Remote sensing and geographic information systems - CiteSeerX

Hyémlogkal Sciences -Journal- des Sciences B^dmlagiqmes,4l(4) August 1996

593

Remote sensing and geographic information systems MICHAEL F. BAUMGARTNER & GABRIELA M. APFL Department of Geography, University of Berne, Hallerstrasse 12, CH-3012 Berne, Switzerland Abstract Remote sensing and Geographic Information Systems (GIS) play a fundamental role in hydrological applications. In this paper the most commonly used processing procedures for remotely sensed data — in particular image processing techniques - and the capabilities of GIS technologies are presented. The aim of this paper is to show the merit of a combination of these tools with hydrological models. An important aspect herein is the use of image processing systems, GIS, database management systems (DBMS) and hydrological models in a integrated analysis system. Télédétection et systèmes d'information géographique Résumé La télédétection et les systèmes d'information géographique (SIG) jouent un rôle fondamental dans de nombreuses applications hydrologiques. Dans cet article, les méthodes de traitement les plus importantes - en particulier les techniques de traitement des images digitales - et les possibilitées offertes par les technologies des SIG sont présentées. Le but de cette présentation est de démontrer l'avantage d'une combinaison de ces méthodes et des modèles hydrologiques. Un aspect important consiste à utiliser des systèmes de base de données et des modèles hydrologiques dans un système d'analyse intégré. INTRODUCTION Remote sensing and GIS technologies are well established tools and are routinely used in applied hydrology. In conventional applications, either results from remote sensing or from GIS analyses serve as input into hydrological models. During the last few years, research has focused on the question of how to integrate both technologies with hydrological models in integrated analysis systems. Remote sensing is presently in transition from a descriptive phase to a quantitative technology. Measurements may be carried out from the ground (field measurements) but the advantages of remote sensing applications in hydrology as a source of spatial information (in opposition to point measurements) becomes more obvious if sensors on air- or spaceborne platforms are used. The sensors measure the spectral characteristics of interest and their variations in time over large areas, providing data input into various hydrological models. Additionally, remote sensing data represent an important input Open for discussion until 1 February 1997

594

M. F. Baumgartner & G. M. Apfl

into algorithms which allow the derivation of hydrological parameters. Remote sensing - especially using spaceborne sensors - offers the advantage of long term (years to decades) temporal and spectral data sets over relatively large regions (local to global scale) and, therefore, of monitoring the (temporal, spectral and spatial) variations of objects at the Earth's surface. This discussion of processing remotely sensed data is focused primarily on data recorded in the optical range (0.4-12.5 /xm) of the electromagnetic spectrum. Generally, two major processing groups may be distinguished: physically-based and statistically-based methods, both in combination with image processing techniques. In most hydrological applications, statistical methods and image processing have proved to be most effective. These techniques do not ask for complex physical measurements which can usually be carried out only for very small areas. Statistical methods are, therefore, applicable over larger areas whereas physically-based methods are preferred in basic research. Visual interpretations are still a meaningful technique for analysing remote sensing data and may even be preferred to digital analysis techniques in certain situations (e.g. in field campaigns in remote regions or in projects where no or only limited computer equipment is available). The most common applications of remote sensing data in hydrology are related to the investigation of cloud cover, impervious surfaces, floods, land use, radiation, rainfall, snow cover, soil moisture and surface temperature to name a few important applications for monitoring and forecasting. Some of these applications are presented here. Additional examples may be found in Engman & Gurney (1991) and Haefner & Pampaloni (1992) which give an overview of the variety of remote sensing applications in hydrology. Geographic Information Systems (GIS) — in addition to remote sensing - have contributed significantly to applied hydrology in state and government monitoring and forecasting projects. GIS technology can provide resource managers and decision makers with tools for effective and efficient storage and manipulation of remotely sensed information and other spatial and non-spatial information (Estes, 1992). Remotely sensed data, effectively integrated within a GIS, can be used to facilitate measurements, mapping, monitoring and modelling activities. Designing and building a GIS database may be an expensive enterprise since it includes the import and entry of data from many different sources (this means that very often data have to be digitized from source documents). Data themes can range from hydrological (e.g. runoff) and climatological (e.g. temperature, precipitation) to data from both point and areal measurements. Remote sensing and GIS technologies do not replace conventional maps for hydrologists but facilitate processing, management and interpretation of all the available data. Remote sensing data coupled to conventional data (e.g. hydrological, climatological, topographical etc.) within a GIS give a digital representation of the temporal and spatial variations of selected variables and can serve as input into hydrological models. Due to the digital character of these data, quantitative analyses can be carried out and data can easily be

