Rezultaty
A processing chain was developed for modelling and forecasting runoff in mountain basins where main contributions to discharge come from snow and glacier melt. The runoff model represents an extension of the semi-distributed Snowmelt Runoff Model (SRM) of Rango and Martinec. The processing chain includes modules for pre-processing of hydrological and meteorological input data at different temporal and spatial scales, and for assimilating snow covered area derived from satellite images in the model.
The runoff forecasting tool runs automatically, utilising numerical meteorological forecast data and satellite sow maps as input. The software was successfully tested in real time in spring 2005, providing forecasts of daily runoff 1 to 7 days ahead in watersheds of the Eastern Alps. The runoff forecasting tool is ready to be implemented in an operational runoff modelling and forecasting system.
The temporal development of the spectral reflectance of snow due to snow metamorphosis and variable contents of impurities, the anisotropic effects in the snow reflectance and the variable reflectance of the bare ground all make precise measurements of the fractional snow cover area (FSCA) challenging. In addition, variable local acquisition and illumination geometry varies due to sun elevation, topography and observation angle. No algorithm so far has been able to give accurate results through the whole melting season. The new algorithm takes all the effects mentioned into account and thereby makes very precise estimates of the fractional SCA.
The concept assumes a day-to-day monitoring of the snow from winter conditions until all snow has melted. The developing spectrum of the snow is both observed by satellite sensors, giving samples of the BRDF, and modelled by including an empirical snow metamorphosis model and a snow impurity model giving full BRDF. Snow impurity development and snow metamorphosis models are used to infer the development of the snow albedo.
The impurity model is taking into account the typical impurity development based on time, land surface cover type and bare-ground fraction. The metamorphosis model follows a similar scheme based on time. The predicted snow spectrum and the local bare ground spectrum are applied in a linear spectral unmixing algorithm to estimate the area fraction of snow (FSCA) and bare ground. By using predicted spectra for the current situation and not a pool of all possible spectra, the result is more accurate and reliable.
The surface temperature of snow (STS) is a geophysical parameter that is relevant for hydrological modelling of the snow melting as well as for climatic monitoring. Retrieval of STS by means of satellite remote sensing is undertaken in the thermal part of the spectrum. Snow surfaces are close to blackbodies, but because of the atmosphere, the brightness temperature observed from the satellite will typically differ from the real surface temperature of the snow cover. The atmospheric attenuation depends on the wavelength of the observed radiation, the length of the atmospheric path between the surface and the satellite, and the chemical composition of the atmosphere.
Methods for retrieving STS from EO data will typically combine observations in different spectral bands and/or from different view angles. We identified Keys algorithm as one of the best single-view techniques for polar atmospheres, and it can be applied on MODIS as well as AVHRR data. The algorithm has been tailored to the snow application and validated using field data and meteorological measurement stations.
Snow wetness is an important indicator in hydrology for snowmelt onset and the development of the snowmelt process. The approach we have developed is to infer wet snow from a combination of measurements of snow temperature (STS) and snow grain size (SGS) in a time series of observations. The temperature observations give a good indication of where wet snow potentially may be present, but are in themselves not accurate enough to provide very strong evidence of wet snow. However, a strong indication of a wet snow surface is a rapid increase of the effective grain size observed simultaneously with a snow surface temperature of approximately 0°C. We have developed an algorithm applying this approach to infer coarse snow wetness classes. The algorithm has been validated using field data and meteorological measurement stations.
Norut IT has developed a set of tools and methods that together are the basis for generating snow covered area maps and snow wetness maps from synthetic aperture radar imagery. The tools are mainly implemented in the programming languages Envi/IDL under a common framework. Parts of the system have been separated as pre-compiled c-routines.
The system is based on using synthetic aperture radar data as input. In addition the system requires input from a digital elevation model and a basic land cover mask (glacier, lakes and forest themes). The system also requires a reference SAR input file for each SAR geometry to be processed.
The core classification of snow is done by applying the Nagler-approach (Nagler and Rott, 2001) where wet snow is classified by using its low radar backscattering as opposed to the higher backscatter by dry snow or bare soil. The current SAR image is thresholded against a reference SAR scene with the same satellite geometry that is known to contain dry snow or bare ground. In addition dry snow is inferred from this by using DEM information (see Malnes et al., 2002). Image pixels above the mean altitude of wet snow that are not classified as wet snow are assumed to be dry snow.
These methods are in them selves generic, and adaptable for several existing and future sensors, as well as for different geographical regions. Today, the system has been limited to certain geographical regions (mountainous parts of Southern Norway), and to certain sensors, and sensor modes (ERS, Envisat and Radarsat). It is, however, relatively easy to extend the system to other regions and other sensors.
