Maa.-57.290 Special Assignment in

photogrammetry, photo-interpretation and remote sensing

The application of remote sensing in rainfall monitoring

Helsinki University of Technology

Liu Quanwei

25 March 1996


Abstract

1 Introduction

2 An simple description of remote sensing

3 Monitoring the rainfall using radar

3.1 Spaceborne radar techniques

3.2 Ground-based radar techniques

4 Monitoring the rainfall using satellites

4.1 Visible and infrared techniques

4.1.1 Cloud indexing techniques

4.1.2 Thresholding methods

4.1.3 Life-cycle methods

4.2 Microwave radiometry

5 Conclusion

References


Abstract

This report reviews past and present uses of radar and satellites in monitoring rainfall where the need is for real-time information. The radar techniques include the use of spaceborne radar and ground-based radar. The satellite techniques include two methods, namely, the visible and infrared techniques and the microwave radiometry. The former is composed of three methods: cloud indexing techniques, thresholding methods, and life-cycle methods. The potential of combining visible, infrared and passive microwave as well as spaceborne radars for rainfall profiling is examined at the end of this report. Before studying the use of these methods for monitoring the rainfall, the basic theory of remote sensing is described.

1 Introduction

Water is one of the most prevalent substances in the Earth-atmosphere system and it is one of the most fundamental to the existence of life. Shortages of fresh water for domestic consumption, as well as suitable water for agriculture and industry have been a critical problem in most parts of the world. Water supply problems already restrict significantly the life of men and the economics of the entire globe.

The study of the water resource and its movement on the Earth is of very practical importance for life on the planet.

Rainfall is one of the most important part of the water resource. It is vitally important to quantify it accurately. Determining the spatial and temporal depth of rainfall input to the Earth is necessary for everyday management of water resources such as rivers and reservoirs, irrigation, and weather forecasting. It is also an essential component of scientific investigation of the hydrologic cycle, the global water balance and large-scale global atmospheric modelling.

Historically. the estimation of rainfall has been accomplished by relatively simple instrumentation that sample the rain by capturing a volume over a continuous or fixed time interval. This instrumentation, commonly referred to as a rain gauge provides a fairly accurate measure of point rates and depths of rainfall. The major shortcoming of this instrumentation is that the measurement is only at certain points. It has been well documented that rainfall on the Earth's surface varies greatly in both time and space. Although there are a vast number of rainfall gauges world-wide, they are not adequate to define the rainfall input for the most needs. The result of this is that rainfall can be measured relatively accurately for small areas with a network of rain gauges, but this approach is not practical for large areas, remote land areas of the globe and for oceans.

Recognizing the practical limitations of rain gauges, scientists have increasingly turned to remote sensing as a possible means for quantifying the rainfall input to the globe. It should be stressed, however, that remote sensing is at present, and will continue a supplement to, rather than a replacement for, more traditional methods of rainfall assessment. The measurement of rainfall by rain gauges is fraught with some problems, but those relatively simple instrumentation will long continue to provide the data against which rainfall assessments by other means must be adjusted. This report reviews past and present uses of remote sensing in monitoring rainfall where the need is for real-time information. It is divided into two subparts--subpart 1 concerning the use of radar, subpart 2 concerning the use of satellites. Before doing so, we devote a separate part 2 to give a simple description on the principle of remote sensing. So to supply a theory basis for this particular application of remote sensing.

2 An simple description of remote sensing

Remote sensing is the acquisition of physical data of an object without touch or contact. It concerns the collection of information related in some way to the Earth' natural resources or environment. Data is collected primarily by satellite and aircraft system involving measurements of the electromagnetic spectrum that can be used to characterize or infer properties of it in conjunction with localized ground-based surveys and measurement. The data are then processed by digital computer or optical techniques to extract information of value of interest. Different sensors can provide unique information about the properties of the surface or shallow layers of the earth.

The electromagnetic spectrum is the basis for all remote sensing. Remote sensing takes advantages of the unique interaction of radiation from the specific regions in the spectrum and the Earth.

There are four basic components of a radiation based remote sensing system (Slater, 1980; Curran, 1985; Cracknell, 1991)

1. radiation source (i.e. the Sun, radar);

2. transmission path (i.e. atmosphere, vegetation canopy);

3. target (i.e. water, soil);

4.sensor (i.e. multispectral scanner, photographic film.).

Each of these plays a significant role in either limiting or controlling what we can measure about the Earth's surface. They will be discussed in turn.

The radiation source most commonly exploited is the Sun. In this case, the characteristic that we measure is the reflected energy from the Earth, but in other applications we measure the energy emitted from the Earth's surface. These applications include thermal infrared and microwave remote sensing. The third commonly used radiation source is radar, in which energy from a limited region of the spectrum is propagated towards the Earth and the reflected or backscattered energy is measured. Remote sensing generally involves only one or several narrow regions or bands of the total spectrum at any one time.

