Articles & News Stories

First results of an integrated flood forecast system based on remote sensing data

Description
First results of an integrated flood forecast system based on remote sensing data
Published
of 13
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  An Integrated Flood Forecast Systembased on remote sensing data Heike Bach, Gudrun Lampart, Wolfram Mauser, and Gertrud Strasser  Abstract - The Integrated Flood Forecast System (IFFS) integrates multisensor remote sensing data within anoperational spatially distributed rainfall-runoff model. The aim is to improve flood forecast through a better estimationof spatial input parameters. Especially the potential of SAR data, used for the determination of topographicinformation and for soil moisture parameterisation, is considered. Optical data are used for the classification of theland cover, which also influences the runoff process. A first demonstration run of IFFS shows the applicability of thesystem and the sensitivity of the model towards soil moisture information, derived from ERS SAR data. I. INTRODUCTIONFloods are among the most severe risks on human lives and properties. The forecast and simulation of floods is essential forplanning and operation of civil protection measures (e.g. dams, reservoirs) and for early flood warning (evacuationmanagement). The economic importance of flood forecasting becomes clear considering that 85 % of civil protection measurestaken by the EC Member States are concerned with floods (EC Report Task Force Water, 1996).Hydrological modelling for flood forecast is widely applied. The models, which are used in practice, developed from simplestatistical models to more physically based systems driven by the rainfall as input. These rainfall-runoff models need thespatial characterisation of the land surface concerning parameters relevant for runoff formation. Static information orinformation with low temporal frequency can be gathered by mapping or using remote sensing data. Temporally highlyvariable land surface parameters required in these models, like soil moisture and snow properties, cannot be measured yet in aspatially distributed way and with the needed temporal frequency (1-3 days).  The new approach of IFFS is to maximise the utilisation of remote sensing data, especially of SAR data. This allows the easytransfer of the methodology to different watersheds and even in regions with limited information on the catchment and poordata availability. Further, the unique possibilities of remote sensing to spatially observe land surface properties should enablethe flood model to better react on the specific environmental situation and through this should improve flood forecast.II. METHODOLOGYIFFS consists of two parts. The first part of the system describes the watershed and is assumed to be temporally static. Thesecond part of IFFS consists of the modelling of the dynamic reaction of discharge on a rainfall input. The hydrological modelused for the translation of rainfall into runoff is a modified version of the SCS method TR-20 [6]. This model is routinely usedin hydrological practice. It has been modified to allow remote sensing inputs in an automated way using GIS tools. A moredetailed description of the methodology of IFFS, as summarised in Fig. 1, will be given using a demonstration case, a flood inthe Ammer catchment in the Alpine Forelands of Bavaria during August 1995.As prime remote sensed data source an interferometrically derived elevation model is used to determine topographicinformation on the watershed. The phase information of a tandem pair of ERS SAR RAW data serves as basis. The imageswere acquired on December 6 and 7 1995. The processed digital elevation model (DEM) has a spatial resolution of 30 m and arelative vertical accuracy of 10 m. This is sufficient for the extraction of flow patterns in hilly and mountainous regions togenerate the hydrological structure of the runoff model (upper part of Fig. 2). Additionally, optical satellite data (LANDSAT-TM) are classified to deliver the land cover distribution [7]. Together with a soil map, the watershed is then classified intohydrologic relevant classes of water storage capacity, so called CN-values (lower part of Fig. 2). Low values mean low runoff potential, whereas areas with high values produce a large amount of direct runoff and thereby contribute strongly to the floodpeak. In addition to remote sensing data, only information on the reservoirs (storage volume vs. discharge curve) is needed forthis static setup of the hydrological model.For the dynamic part of the system, which deals with a specific flood event, rainfall information is required as driving variable.In addition, also soil moisture information is of relevance for runoff modelling because it determines the extent of saturation of the watershed   and thereby the partitioning of rainfall into surface runoff and infiltration. The   same amount   of rainfall, whichnormally does not lead to a significant increase in water level, can cause a severe flood, if the soil has already been filled withwater and the storage capacity is close to zero.  Because information on the actual soil moisture distribution is normally missing, it is estimated in hydrological models byutilising an antecedent precipitation index, that is derived from rainfall measurements of the preceding days, or a moistureindex, that is derived from actual discharge values. However, this parameter can not reflect adequately the full temporal andspatial variability of soil moisture. Large errors in flood forecast may occur as a result. Therefore SAR data is used in the IFFSto derive soil moisture distributions to improve the antecedent moisture characterisation of the watershed. With thisinformation, the actual water storage capacity of the soil is determined as input into the rainfall-runoff model.The model to determine soil moisture from ERS data is based on a careful empirical analysis of the influencing factors on theradar backscatter signal. These influences can be summarised under three categories: topography, vegetation, soil type andwater content of the surface (top soil moisture content).The influence of topography is considered through the correction of the illumination. Also special image geometry effects inSAR images have to be corrected. For this, first the pulse timing and the local incidence angles has to be derived of each pixelthrough a reconstruction of the satellite orbit and using a DEM. A spatially weighted resampling method, which takes intoaccount the special SAR-geometry and conserves the backscattered energy for each radar pulse in combination with a cosinescorrection is applied to convert the measured backscatter intensity to equivalent flat terrain backscatter [3]. This method forgeoreferencing and correction of illumination effects in SAR images of hilly areas was applied to two ERS1/2 images (23 and24 August 1995) acquired shortly before a flood period. Fig. 3 shows a subarea of the ERS SAR scene before (left) and aftercorrection (centre).In a next step, the corrected ERS images were used for the derivation of soil moisture information prior to the flood in August1995. The basis of the approach for surface soil moisture determination from SAR is an algorithm developed for ERS data [1]which was already successfully used in a series of applications [4. 