NORUT IT intends to exploit its parts of the Envisnow snow parameter retrieval system in future applications and for commercial applications.
Methods and software have been refined to enable automated snow mapping for imaging radar (synthetic aperture radar, SAR) and optical sensors. The processing steps include terrain-corrected geocoding for any kind of terrain (if digital elevation data area available), classification of snow covered area based on backscattering (radar) or multi-spectral reflectance signatures (optical), and transformation of the data into various types of map projection. SAR enables observation of the Earth´s surface independent of clouds and daylight, thus being of interest for operational applications in hydrology and water management.
A limitation of the presently available SAR sensors is that they enable detection only for melting snow. The methods and software were successfully tested for SAR on board the satellites Envisat, ERS-1, ERS-2, Radarsat and SIR-C/X-SAR. For optical sensors the snow classification tool was applied and tested for medium resolution (MODIS, MERIS) and high resolution sensors (Landsat, ASTER, etc.).
Snow cover maps for SAR and optical imagery show some systematic differences because of different observation geometries, sensor resolutions and sensitivities to snow physical properties. Therefore a procedure was developed for homogenizing the snow cover maps from the different sensors. The developed software enables the exploitation of a wide range of sensors for snow mapping and is suitable for operational applications in real time.
Snow water equivalent (SWE) is the key snow parameter of interest for hydropower production, flood management and climate studies. So far, there are no operational SWE-methods. Norut IT has developed an innovative delta-K interferometric SAR method for retrieval of SWE. It has been demonstrated on ERS data using a summer and a dry snow winter scene. Averaging with 10km x 10km resolution gives sub-degree standard deviation on phase estimate. 1-degree phase change corresponds to 10 cm snow. This phase term is for dry snow proportional to the snow water equivalent. SWE, and can hence be derived.
The methodology is based on natural permanent scatterers and dry snow conditions. In order to resolve phase wrapping ambiguities caused by refraction in dry snow, a band splitting method (delta-K) has been implemented. This results in 36 m unambiguous snow depth.
A new time series of ASAR SLC data is presently collected over the Altevatn area, northern Norway, and the methodology will be tested on these data.
We believe that the potential for the delta-K interferometric technique for SWE estimation has a huge future potential, and will, if successfully demonstrated be very interesting for hydrological users and hydropower production users. Norut IT intends to exploit the method in future projects, and for future applications.
A complete algorithm to generate soil moisture maps from SAR data in mountainous region has been developed and implemented. The algorithms includes the following phases:
- Extraction of the area of interest from SAR image.
- Calibration of backscattering coefficient taking into account the local incidence angle by means of the digital elevation model (DEM) and orbital parameters.
- Geocoding of the image.
- Masking of the SAR image for shadow and layover.
- Overlaying of a vegetation map, obtained from ground data, optical images or cross-polarized SAR data, to exclude forest areas from the analysis and to take into account the attenuation effect of the grass where it is present.
- Masking of image for forests.
- Correction for the effect of herbaceous vegetation on the backscattering coefficient by using a Discrete Element Radiative Transfer Model.
- Extraction of soil moisture at pixel scale by using an inversion approach based on a Neural Network trained with experimental or model generated data.
- Generation of soil moisture maps by aggregating a suitable number of pixels.
The algorithm has been tested on two test areas in North Italy: a flat agricultural area (Scrivia watershed) and a mountainous site (Cordevole watershed). In both cases, up to five levels of soil moisture between 10-15% and 45% have been identified, with a mean error of the 10%.
Electromagnetic models simulating backscattering from natural media are fundamentals, for interpreting experimental data, performing sensitivity analysis and retrieving the unknown parameters. Two different codes have been implemented in Matlab.
In one approach propagation and scattering in snow are described by using the Dense Medium Radiative Transfer theory (DMRT) assuming that the particle size is small compared to the operational wavelength of sensor and that the effective propagation constant has a small imaginary part compared with its real part. In an alternative approach, based on the Strong Fluctuation Theory (SFT), the inhomogeneous layer of snow is modelled as a continuous medium with the scattering effects taken into account by making use of random fluctuations of permittivity. The latter ones can be described by a correlation function, with the variance characterizing the strength of the permittivity function of the medium and correlation length corresponding to the scales of the fluctuations. An effective permittivity is used to characterize the randomness and scattering effects.
Both models have been validated with experimental data. The sensitivity analysis carried out as a function of sensor and medium parameters has pointed out the following:
- Backscattering from dry snow increases as a function of observation frequency, snow depth, snow water equivalent and particle radius, and shows a maximum as a function of volume fraction.