The transmission path of the electromagnetic spectrum involves the atmosphere and this has a significant impact on which parts of the spectrum can be used. Specific gases in the atmosphere selectively affect the amount of energy that transmitted, and this leads to a concept known as an atmosphere window. An atmosphere window is a wavelength band in which the atmosphere has little or no effect on the intensity of the Sun's radiation or reflected radiation from the Earth. Particle matter such as smoke or dust can also affect the transmission path by scattering or absorbing radiation over the entire spectrum.

The target is the subject of any observation as well as any other objects within the field of view of the sensor. Modern remote sensing is based on interpreting measurable variation in spectral, temporal and spatial characteristics of the Earth. The spectral characteristics of the target are the unique spectral reflectance for specific Earth features. Considering the spectral reflectance of water, probably the most distinctive characteristic is the energy absorption at reflected infrared wavelengths. Locating and delineating water bodies with remote sensing data is done most easily in reflected infrared wavelengths because of this absorption property. The temporal characteristics are any factors that change the spectral characteristics of a feature over time. For example, the spectral characteristics of many species of vegetation are in a nearly continual state of change throughout a growing season. These changes often influence when we might collect sensor data for a particular application. Another example is the use of day-night or seasonal thermal data to infer information about the target. Thermal effects influence virtually all remote sensing operations. These effects normally complicate the issue of analyzing spectral reflectance properties for Earth resources. However, temporal effects might be the keys for gleaning the information sought in an analysis. For example, the detection of the process of change is predicted on the ability to measure temporal effects. Spatial characteristics that we use involve shapes and relative sizes of objects. Spatial effects also refer to factors that cause the same types of features at a given point in time to have different characteristics at different geographic locations.

The type of sensor is perhaps the only characteristic of remote sensing over which the user can has some control. Careful matching of the sensor to the problem can ensure that the results of the study will be useful and easily quantifiable. The most commonly used sensors will be discussed in turn.

Gamma radiation remote sensing is based on the attenuation of natural terrestrial gamma radiation by soil water or a layer of snow. The general procedure is to determine a background measurement for no snow or dry snow conditions. A subsequent measurement in the presence of snow or increased soil moisture will reveal an attenuated flux which can be related to the snow water equivalent or change in soil moisture.

Aerial photography has a long history of use in environmental management. The first aerial photographs used the visible portion of the electromagnetic spectrum. Advances in photography and films have led to the capability of making images of other parts of the spectrum with much interest in the near-infrared and thermal refines. For example, photographic images have provided information on sediment plums, erosion features, discharges from pipes, and spills. Close-range stereophotographs have been used to study erosion and gully formation.

Multispectral scanners are instruments that measure the spectral reflectance of narrow wavelength bands and record the information electronically. This technique involves measuring simultaneously the spectral response of the landscape in two or more narrow wavelength bands of the electromagnetic spectrum. Multispectral classification of this data is then used to discriminate objects based on the characteristic of reflectance. Multispectral analysis has developed from early system, using two or more cameras with different lens filters to make images. The Landsat satellites have been providing four spectral bands of the Earth's surface with the Multi Spectral Scanner, and more recently with the Thematic Mapper. Multispectral scanner techniques have two major advantages over photographic methods. First, instruments can be designed to measure very narrow wavelength bands, and second, the digital form of the data enables the use of rapid and sophisticated analysis and classification techniques. Multispectical data have been analysed in many ways to produce information with potential use for the natural resources modeller: vegetation biomass, soil type, vegetation type, snow cover, water area, impervious area and various water quality parameters.

Thermal sensors directly measure the emitted thermal energy of the Earth's surface. Surface temperature changes are the result of the balance of radiant, latent, sensible and ground heat fluxes. Analyses of remotely sensed thermal data can be used to develop maps of the environmental conditions of the Earth's surface. In general, thermal sensors are used to measure variations in temperature across the landscape. One infers information about the properties of the landscape that affect temperature change. Examples of the use of thermal data are to estimate evaporation, soil moisture, drainage patterns, ground water seepage zones, canopy temperatures, and thermal plumes from thermoelectric power plants or industrial sources.

Microwave sensors can directly measure the dielectric properties of the Earth's surface. Any changes in these properties directly affects the reflectivity or emissivity measured by microwave systems. The dielectric property of the Earth's surface layer is in turn strongly dependent on the moisture content. Measurements in the microwave region of the electromagnetic spectrum can be related to the moisture content of the soil surface layers. Similar relationships exist for snow. The physical relationships between moisture, dielectric properties and microwave response, together with the ability of microwave sensors to penetrate cloud cover, make microwave sensors a useful all-weather sensors to measure the moisture of the Earth's surface. Meanwhile, active microwave systems (radar) send out an energy pulse and measure the reflected pulse and the Earth's naturally emitted microwave radiation. Active microwave and passive microwave systems have been flown on aircraft and satellites. Examples of the use of microwave data are to estimate soil moisture, vegetation type, snow water equivalent, condition of snow pack, frozen soil and sea ice.

Laser remote sensing is a rapidly expanding research area with potential application on natural resource models. The principle involves projecting a narrow beam of coherent visible or near-infrared light and measuring the reflected energy with a photomultiplier tube to determine the distance between the laser system and the object of interest or to measure the backscatter from aerosols or the land surface. Airborne laser systems have been used to collect data for topographic maps and especially data for topographic surveys where high-density topographic information is required.