5]. The algorithm is based on the empirical compensation of the influence of vegetation and soil-surface scattering, organic matter content and density of soils on the backscattering signal,which in the model consists of a dielectric part and a geometric part.The correlation of radar backscatter and measured soil moisture showed different behaviour depending on land cover type. E.g.a barley field always showed lower backscatter coefficients compared to a maize field, if both fields exhibit the same soilmoisture and soil type. This variation of backscatter was interpreted as a result of different roughness values depending on theobserved land cover class. The basic assumptions of the approach to take into account vegetation roughness are taken fromfindings, which were gathered with the MIMICS (Michigan Microwave Canopy Scattering Model ) model [8]. The MIMICSmodel when run for a bare soil surface and a C-band VV, 23 o  incidence angle configuration in general shows a parabolic  increase of backscatter with increasing dielectric constant (DC). Backscatter converges towards a saturation level withincreasing DC. Different RMS-roughness values of the soil surface shift this general dependency of backscatter vs. DCtowards higher backscatter values with increasing roughness. A RMS-roughness value of 0 cm shows the smallest backscattervalues, a saturation level for the roughness influence on the signal is reached at an RMS-value of 2.4 cm. Beyond this value theinfluence of the surface roughness on the DC signal therefore vanishes.Since vegetation cover adds roughness to the surface as much as soil-roughness elements do, its influence on the DC retrievalaccording to MIMICS should also consists in an offset value added to the basic relationship between DC and backscatter. If this offset can be determined empirically and assigned an equivalent soil-RMS-value, it can be compensated by adding anappropriate vegetation dependent offset. In the model the offset is determined in a way, that it converts all vegetated surfacesinto equivalent bare soil surfaces with a surface roughness of 2.4 cm. This eliminates the influence of surface roughness as hasbeen stated before.To quantify the species dependent offset values, multidimensional statistical analyses of the backscatter signal in relation toground-truth measurements were carried out. The measurements consisted of the average ERS-backscatter of the observedfield, the measured soil moisture of the upper soil layer, soil type, land cover and vegetation type, vegetation height andbiomass. During the calibration phase of the model only flat fields were considered to reduce topographic effects. Thecorrected signal can then be used to retrieve surface soil moisture (Fig. 4).The analysis of the ground truth data has shown, that this correction approach is not sufficient for grassland. For grassland, thebackscatter signal is also influenced by the vegetation structure. Meadows change their phenotype not only within a vegetationperiod, but also through cultivation activities of the farmers. The structure change can be quantified by dry biomassmeasurements in the fields. Using biomass measurements of meadows in the test-sites, the radar signal can be corrected.The last missing step is the conversion of the dielectric constant of the surface soil underneath the vegetation into soil-moisturecontent. The value of the dielectric constant is dominated by the water content of the soil, because the dielectric constant of water is approximately 20-30 times higher than that of the soil matrix. Thus the dielectric constant of the soil-water mixturechanges strongly with the soil water content. The soil moisture retrieval model works well for annual vegetation if all theancillary data are available with the appropriate accuracy. The necessary information on land cover was also used to determinethe forested areas and settlements, where soil moisture can not be determined with C-band SAR data. The resulting top soilmoisture distribution for August 24 th  1995 is illustrated in Fig. 3 (right). Soil moisture values vary between 40 and 55 Vol. %.  III. RESULTSTest runs were conducted for the flood event in August 1995 using interpolated rainfall fields as input into the model. Fig. 5shows the rainfall which was measured at one meteorological station together with the observed runoff at the gauge station atthe outlet of the Ammer catchment. In the top section of this graph also two model results are shown. The upper curve showsthe modelled runoff under the assumption that the moisture conditions prior to the flood are wet, which means close tosaturation. The lower curve is modelled under the assumption that the soil is dry and thus has a high storage capacity for water.These two assumptions are the two extremes concerning soil moisture conditions. They have been chosen to demonstrate thehigh impact of the soil moisture on the model result and to illustrate the sensitivity and range of the modelled dischargedependent on the antecedent moisture conditions.Further, also the ERS observations are indicated. Images of August 23 rd  and 24 th  were available. The soil moisture maps, whichwere calculated for each image separately, were averaged and mean values for each subwatershed were calculated. Thisinformation was used to derive the antecedent moisture conditions for the next model run. In the actual version of IFFS as inthe standard version of the applied rainfall-runoff model, only three moisture conditions can be assumed (wet, dry and normal).Therefore as a first approach a classification of the soil moisture map was performed assigning soil moisture values below 30% to dry, up to 50 % to normal and above 50 % to wet conditions. In a extended version of IFFS it is planned to directly usethe derived soil moisture values in the hydrological model. The bottom section of Fig. 4 shows the result of the modelledrunoff using the soil moisture conditions derived from the ERS observations. The hydrograph is compared to the measureddischarge and both coincide very well in amplitude as well as shape. This is achieved without any calibration of the rainfall-runoff model.IV. DISCUSSIONThe demonstration run shows the potential of IFFS. Except for the rainfall measurements, the soil map and additionalhydrologic information like reservoir data, all information and input data are remotely sensed. It should explicitly bementioned here that no area-dependent calibration of the flood model has been conducted in calculating the results of Fig.5.Since IFFS is primarily based on remote sensing data, it can be transferred and applied also in regions were information on the
Search
Similar documents
View more...
Tags
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks