Remote sensing in hydrology

Remote sensing in hydrology

Journal of Hydrology, 100 (1988) 239-265 Elsevier Science Publishers B.V., Amsterdam 239 Printed in The Netherlands [41 REMOTE S E N S I N G IN HYD...

2MB Sizes 0 Downloads 58 Views

Journal of Hydrology, 100 (1988) 239-265 Elsevier Science Publishers B.V., Amsterdam

239 Printed in The Netherlands

[41

REMOTE S E N S I N G IN HYDROLOGY

GERT A. SCHULTZ

Institute for Hydrology, Water Resources and Environmental Engineering, Ruhr University Bochum, P.O. Box 102148, 4630 Bochum 1 (F.R.G.) (Received J a n u a r y 27, 1988; revised and accepted February 23, 1988)

ABSTRACT Schultz, G.A., 1988. Remote sensing in hydrology. J. Hydrol., 100: 239-265. The '~Electronic Age" offers new and attractive opportunities to hydrologists for remote sensing (RS) of hydrological data. A discussion of hydrologically relevant platforms and sensors and the type of electromagnetic signals used by such sensors is followed by a n analysis of the structure of mathematical hydrologic models which use RS information either as input or to provide a basis for model parameter estimation. Three examples of RS application in hydrological modeling are given: (1) model parameter estimation with the aid of multispectral Landsat satellite data; (2) computation of historic monthly runoff for design purposes with the aid of a lumped system model using NOAA infrared satellite data as input; and (3) real-time flood forecasting applying a distributed system model using radar rainfall measurements as input. F u r t h e r applications of RS information in hydrology are discussed in the field of evapotranspiration, soil moisture, rainfall, surface water, snow and ice, sediments and water quality. A brief discussion of RS data availability and the hardware and software required is followed by a n assessment of future opportunities. The potential of passive and active microwave sensors for hydrological applications is emphasized.

INTRODUCTION

Most of us would be prepared to accept, or at least to treat seriously, a report t h a t vast groundwater resources had been discovered beneath the Sahara Desert, through the aid of satellite data - even if the event is extremely unlikely in the present state of technology. When, however, hydrologists working in the field of remote sensing explain to their "classical" colleagues that data obtained from remote sensing can significantly improve our knowledge of precipitation, runoff, evaporation, soil moisture, snow, etc., the classical hydrologist, typically, accepts the information but continues to do what he always did, i.e. to squeeze out information from often more or less useless conventional field data by the most sophisticated mathematical techniques. Many hydrologists still seem to believe in the possibility of overcoming the "garbage in-garbage out" problem with the aid of mathematics instead of looking for better "garbage in".

0022-1694/88/$03.50

© 1988 Elsevier Science Publishers B.V.

240 In the fifties the Swiss watch making industry, in the sixties the German photographic industry, in the seventies the European electronics industry swept away the new developments - are hydrologists going to do the same in the eighties? ~'I find it difficult to believe" wrote Eamonn Nash (pers. commun.) recently " t h a t we should enter the third millenium after Christ measuring rainfall in little buckets and guessing evaporation". The crucial problem in hydrology, in contrast to hydraulics, is the fact t h a t we almost always do not have adequate data to describe quantitatively a hydrological process with sufficient accuracy. To overcome this inadequacy we usually apply one of the following remedies: (a) collection of more, and more reliable conventional data; (b) application of more sophisticated mathematical techniques; and (c) use of new data acquisition techniques such as remote sensing. The difficulty with (a) is that one has to wait for what are usually still point measurements while hydrologists are usually interested in areal values. Remedy (b) may extract more information from available data but it does not help to discover the type of data inadequacy so that it may be corrected. These techniques may sometimes introduce errors of the same order of magnitude as those they intend to correct (e.g. Kalman filters, due to their complex mathematical structure, are used merely in combination with rather simple hydrological models and have to assume certain error distributions (e.g. normal distribution for measurement errors) which is not correct for most hydrologic measurement devices). The third remedy (c) which is relatively new, while it does not solve the problem of inadequacy of hydrological data, shows far more promise than approaches (a) and (b) - judging from what has already been achieved. This is due to the fact t h a t RS techniques offer: (1) area measurements instead of point measurements; (2) all information is collected and stored at one place; (3) rather high resolution in space and/or time is achieved; (4) data becomes available in digital form; (5) data acquisition devices do not interfere with the process being observed; (6) information may be obtained from remote areas of the earth where otherwise no measurements would be taken; and (7) after installation of RS networks, observations on hydrological quantities can be obtained quite cheaply. There are however also significant drawbacks in obtaining and using RS data. Some of these will be discussed later. However, the main difficulty usually lies in calibrating the electromagnetic signals in hydrological terms. REMOTE SENSING PLATFORMSAND SENSORS Remote sensing (RS) is based on the observation of a target or a process from a distance, in contrast to ~'in situ" measurements in which measuring devices are in touch or immersed in the observed system and/or process. For hydrological investigations, electromagnetic energy is the most important medium through which the observations are made and it only will be discussed here.

241 Hertz

Frequency I 0s

1012o 1011~ 10'1s 10~7 10'1s lOllS 10114 10 t13 1(~12 10111 10110 10~S

I I I ~:>, t

I;1

~;>, I

x

I [

o

E

c

L

]

o

~

I

E

lnm

1/Jm

>

I

Meters 10L111{~-1o101-s lOl-S 10l-7 lol-s 101-, 101-4 101-3 1

10-2

10 [-,

0

i

lmm

m

"1- rr

10 I1

'1

10 =

Wavelength

Transmission through atmosphere

L Total X-ray gamma imaging ray counts, Gamma ray spectrometry

per

second)

II I~t~

[

[

1017 lOle ( c y c l e s

Atomic absorption speetrophotometry, Mechanical line scanning

Imaging, Radiometry, single and spectrometry, multi-lens thermography cameras, various film emulsions, Multispectral photography

Passive microwave radiometry, Radar ~maglng

Electromagnetic sensing

Principal techniques for environmental remote sensing

Fig. 1. Hydrologically relevant portion of the electromagnetic spectrum(fromRitchie and Engman, 1986). A distinction is made between *'passive" and '*active" RS. In the former only natural energy emitted from the target is measured, while in the latter artificially generated signals are transmitted and the proportion of the return signal is measured. Conventional air photography belongs to the passive techniques while weather radar applies to the active. In principle, the complete electromagnetic spectrum (Fig. 1) could be used for RS, since electromagnetic radiation is emitted from all bodies having temperatures above absolute zero. The frequency (or wavelength) distributions are determined by the physical characteristics of the bodies (water, soil, vegetation, clouds, etc.). Similarly, a combination of signals observed in various spectral bands from a target under consideration allows, with more or less accuracy, an inference of which type of body is observed. Sensors have been developed which provide information in a specific spectral band, e.g. visible, infrared, near infrared, microwave, etc. The electromagnetic spectrum is shown in Fig. 1 along with some of the hydrologically relevant spectral bands. In the lower portion of Fig. 1 the transmission of the various frequencies through the atmosphere is shown such that white colour means high transmission while black means no transmission. The white sections are called '*atmospheric windows". Frequencies appearing completely black cannot be used for RS from space. RS-Platforms to which the RS sensors are attached include (Farnsworth et al., 1984): (1) ground observation platforms (masts, towers, vehicles); (2) balloons; (3) aircraft and remotely piloted vehicles; (4) rockets; and (5) satellites. We distinguish between polar orbiting and geostationary satellites