- At the Envisat ASAR frequency (C-band) the contribution to total backscattering of a layer of dry snow 2 meter deep is very low and close to the threshold of SAR sensitivity.
- For a soil covered by wet snow the backscattering coefficient decreases as a function of snow wetness with a trend that is a function of frequency.
An Algorithm for the retrieval of dry snow water equivalent (SWE) and snow depth (SD) by using multi-frequency radiometric data from satellite and artificial neural networks (ANN¿s) has been developed and tested by using data at 19 and 37 GHz. The algorithms have been tested by using data from the Special Sensor Microwave Imager (SMM/I) collected for long periods of times (years) over Finland, and from the Advanced Microwave Scanning Radiometer (AMSR-E) acquired over a large in Norway) in the 2002/2004 period.
The results obtained have been compared with those obtained using other approaches on the basis of the root mean square error (RMSE) and the regression coefficient. It has been shown that the ANN based technique gives significantly better results than other approaches, such as those based on semi-empirical models or on iterative inversion. The developed technique, which is suitable for near real-time applications, can be very useful when periodical ground measurements are collected in a few stations only, and no information is available from any areas in between.
Frequent mapping of snow parameters, like snow cover area (SCA), is important for applications in hydrology and climatology. The objective is to analyse on a daily basis a time series of optical and Synthetic Aperture Radar (SAR) data together producing sensor-independent products. A multi-sensor SCA product has been defined. A prototype production line has been developed to automatically perform data retrieval, pre-processing, parameter retrieval and product generation.
The approach is to analyse each satellite image individually and then first to combine them into a day product. How each image contributes to the day product is controlled by a pixel-by-pixel confidence value that is computed for each image analysed. The confidence algorithm may take into account information about the local observation angle/IFOV size, probability of clouds, prior information about snow state, etc. The time series of day products are then combined into a multi-sensor/multi-temporal product.
The combination of products is done on a pixel-by-pixel basis and controlled by each individual product/pixels confidence and a decay function of time. The multi product is then to represent the most likely status of the SCA.
Norut IT has developed a set of methods and software for geocoding and calibration of data from satellite borne Synthetic Aperture Radars (SAR). The software is adapted to the Envisnow processing system, and works automatically for specified data from the Envisat ASAR, ERS-1 and ERS-2 and Radarsat instruments. The tools are mainly implemented in the program languages Envi/IDL under a common framework. Parts of the system has been separated as pre-compiled c-routines. The output is a geocoded and calibrated SAR image on a selected projection.
The system is based on using SAR data as input. In addition the system requires input from a digital elevation model of the area covered by the SAR image.
The methods are in it selves generic, and adaptable for several existing and future SAR sensors, as well as for different geographical regions. Today, the system has been limited to certain geographical regions (mountainous parts of Southern Norway), and to certain sensors, and sensor modes (ERS, Envisat and Radarsat). It is, however, relatively easy to extend the system to other regions and other sensors.
NORUT IT intends to exploit the geocoding software in future applications and for commercial applications. A continuous development of the software, adapting to new SAR instruments and new locations are foreseen.
Snowmelt is a significant contributor to spring floods in Norway. Updated information of the snow conditions is therefore of major importance to the national flood forecasting. At the Norwegian Water Resources and Energy Directorate (NVE) daily flood predictions are carried out based on runoff simulations using the HBV-model. The HBV model is a lumped rainfall, runoff model which uses precipitation and temperature as input parameters.
In order to improve the spring flood prediction a study testing operational use of satellite-observed SCA (snow covered area) in the HBV-model is carried out. The study includes
- Calibration of HBV-models against both discharge and SCA, and against discharge only, and
- Updating of the HBV-models based on satellite observed SCA.
The results show that the HBV-models calibrated against SCA in addition to discharge simulate discharge as well as models calibrated against discharge only. A success rate of 28 percent was found for the updates of the model. The success and failure of the updates was quite random. However, a weak tendency of higher success rate at large SCA values was found.
A new dynamic snow distribution model was also implemented in the HBV-model. This applies a gamma distribution in which the parameters are functions of the number of accumulation and melting events. In this was the modelled spatial distribution of SWE (snow water equivalent) more closely follows the observed spatial distributions of SWE. HBV-models with the new snow distribution model predicts discharge as good as the traditional snow distribution model, predicts SCA better and gives a consistently improved prediction of the discharge when updating the model for satellite derived SCA.
Norut IT and Norsk Regnesentral have developed the EnviSnow system prototype, which make it possible for the users to utilize snow/soil parameter algorithms to retrieve snow and soil parameters from EO data (SAR and optical satellite data).
The project partners operate the system during the project lifetime, and it will be used during the demonstration of the project results. But the system (or at least parts of the system) will also be useful for the partners after the end of the project.
After the end of the project the EnviSnow system will be used by:
- End Users (like NVE) to generate products used in hydrological models
- Research Institutes (like Norut IT) in their daily research activities
The system is based upon the Linux operating system and OGC standards. Parts of the system have been separated as pre-compiled java-routines.
Norut IT intends to exploit the Envisnow system (or at least parts of it) in future applications and for commercial applications.
The main components developed are:
- Production line including all the processing algorithms in the project (based on ENVI)
- Production storage including databases and catalogue service (based on GIN)
- Web visualizing
Production line:
The framework (based on ENVI) is intended for automatic processing of remote sensing and in situ datasets for the extraction of snow related parameters. A simple controller controls the production line where one dataset is processed at the time, running the process through the necessary steps. In addition to the actual processing line, ENVI functionality is made available through the ENVI menu.
The framework is flexible in that it puts few restrictions on the application developers, and makes it possible to plug in new methods and data formats without having to change the framework.
It offers support for putting processed products into the Production storage. Remote: This processing mode is intended for the daily production of data. The data provider sends an email to the system when new data are ready for download. These data are then automatically downloaded (FTP) and processed. Local: This processing mode is intended for reproducing results from locally stored data, e.g., when an algorithm has been improved. A list of local datasets is then processed.
Production storage:
The Catalog and Storage Service required by the EnviSnow system, is based on the GIN (Geographic Information Network) software developed at Norut IT. The GIN software is the result of R&D work at Norut IT during the last couple of years. It has now reached a sufficient level of maturity and stability to be useful for other research projects like EnviSnow. At the same time experience gained in EnviSnow will influence further evolution of GIN.
In order for the user to save a product in the EnviSnow database, metadata always have to be filled in. The system will not allow a product to be saved in the database without metadata. The project has defined some common attributes for all data sets (metadata).
In addition a GIN-based Web Map Service (WMS) is developed.
Web visualizing:
The results will be made available as files for visualising in a web-based simple viewer. A simple WMS client is developed, that make selected data in the Production Storage available to public users as simple images (shape files).
The results from the hydrological models (HBV) is also saved and viewed in the same way as the other geographical results in the EnviSnow system.
A new snow distribution model has been implemented in the Swedish HBV model. The new distribution model takes into account that the spatial distribution of snow changes during the snowy season, from a very skewed at the start of the accumulation period, a steadily decreasing skew in during the accumulation period and an increasing skew during the melting period.
The shape of the distribution of snow is believed to be important for the development of snow free areas during melting. The distribution of snow is at all times gamma distributed, but with the shape parameter as a function of the present amount of snow. The increase of snow-free areas as a response to a melting event is analytically linked to the parameters of the distribution.
The approach was implemented in the HBV model and tested for two alpine basins in Southern Norway. The new approach gave similar results as with using the traditional distribution model. However, when deviations between modelled and remotely sensed snow coverage were taken into account, significant improvements was observed for one of the basins.
The method of the assimilation of snow cover area (SCA) satellite observations into the hydrological model was developed. The method is based on the correction of the simulations of SCA by hydrological model to agree with the observed SCA.
The hydrological model is improved to simulate the SCA by using the elevation and slope information from the digital elevation model (DEM). The elevation has an effect on both snow accumulation by lower air temperatures and more precipitation on higher elevations and snow melt by delay in melting on higher elevations due to the lower air temperatures. The elevation effect is simulated in the improved hydrological model in order to simulate the spatial variability of the accumulation and melting of the snow cover that allows simulating the snow cover area.
The method of data assimilation in the hydrological model is improved by adding the assimilation procedure of SCA observations to the assimilation of other data (such as discharge, water level and snow water equivalent). The task of data assimilation procedure is to correct the simulation of the hydrological model to agree with all these observations. The SCA observations are used to correct the simulated amount of the snow water storage to agree with the real snow water storage, which is not observable due to the spatial variability of the snow cover properties. Using the SCA observations in the hydrological model is the large step ahead towards the spatially distributed snow cover simulation.
The effect of the SCA observations on the accuracy of the forecasts was tested by making forecasts over spring period with and without SCA observations. The SCA observations were avail-able for the springs 2001-2005.
The observations from springs 2001-2003 were used for the calibration of the model. The data form the spring 2004 was used to test the effect of the SCA observations on the accuracy of the forecasts. The hydrological model was run in two versions with and without using the SCA observations and the simulated flood forecasts were compared. Since spring 2005 the developed method is used in the Finnish operational hydrological forecasting system SYKE-WSFS. The real time flood forecasts are made for the territory of whole Finland, including cross-boundary watersheds, total of 390000km{2}.
Description:
A method to estimate the real fraction of snow covered area (SCA) using
Terra/MODIS-data is described here. The method is based on a semi-empirical reflectance model, which expresses the reflectance from target area as a function of three reflectance contributors (wet snow, dense coniferous forest and snow-free ground), apparent forest transmissivity and SCA. In addition to MODIS, the method can be adapted to any optical sensor. The method is especially designed to be applied in boreal forest zone. The forest canopy does not hamper the estimation: the apparent transmissivites just have to be estimated using MODIS data under full dry snow conditions.
Dissemination and use potential:
The produced SCAs may be disseminated as numerical files or thematic maps.
At the moment, as the method is operatively applied in Finland, the results are presented related to Finnish natinal coordinate grid, but also other coordinate systems can be used, depending on the application area.
The use potential is high, as the method can be applied to work with any calculation area larger than MODIS pixel (0.5km x 0.5km) and anywhere in boreal forest zone.
Key innovative features:
- The relatively simple reflectance model is easy to apply for any optical sensor
- The method does not need auxiliary land use data except information on water areas (which are very easy to classify using summertime satellite images)
- A novel method where forests are concerned via forest transmissivity, which is calculated using the same kind of satellite data as, is used for the SCA-estimation.
Current status:
The method will be tested using in situ SCA-information at wether stations and snow courses in Finland. Good results are expected, as SCA-estimates produced by applying NOAA/AVHRR-data are proved to be accurate, when validated against those in situ observations.
Use of the method:
The method now used to provide sca to be assimilated with Finnish national hydrological modelling system.When applied outside Finland, hydrologists are able to use the Method to produce SCA-estimates in a way most beneficial for their needs.
A simulation exercise has been performed in order to study the temporal development of snow-covered area and the spatial distribution of snow water equivalent (SWE).
Special consideration has been paid to how the properties of the spatial statistical distribution change as a response to accumulation and ablation events. A distributed rainfall-runoff model at resolution 1 1km{2} has been run with time series of precipitation and temperature fields of the same spatial resolution derived from the atmospheric model HIRLAM.
The precipitation fields are disaggregated and the temperature fields are interpolated. Time series of the spatial distribution of snow water equivalent and snow-covered area for three seasons for catchments in Norway is generated. The catchments is of size 3085km{2} and two rectangular sub-areas of 484 km2 are located within the larger catchments.
The results show that the shape of the spatial distribution of SWE for all three areas changes during the winter. The distribution is very skewed at the start of the accumulation season, then the skew decreases during the accumulation season, and as the ablation season sets in, the spatial distribution again becomes more skewed with a maximum near the end of the ablation season. For one of the sub-areas, we find a consistently more skewed distribution of SWE, which is addressed to higher variability in precipitation.
This indicates that observed differences in the spatial distribution of snow between alpine and forested areas can be a result of the differences in the spatial variability of precipitation. The results obtained from the simulation exercise are consistent with a new approach of modelling the spatial distribution of SWE as summations of a gamma distributed variable.
The further development of the hydrological model and data assimilation algorithm for the Finnish flood forecasting system (SYKE-WSFS) was finished in 2004. The effect of the SCA observations on the accuracy of the forecasts was tested by making forecasts over spring period with and without SCA observations. The accuracy was tested on springs 2003-2005. The hydrological model was run in two versions with and without using the SCA observations and the simulated flood forecasts were compared in 14 different discharge measurement points. When the final version of the data assimilation algorithm was applied, the accuracy of the forecasts improved considerably or slightly in 1-3 forecast points, depending on the spring and the accuracy did not decrease in any of the forecast points.
Since spring 2005 the developed method is used in the Finnish operational hydrological forecasting system SYKE-WSFS. The real time flood forecasts are made for the territory of whole Finland, including cross-boundary watersheds, total of 390 000km{2}. The forecasting system provides forecasts for over 1200 points on lakes and rivers in Finland. The forecasts are provided for public by web pages in http://www.environment.fi/waterforecast where forecasts are updated several times per day. The hydrological forecasting system SYKE-WSFS amongst the other hydrological variables shows also simulated forecasted and observed SCA for over 1200 simulated points. The maps of simulated and observed SCA for the territory of Finland are produced daily.
Wyszukiwanie danych OpenAIRE...
Podczas wyszukiwania danych OpenAIRE wystąpił błąd
Brak wyników