin September, we define thequantity D as

D=(L\T/L\S)Apr -(L\T/L\S)sep .... (9)

for each grid box in Rajasthan and Gujarat. Figure 8,which is a histogram of D, shows that all values of D

are positive, indicating an overall decrease in the rateofland surface heating in September as compared toApril. This could be mainly due to the soil moistureinput through the total rainfall during Jlolne-Septem­bel' 1991. In addition,other effects such as vegetationcover, biomass and wind-induced cooling could also

6 AT (K)

•••••

, I ••35 '0 45

IN5AT GRAY COUNT OIFf. CAT J. K

•••

•••

••

••

•••••• •

..L30

0-05

the region has remained more or less similar, though ithas shrunk in area to some extent. Note that there is a

small narrow strip oflower L\Tvalues of2-3 K (yellowcolour) embedded in this large region. This could bepossibly due to irrigation from the Indira Gandhi Ca­nal.

The next L\Trange of2-3 K (yellow colour) occu­pies a part of eastern Rajasthan and most part ofnorth Gujarat, Saurashtra and Kachchh. The lowestrange of L\T= 1-2K (brown colour) occupies the rem­aining part of eastern Rajasthan, south Gujarat, coa­stal Saurashtra and along the Indus river showing itscharacteristic shape. These regions would most likelybe having good amount of vegetation.

200~ 0.25

1401-

"...

~ 180 f 0.20on~z::>

8 I- 0.15> 160..a:'"

.~ I- 0.10z

162 INDIAN J RADIO & SPACE PHYS, JUNE 1993

Q;.D

E 15:Jc

c-u~10::loWa::u,

5

01 0.5 10 15 2.0 2.5o (RANGE), K/kWH m-2

Fig. 8-Histogram of D

be present. For the present study, we do not have anyground truth data on soil moisture, and, therefore, wehave made a crude assumption that soil moisture in-put during the period April-September 1991 is direc-tly proportional to the total rainfall during the south-west monsoon period (June-September 1991), Thisassumption implies that both evaporation and run-off have been neglected,

Since the rainfall data available over Rajasthan arevery scanty, while the data for Gujarat are fairly detai-led (see Fig. 2), we pay our attention, for the timebeing, to the eleven grid boxes over Gujarat only (i.e,box Nos. 23, 24, 25, 26, 28, 29,30,31,32,33 and 34) asshown in Fig. 2 and study correlations with averagerainfall only for these boxes.

The quantity D. as defined earlier may be consi-dered as representative of the cooling due to soil mois-ture input. Thus, we should expect a positive correla-tion between D and soil moisture as parametrized bythe total rainfall. This can indeed be seen from Fig. 9which gives a scatter plot between D and average rain-fall for each grid box over Gujarat. The data pointsindicate two separate regression lines A and B. Eightdata points out of the total eleven are highly correla-ted with total rainfall with a correlation coefficient of+ 0.96 and follow regression line (A) which passesvery near the origin, while the remaining three datapoints for box Nos. 32, 33 and 34, follow a distinctlyseparate regression line (B) which has a large inter-cept on the 0 axis. In other words, the line B indicatesthat for the same amount of the average rainfall, thebox Nos. 32,33 and 34 give higher values of D or theseboxes indicate lower l1T!l1S values in September ascompared to other boxes. This could be due to thepresence oflarge amount of vegetation in these boxesduring September.

Referring to the INSAT-l1Timage for September1991 (Fig. 7), it can be clearly seen that the locationscorresponding to these three boxes do indicate verylow l1Tvalues (1-2 K) suggestive of vegetation. The

B

A

.23

~~O--~--~3~O~~--~5~O~~--~7~O~~--~90AVERAGE RAINFALL ,em

Fig. 9-Scatter plot of D (in K/kWH m - 2) versus season's totalrainfall (in em) averaged for each grid box [The number given toeach data point refers to the grid box number as shown in Fig. 2.]

amount of vegetation in these three boxes during Sep-tember was also verified to be large from the DroughtBulletin No.9 for Gujarat state 1 I. In comparison, theremaining boxes show relatively less vegetation. Ifquantitative information on NDVI were available foreach grid box for September, it would have been pos-sible to study multiple correlation between D, NDVIand rainfall. It would be useful to see how /1Tj/1Svalues in September are related to the soil moisturecondition. We, therefore, examine the correlation be-tween /1T] /1S for September and the average rainfallfor all grid boxes in Gujarat. A scatter plot (notshown) between these two quantities gave a negativecorrelation with a non-linear relationship. A log-logscatter plot was, therefore, made (Fig. 10). This showsthat as in the previous case (i.e. D vs. rainfall) the datapoints for the eight grid boxes of Gujarat seem tofollow a single regression line with a correlation co-efficient of - 0.89, while, the data points for the gridboxes 32,33 and 34 fall quite apart showing the effectof increased vegetation in these locations after themonsoon. It may be noted here that for this case, inst-ead of 11T! l1S, 11T alone could also be used, since l1S issame for all grid boxes. Only for comparison betweentwo separate time periods, l1T!l1S is necessary. Theabove result is also consistent with the fact that higherrainfall or higher soil moisture would lead to lowervalues of l1T and vice-versa.

,

PATHAK et al.: SOIL MOISTURE ESTIMATION USING INSAT-VHRR INFRARED DATA 163

1.00

•25\I)<I-I-<I

0.9001 •.2 26,30

• 33 •31

32

340.8

1.0 1.5

log R2.0

Fig. IO-Scatter plot oflog (~T/~s) of September 1991 versus logR, where R is the average rainfall [Each data point is numbered as

in Fig. 9.]

4 DiscussionThe present study was undertaken to study the fea-

sibility of estimating area-averaged soil moisture con-ditions using INSA T- VHR R infrared data. In part-icular, the INSAT data were used to estimate mid-morning AT/AS, which gives a good signature of soilmoisture according to the earlier studies ':". Since noground truth data on soil moisture were available,total seasonal rainfall was used for the analysis. Thisassumption implied that both evaporation and run-off are negligible, which is not true. The present studyalso did not include some of the relevant features likepresence of forest cover, irrigated areas, mountainterrain, areas near to sea coast etc. all of which wouldaffect ATin different ways. Some of the results of thepresent analysis might have been affected by thesefactors. It would be more appropriate to use the mois-ture balance approach 12 using observed/climatolog-ical potential evapotranspiration values for estimat-inz soil moisture conditions for correlating withI:> .AT/AS values. This would result in an empirical rela-tionship between soil moisture and AT/AS. This app-roach is being tried with the present data.

5 Summary and conclusionsThe main results of the present work are summari-

zed as follows.

(i) The INSAT-AT distrib~t!on over th~ test site forthe dry pre-monsoon conditions ~f Apnl .1991 ro~g-hly resembles the climatological distri bution o.f a,?d,semi-arid and sub-humid regions on a qualitativebasis, wherein the highest AT values correspond toarid regions, while medium and low ~Tva.lues corres-pond to the semi-arid and.sub-hu~lld reglOn.s, resp~-ctively. This result is consistent with t~e baSIC phYSI-cal relationship that exists between soil moisture andland surface temperature change. Further, theobser-ved AT values are found to be consistent with theclimatological data on soil temperature .(ii) The INSA T-AT data for April are fo~nd to benegatively correlated with the correspo.ndl?g NDVIdata, showing the effect of evapotranspiration by ve-getation.(iii) The INSAT-AT values for the relatively wetpost-monsoon conditions of September 1991 are fo-und to be somewhat lower than those in April. Thenormalized difference D, as defined in Sec. 3 on al" x 10 lat.rlong. average basis, is found to be predom-inantly positive. This is considered to be indicative. ofsoil moisture input due to the total monsoonal ram-fall. This is also borne out from the observed high posi-tive correlation between D and total rainfall averagedfor each lOx 10 lat.rlong. grid box of Gujarat. Thedata for three grid boxes show slightly different beha-viour, which is tentatively explained to be due to theeffects of vegetation and wind speed etc.(iv) The INSAT-ATvalues for September show anegative correlation with the total rainfall averagedfor each grid box. In this case also, the data for thesame three grid boxes, as in the previous case, showslightly different behaviour which could be due to theeffects of vegetation and wind speed. The observednegative correlation is consistent with the expectedrelationship between ATand soil moisture. The pre-sent study is based on very limited data and the me-thod of analysis is rather approximate with a numberof simplifying assumptions. However, the results doindicate that mid-morning change in the land surfacetemperature derived from INSAT-VHRR ?ata~an beused to estimate large-area averaged soil moistureconditions. Further work, using actual observed gro-und truth soil moisture data or estimated soil moistureusing moisture balance approach as well as other =.vant measurements like soil temperature, AS, windspeed, vegetation biomass (NDVI), rainfall, etc. wo-uld be necessary to gain better insight into the prob-lem.

AcknowledgementThe authors are grateful to Dr P S Desai, Head,

Meteorology and Oceanography Division, SAC, Dr

164 INDIAN J RADIO & SPACE PHYS, JUNE 1993..Baldev Sahai, Group Director, Remote Sensing App-lications Group, SAC, and Prof. B Subbaraya, Prog-ramme Director, liD GBP, for their keen interest andencouragement in this work. Thanks are due to Dr VK Agrawal for helpful discussion. They sincerely ack-nowledge the active support provided by the staff ofAES, SAC, Ahmedabad, in recording the INSAT-VHRR data using the DRDS facility and also the helpprovided by the staff of the Directorate of Agricultu-re, Gujarat State, for supplying detailed rainfall dataof Gujarat.

ReferencesI Kahle A B, Gillespie A R & Goetz A F H, Geophys Res Lell

(USA). 3 (1976) 26.2 Watson K, Pohn H A & Offield T W, Proceedings ofEighth

International Symposium on Remote Sensing 0/ Environment.held at Michigan, USA, 1972, p. 122.

3 Wetzel P J, Atlas D& Woodward R H, J Clim & Appl Meteorol(USA), 23 (1984) 375.

4 Wetzel P J & Woodward R H, J Clim & Appl Meleorol(USA),26 (1987) 107,

5 Tucker C J, Remote Sensing Environ (USA), 8 (1979) 127.6 Agroclimatic Atlas of India. India Meteorological Departm-

ent, 1978.\1 Mani A, Handbook ofSolar Radiation Data/or India (Allied

Publishers Pvt. Ltd, India), 1980,8 Prince J C, Remote Sensing Environ (USA), 13 (1983) 353,9 Climate Variations, Drought and Desertification. WMO No,

653, World Meteorological Organization, 1985.IO Goward S N, Cruickshanks G D & Hope A S, Remote Sensing

Environ (USA). 18 (1985) 137,\.1 J Drought Bulletin for 24 Sept,-07 Oct, 1991 for Gujarat and

Rajasthan States. National Remote Sensing Agency, Hydera-bad, 1991.

12 ThoranthwaiteCW & Mather J R, The Water Balance, Clima-tology VIII (i): Drexellnst. of TechnoL, NJ (USA), 1955, p.104.


Recommended