242

~11~

I

1

I

~

METEOSAT [ESA)

"~LANDSAT

4

GOES-F GOES-E

GOES-W 135 ° W

Fig. 2. Operational earth observation satellites, put in orbit to observe global weather patterns, geologic features and agricultural productivity (adapted from NESDIS programs, NOAA satellite operations, edited by J.H. McElron, March 1985).

(Fig. 2). While polar orbiting satellites follow elliptical orbits (crossing near the poles) geostationary satellites are positioned always above the same point of the earth (e.g. Meteosat 0° longitude, 0° latitude). There are several different RS sensors providing hydrologically relevant information: (a) Aerial photography, which is the most common form of RS, uses the visible (VIS) and near infrared (NIR) regions of the electromagnetic spectrum. (b) Scanning radiometers use a revolving or oscillating part of the instrument (e.g. mirror) to scan the earth beneath, thus building up strips of data (Fig. 3), while being carried along the flight path (of an aircraft or satellite). (c) Spectrometers, in which the incoming radiation is selected and disposed by means of prisms, gratings, mirrors or filters to provide multispectral data for detailed spectral signature analysis. (d) Microwave radars, which are active sensors in contrast to the previously discussed devices. They measure the reflected echoes of radiation emitted from the device itself. Due to their cloud penetrating capability they are suitable for all weather earth observation. Among the many different spectral bands by which data can be acquired by the various sensors, the visible (VIS), infrared (IR), water vapour (WV) and microwave (MW) spectra are most important, the latter both, passive and active. In Table 1 several hydrologically relevant platforms and sensors are presented along with the relevant spectral bands.

243

Opt i

;can mirror oscillates

185 km (Landsal)

NORTH WEST

/

~

6 Lines/scan/band

/___~EAST/ Pathof /spacec~/aft t rave,

SOUTH

/

/

Fig.3.Satellitewithscanningradiometer(polarorbiting). Since all information is obtained in the form of electromagnetic signals, the first task is to evaluate which type of signal is most suitable for which type of hydrological parameter or variable. In any case a calibration procedure with the aid of ground t r u t h is necessary. A n o t h e r feature which has to be carefully observed is the resolution in time and space which can be obtained with different equipment. Rapidly varying processes in small areas require high resolution, in both time and space. This is seldom available. However, ground-based radar measurements yield rainfall data with a high resolution in time and space. Air photography and some types

12 h 1 ~ 18 d

2.5d*

30rain (!)

Satellites: NOAA series, TIROS N Landsat

SPOT

GOES, GMS, Meteosat

Spectral bands

10 20 20 20 5

x x x x x

10m (!) 20m 20m 20 m 5kin (SSP)

900 x 900 m TM 30 x 30 m (MSS 54 × 76m)

VIS, IR VIS: 0.454}.52 #m (blue) 0.524}.6 pm (green) 0.634}.69 #m (red) 0.76-0.9 pm (NIR) 1.55-1.75 gm (IR) 10.4-12.5#m (therm. IR) 2.08~2.35 gm (IR) 0.514).73 ttm 0.5&59#m 0.61-0.68 #m 0.794).89 #m (NIR) VIS, IR, WV

[ Can be adapted to the hydrological problem ] X band (3 cm) Less t h a n C band (5 cm) 1 × lkm S band (10 cm)

Space (pixel size)

* This good figure can be achieved by tilting the sensors. VIS - visible; NIR near infrared; IR = infrared; WV = water vapour.34

Few minutes!

Time

Resolution in:

Air photography Radar

RS platform

Information on remote sensing platform and sensors

TABLE 1

Geo-synchronous

Polar orbiting

In orbit: Polar Orbiting Polar orbiting

Ground based

Remarks

t~

245 of satellites (e.g. Landsat or SPOT) produce data with high resolution in space but low in time. Other platforms (e.g. GOES, GMS or Meteosat satellites) produce high resolution data in time (0.5 h repetition rate) but low resolution in space (5 × 5 km pixels in the infrared band; a "pixel" is a picture element). MATHEMATICALMODELSUSING RS INFORMATION Most "classical" mathematical models used in deterministic hydrology use the theory of linear systems. An input, I(t), (e.g. effective rainfall) is transformed into an output, A(t), (i.e. flood hydrograph) through a transfer function, h(t), (i.e. a unit hydrograph). The procedure is expressed by the convolution integral:

A(t) = i I ( t - z) h(O dr

(1)

,J

0

The transfer function contains model parameters which can be evaluated either by a calibration procedure using one or more pairs of I(t) and A(t) functions, or by using previously established relations with some significant catchment characteristics. Such models are, typically, "lumped" with a single input and a single output, both being hydrometeorological variables. The use of such models requires: (1) measurement of model input; (2) development of the mathematical structure of the model; and (3) estimation of the model parameters either by a calibration procedure (e.g. as in the unit hydrograph method) or by estimating the parameters from catchment characteristics. If data from RS sources are to be used, in principle, the same procedure is to be followed. There are, however, some significant differences. The main problem consists in the fact t h a t R S sensors never measure hydrological data. They can measure only electromagnetic signals in certain spectral bands emitted from bodies on the earth or in the atmosphere (passive RS) or reflected from such bodies (active RS). The hydrologist using RS data is faced with the fact that he usually does not have information, how these electromagnetic signals can be converted into hydrological information. Sometimes, but seldom, this task is rather easy; the areal extent of snow cover on the earth can be measured with the aid of a sensor which measures thermal IR frequencies (if the sky is cloud-free). This is possible because snow is usually colder than the environment and because thermal IR signal intensities represent temperatures. If, however, there are clouds, thermal IR measures the cloud top temperature which may be used as an indicator of rainfall potential. This shows that the same electromagnetic signal may be used for the estimation of completely different hydrological variables. The hydrologist intending to build a mathematical model incorporating RS information has to find out first, which observed electromagnetic signal (i.e. frequency or "spectral band" e.g. visible, IR, etc.) is relevant for which hydrological variable. This is a difficult task since the knowledge available in this field

246

is not much advanced yet. Often one just assumes such relationships, builds a model accordingly and checks how well it functions. Perhaps the most simple model structure which could be considered would be a regression type model connecting the RS signal (from one spectral band only) directly with the hydrological variable. In fact, at the start of RS applications in hydrology such models were used and sometimes still are. A little more complex model would follow the theory of linear systems transforming model input (again from one spectral band only) into output with the aid of the convolution integral, eqn. (1), which requires a transfer function. This tranfer function has to be specified by the model builder and calibrated with simultaneous input (electromagnetic) and output (hydrological) data. This procedure is analogous to the Unit Hydrograph technique. This method represents a single input-single output model of the lumped type. It has been found that often the information obtained from one spectral band only (e.g. IR) is not sufficient for the estimation of a hydrological variable, but that the simultaneous information from two spectral bands (e.g. IR plus VIS) contains enough information for this purpose. In this case a model should be constructed which transforms simultaneous information from two electromagnetic input variables into one hydrological output variable. This model type represents the simplest form of a multiple input-single output model of the lumped type, an example of which is given later. In several cases input data are used from up to seven spectral bands (e.g. Landsat TM) as model input which requires much more complex multiple input-single output models.

I

Input

f

data

/ ~

from:J

"''--.~,

~ p l a t f o r m

L~'l-.~-~-

/~-~

~

C.eieosot:

I n f o r m a t i o n from RS sources: Spectral band :

1, 2

. n (e.g. n = 7 (Londsot})

Pixel

:

1.2

. m {e.g. m = 10 000 p i x e l s of I sq. km each (NOAA)}

increment:

1,2

--Time

T (e.g. 60 intervals of

1/2 hour each ( M e t e o s a t ) }

F i g . 4. R e m o t e s e n s i n g d a t a f o r h y d r o l o g i c a l m o d e l s .

247

Since lumped models cannot take into account the resolution of hydrometeorological variables in space (e.g. within a river catchment), it would be advisable, if high model accuracy is required, to use distributed system models in connection with RS data, provided these are available with the required spatial resolution. A complex model should make use of all information available with a high resolution in space and time. Figure 4 gives an indication of this situation, where each sensor of a platform collects data for each pixel in the catchment area. If, for example, a geostationary satellite takes observations every half hour in 3 spectral bands with pixel size 5 x 5 km, on a catchment of 25,000 km ~, for an event of 3 d duration, the following amount of information has to be processed: 3 spectral bands x 1000 pixels x (3d x 48 images d 1) = 432,000 pieces of data. Furthermore, a model might have a separate transfer function for each pixel and each sensor, i.e. in the above example 3000 transfer functions. After many centuries during which hydrologists were faced with the problem of not having enough information obviously in this case there is a surplus of information, part of which is obviously not needed. The question therefore arises, how much information is actually needed and what type or structure of mathematical models would be appropriate to handle the data. Certainly, a single input-single output lumped model would not adequately exploit the available information. One should use distributed models which make full use of the space resolution of which the technology is capable. Furthermore as the information from one spectral band would not be sufficient for most hydrological purposes, more spectral bands should be used, thus requiring multiple input-single output models. In order to reduce the immense amount of data, groups of pixels (segments) with hydrologically similar features could be lumped together (Groves et al., 1985). The combination of data from RS and GIS (Geographical Information Systems) sources obviously is of advantage. In any case, the development of remote sensing-based hydrological models is still in its infancy. Very little work in this field has yet been done (Groves et al., 1985; Fortin et al., 1986; Schultz, 1986). Not too much has changed since Peck (1981) wrote " . . . most hydrologic models . . . do not have a significant potential for using RS observ a t i o n s . . . Hydrologic modeling can be improved through the development of a new generation of models or subroutines for existing models which recognize the characteristics of the new RS capabilities". ESTIMATION OF MODELPARAMETERSBY REMOTE SENSING While the structure of scientifically based hydrologic models should be transferable from one river basin to another, the model parameters must be recalibrated for each catchment. If lumped models are used, these parameters

248

usually do not have much geophysical significance. In the next chapter an example will be given of such a model using RS data for parameter determination as well as for input estimation. For more complex models, which would normally be distributed, and which would reflect geophysical processes more faithfully, the model parameter values may be obtained from catchment characteristics measured from maps, except of course that one could not in this way obtain any indication of seasonal variation. Remote sensing by air photography or satellites may be very useful in such cases. In a study by a group of hydrologists from various U.S. institutions (Rango et al., 1983) it was shown t h a t the accuracy of land-use classification obtained from Landsat imagery was around 950, which is at least as accurate as could be obtained by conventional means. Furthermore it was shown that for watersheds greater than 30 km 2 the Landsat approach was more cost-effective. The high resolution in space of SPOT (10 × 10 m) and Landsat 5 observations (Table l) enable detailed information on soil type, soil cover, vegetation, land use, drainage density and other factors to be obtained, thus permitting the use of high resolution distributed system models. If grid cell models are used, it is advisable to choose the cells in the same size as the RS pixels or multiples thereof. The important parameter of slope, whether of the river channels or overland in the catchment, may be obtained with the aid of SPOT data, since this satellite may be tilted to make observations from different angles. The RS techniques for model parameter estimation take advantage of the high resolution in space while the usually corresponding low resolution in time (Table 1) is not relevant since model parameters do not change rapidly. Seasonal changes of model parameters can, however, be determined by multitemporal RS techniques. Methods identifying hydrological model parameters depending on catchment characteristics with the aid of RS data follow usually two consecutive steps. In a first step it is necessary to find out, which spectral band or combination of different bands from which sensors (and platforms) is relevant for the estimation of these parameters. In a second step a methodology has to be developed which allows to manipulate the RS data such t h a t the result provides the hydrologically relevant information. Figure 5 (left) shows the result of the manipulation of data from four different spectral bands of Landsat Fig. 5. L a n d - u s e c l a s s i f i c a t i o n (left) a n d v e g e t a t i o n i n d e x ( r i g h t ) in t h e R u h r R i v e r B a s i n ( G e r m a n y ) as b a s i s for h y d r o l o g i c m o d e l p a r a m e t e r e s t i m a t i o n ( L a n d s a t TM, A p r i l 25, 1984). (left) S u p e r v i s e d l a n d use c l a s s i f i c a t i o n . Colours: 1 w a t e r ; 2 - forest; 3 = c i t y / i n d u s t r y ; 4 residential; 5 - bush es; 6 - p a s t u r e ; 7 - c r o p l a n d ; a n d 8 = b a r e soil. ( r i g h t ) P a r a m e t e r s c h a n g i n g w i t h v e g e t a t i o n m a t u r i t y . O 0.5 - no v e g e t a t i o n ; 0.5-0.7 - n a t u r a l v e g e t a t i o n ; a n d 0.7 1 agricultural vegetation. Fig. 6. R a i n f a l l i n t e n s i t i e s ( i s o h y e t s ) from g r o u n d t r u t h ( r i g h t ) v e r s u s c l o u d t op t e m p e r a t u r e s from s a t e l l i t e i m a g e r y (left). S o u t h e r n G e r m a n y ( R h i n e a n d D a n u b e R i v e r C a t c h m e n t s ) . C ol ours : g r e e n = w a r m , > - 30°C; l i g h t b l u e = m e d i u m , - 30 to - 40°C; d a r k b l u e = cold, 40 to - 50°C; a n d m a g e n t a = v e r y cold, < - 50°C.

249

Time: 6 Aug. 1978, 4:55 pm

6 Aug. 1978, 4:45 - 5:15 pm

Time: 6 Aug. 1978, 5:55 pm NOAA Infrared Image

6 Aug. 1978, 5:45 - 6:15 pm Isohyets from Ground Truth

250

(a)

(b)

l

iProbobillty

,

w~--

~

~--

~ ---•

~

~

_/. , ~.

ba r e soil

f l/ ,-

/- b u s h e s

_ os oro

-'~.~,~r

7A~,J- f o r e s t

Fig. 7. Land-use classification with the aid of satellite data from two spectral bands. (a) Clusters of pixel reflectance intensities in a two-dimensional spectral feature space; and (b) probability density for eight classes of land-use in a two-dimensional spectral feature space.

TM. The image presents the areal distribution of eight different classes of soil type, land cover and land use in the Ruhr river catchment area in Germany. The different land-use classes are coded in "false colours". In the north one can see the city of Soest, further south areas which are mainly cropland and bare soil, interrupted by a highway and occasional residential areas. The black area represents the Moehne lake dammed up by a concrete dam. F u r t h e r south the land cover consists mainly of forests and bushes. The image on the right of Fig. 5 shows the same scene but the RS data are manipulated in a different way (by mathematical operations using only two spectral bands) in order to identify the vegetation index relevant for runoff particularly from agricultural areas. Since the images were t aken in April we still see several fields without vegetation, a situation which will change significantly during the summer, thus changing also the overland flow conditions. Neither image shown in Fig. 5 represents a direct Landsat image in one spectral band. The images of Fig. 5 were obtained as composite images by combining the information from four (two) different spectral bands with the aid of a special mathematical procedure: Different land-use types can be discriminated in the so-called spectral feature space, i.e. from simultaneous consideration of information obtained from two or more spectral bands. This is due to the fact that different objects reveal different but typical features in the various spectral bands. Caused by object variations and noise they do not appear as points but as a randomly distributed cluster in the feature space. Figure 7a shows a two-dimensional spectral feature space with observed intensities (by sensors) of reflectance obtained from a large number of pixels of an image. The statistical approach used in order to discriminate various land-use classes in the feature space is the so-called ~'supervised classification". In this technique first the number and type of land-use classes of interest are specified

251 (e.g. eight in Fig. 5a), then typical areas representative for each class are selected either on maps or in nature (training areas) and the corresponding pixels of the image identified in all relevant spectral bands. Their statistical class characteristics are then determined in a multidimensional feature space. Under the assumption that the probability function for the pixels in each land-use class is represented by a multivariate normal density function it is possible to compute these functions as well as their parameters (means, covariances, etc.). Figure 7b shows the result of this procedure for eight land-use classes in a two-dimensional spectral feature space (Lillesand and Kiefer, 1979). The land-use classification for each pixel can now be performed with information contained in Fig. 7b, i.e. the probability obtained for a pixel to belong to any of the defined land-use classes is computed with the aid of the known probability functions (Fig. 7b). The pixel then is classified into that class for which this probability is maximum. For this decision a maximum likelihood estimator is applied. The use of composite images as shown in Fig. 5 for hydrological purposes is manifold. The first and probably best-known technique was presented by Ragan and Jackson (1980), who used such images to identify the curve numbers CN in the well known SCS model (developed by the U.S. Soil Conservation Service) for the computation of runoff volumes from rainfall and catchment characteristics. These curve numbers depend on soil type as well as on land cover, vegetation and land use, all of which were identified with the aid of satellite imagery. In the hydrological model the curve numbers combine the effect of parameters representing interception, infiltration and surface storage processes. Other models use information as given in Fig. 5 in a different way, e.g. by inclusion of more information such as drainage density or seasonal parameter variations obtained from multitemporal imagery. A disadvantage of models like the SCS model consists in the fact that the information of composite imagery is used as lumped data for the whole catchment (in form of curve numbers), although this information is available with a high resolution in space. In the research program of the author's team a distributed system model is developed, taking advantage of the high spatial resolution of Landsat TM imagery (Fig. 5). This way it becomes possible to use the relevant parameter values for each pixel within a catchment thus, hopefully, increasing the accuracy of model performance. The application of such a model for real-time flood forecasting is presented in a subsequent chapter. A REMOTE SENSING-BASED LUMPED SYSTEM MODEL As discussed in the chapter on mathematical models we can expect, in the long run, to see rather complex RS based hydrological multiple input-single output models of the distributed system type making full use of the information of the various spectral bands and the high resolution in space and time.

252 At the beginning of a development one usually starts with simple techniques. Therefore also in the field of RS-based hydrologic models most approaches applied lumped system models using data from just one spectral channel as input. Even this was not too simple as the following example shows. In the field of water resources planning we are often faced with the problem of design of a water management system where no or almost no hydrological data are available. For reliable designs we need, however, long time series of hydrologic data (e.g. monthly runoff volumes for reservoir design purposes). While, during the planning period only hydrological data for one or two years can be collected (which is too short a period for system reliability assessment) it is possible to overcome this problem of too short data time series with the aid of longer time series of satellite information. A mathematical model had to be developed which relates satellite information to runoff values. At the start of the model development no such model existed and it was not clear whether existing satellite data would be suitable for this purpose at all. The required long time series of satellite information was available from the NOAA series ( ~ 15 years) and IR data seemed to be most promising for this purpose. In the chapter on satellite-based mathematical models various approaches of differing complexity were discussed. Here a simple model (single input single output lumped model) was chosen since there were still so many uncertainties. The first problem to be solved arose from the fact that only part of the IR information was relevant (cloud top temperatures < 45°C) and that the influence of the relevant temperature ranges obviously increased with decreasing temperatures (of cloud tops) in a nonlinear mode. After many trials the RS data from the IR band were manipulated in order to form the model input B, according to eqn. (2) where the nonlinearities are quantified by the asj-values (Striibing and Schultz, 1985): B~ = P

k=

1j

1

where: Cjk, - fraction of catchment area in temperature range j on image k of day i: asj - weight coefficient for temperature r a n g e j in season s; p - number of satellite images per day (e.g. 2 from NOAA): and r = number of relevant temperature ranges. B, may be interpreted as mean daily temperature-weighted cloud cover index over the catchment area of interest. Since simple correlation models were not successful the input obtained by eqn. (2) was transferred into runoff (daily indicators) with the aid of the convolution integral, eqn. (1), in the following form:

Q(t)

i B(t-

r) h(z) d~

0

or in discrete form (runoff indicator for day m):

253

Fig. 8. Meteosat IR image of Europe showing rain clouds.

Q,, :

~ z

B,, ,

h, At

(3)

I)

w h e r e n = n u m b e r of o r d i n a t e s of the discrete t r a n s f e r f u n c t i o n h. C o m p a r e d to c o n v e n t i o n a l h y d r o l o g i c a l l u m p e d models h e r e a difficulty consists in the fact t h a t in a c a l i b r a t i o n p r o c e d u r e not only the s y s t e m f u n c t i o n k, in eqn. (3) h a d to be d e t e r m i n e d but also the w e i g h t i n g coefficients asj of the i n p u t v a r i a b l e in eqn. (2). F o r this p u r p o s e a trial and e r r o r t e c h n i q u e was applied u s i n g as q u a l i t y c r i t e r i a the m i n i m u m s q u a r e d e v i a t i o n b e t w e e n c o m p u t e d and o b s e r v e d runoff. It s h o u l d be n o t e d t h a t the s y s t e m f u n c t i o n as well as the asj v a l u e s are different for four s e a s o n s of the year. T h e r e q u i r e d m o n t h l y r u n o f f v a l u e s w e r e o b t a i n e d by a d d i t i o n of the r e l e v a n t daily r u n o f f i n d i c a t o r s of eqn. (3). The s i m u l t a n e o u s g r o u n d t r u t h and s a t e l l i t e d a t a of a s h o r t period (1 year) were used for model c a l i b r a t i o n . After this it was possible to r e c o n s t r u c t r i v e r flows, b a c k w a r d s in time, on the basis of s a t e l l i t e d a t a alone, i.e. from p r e v i o u s l y r e c o r d e d IR values. F i g u r e 8 shows an IR i m a g e of E u r o p e o b t a i n e d from M e t e o s a t . T h e model uses this type of i n f o r m a t i o n in digital form.

254 Mean monthly runoff

{m3/s) 35,0

observed"

30.0

25,0 compL 20.0

Calibration perlo( 15.0

10,0

5.0

10,2

2

,

t

6 ,9,~6

8

,0,2

2 ~

~

6

8

~0

Month

1977 - -

Fig. 9. Monthly runoff values computed with the aid of IR satellite imagery of NOAA (Save River, France).

The model was applied to four catchments in Sout hern France. Figure 9 shows an example of the results for the river Save for the two years 1976/77. The simultaneous ground t r u t h and satellite data of the first year were used for model parameter calibration. Then the whole time series of monthly runoff values was computed with the aid of satellite data alone (period after 10/1976). Although the results are by no means perfect, this reconstruction of historic runoff values with the aid of satellite data may be a great help for planning purposes. Some of the deviations in Fig. 9 are due to the fact t hat NOAA satellites produce only two images per day, thus sometimes missing rainfall events. If GOES or Meteosat data were used (48 images per day) this problem would be overcome. The time series with available satellite imagery are, however, much shorter for these satellites. A DISTRIBUTED SYSTEM MODEL FOR FLOOD FORECASTING USING REMOTE S E N S I N G

DATA AS INPUT

While in the previous chapter a lumped system model was demonstrated for the estimation of historic runoff sequences, in this section a distributed system

255

%

SIZe

29

275

270

25 L

72 70

65

60

~o

Fig. 10. Isohyets from radar rainfall measurements, Gfinz River catchment. Grid of polar coordinates (7-8-1978, 14.45-15.00 h).

model applied to real time flood forecasting will be shown. The model input consists of radar rainfall measurements obtained from a ground based C-band weather radar of the German Weather Service.

Hydrologic model The mathematical model consists of three component models: (a) model transforming the observed radar echo into rainfall; (b) stochastic model for real-time rainfall forecasting; and (c) deterministic model transforming the (observed + forecast) rainfall into a runoff hydrograph (rainfall-runoff model). The model (a) works on the basis of the classical radar equation (Attmannspacher and Riedl, 1979). The model (b) is a stochastic model which is based on conditional probabilities for various rainfall types. On the condition of how much it has been raining until the time of forecast, the amount and duration of rainfall until the end of the precipitation event is forecast. The model (c) is a linear distributed system model working on a grid cell

256 Mean romfall {mm) 3,2 ¸

2,~

1,5

J

0 ~--01

16 08.

32

LI8

+ 08.08

61z,

~ 09.08

tTime

Time (hours) 80 95 112 1" 10,C8. -' 11 08 78

of forecgsf

Runoff { m3/s)

80.

64/

~

',r-observed

'Xt'~, f ° r e cas

z.8 32

0

" t,

0

4-0708

ld6

i 5~

32

418

.o8.o8.

~ o9.o~.

8'0

{

Time (hours) T I lID 9T5 115

~o.o8.. ~ ,.o8~

~LTime of forecos't Fig. 11. Flood f o r e c a s t (G/lnz River) based on radar rainfall measurements (at the end of the storm).

basis with polar coordinates of which the radar forms the center. The model makes use of the high resolution in space and time of radar rainfall measurements-(Table 1!). The pixel size is 1 km × 1° of arc (Fig. 10), the time scale can be adapted to the problem. In the example 0.25 h intervals were chosen. The model involves separation and evaluation of the two types of storage effects in a drainage system, i.e. travel time of a flood wave (translation) and storage attenuation (retention) in two consecutive steps for each grid cell (Schultz, 1969). The model input, rainfall, obtained from echo measurements of the weather radar after calibration, is superimposed on the catchment area and the grid cells as shown in Fig. 10 every quarter hour (Klatt and Schultz, 1985). The model then computes a flood hydrograph forecast in real time.

257

J ~ean ramfall (ram) 32

Pu = probobd4y of non-exceedance

~bserved /

2~

"~ I Jl

12

~ forecast ralnfalt I ~ p =099

Iiii,

O.B

l]

o u" °.9°

Time(hours)

I

O-

18

32 ,~ o8o~

,,L--o,~.o~

~8 5,. ; 09.08.

}

so ~oo8.

96 11~ ~ 11o~.%

LT~rne of forecast

80-'

Runoff

(m3/s)

//~

observed

5/.-

o,9o Z.5 32

i

15 0

I

I

I~

Time (hours}

312

i

/18

I

d.~

J'-- 07' 0B----J~ - OB,08...... t 09 08 ~ - ~--Tlme of forecast

I

810

i

I

IL

9'5 112 10.08. ~' 11.08,78 -

Fig. 12. Flood forecast (Gtinz River) based on radar rainfall measurements and forecast rainfall.