For the sake of competence to introduce the remote sensing system, the platforms for supporting sensors should be mentioned. Platforms can vary from ground-based supports to aircraft and satellites. Generally, truck-mounted, ground-based and aircraft systems are used in sensor development to verify design characteristics. Truck and aircraft systems are too limited in their potential coverage and too expensive to be considered for operational remote sensing except for very limited and special applications, the goal is eventually to get the sensor mounted on free- flying satellites. When this has been accomplished, the potential exists for obtaining large quantities of data over large areas of the globe, for extended periods of time at relatively low cost.

Once data have been collected from a remote sensing system, the user must then interpret the data for his specific application. Generally, the data are available in three forms: as an image analogous to an aerial photograph, in an analogue format, and in digital format typically as a computer-compatible type. These three data are interchangeable as imagery. Imagery can be digitized and digital data can be processed as an image or a computer-produced map; analogue data can be developed from one or the other, or used to produce digital data or an image. Usually, photometric accuracy is not preserved in analogue data and in conversion between analogue and digital forms unless calibration data are also provided and great care is maintained to preserve accuracy. The conversion from imagery or analogous data to digital data, and vice versa, is based on separating the measured values of reflectance into binary increments, usually based on byte word lengths. Once the data are in digital form there are many instruments and computer systems that can perform a myriad of useful analyses to help the user interpret the data. Imagery from specific spectral bands can be produced. Computer system can also perform different types of classification procedures and compute different statistic of the scene. In addition the data may be included in numerical hydrologic models (Cracknell and Hayes,1991).

3 Monitoring the rainfall using radar

Radar is a family of active remote sensing systems, most of which operate in the microwave (1mm-1m) region of the electromagnetic spectrum. Some radars have been designed for operation on the ground, and others on aircraft and even satellites. In rainfall monitoring ground-based microwave radar have been strongly predominant. Hence, we will focus our attention on these systems.

3.1 Spaceborne radar techniques

Radar measurement of rainfall is based on the Rayleigh scattering caused by the interaction of rain and the radar signal (Cracknell et al. 1991). A formula expressing the relation between the power of the radar pulse returned to the antenna and the average power returned from the rainfall, the dielectric constant of rainfall and the wavelength exists in this literature, assuming the rainfall fills the radar pulse volume.

Because the radar interacts primarily with the rain, this formula can identify the presence of rain below the clouds and theoretically, at least, provide an opportunity to measure rainfall rates without several restrictive assumptions.

Spaceborne radar has not yet been used for estimating rainfall for the simple reason that there is no existing satellite radar with suitable frequencies for rainfall. What has been produced is a fairly impressive body of information based on ground -based radar and the theory and planning of satellite programmes for measuring rainfall (see also 4.2 ).

Meneghini et al(1983) have proposed using a spaceborne radar for estimating rain rates on the attenuation of the radar signal by the rain. They propose using the surface as a reference target for determining the path averaged rain rate that is independent of the reflectivity/rain rate law and the radar calibration.

The potential for using dual-wavelength radar for measuring rain rates has been proposed by Goldhish(1988). He analyzed the algorithms necessary to retrieve rain rate profiles from a spaceborne radar operating at 13.6 and 35 GHz by comparing the results with simulated rain rate profiles. It was shown that the backscatter method at 13.6 GHz is useful near the top part of the rain but that at 35 GHz is not useful nor is the attenuation approach. The dual-wavelength approach yielded acceptable accuracies for rain from about 0.75 to 3 km above the Earth. He concluded that no single technique gives rise to a panacea in the making of accurate rain measurement and that difficulties exist with each method.

Eastwood et al (1990) proposed a spaceborne radar system for mid-latitude orbiting precipitation and cloud mapping. Radar is designed to utilize a narrow, dual- frequency beam, electronically scanned antenna. This system can provide a 3- dimensional rainfall dataset that covers *60º latitude band of the globe. It can achieve 4 km spatial resolution and 300 km cross-track swath. Vertical resolution of 500 m is achieved by short-pulse transmission. It is expected that such system can measure rain rates up to 100 mm/hr for precipitation at the cloud base, surface precipitation up to 20 mm/hr, and cloud reflectivities as low as -39 dBz. By averaging over 100 independent samples, the signal reflectivities can be estimated to better than 20%. Other rain characteristics, such as height, thickness, and cell size, can also be extracted from the data.

For testing the accuracy of rainfall and rain rates that a spaceborne radar can estimate, Meneghini et al (1990) conducted an airborne rain measuring experiment using a 2-wavelength radar and a 3 channel radiometer. The results indicated that backscattering and attenuation methods are complementary and can be used to extend the effective dynamic range of the radar. A dual-wavelength radar should further improve accuracy by providing independent estimates of the rain rate.

3.2 Ground- based radar techniques

Although spaceborne techniques are still mainly experimental or under development, ground-based radar has been used increasingly for operational rain forecasting since early World War Two, as well as providing valuable research results for future ground and spaceborne systems.