Remote sensing and geographic information systems

595

updated and specific selections of data can be displayed or printed as a conventional map. Remote sensing and GIS can be used separately or in combination with hydrological models (whereas selected hydrological modelling can be done within the GIS). In the case of a combined application, an efficient, even though more complex, approach is the integration of remote sensing data processing, GIS analyses, database manipulations and models into a single analysis system. Such an integrated analysis, monitoring or forecasting system (based on remote sensing, GIS and DBMS technologies) requires the hydrologist to understand not only the hydrological problem but also the available technologies yet without being a computer expert. Such an integrated analysis system enables the hydrologist to focus on solving the problem rather than spend time on technical difficulties. Solving a hydrological problem with such an integrated system may be summarized as follows: — —

recognition of a specific hydrological problem measuring the necessary variables by means of remote sensors and conventional, terrestrial methods — design of a GIS including data layers, attribute information, etc. — building up a database for managing all the data — processing the data using techniques such as image processing, GIS analyses, regression and correlation analyses, etc. for deriving model-relevant variables — selection or design of an appropriate hydrological model — input of variables into the model, model computations and error analysis — comparison of the results from model computations with actual remote sensing and ground data (determination of accuracy, updating the model computations, model calibration, etc.). Up to now, technological and educational difficulties have prevented a fully integrated approach. To date, in conventional hydrological applications, either remote sensing or GIS techniques have been used separately from the modelling process. Only in a few examples are the three techniques (remote sensing, GIS and hydrological modelling) used in a combined, integrated approach. The following sections present summaries of the most important image processing and GIS techniques. Furthermore, their integration with hydrological models is discussed by using a practical example. New technologies as presently developed in research are summarized in a concluding section.

PROCESSING O F R E M O T E L Y SENSED DATA In many applications, a visual interpretation of remotely sensed data may be adequate. Resulting maps or similar products can be used as a stand-alone product and as input for further analysis in a GIS or a hydrological model. However, the aim of this section is to discuss the digital capabilities of these tools since they are the most commonly used ones. For digitally processing

596


remotely sensed data recorded in the optical range of the electromagnetic spectrum, fundamental physical knowledge concerning the path of radiation from the sun to Earth and to the sensor is necessary. The radiation is influenced by variations in the spectral distribution, by the atmospheric (water vapour, aerosols) and meteorological (temperature, wind, etc.) situation, and by the reflection, absorption and emission properties of objects at the Earth's surface (spatial, spectra! and bio-physical conditions) as well as by the observation geometry under which the measurements are made (Schanda, 1986; Asrar, 1989). A different situation is encountered by using active sensor systems such as radar (e.g. Doppler radar, Synthetic Aperture Radar, Shuttle Imaging Radar) or lidar. Since the majority of systems used in applications cover the optical range of the electromagnetic spectrum, the discussion of processing methods focuses on these latter frequencies. For processing and analysing remotely sensed data in hydrological applications, several concepts can be distinguished, namely multispectral, multitemporal, multisensor, multivariate and near-real time concepts. Furthermore, analysis approaches are scale-dependent. Local, regional or global/continental approaches can be distinguished. At a local scale (1:25 000 or less), groundbased and airborne sensor systems predominate. The use of spaceborne data is often limited at such scales to conventional photographic products from manned spacecraft and imagers on satellites such as SPOT (Système Pour l'Observation de la Terre). Sensor systems on Earth resources satellites (Landsat, SPOT, etc.) are the basis for local and regional (1:25000 to 1:500000) scale analyses. Weather satellite data are preferred for regional scale studies (e.g. from the National Oceanic and Atmospheric Administration (NOAA) with the Advanced Very High Resolution Radiometer (AVHRR) to global scale studies (e.g. from GOES/Meteosat). In general, the hydrological application decides which sensor system should preferably be used. One problem with today's spaceborne sensors is that high temporal and high spatial resolution are technically not possible at the same time. The only way to overcome this problem is to combine several sensor systems (multisensor analysis concept) which may become very expensive due to the data pricing policy of the separate space agencies (or their representatives). For long term monitoring (multitemporal analysis concept) programs, only NOAA-AVHRR data (and similar future systems) offer an acceptable compromise concerning spatial and temporal resolution. Preferably, these data are calibrated from time to time by data with higher spatial resolution. A combination of spatial and spectral information systems (the imaging spectrometers e.g. AVIRIS) on future air- and spaceborne platforms will improve the multispectral analysis concept. Remote sensing data are most commonly used in combination with other data (multivariate analysis concept) as e.g. topographical (digital terrain models), meteorological and hydrological information. A limitation in today's application of remote sensing data is the lack of a near-real time access. Especially in forecasting and