Results F l o o d f o r e c a s t s w i t h t h e a b o v e m e n t i o n e d model ( w i t h o u t the rainfall forec a s t i n g part of the model) made at a time after the end of rainfall yielded good results in a river b a s i n in s o u t h e r n G e r m a n y (Fig. 11). If, h o w e v e r , the f o r e c a s t h a s to be made during the storm, the rainfall f o r e c a s t i n g r o u t i n e has to be used w h i c h renders the results u s u a l l y less a c c u r a t e (Fig. 12). FURTHER FIELDS OF APPLICATION OF REMOTE SENSING IN HYDROLOGY It w o u l d n o t be possible, here, to give a c o m p r e h e n s i v e o v e r v i e w of all possible RS a p p l i c a t i o n s to h y d r o l o g y . The idea is rather to give an introduction to basic principles and t h e n a few e x a m p l e s of h o w RS can be applied to

258 the solution of some hydrological problems. This has been done in the preceding chapters for model parameter estimation, reconstruction of historic monthly runoff values and real time flood forecasting. Now just a few of the more important RS applications in other fields of hydrology shall be mentioned very briefly. More information is available in the literature (Farnsworth et al., 1984; Herschy et al., 1985: Goodison, 1985; Johnson, 1986~ Collinge, 1987).

Rainfall A very promising RS approach for rainfall estimation, radar rainfall measurements, were discussed in a previous chapter. In Western Europe the various national weather services are presently setting up a continental weather radar system on the basis of C-band radars. In the United States a similar project (Nexrad) is established based on S-band radars (Table 1). Several techniques have been developed using satellite data for the estimation of area and intensity of rainfall (Barrett and Martin, 1981). Images from geostationary satellites (Table 1) in the visible and infrared bands are mostly used. As an example of a correlation between IR imagery and rainfall intensity Fig. 6 (colour plate) shows infrared images of a NOAA satellite and simultaneous isohyets derived from conventional rainfall measurements on the ground in southern Germany. The relationship between IR data and rainfall intensity cannot, however, be expressed by simple correlation models. The methods applied consist of ~cloud indexing" and ~life history" techniques as well as bispectral methods using visible and infrared data as model input. An operational technique for convective rainfall estimation based on GOES infrared data was developed in the U.S.A. (NOAA-NESDIS) by Scofield and Oliver (1977). An interesting combination of radar and satellite data for rainfall estimation can be found in the British FRONTIERS project (Collinge, 1987).

Evapotranspiration Evapotranspiration (ET) is one of the most significant parameters of the water balance, and one of the most difficult to measure. Even with the aid of RS data no direct information of ET can be obtained. Surface parameters, however, which affect the moisture flux from the land surface, such as solar radiation, albedo, vegetation cover, surface temperature and soil moisture can be estimated from RS information. Many of the approaches using surface temperature data work on the basis of the energy balance equations and use several readily available meteorological data. Some more recent methods use both, surface temperature and moisture data. The mathematical models following this principle with the aid of RS data, use temperature estimates obtained from thermal IR and soil moisture from microwave (MW) observations. IR data are usually obtained from satellites,

259 while MW data are observed from airplanes since no MW sensors are flown on satellites which fulfill the requirements of hydrology. P a r t i c u l a r problems result from the fact that in applying such techniques we have to use different approaches for potential evaporation estimations, actual E T from bare soils and E T from vegetation canopies. For more information about the models the reader must be referred to the literature (Rosema, 1981; Seguim 1983). Soil moisture Soil moisture is a very important hydrologic parameter which is difficult to measure over larger (catchment) areas. Therefore the use of RS techniques for quantifying this parameter has been studied for over 10 years. Visible and IR data have been used, but the most promising approaches are based on passive and active microwave data. A major problem arises from the fact that RS sensors give information only on the top layer of the soil, while for hydrologic processes one is interested in the soil moisture down to about 2 m below the surface. Therefore two problems have to be solved: (a) estimation of soil moisture properties at or near the surface; and (b) inference from the information obtained under (a) to soil moisture profiles down to about 2 m. Passive microwave sensors measure the thermal emission from the soil which is proportional to the soil temperature. The constant of proportionality depends strongly on the soil moisture content. The MW emission depends on soil texture, surface roughness and vegetation. With the aid of VIS and near IR data these influences can be estimated in order to correct the MW observations for vegetation effects. It has been demonstrated t hat for various reasons a 21 cm radiometer is best for soil moisture observations, which allows a soil penetration depth of 2 5 cm. Active MW (radar) measures the backscattered energy of an emitted electromagnetic signal. Compared to passive MW incidence angle and roughness factors are more important whereas vegetation is less so. C-band radars (Table 1) seem to represent the most suitable wavelength for these purposes. Generally speaking MW applications are faced with the problem of space resolution versus sensor size. Small sensors (e.g. on Nimbus satellite) have a r a t h e r bad resolution (150km) while a good resolution can be obtained from sensors which are presently still too big to be flown on satellites. Therefore MW use is presently limited mainly to airplane or ground-based platforms. Models have been developed which correlate these measurements of the top-layer soil moisture with the moisture profile further below. Such models enable a soil-moisture profile time series to be computed from one known profile (at the beginning) and a following sequence of MW measurements, which is also important for rainfall runoff modeling. A lot of work has been carried out in this field and is described in the literature (Wang et al., 1983; Schmugge, 1987).

260

Groundwater At present there is no direct method available allowing to evaluate groundwater resources from RS data. Only indirect information can be obtained from RS sources such as: (1) likely areas for the existence of groundwater (GW); (2) indicators of the existence of GW; (3) indicators of regions of GW recharge and discharge: and (4) areas where wells might be drilled. These indicators are based mainly on geologic and geomorphologic structures or on multitemporal observations of (past) surface water and on the transpiring vegetation. Landsat VIS and IR data are preferred for these purposes. Soil water observations can be the basis for estimation of groundwater resources. Also multitemporal imagery in the thermal IR band indicating temperature changes may provide information on GW (areas containing GW are warmer than the environment in certain seasons of the year). Allewijn (1987) developed a regional groundwater flow system model in which he combines conventional data with RS data. In a first phase preliminary infiltration and exfiltration components are delineated with the aid of Landsat-based land cover maps. Many regional GW exploration studies (e.g. in Africa, Australia, India) made extensive use of RS, particularly Landsat satellite data (Farnsworth et al., 1984). Also methods which are standard practice nowadays can be considered as RS techniques such as measurements of electrical resistivity or by electromagnetic and magnetometric as well as seismic sensors.

Surface water RS imagery used as maps may serve as basis for the inventory of water bodies such as lakes, dams, rivers, swamps, flooded areas, etc. Multitemporal images can detect changes in the extension of water bodies. Multispectral air photography is used as well as satellite imagery with good spatial resolution (Landsat TM, SPOT). Possibilities of the estimation of historic runoff volumes (monthly values) with the aid of satellite information as well as real-time flood forecasting on the basis of weather radar measurements were demonstrated in previous chapters.

Snow and ice Water stored in snow and ice represents a valuable resource e.g. for hydropower production or irrigation during the melting season. Important parameters are the areal extent of snow cover and the snow water equivalent. The areal extent of snow can be observed by satellites in the VIS and NIR bands. Due to the dependency between space and time resolution (Table 1) for small catchments (< 200 km 2) Landsat data are advisable, for areas > 200 km 2

261 NOAA, and for areas > 1000 km 2 Meteosat data may be used. In the West of the U.S.A. snow cover monitoring by RS is already operational. The problem of VIS and IR wave lengths is, however, the fact t hat they cannot penetrate clouds. Therefore significant research effort has been devoted to the use of microwave measurements (active and passive) which can penetrate not only clouds but also into the snow. This gives rise to possibilities of estimating snow water equivalent. Here, however, various parameters influence the measurements such as grain size, wet/dry snow quality, etc., and the bad spatial resolution of satellite MW sensors has also to be considered. A more satisfactory RS technique for the estimation of snow water equivalents consists in the use of airborne gamma ray sensors. The "Snowmelt Runoff Model" (SRM) computes runoff volumes from snowmelt and uses as input snow cover data t h r o u g h o u t the snowmelt season (Martinec et al., 1983) which may be obtained from the above mentioned RS sources. These data are used in order to construct snow-cover depletion curves, which are used to estimate daily snow-cover extent. The model uses the location of snow cover within a catchment in applying a degree-day index equation for generating daily snowmelt depth. Temperature and precipitation measurements are also required for the application of the SRM. SEDIMENTS AND WATER QUALITY Sediment movement and deposition in water bodies as well as water quality parameters cannot, as yet, be monitored in a very detailed way by RS devices. There are, however, many applications for special problems. Thermal pollution as well as pollutants with temperatures differing from the water temp er a t ur e can be identified with the aid of airborne IR sensors and active MW, a technique which is already operational. Oil spill monitoring by RS (e.g, by multispectral scanners, laser or side-looking radar) is also an operational technique in some parts of the world. Algae, chlorophyll, aquatic life parameters, etc., can be observed by various RS devices since they change water colour, temperature or surface characteristics. Most applications of RS water quality monitoring can be found in lakes, dams, bays and estuaries. For rivers this is more difficult but airborne active MW devices have been successfully applied for the detection of polluters. Coastal sediment concentrations were measured for various concent rat i on classes in Fundy Bay with Landsat imagery (Munday et al., 1979). Total sediments as well as suspended sediments (e.g. in Lake Chicot) were observed with RS in lakes. The propagation of silts with the advancement of the Nile flood in the Assuan Dam was observed with the aid of Landsat imagery (Johnson, 1986). DATA AVAILABILITY,HARDWARE. SOFTWARE Data can be obtained from space agencies (e.g. ESA, NOAA), from public or commercial institutions disseminating RS data, or they have to be collected

262 (e.g. with the aid of airplanes and relevant sensors). In any case costs are involved, in some cases quite high costs (e.g. Landsat, SPOT). These data have to be processed either in analog or digital form. In the latter case which is much more suitable for hydrological purposes image processing devices are required in connection with host computers. For some applications PCs are appropriate but for more sophisticated techniques larger (and more costly) equipment has to be used. This can be done only if suitable software is available, which sometimes is sold together with the hardware but mostly hydrologists have to develop their own specific software particularly for the application in mathematical modeling (e.g. as model input or for model parameter determination). FUTURE PROSPECTS Looking at the development of hydrology during the last decades we can observe new and serious long-term developments, e.g. stochastic hydrology, and short-term fashions which disappear almost as soon as the new colours in ladies' fashions, e.g . . . . (I don't want to lose friends!) . . . The application of remote sensing techniques in hydrology certainly belongs to the long-term developments. We can trust th/tt every hydrologist starting his career after the year 2000 will be faced with RS in some way or other. RS application in hydrology did not start until about the year 1970 and is still in its infancy. We can see, however, the lines along which it will probably develop into the coming millenium. The most spectacular developments can be expected from RS platforms and sensors. In Western Europe and large areas of the U.S.A. there will be a rather dense network of weather radars allowing hydrologists to obtain rainfall intensities in real time with very high resolution in space and time for almost every river catchment. This may form the basis for permanent real-time flood and flow forecasting. For the 1990s great developments in space technology are planned. International space agencies will cooperate (NASA, NOAA, ESA) to establish a manned space station accompanied by polar platforms and satellites (such as the European Columbus programme). Such satellites as ESA's ERS-1, Canada's Radarsat or N-ROSS, AMSU, and TOPEX of the U.S.A. will carry advanced sensor systems of great relevance to hydrology, particularly microwave sensors. These sensors (active and passive) can penetrate clouds, thus making measurements ~'weather independent", and will be most useful for better estimation of soil moisture, evapotranspiration, snow water, precipitation and water quality. Presently satellite sensors offer good resolution either in space o r in time (Fig. 13). Sensors with good resolution in space (Landsat, SPOT) provide information on slow hydrologic processes such as snowmelt, ice, land use, model parameters. Those having a good resolution in time (GOES, Meteosat, GMS) allow the monitoring of dynamic processes such as rainfall, runoff and

263 TIME 5--~-1month

W O RSD

< i I'-

\

Z

,'X

,,,

T I RO S ~X"~

\\\\\"q \

D .J o ILl tY

2-

32-

X795

"..._ U i I " lira

I

I

t

I'~ ~

1

2

3

5

8 10

2

3

5

8 102

RESOLUTION

2

3

B 103

,I 2

3

5

8 10Am

IN SPACE

Fig. 13. Resolution in time and space of existing remote sensing devicesand expectedfuture trends. floods, but only for larger areas, since the space resolution is not high. What hydrologists need, however, would be a good resolution in space and time. Fig. 13 presents information about this dependency and the expected future developments (arrows). The solution of this problem may be overcome in either of two ways: (1) Geostational"y satellites (e.g. GOES, Meteosat) which already provide a good resolution in time should be equipped with sensors producing a much better space resolution. (2) Polar orbiting satellites (e.g. Landsat, SPOT) already providing a good resolution in space should be flown in pairs or multiples with a reasonable time shift, thus producing an acceptable resolution in time. If funding will not be cut too drastically, the second alternative seems to be the more realistic one. F u r t h e r m o r e we can expect information in more spectral bands from these sensors (e.g. 20 channels in platforms of NASA's EOS program). While hydrologists, unfortunately, did not have much opportunity to influence these technical programmes, they will certainly find a large and

264

challenging field of activity in developing better theory and better mathematical models on the basis of RS information. From existing RS data one could extract much more information than has been used until now (e.g. future multichannel, multitemporal, more sophisticated models). The information which we hope to obtain in the future from RS sources opens even wider and more promising fields of hydrologic activity, maybe within the framework of global research programmes such as IGBP or GEWEX. Despite the positive presentation of present and expected future developments in RS applications to hydrology, it should not be forgotten that at present there are barriers to a wider use of these techniques such as data access, nonavailability of hardware and/or software, lack of qualified personel, etc. These problems should and can be overcome by a joint effort of those developing RS techniques and those hydrologists who are expected t~ be the future users. The advent of remote sensing and space technology has added a new dimension to hydrology, offering opportunities which are not only very useful and progressive but, to the pleasure of the hydrologists involved, also most fascinating and rewarding. REFERENCES Allewijn, R., 1987. Regional hydrological systems analysis using remote sensing data and a geographic information system: applied to groundwater modelling of the Roermond area (the Netherlands). Proc. EARSeL Symp., Eur. Assoc. Remote Sens. Lab., EARSEL Publ., 18pp. Attmanspacher, W. and Riedk J., 1979. Radar area precipitation measurements as basic data for hydrological purposes. Proc. Symp. Workshop Digital Radar Reflectivity Process., Edmonton, Alta. Barrett, E.C. and Martin, D.W., 1981. The use of satellite data in rainfall monitoring. Academic Press, New York, N.Y. Collinge, V.K. and Kirby, C. (Editors), 1987. Weather radar and flood forecasting. Wiley, New York, N.Y. Farnsworth, R.K., Barrett, E.C. and Dhanju, M,S., 1984. Application of remote sensing to hydrology including groundwater. Int. Hydrol. Programme, IHP-II Proj. A.1.5, UNESCO, Paris. Fortin, J.P., Villeneuve, J.P., Guilbot, A. and Sesuin, B., 1986. Development of a modular hydrological forecasting model based on remotely sensed data, for interactive utilization on a microcomputer. In: A.J. J o h n s o n (Editor), Hydrological Application of Space Technology. IAHS Publ., 160:307 319. Goodisom B.E. (Editor), 1985. Hydrological Application of Remote Sensing and Remote Data Transmission. Proc. Hamburg Syrup., IAHS Publ. No. 145. Groves, J.R., Ragan, R.M. and Clapp, R.B., 1985. Development and testing of a remote sensing based hydrological model. In: B.E. Goodison (Editor), Hydrological Application of Remote Sensing and Remote Data Transmission. IAHS Publ., 145: 601~612. Herschy, R.W., Barrett, E.C. and Roozekrans, J.N., 1985. Remote sensing in hydrology and water management. Fin. Rep. of a Contract Between the Eur. Space Agency and the Eur. Assoc. Remote Sens. Lab. (EARSeL), Working Group 10, ESA Contr. No. 5769/A24/D/JS (Sc). Johnson, A.J. (Editor), 1986. Hydrologic Application of Space Technology. Proc. Cocoa Beach Workshop. Fla., IAHS Publ. No. 160. Klatt, P. and Schultz, G.A., 1985. Flood forecasting on the basis of radar rainfall measurement and rainfall forecasting. In: B.E. Goodison (Editor). Hydrological Application of Remote Sensing and Remote Data Transmission. IAHS Publ.~ 145:307 316.

265 Lillesand, T.M. and Kiefer, R.W., 1979. Remote sensing and image interpretation. Wiley, New York, N.Y. Martinec, J., Rango, A. and Major, E., 1983. The snowmelt- runoff model (SRM) User's Manual. NASA Ref. Publ. 1100, Washington, D.C., ll0pp. Munday, J.C., Jr., AlfSldi, T.T. and Amos, C.L., 1979. Bay of Fundy verification of a system for multidate Landsat measurements of suspended sediment. In: M. Deutsch, D.R. Wiesnet and A. Rango (Editors), Satellite Hydrology. Fifth Annu. William T. Pecora Mere. Symp. Remote Sensing, Sioux Falls, S.D., AWRA, pp. 622 640. Peck, E,L., McQuivey, R., Keefer, T., Johnson, E.R. and Erekson, J., 1981. Review of Hydrologic Models for Evaluating Use of Remote Sensing Capabilities, NASA CR 166674. Prep. for Goddard Space Flight Center, Greenbelt, Mld. Ragan, R.M. and Jackson, T.J., 1980. Runoff synthesis using Landsat and SCS Model. J. Hydraul. Div., ASCE, Vo]. 106 (HY 5). Rango, A., Feldman, A., George, T.S., III. and Ragan R.M., 1983. Effective use of Landsat data in hydrologic models. Water Resour. Bull., 19(2): 165-174. Ritchie, J. and Engman, E.T., 1986. Remotely sensed data for natural resources models. Environ. Conserv., 13(3): 203 210. Rosema, A.A., 1981. Thermal sensing of soil moisture, evaporation and crop yield. In: A. Berg (Editor), Application of Remote Sensing to Agricultural Production Forecasting. Publ. for the Comm. Eur. Communities, Balkema, Rotterdam. Schmugge, T., 1987. Remote sensing applications in hydrology. Rev. Geophys., 25(2): 148 152. Schultz, G.A., 1969. Digital computer solutions for flood hydrograph prediction from rainfall data. In: The Use of Analog and Digital Computers in Hydrology. Proc. Tucson Symp., 1:125 137, IAHS Publ. No. 80. Schultz, G.A., 1986. How does remote sensing information influence the structure of hydrologic models? In: H.W. Shen, J.T.B. Obeysekera, V. Yevjevich and D.G. Decoursey (Editors), Multivariate Analysis of Hydrologic Processes. Proc. Fourth Intern. Hydrol. Symp., Hsieh Wen Shen. Eng. Res. Center, Colorado State University, Fort Collins, Colo., pp. 450 464. Scofield, R.A. and Oliver, V.J., 1977. A scheme for estimating convective rainfall from satellite imagery. NOAA Tech. Memo. NESS 86, Washington, D.C., 47pp. Seguin, B. and Itier, B., 1983. Using midday surface temperature to estimate daily evaporation from satellite thermal IR data. Int. J. Remote Sensing, 4(2): 371 383. Strtibing, G. and Schultz, G.A., 1985. Estimation of monthly river runoff data on the basis of satellite imagery. In: B.E. Goodison (Editor), Hydrological Applications of Remote Sensing and Remote Data Transmission. IAHS Publ., 145:491 498. Wang, J.R., O'Neill, P.E., Jackson, T.J. and Engman, E.T., 1983. Multifrequency measurements of the effects of soil moisture, soil texture and surface roughness. IEEE Trans. Geosci. Remote Sensing, GE-21, pp. 44 50.