A basic distinction within microwave radar is coherent and non - coherent systems. Non-coherent systems lack stable transmitter frequency, and are used principally to observe the location and pattern of echoes, to measure the intensity of back- scattered signals, and, perhaps, to detect pulse-to-pulse changes in signal strength so as to provide estimates of the relative motion of the targets. Coherent systems use very stable transmitter frequencies, permitting these 'Doppler' systems to measure very precisely the shift in microwave frequency caused by moving targets. Unfortunately these coherent radar are beset by two difficulties which have effectively limited their applications. These difficulties are:

1) There is a restriction on the range of Doppler radar resulting from range and velocity ambiguities where a high pulse repetition frequency is used. Many Dopplers operate out to only several tens of kilometres.

2) Doppler radar only measure the line of sight velocity component of the targets, hence, relatively sophisticated equipment and procedures are required to yield 2D or 3D patterns.

The use of non-coherent radar in rainfall monitoring include the following (Browning, 1978).

1) Qualitative determination of the dynamical structures of clouds and precipitation structures, e.g. assessment of echo shape, and patterns of precipitation associated with different weather systems such as mid-latitude depressions and severe storms.

2) Short-period forecasting of rainfall, involving simple extrapolation, or the expected development and the decay of rain systems, taking account of the particular topography of the forecast area.

3) Quantitative measurement of precipitation. It is this point with which we shall be most concerned here.

Radar systems have become quite widely used in rainfall monitoring especially at the mesoscale, operating out to ranges of the order of 100 km, although even here research and development has outweighed fully operational application.

These different operating modes are possible when the aim is to scan and evaluate rainfall rates over an area centred on a radar facility. Of these, two are impracticable operationally because they depend on relationships between rainfall rate and attenuation of radar signal, either separately, or in conjunction with signal reflection from the areas of rain: they require large numbers of remote targets placed around the edge of the area, adding to problems of system cost and security. Hence, the mode which employs a combined receiver/transmitter for signal propagation and evaluation of energy reflected back to it from rain in the target area is of more general use.

Given appropriate knowledge of the technical specification and performance of a radar installation the radar equation basic to most radar studies of rainfall may be represented as

P=cZ/sqre(r)

Where P is the mean received power from the target, c is the speed of light (constant), Z is a reflectivity factor specific to the type of precipitation in the target areas, and r is the range between the radar and the reflecting shower of rain. Of special significance is the reflectivity factor, which represents a set of physical conditions prone to considerable variation both from time to time and from place to place. Unfortunately, it is usually necessary to evaluate this factor empirically on a region-by-region basis over lengthy periods of time. Inevitably, its use leads to operational errors in the evaluation of rainfall, especially on short term bases. Much better results may be obtained by relating the average reflectivity observed over a period in the environment of a well-exposed rain gauge to the rain recorded by that gauge, the recorded amounts being telemetred to the radar and the relationship calculated at such as hourly intervals. Unfortunately, such refinements are not always possible. In any case, the key assumption is that, for different latitudinal zones, and different types of rain, the radar signal is proportional to the rate of rainfall divided by the square of the range.

Radar has a unique capability to observe the aerial distribution of precipitation at frequent intervals, giving information from quite large areas to a single centre with a minimum of telemetry requirements. However, radar techniques have yet to become the standard method of supplying the user with the mesoscale rainfall data they desire. A plethora of practical problems explains most, if not all, of the remaining consumer resistance. These problems include and involve the following:

1) Signal fluctuation from rain. The radar signal from rain is the sum of the signals from all the raindrops distributed, and moving at random, in each pulse volume. Not surprisingly the result is a strongly fluctuating signal, whose average must be established and interpreted with care.

2) Variations of reflectivity in the vertical. Since the surface of the Earth is curved, and radar beams travel in relatively straight lines, they sample the atmosphere at progressively greater heights as they move away from the transmitter. For a horizontally operating radar the bottom of a beam rises to 0.1 km at a range of 100 km. Since precipitation layers are usually quite shallow this effect, more than any other, restricts the radar range, and limits the accuracy of estimates of rainfall intensity at some distance from the radar facility. In middle/high latitudes a further related problem is the low height of the melting layer, in which the reflectivity factor may be enhanced ten times by changes in dielectric constant and fall speed, and aggregation of wet snowflakes. No adequate method has been devised yet to correct for the resulting bright band echoes.

3) Effects of screening and ground clutter. Radar beams are reflected not only by cloud drops and / or raindrops, but also by upstanding features on the ground, e.g. tall buildings and relief. The usual solution for this problem is to map the strong, permanent echoes caused by such features, and subtract them from the operational echo pattern, and interpolate isohyets across them. This problem, often in combination with the one which preceded it, places significant restrictions on the utility of radar in hilly or mountainous terrain.

4) Calibration of radar echoes in terms of rainfall intensity. Although estimates of rainfall intensity can be made in the absence of calibration gauges, such gauges should be operated whenever possible. Experience in field tests have further shown that there is a marked improvement in accuracy if the method of calibrating the radar against a value of rainfall is obtained from a gauge within the actual rain area itself.

5) Staffing needs. Adequate staff training and supervision is vital if a radar facility or network is to operated efficiently. Conditions for this are less than optimal in many countries, especially in lower latitudes where the need for improved rainfall intelligence is greatest. The associated problems often become particularly actuate when foreign technical assistance ceases. This is an important fact of scientific life which, perhaps, only an independent witness is able to stress. For example, in the late 1960' s, when the former Sevien Union technical specialists withdrawn from China due to political reasons, the Chinese radar systems had almost collapsed for several years.

6) System costs and cost - effectiveness. It is in this key area of evaluating the utility of radar that generalisations are most difficult to make.

Especially on grounds of cost - effectiveness there would appear to be a strong case for widespread employment of radar for rainfall monitoring, since more qualitative information such as system growth cycles and movement can be obtained at the same time in support of studies of rainfall behaviour. However, there are very few localities in the world in which continuously recording rain gauge networks are dense enough to meet the minimum requirements.

In conclusion, then, the situation is this: radar can indeed provide valuable rainfall data, especially when integrated with a network of calibration gauges, but almost all present users of rainfall data have to cope at present with small to tiny fractions of the data that radar systems would yield at break-even points, established in comparison with conventional gauge networks. Thus it is generally difficult to convince funding agencies of the need for the much greater volume of data that radar could provide, especially in view of the substantial increase in costs that would be incurred over and above actual present expenditure on rainfall monitoring.

The need for improved rainfall data, in the short to medium terms, at least, must be fulfilled by other remote sensing means. Satellites offer more hope in these time frames (Curran,1985).

4 Monitoring the rainfall using satellites

Although the first weather satellite was launched in 1960, it was not until 1966 that the first operational weather satellite system was inaugurated using two polar- orbiting satellites of a ''cartwheel'' variety. Not surprisingly, attention in the early years of satellite meteorology was focused on those atmospheric phenomena which could be observed relatively directly in the visible and infrared wavebands such as cloud types and systems rather than others which could be assessed relatively indirectly or inferentially such as rainfall rates and distributions. Thus the earliest research which proposed a method for the systematic evaluation of rainfall from the cloud contents of weather satellite images appeared as late as the tenth year of satellite operations. In this and following researches, it was assumed that satellites were, in effect, in competition with conventional systems in the monitoring of precipitation. However, direct measurement of rainfall from satellites for operational purposes has not been generally feasible because the opacity of clouds prevents direct observation of the precipitation with visible, near-infrared and thermal infrared sensors. But improved analysis of rainfall can be achieved using both satellite and conventional ground-based data. Satellite data are most useful in providing information on the spatial distribution of potential rain-producing clouds. Useful data can be derived from satellites used primarily for meteorological purposes, including polar orbits such as NOAA-N and DMSP, and geostationary satellites such as GOES, GMS and Meteosat, but their visible and infrared images can only provide information about the cloud tops. However, since these satellites do provide frequent observations, even at night with the thermal sensors, the characteristics of potentially precipitating clouds and the rates of changes in cloud area and shape can be observed. From those observations, estimates of rainfall can be made which relate cloud characteristics to instantaneous rain rates and/or rain total over time (Barret, 1988).

The approaches used in making quantitative estimates of rainfall by using satellites can be divided into two streams:

4.1 Visible and infrared techniques

The availability of meteorological and Landsat satellite data has produced a number of techniques for extracting the most important information on rainfall from satellite imagery of clouds in the visible and /or infrared wavebands (Barret, 1988). These techniques have led to the development of three dominant approaches: the cloud indexing approach, the thresholding approach, and the life-cycle approach. Cloud indexing, which is time independent, identifies different types of rain clouds and estimates the rainfall from the number and the duration of clouds or their area. Thresholding techniques consider that all clouds with low upper-surface temperature are likely to be rain clouds. Life-cycle methods are time dependent and consider the rates of changes in individual convective clouds or in clusters of convective clouds. All these methods are essentially empirical in that they use statistical coefficients based on historical cloud and ground measured rainfall. They will be addressed in the following subsections.

4.1.1 Cloud indexing techniques

These were pioneered and have been developed by Barrett (1986) and other scientists. Satellite cloud images are ascribed indices relating to cloud cover, and the probability and intensity of associated rain. The techniques rely on visible and infrared data to characterize a cloud type or temperature which is then related to rainfall via empirical relationships. Different methods have been used to calibrate the indices to give rainfall estimates.

The Earthsat method (Moses and Barrett, 1986) is an operational rainfall estimation scheme that has been developed to provide input to crop yield models and commodity forecasting systems. The Earthsat method uses a regression approach to estimate 6 hour precipitation from cloud temperature and empirical information for the major crop-growing regions of the world. The basic regression equation can be found in this literature.

The result of the regression calculation can be further modified by consideration of synoptic station reports. Experienced meteorologists can usually improve upon these estimates by improving the vertical motion fields through interpretation of the satellite imagery.

The Bristol method uses an empirical relationship between satellite determined cloud indices, climatic indices dependent on the mean monthly rainfall, and 12-hour rainfall totals. A family of curves has been developed by some researchers (e.g. Barrett, 1981) in tropical and mid-latitude zones. These studies indicated a consistent increase in precipitation amounts from dry to hot-humid climates, but also indicated that higher-intensity rain clouds could not often be differentiated from lower-intensity rain clouds. Hence there is a need to treat each pixel location separately and in the light of climatological information.