597

environmental monitoring programs, data must be received and processed within 24 hours (or even faster). Generally, the processing scheme includes procedures for preprocessing, processing and postprocessing of remote sensing data which are usually summarized in the expression "image processing" (Swain & Davis, 1978; Schowengerdt, 1983; Lillesand & Kiefer, 1994). Image processing is often used in several ways. In this context, all the necessary procedures for treating image data from the raw data to the final result including, for example, normalization, calibration, geocoding and classification are included in the expression image processing. All remotely sensed data have to undergo a certain degree of preprocessing, i.e. data must be calibrated (e.g. for deriving absolute data), normalized (e.g. for solar zenith angle and satellite position), atmospherically corrected (e.g. by using such models as Lowtran and balloon sounding data) and geometrically corrected (sensor and target characteristics e.g. skew and panoramic distortion), and geocoded (related to a reference system such as UTM). The processing methods may be subdivided into two major groups: physically-based and statistically-based procedures. The borderline between the two groups is somewhat artificial which means that in many applications procedures from both groups are used in combination. The physically-based procedures are preferred for developing the fundamental principles for future applications. They are focusing on modelling the interaction between solar radiation and the Earth's surface. These models are validated using measurements of objects at the Earth's surface with known physical characteristics. Such models are then used to determine the unknown characteristics of an object based on the known (measured) spectral signatures (Asrar, 1989). Measuring and determining all the necessary model variables and parameters leads to expensive field campaigns. Furthermore, such model input can be determined only for very small regions since using (permanent) equipment over large regions is not feasible which means that the advantage of collecting spatial information is lost. An example may be the atmospheric correction which can be carried out by using balloon sounding data. The corrections are useful for a very close region surrounding the location of the sounding and for the time when the sounding took place. Especially in densely populated regions, the temporal and spatial variations of the atmospheric situation are high. A correction for larger regions and for remote sensing recordings which were not made during the same time as the sounding can lead to large deviations from the actual situation. The second group for processing remote sensing data, the statistical methods, are the most commonly used techniques in remote sensing applications (Haralick, 1976; Duda & Hart, 1973). The basis of statistical methods is that various objects at the Earth's surface are reflecting, absorbing and emitting radiation depending upon their type. An object type is not represented by a single reflection or emission value but is characterized by a cluster of values

598


with a specific range in each spectral band. Each measured value (pixel) is represented by a vector in the n-dimensional feature space where measurement axes (feature space) correspond to the spectral bands of the sensor system. It is the aim of the statistical methods to find algorithms for classifying each pixel — according to its specific reflection and emission values - into a predefined object class. The classification is carried out by finding a decision rule (classification algorithm) which subdivides the feature space into decision regions (object classes) based on the statistical distribution of the pixel values. Each pixel is classified into an object class (or category) according to its specific spectral characteristics using the mean, standard deviation and covariance of each class. These parametric procedures are based on a probability distribution function (Gaussian) for each category. The parameters for these functions are derived automatically or manually by using known training data sets and serve as input into the classification algorithm. The training data are usually based on ground truth, i.e. in parallel with the remote sensing data acquisition, samples of regions with known characteristics are collected during field campaigns. A first step in using parametric classification procedures is feature extraction or selection allowing the reduction of redundant information (e.g. by principal component analysis) and the transformation of the feature space for an enhanced separability of the classes. The second step involves the classification of all the vectors into a specific object class which can be carried out using either classification by supervised or by unsupervised learning techniques. Classification based on supervised learning techniques is by far the most commonly used. It is performed in two steps: — training: definition of the number of useful classes which is scene and scale dependent; definition of the probability distribution for all the classes by using training data; determination of the classification algorithm; and — classification: each pixel is classified into a land cover class using the probability distribution from training data sets and using the predefined classification algorithm. Minimum Distance to Mean, Mahalanobis Distance and Maximum Likelihood are the most commonly used algorithms. Hierarchical statistical processing procedures (Itten et al., 1985) represent a more refined approach for classifying remote sensing data. Instead of processing all spectral bands with a large number of categories in one step, the number of spectral bands and categories is adapted to a specific problem (Fig. 1). Depending on the problem, several classifications - each resulting in a bit mask (thematic maps) - are carried out. It should be noted that the more complex the problem, the more numerous the classification steps and the more sophisticated the classification algorithms must be. After such a multi-step classification, all the resulting bit masks are combined into the final classification chart (or map).


599

water

3,4,5

construction sites, concrete surfaces

cereals

other

residental (low density)

residentai (high density)

Fig. 1 Dendrogram for a multi-step hierarchical, binary classification of remote sensing data (PPD = Parallelepiped; MLI = Maximum Likelihood; PC A = Principal Component Analysis; NDVI = Normalized Difference Vegetation Index) (after: Itten et al, 1985)

Post-processing procedures are the final steps in the processing of remote sensing data. The resulting thematic maps can be filtered for noise reduction, converted from raster to vector for exporting the data to a GIS or a (hydrological) model (if not used in a raster GIS), or graphical (charts and maps) and statistical (tables) output can be produced. Processing, analysis and interpretation are based on the skills and the knowledge of the hydrologist with his/her physical and geographical background, i.e. in applications, no fully automated, interpreter-independent system exists today. Automated preprocessing and processing procedures can improve the analysis of remote sensing data, especially for time series evaluations where time is a critical factor (user-based interpretations are very time consuming). In most of the present applications - beside the spectral information - no information concerning syntactic or semantic neighbourhood relationships to other objects or information on object size, shape, orientation, texture, pattern or location is used. Using structural analysis methods, a priori knowledge of objects is included in the evaluations, i.e. a pixel is not treated separately but including its context (Nagao & Matsuyama, 1980; Matsuyama, 1987). Semiautomatic procedures are used only in a few examples, e.g. for geocoding and classification using approaches related to texture analysis and pattern recognition, and static as well as dynamic thresholds.

600


T E M P O R A L AND SPATIAL ANALYSIS USING G E O G R A P H I C INFORMATION SYSTEMS A GIS is a system which is designed to collect, store, update, manage, manipulate, analyse and represent graphical and non-graphical spatial data. More precisely, the analysis procedures allow the user a wide variety of investigations such as: — proximity analyses, neighbourhood operations (e.g. identifying objects within a certain neighbourhood fulfilling specific criteria) — to show relationships between data sets within such a neighbourhood — temporal operations and analyses — to generate new information by combining several data layers and attributes (e.g. by splitting or aggregating etc.). Furthermore, a GIS supports the import of external data (such as remote sensing), and the selection and transfer of data into application-oriented, analytical models (Antenucci et al., 1991). Basically, the data in a GIS can be subdivided into graphical and non-graphical data. Graphical data are digital descriptions of map features containing coordinates, symbols and rules. Each map feature is separately stored in a layer and all the layers are geo-referenced enabling comparisons and evaluations between the different layers. For the description of the map features, several types of graphical elements are used such as points (including nodes), lines (including line segments, arcs etc.), regions (polygons), pixels and symbols. Non-graphical data (attributes) describe the characteristics, qualities or relationships of map features at a specific geographical location. Four attribute types are distinguished: attributes describing the quality and quantity of data; geographically referenced data at a specific location (excluding the map features themselves); description of spatial relationships of map features; geographical indices linking attribute data with graphical data via geocodes. Graphical data can be stored in a raster or vector format, or some hybrid of the two. In hydrological applications, raster-based GIS are often preferred to vector-based systems. The reason for this situation is that, very often, distributed parameter models are used where the basin is partitioned into unit elements of homogeneous hydrological parameters. In combination with the pixel structure of remote sensing data, the raster-based GIS is an ideal solution. Many investigations such as vertical analyses (i.e. logical operations) between several layers — or even more general — all geometrical and overlay operations can more easily be performed in the raster domain. A disadvantage of this data format is that a large amount of data has to be stored and, therefore, computations are considerably slowed down. An advantage of the vector format is that network analyses and topological operations are preferably carried out in the vector domain and that storage capacity is significantly lower than in the raster format. Vector-based systems are more commonly used in applications compared to raster-based systems (Allen, 1994). Hybrid systems do not make a distinction between raster and vector format but the number of applications


601

is rather limited. Using two different data structures (raster for image data and vector for GIS analyses) brings the problem of converting data from one format into another. The accuracy of this conversion is critical: since it is very often a black box process, data should only be converted once otherwise uncontrollable deviations can occur (Baumgartner & Apfl, 1994). GIS have successfully been used in hydroîogïcai applications as standalone systems (as database for planning and management, and for GIS internal, hydrological modelling) or in combination with hydrological models. The synergism between G1S and remote sensing enables hydrologists to model temporal and spatial variations of hydrological processes efficiently. The integration of GIS and remote sensing with hydrological models and, additionally, database management systems (DBMS) is today the technically most advanced and applicable approach.

INTEGRATIVE ANALYSIS APPEOACH As has been shown, an interesting synergism between remote sensing and GIS exists, but due to the format problems addressed above, the user cannot yet take full advantage of this synergism (Frank et al., 1991; EMers et al., 1991; Estes, 1992). Therefore, inmost of the hydrological applications, either remote sensing or GIS techniques are used. In addition to the technological difficulties, only a few people have crossed the borders of their initial fields of specialization into other ones. Now, since microcomputers are available for a reasonable price, the situation is beginning to change (Fabbri, 1992). Until a fully integrated geographical information system (IGIS) is available (Ehlers et al., 1991), i.e. all necessary software modules (image processing, GIS and DBMS, hydrological models and graphical output) can be used integratively, several technical and scientific impediments still need to be solved. Consequently, the user has to continue working with today's antiquated tools (Frank etal., 1991). This means that a system must be developed which allows the user to process data and to transfer results from one module to another without being concerned about the data structure, and which supports the user in handling all data as well as directing the user through the high number of processing procedures. Ehlers et al. (1991) show an intermediate solution of a partially integrated analysis system based on two different data formats (i.e. raster and vector) but with a sophisticated user Interface which helps the interpreter navigate through a large variety of different software packages, processing procedures, data formats and analysis techniques. Presently, not many such partially Integrated systems exist. In several papers, the Integration of image processing, GIS and hydrological models is discussed (see examples in Fabbri, 1992; Sharma & Anjaneyulu, 1993) but In most of the articles It is not clear how data are transferred from one software module to another and whether a sophisticated user interface Is required. An example of a widely known and

602


applied system is ILWIS (Integrated Land and Water Information System) by ITC (International Institute for Aerospace Survey and Earth Sciences) (ILWIS, 1990). Another interesting approach is reported by Carroll (1995) designed at the US National Weather Service which can be used by hydrologists responsible for generating operational river and flood forecasts, water supply forecasts, and spring snowmelt flood outlooks. As a third example, the Environmental Data Atlas for the Savannah River Site project (Cowen et al., 1995) may be mentioned. For a detailed description of the philosophy of such a partially integrated system, the ASCAS (Alpine Snow Cover Analysis System) by Baumgartner & Rango (1995) is used and will be discussed in the next section.

ASCAS - AN EXAMPLE O F A PARTIALLY INTEGRATED SYSTEM ASCAS is designed for monitoring alpine snow cover variations, snowmelt runoff simulation and forecasting, and for estimating the effects of a possible climate change on snow cover and snowmelt runoff (Baumgartner & Rango, 1995). ASCAS is menu-driven and partially integrates, as a non-centralized configuration, five software modules: image processing, GIS, relational DBMS, hydrological modelling and graphical representations. In ASCAS, menus guide the user through the variety of processing procedures without being a computer expert, i.e. without being concerned about programming, operating systems, different data formats and network characteristics. The user can simply concentrate on the application by making selections within the menu. With this technical simplification, the user can focus on the fundamental knowledge of the algorithms (e.g. geocoding, classification etc.) involved in the whole process. Several software interfaces guarantee the data exchange between the different modules. The system is implemented on a microcomputer which is connected to a LAN (local area network) and the Internet for importing specific data (e.g. meteorological and climatological data from national weather service databases). The image processing module allows the analysis of remote sensing data - in this case NOAA-AVHRR data — for deriving time series snow cover maps (Baumgartner, 1990) based on algorithms as described above (classification by supervised learning). For the comparison of time series of snow cover maps and for their integration in the GIS module, geocoding plays a fundamental role since monitoring is carried out in river catchments of different countries which each have their own projection system (e.g. Lambert, GaussKraeger, Oblique Mercator or Albers). The geocoding process has to be applied and supervised very carefully since it is often a black box procedure. The resulting snow cover maps (bit maps) are converted to a vector format and integrated in the GIS. In the GIS module, graphical data (layers) have to be digitized from


603

Table 1 Topological data (layers) in the ASCAS-GIS (code 1 - digitized from topographical maps; code 2 = derived from satellite data; and code 3 = derived from codes 1 and 2) Name

Code

Type

Description

elev_zone_ls elev_zone_2s elev_zone_3s elev_zone_4s elev_zone_5s elev_zone_6s elev_zone_7s elev_lines_s aspect4_rs aspect4_ts asp_elev_rs asp_elev_rs snow_061183 snow_281283 snow_140284 snowj)60384 snow_140384 snow_150484 snow_100684 snow_070784 snow_061192 snow_261292 snow_130293 snow_100393 snow_180393 snow_200493 snow_050693 snow_290793 snow_200893 basins_s lakes_s rivers__s climate_s gauge_s sc_as_el_01rs sc_as_el_01ts

3 3 3 3 3 3 3 1 1 1 3 3 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 3 3

region region region region region region region line region region region region region region region region region region region region region region region region region region region region region region region line point point region region

elevation zone 1 : 3000 m a.s.l. (Switzerland) elevation lines, 500 m equidistance (Switzerland) aspect classes (NE, SE, SW, NW), Rhine (Switzerland) aspect classes (NE, SE, SW, NW), Ticino (Switzerland) aspect and elevation classes, Rhine (Switzerland) aspect and elevation classes, Ticino (Switzerland) snow cover map of 061183, Alps snow cover map of 281283, Alps snow cover map of 140284, Alps snow cover map of 060384, Alps snow cover map of 140384, Alps snow cover map of 150484, Alps snow cover map of 100684, Alps snow cover map of 070784, Alps snow cover map of 061192, Alps snow cover map of 261292, Alps snow cover map of 130293, Alps snow cover map of 100393, Alps snow cover map of 180393, Alps snow cover map of 200493, Alps snow cover map of 050693, Alps snow cover map of 290793, Alps snow cover map of 200893, Alps river basins (Switzerland) lakes and reservoirs (Switzerland) rivers (Switzerland) climate stations (Switzerland) stream gauges (Switzerland) snow cover_aspect_elevation chart 1, Rhine (Switzerland) snow cover_aspect_elevation chart 1, Ticino (Switzerland)

topographical maps or imported from other databases. Table 1 gives an overview of a possible list of layers to be used. For each layer, an attribute can be added giving topology-specific meta-information. Such attributes do not contain data but they give general information related to the specific layer and show the path and file names where data can be found in the database (Table 2). By splitting two or more layers as e.g. "snow cover map", "basin boundaries", "elevation zones", "aspect classes", etc., the snow coverage in a basin, elevation zone or aspect-elevation class can be determined. Furthermore, snow accumulation and ablation charts can be produced by superimposing a certain number of snow cover maps during a selected time period. By overlaying a time series of snow cover maps with a digital elevation model (DEM), the snowline variations within a selected basin can be derived. All this information can be used for climatological and hydrological interpretations of snow cover variations.


604

Table 2 Topology-specific meta-information for the attribute "streamgauge" Attribute

Value

Explanation

iayer name name2 station no longitude latitude elevation aver altitude basin size giac cover Q aver Feb Q aver Jun Q aver year path name

stream gauge s Rhine-Felsberg Rhine S0261 755880 190120 562 2010 3249 3 240/23 61/23 117/23 c:\ranoffts 1993 q

layer name (s = Switzerland) name of stream gauge main river station number (s = Switzerland) longitude of stream gauge [km] (Swiss Proj.Sys.) latitude of stream gauge [km] (Swiss Proj.Sys.) elevation of stream gauge (m. a.s.l.) average altitude of basin (m a.s.l.) total area of basin (km2) area covered by glaciers ( % of total area) average runoff, Feb. (m3 sec"1), interval (years) average runoff, June (m5 sec"1), interval (years) average annual runoff (m3 sec"1), interval (years) path to database and according file name

In a GIS-exteraal relational database, a set of data (e.g. daily values for temperature minima and maxima, precipitation, snow depths, new snow, runoff, date of satellite recording, equator crossing of satellite, etc.) can be stored, then transferred to the GIS, and linked to specific layers and selected objects within the layers. In addition to the temporal and spatial analyses with the GIS, these data are used in the hydrological module for snowmelt runoff simulations and climate change scenarios (as is the case in ASCÀS). Depending on the computer environment, such a GIS-external DBMS can significantly facilitate computations (lower CPU time). The major part of the data are supposed to be imported via network connection to other databases (e.g. national weather services). So far in this paper, only the conventional approach in using remote sensing or GIS techniques has been discussed. As a new approach, ÀSCAS uses the results derived in the image processing and the GIS module — in combination with data from the DBMS module - as input to a hydrological model. In ASCAS, the SRM (Snowmelt Runoff Model) (Martinec etal, 1994) is used for computing snowmelt runoff and climate change scenarios. As input variables to SRM, daily snow cover depletion data (derived from remote sensing data in the image processing and GIS module), daily minimum and maximum temperatures and precipitation data (all extracted from the DBMS module) are necessary. SRM computes the daily snowmelt runoff for mountainous basins based on the degree-day method. With SRM, important information for hydroelectric power companies and irrigation management (time and amount of peak flow) can be produced. Furthermore, SRM offers the possibility to compute the influence of temperature and precipitation changes (related to a possible climate change) on snow cover and runoff. Daily minimum and maximum temperatures can be decreased or increased by a selected number of °C whereas daily precipitation data can be increased or decreased by a selected percentage according to a specific climate change scenario. The climatological analyses with ASCAS


605

(Baumgartner, 1995) are carried out In several alpine basins (the larger ones are the Durance, Inn, Rhine, Salzach, Ticino). It has been shown that changed weather patterns over Europe since the beginning of the 1980s — especially in the winter half year — influence the snow cover distribution in the Alps and the surrounding lowlands. During the last decade, significant deviations from the long term mean snow coverage (i.e. the last 60 years) have been detected. The number of snow days (down from 40 to a few single days) and the snow depths at lower elevations ( < 1200-1500 m a.s.l.) have drastically decreased. The comparisons of snow cover accumulation and ablation charts have shown that the snowline in winter is now significantly higher (1200-1500 m a.s.l.) than at the beginning of the 1980s (500 m a.s.l.). Additionally, snow cover depletion in spring occurs much earlier (by two to three weeks) which has considerable influence on hydroelectricity production, irrigation and tourism.

CONCLUSIONS The advantage for the user of such an integrative use of remote sensing, GIS, DBMS and hydrological models — as has been shown with ASCAS - is the easy handling (import, export, transfer), processing, analysis and interpretation of the data without being a computer expert. Especially for repetitive investigations, an integrated system shows significant advantages compared to a separate use since import/export and transfer interfaces function on an operational basis and the interpreter is guided through the system all of which means an increased productivity. In the near future, more complex multispectral, multitemporal and multivariate data sets of different origin will be available. The needs for long term monitoring purposes are a high temporal (12 hours) and a moderate spatial (100-250 m) resolution as well as consistent data systems (sensors) and data sets. The imaging spectrometer - especially the moderate (MODIS by NASA) or medium (MEMS by ES A) resolution systems — will improve the collection of remote sensing data. In addition to a high temporal and a moderate (medium) spatial resolution, these systems offer an excellent spectral resolution which improves the utility in hydrological applications. For a comparison of global data sets, procedures for geocoding (including digital elevation models for a three-dimensional correction), calibration, normalization, and atmospheric correction must be standardized. Furthermore, processing algorithms have to be adapted for more synthetic, integrative and automated analyses including contextual pixel analysis (Nagao & Matsuyama, 1980). Present research shows that with classifications based on neural networks equal accuracies can be reached as with traditional statistical, parametric algorithms (Paola & Schowengerdt, 1995). Although these methods can easily be implemented, their major disadvantage is their slow and sometimes inconsistent learning phase (due to obscure input parameters). Research in GIS technologies focuses on object-oriented design (Raper & Livingstone, 1995;

606

M. F. Baumgartner

& G. M. Apfl

Zhan & Buttenfield, 1995) and how spatial data analysis tools can be developed to match the way users perceive their domains (Burrough & Frank, 1995). For the hydrological (and other) applications of image processing of remotely sensed data and spatio-temporal analyses with GIS, a complete integration of image processing, GIS and DBMS is necessary. In addition, digital (soft copy) photogrammetry, parts of the computer science and of digital cartography should be integrated leading to the new "iconic informatics" science (Li Deren, 1992). Future interpretation systems (expert systems) based on artificial intelligence could use a priori expert knowledge which is stored in a database and is updated with the solution proposed by the interpreter and the new findings (cognition). Remote sensing with its huge amount of data, repetitive procedures (time series analyses) and necessity of large physical and geographical a priori knowledge is an ideal field for the applications of expert systems (Matsuyama, 1987; Asrar, 1989).