The BIAS (Bristol/NOAA interactive system) method has been developed on the basis of Bristol method (Barrett et al ,1986). For details please refer to this paper.

It is the cloud indexing types which have, in one form or another, shown most flexibility and which have yielded the first results in support of operational rainfall monitoring programme.

4.1.2 Thresholding methods

There exist two types of threshingholding methods, one is based on cloud top brightness, the other is based on cloud top temperatures.

The cloud brightness techniques depend on the assumption that precipitating clouds are often brighter than others, so that automatic brightness thresholding can be used to map rainfall from satellite visible imagery, and that calibration of rainfall amounts can be achieved using raingauge and/or radar observations. This technique uses daytime visible and thermal infrared image pairs to establish rain or no-rain thresholds. These are then used to reduce possible misinterpretation of rain clouds from the night-time thermal infrared imaginary above. The cloud temperature technique is based on temperature thresholding of thermal infrared imagery analyzed by computer to identify potential rain clouds.

4.1.3 Life-cycle methods

Some life-cycle methods are designed to provide rain estimates from any type of convectional clouds by taking into objective consideration of the growth or dissipation of individual clouds with time. This approach implicitly recognizes that convective clouds exhibit different rainfall intensities during their growth and dissipation cycle.

The Woodley-Griffith techniques was developed initially to predict rainfall over south Florida as part of the Florida area cumulus experiment (Criffith et al .1978). This method uses an empirically derived relationship between calibrated ground- based radar echoes and geostationary satellite imagery of cloud areas. A time-cycle relationship between the radar echo area and the cloud area is developed for discrete time intervals during the lifetime of the cloud. The relationship used in the Woodley- Griffith technique can be found in this paper.

A family of life-cycle techniques with the more specific purpose of evaluating and monitoring high-density events has been developed from the work of Scofield and Oliver (1977). By originally using half- hourly rainfall amounts for convective systems from tropical air masses, an analyst can then use a decision tree to make rainfall estimates at different points. This technique is divided into three parts: first, the active portion of the convective system is delineated; second, an initial estimate of rain rate is made from thermal infrared image alone; then, third, the changes in two consecutive images (visible and thermal infrared) are evaluated to find clues that would indicate heavier rainfall.

Scofield (1986) has developed a series of seven convective and five extratropical cloud categories that can be used to help meteorologists improve their estimates of heavy precipitation across a range of different weather situation. The categories have been developed from satellite data, ground radar, surface and upper-air data, and from precipitation characteristics. Each category is based on the life cycle of the cloud pattern and cloud-top temperature changes as well as the other information.

All of these methods used in the past involve manual interpretation of imagery from meteological satellites. Because of the factor of human interaction, the number of images that can be handled daily is limited. The human interaction factor also makes the analysis procedure subjective. Snijders (1991) undertook a study to evaluate these three methods to monitor rainfall over West Africa using automated processing of full-resolution digital data, being 2.5 km by 2.5 km for the visible channel and 5 km by 5 km for the thermal infrared channel, from the Meteosat satellite. The purpose of this study is to make a direct comparison of these techniques for the same area and period of time. The results indicated that none of the techniques yielded better results than the others, but there were distinct differences in the performance. The cloud indexing method and threshold techniques performed best at the lower latitude, while the life cycle techniques performed best at the higher latitudes.

There are many variables effecting the radiances observed by these methods, such as sun elevation, satallite-sun azimuth angle, layered clouds, reflectivity of underlying surfaces, snow and ice surfaces, shadows etc. Eliminating errors caused by these is possible in some cases with statistical or physical methods. Some of them are so complex, that they have to be neglected in a real time application. It is time consuming to analyze a raw satellite image affected by these errors. Pylkkö and Aulamo (1991) discussed a method to ''teach'' the computer to do the analyzing. They used data from satellite images to find ways to automatically analyze the probablity of rainfall. The first results show that the Meteosat data can be used to analyze rainy areas in synoptical scale. These products definitely supply useful information to help monitor rainfall.

4.2 Microwave radiometry

Microwave techniques have a great deal of promise for measuring rainfall and have been employed for many years to provide rainfall datasets because of the potential for sensing the rain itself and not a surrogate of rain such as the cloud type. Microwave radiation with wavelengths of the order of 1 mm to 5 cm results in a strong interaction between the raindrops and the radiation. This is because the drop size is comparable with the wavelength.

Passive microwave radiometers on the Electrically Scanning Microwave Radiometers (ESMR-5, 19.35 GHZ) have yielded measures of naturally-emitted microwave radiation from the surface of the Earth, and the water content of the atmosphere. ESMR-5 data have been processed to give, among other outputs, maps of instantaneous precipitation intensities. Passive microwave approaches are being developed further through experimentation .

Allison et al (1975) have shown that ESMR data is very useful in delineating areas of rainfall over the oceans and gives qualitatively good comparisons with rainfall data. Wijheit et al (1977) used a radio active transfer theoretical model which include scattering effects and concluded that, despite the difficulties in interpreting rain rate, it can be used to estimate the higher rainfall rates. This technique has also been used to estimate weekly, monthly and annual rainfall maps for the major ocean areas.

Some scientists have opened up a new range of possibilities for the development of rainfall algorithms over land. A more recent rainfall algorithm is based on the difference in measured brightness temperatures at two frequencies, as suggested by Gordy (1984). This algorithm is based on the relationship between emissions and frequency, which decreases for most surfaces but increases for dry snow, old sea ice and in the presence of scattering caused by raindrops. In later work, Ferraro et al (1986) developed a classification approach for identifying rain as well as other geophysical features. Their classification scheme is based on the difference in two vertical polarized frequencies plotted against their average.

The polarization algorithm suggested by Grody (1984) is based on the collocation between two polarization at the same frequency so that the surface emissivity effects are minimized and the precipitation effects enhanced. For details about this algorithm refers to this paper

Although the relationship between the rain rate and the microwave response is fairly well understood. there are several limitations in this approach. One of the more serious is the unknown depth of the liquid rain layer. especially in areas where warm rain forms without an ice phase. The effect of cloud density which is made up of small droplets less than 50*m in diameter has not been accounted for and is a potential source of error that could perhaps be solved with multifrequency measurements. There is also work to be done on improving the modelling of the scattering model to represent more realistic ice particle geometries. But the design of rainfall retrieval algorithms over land is fraught with difficulties, the worst of which lead to ambiguities. Furthermore, passive microwave data is less easily and less frequently available Finally these problems do not begin to address the more fundamental problems of measuring rainfall. These include the great spatial and temporal distribution of rainfall and the fact that instantaneous rain rates are being measured when, in fact, what is needed is the integrated rainfall volume over some time period. Therefore, combining visible, infrared and passive microwave data may be the best satellite or satellite improved rainfall monitoring method. So the trispectral technique is the next step in the development of rainfall algorithms.

In the first examination of the potential of combining visible, infrared and passive microwave, a hybrid approach (Barrett and Kidd, 1987) investigated the use of scanning microwave multichannel radiometer (SMMR) passive microwave data in support of BIAS discussed in 4.1.1 for rainfall monitoring in several areas of the globe where they could overlay the SMMR data on to the BIAS images. Although not a definitive study, it was concluded that the passive microwave data could be useful in helping to

1) locate the leading and trailing edges of rain areas.

2) confirm the extent and organization of rain areas especially if no visible/infrared data were available.

3) Locate heavy-rain areas where cumulonimbus cells are located within stratiform clouds

4) locate areas of rain that could not normally be identified by BIAS.

More recently, the potential of the combination of spaceborne radars and passive sensors for rainfall profiling have been examined (Wilheit, 1990). The general consensus is that radar/passive instrument complement can provide a better characterization of the rain systems. It is much easier to achieve a wide swath for the measurements with a radiometer than with a radar.

Weinman, J. A. et al (1988) developed an algorithm to seek rainfall profiles from multifrequency, dual polarization passive radiometers operating in combination with a radar that operates under condition that produce significant extinction. The algorithm was solved that yield a robust solution of the radar equation subjected to constraints imposed by measurements of dual polarized multifrequency microwave radiances. Using simulated data, it was shown that a single frequency radar operating from space in conjunction with a multispectral radiometer may be able to provide cost effective measurements of global tropical rainfall distributions.

Present plans for the Tropical Rainfall Measurement Mission (TRMM) will have both passive and active sensors in its instrument package (Simpson, 1991), it consists of a closely knit combination of a multichannel dual polarized passive microwave radiometer with a 14 GHz rain radar and a six-channel visible/infrared instrument. These instruments can be co-registered and their data used in combination to optimize the rain product. The orbit elevation has been chosen as 350 km to obtain high resolution, which will be about 24 km, 4 km, and 2 km at nadir for the passive microwave 19 GHz channel, the radar, and the visible/infrared respectively. The surface swath is 220 km for the radar and 600 km for the other instruments. In order to document the diurnal rain variability, the orbit is inclined at 35º so that the overpasses occur at different local times on successive days. Algorithms using the radar and passive microwave together are progressing well.

Weinman et al (1988) have shown that the passive microwave results may be used to constrain the radar equation, obtaining good vertical profiles of rain rates as well as surface values.

5 Conclusion

With the study above we can conclude that remote sensing has a wide and effective application in monitoring the rainfall, although there exist some open questions to be solved. Surface radar, or satellite visible, infrared or passive microwave systems are employed to provide the basic data for monitoring the rainfall. Surface radar has the advantage of being physically direct, but has limited range and is subject to a multiplicity of often unhelpful physical influences. Satellite visible and infrared systems provide globally comprehensive data, but rainfall distributions and amounts can only be established inferentially from their data. Satellite passive microwave systems function promisingly over sea surfaces but not satisfactorily over land. It would seem that a complex system of interrelated sensors of several different types would yield the best results, and provide maximum flexibility.

It appears likely that considerable progress will continue to be made on improving our ability to measure rainfall and rain rates from remote sensing instruments. It is also expected that our capabilities for doing this from space will also develop rather rapidly, although no single approach or wavelength will be the answer. It is increasingly obvious that progress will be made using multifrequencies and multipolarizations, as well as a host of non-spaceborne instrumentation such as atmospheric soundings, ground data, etc. It also seems obvious that, owing to the inexact nature of the science, the use of interactive computers and expert systems will be incorporated to assist analysis in the more subjective aspects of estimating rainfall.

References

Allison, L.J. et al.(1975) Tropical cyclone rainfall as measured by the Number 5 ESMR. Bull. Amer. Soc., 55:1074-1089

Barrett, E. C. (1981) AGRISTARS state *: Development and application of the Bristol method for improved rainfall monitoring for mid-latitude use. Final Report to US Dept of Commerce, Washington DC, Remote sensing Unit, University of Bristol.

Barrette, E. C., Beaumont, M. J., Harrison, a. and Richards, T. S. (1986) BIAS S users guide. Final Report to US Dept of Commerce, Washington DC, Remote sensing Unit, University of Bristol.

Barrett, E. C., D'Souza, G., Power, C. H. and Kidd, C. (1988) Toward trispectral satellite rainfall monitoring algorithms. In Tropical Precipitation Measurements, Proc. Int. Symp., Tokyo, Japan, NASS/NASDA (eds J. S. Theon and J. Fugono), A.Depak, Hampton, VA, pp.285-292

Cracknell. A. P. and Hayes. L. W. B. (1991) Introduction to Remote Sensing, Lodon: Taylor & Francis.

Curran. P. J. (1985) Principles of remote sensing, New York. Longman group limited.

Eastwood, I. and Kellogg, K. (1990) Spaceborne Radar for Rain and Cloud Measurements--A Conceptual Design, present at the 1990 international Geoscience and Remote Sensing Symposium (IGARSS'90), Washington, USA, May 20- 24,1990

Ferraro, R. J., Grody, N. C. and Kogut, J. A. (1986) Classification of geophysical parameters using passive microwave satellite measurements. IEEE Trans. Geosci. Remote Sensing GE-24, 1008-1013

Goldhirsh, J. (1988) Analysis of algorithms for the retrieval of rain-rate profiles from spaceborne dual-wavelength radar. IEEE Trans. Geosci. Remote Sensing GE-26, 98-114

Grody, N. C. (1984) Precipitation monitoring over land from satellite by microwave radiometry. Present at the Intertional Geoscience and Remote Sensing Symposium (IGARSS'84), Strasbourg, France, ESA SP-215. pp417-423

Meneghini, R., Eckerman, J. and Atlas, D. (1983) determining rain rate from a spaceborne radar using measurements of total attenuation. IEEE Trans. Geosci. Remote Sensing GE-21,34-43.

Meneghini, R., Kozu, T., Kumagai, H., and Boncyk, W. C. (1990) Analysis of Airborne Radar and Radiometer Rain Measurements and Their Relationship to Spaceborne Observations, present at the 1990 international Geoscience and Remote Sensing Symposium (IGARSS'90), Washington, USA, May 20-24,1990

Moses, J. F. and Barrett, E. C. (1986) Interactive procedures for estimating precipitation from satellite imagery. Hydrologic Applications of Space Technology. Proc. Cocoa Beach Workshop, Florida, August 8-12,1985.

Pylkkö, P. and Aulamo, H. (1991) Using Satellite Images to Estimate Rainfall in North Latiyudes, present at the 1991 international Geoscience and Remote Sensing Symposium (IGARSS'91), Espoo, Finland, June 3-6,1991

Simpson, J. and Theon, J. S. (1991) The Tropical Rainfall Measuring Mission (TRMM) and Its Role in Studies of Climate Variations, present at the 1991 international Geoscience and Remote Sensing Symposium (IGARSS'91), Espoo, Finland, June 3-6 , 1991

Slater, P. N. (1980) Remote sensing, Optics, and optical systems.USA: Ddison- Wesley Publishing Company.

Snijders, F. L. (1991) Rainfall monitoring based on Meteosat data-a comparision of techniques applied to the Western Sahel. International Journal of Remote Sensing, Vol. 12, No. 6.

Spencer, R. W., Goodman, H. M. and Wood, R. e. (1988) Precipitation retrieval over land and ocean with the SSM/i, Part 1: Identification and characteristics of the scattering signal. J. Atmos. Oceanic Tech. 2, 254-263

Weinman, J. A., Kummerow, C. D. and Atwater, C. S. (1988) An algorithm to derive precipitation profiles from a downward viewing radar and a multifrequency passive radiometer, present at the 1988 international Geoscience and Remote Sensing Symposium (IGARSS'88), Edinburgh, U K, September 12-16,1988

Wilheit, T. T. (1990) Error Model for Combined Radar and Radiometer Measurements of Rain from Space, present at the 1990 international Geoscience and Remote Sensing Symposium (IGARSS'90), Washington, USA, May 20-24,1990


Any comments or suggestions regarding this assignment should go to Liu Quanwei