REFERENCES Allen, J. A. (1994) Spatial data: data types, data applications and reasons for partial adaption and non-integration. In: Environmental Remote Sensing from Regional to Global Scale, ed. G. Foody & P. Curran, 214-222. Wiley, Chichester, UK. Antenucci, J. C , Brown, K., Croswell, P. L., Kevany, M. J. & Archer H. (Eds) (1991) Geographic Information Systems - A Guide to the Technology, Van Nostrand Reinhold, New York, USA. Asrar, G. (Ed.) (1989) Theory and Applications of Optical Remote Sensing, Wiley Series in Remote Sensing, Wiley, New York, USA. Baumgartner, M. F. (1990) Snow cover mapping and snowmelt runoff simulations on microcomputers. In: Remote Sensing and the Earth's Environment, European Space Agency (ESA), Paris, France. SP-301, 77-85. Baumgartner, M. F. (1995) Use of digital satellite image processing, geographic information systems, and modelling in snow hydrology and climatology. Tech. Rep. RS-11, Dept. of Geography, Univ. of Berne, Switzerland. Baumgartner, M. F. & Apfl, G. (1994) Towards an integrated geographic analysis system with remote sensing, GIS and consecutive modelling for snow cover monitoring. Int. J. Remote Sens. 15(7), 1507-1518. Baumgartner, M. F. & Rango, A. (1995) A microcomputer-based alpine snow cover analysis system. Photogramm. Engng Remote Sens. 61(12), 1475-1486. Burrough, P. A. & Frank, A. U. (1995) Concepts and paradigms in spatial information: are current geographical information systems truly generic? Int. J. Geog. Inf. Sys. 9(2), 101-116. Carroll, T. R. (1995) GIS used to derive operational hydrologie products from in situ and remotely sensed data. In: Geographical Information Systems in Assessing Natural Hazards, ed. A. Carrara & F. Guzzetti, 335-342. Kluwer, The Netherlands. Cowen, D. J., Jensen, J. R., Bresnahan, P. J., Ehler, G. B., Graves, D., Huang, X., Wiesner, C. & Mackey, H. E. (1995) The design and implementation of an integrated geographic information system for environmental applications. Photogramm. Engng Remote Sens. 61(11), 1393-1404. Duda, R. O. & Hart, P. E. (1973) Pattern Classification and Scene Analysis, Wiley Interscience, New York, USA. Ehlers, M., Greenlee, D., Smith, T. & Star, J. (1991) Integration of remote sensing and GIS: data and data access. Photogramm. Engng Remote Sens. 57(6), 669-775. Engman, E. T. & Gurney R. J. (1991) Remote Sensing in Hydrology, Chapman & Hall, London, UK. Estes, J. E. (1992) Remote sensing and GIS integration: research needs, status and trends. ITC Journal 1992-1, 2-10. Fabbri, A. G. (1992) Remote sensing, geographic information systems and the environment: a review of interdisciplinary issues. ITC Journal 1992-2, 119-126. Frank, A. U., Egenhofer, M. J. & Kuhn, W. (1991) A perspective on GIS technology in the nineties. Photogramm. Engng Remote Sens. 57(11), 1431-1436. Haefner, H. & Pampaloni, P. (1992) Water resources. Int. J. Remote Sens. 13(6&7), 1277-1303. Haralick, R. M. (1976) Automatic remote sensor image processing. In: Digital Picture Analysis ed. A. Rosenfeld, Topics in Applied Physics, Vol.11, 5-63. Springer, Berlin/Heidelberg, Germany. ILWIS (1990) ILWIS: Integrated Land and Water Information System, User's Manual (Version 1.2), Int.


607

Inst, for Aerospace Survey and Earth Sciences (ITC), Computer Department, Enschede, The Netherlands. Itten, K. I., Staenz, K., Baumgartner, M. F. & Johnson, T. (1985) A stepwise hierarchical multi-binary approach in TM landuse classification. In: Proc. 5th IEEE/IGARSS Symp. (Amherst, Massachusetts, USA), 965-969. Li, Deren (1992) From photogrammetry to iconic informatics. ZPF - Zeitschrift fir Photogrammetrie unci Fernerkundung 1, 5-9. Lillesand, T. M. & Kiefer, R. W. (1994) Remote Sensing and Image Interpretation (3rd edn), Wiley, New York, USA. Martinec, J., Rango, A. & Roberts, R. (1994) The Snowmelt Runoff Model (SRM) User's Manual, ed. M. F. Baumgartner, Geographica Bernensia P-29, Dept. of Geography, Univ. of Berne, Switzerland. Matsuyama, T. (1987) Knowledge-based aerial image understanding systems and expert systems for image processing. IEEE Trans. Geosci. Remote Sens. GE-25(3), 305-316. Nagao, M. & Matsuyma, T. (1980) A Structural Analysis of Complex Aerial Photographs, Plenum Press, New York, USA. Paola, J. D. & Schowengerdt, R. A. (1995) A review and analysis of backpropagation neural networks for classification of remotely-sensed multi-spectral imagery. Int. J. Remote Sens. 16(16), 3033-3058. Raper, J. & LivingstoneD. (1995) Development of a geomorphological spatial model using object-oriented design. Int. J. Geog. Inf. Sys. 9(4), 359-384. Schanda, E. (1986) Physical Fundamentals of Remote Sensing, Springer, Berlin/Heidelberg, Germany. Schowengerdt, R. A. (1983) Techniques for Image Processing and Classification in Remote Sensing, Academic Press, Orlando, USA. Sharma, S. K. & Ajaneyulu, D. (1993) Application of remote sensing and GIS in water resources management. Int. J. Remote Sens. 14(17), 3209-3220. Swain, P. H. & Davis, S. M. (Eds) (1978) Remote Sensing: The Quantitative Approach, McGraw-Hill, New York, USA. Zhan, F. B. & Buttenfield, B. P. (1995) Object-oriented knowledge-based symbol selection for visualizing statistical information. Int. J. Geog. Inf. Sys. 9(3), 293-316.