Assessment Of Applying Different Datasets For Distributed Hydrological Simulation In Hanjiang River Basin

Hydrological simulation is based on observation. In the past, input data for hydrological simulation was achieved mainly from site observations and field investigations. In recent years, because of the rapid development of remote sensing technique and it’s combining with geographic information system(GIS) technique, continuous monitoring for surface parameters changes over large areas becomes possible. Change course of surface condition is reflected through long time series remote sensing data. A great deal of remote sensing data has been used to reverse surface parameters for land water cycle. That is a great support for hydrologic models. However, many remote sensing data products can’t be used directly or fully, because of the shortcomings of remote sensing technique. Though it is an opportunity for hydrologic models, a new task is coming. That is to discriminate remote sensing data for hydrological simulation. Development of hydrological science in the future mainly depends on available hydrological data and its accuracy. To discriminate remote sensing data, make full use and control of it, which makes it coupling with hydrologic models and GIS technique better, will be an important future direction of hydrologic models.Hanjiang river basin, which is located in the south of China and is covered with several vegetation types, was chosen for the study. Application of different datasets for distributed hydrological simulation was assessed using BTOPMC. Main research works and achievements were introduced as follows.(1) Based on three DEMs of HYDRO1 K, SRTM3(version 4) and ASTER GDEM, terrain information of Hanjiang river basin was extracted using the terrain module of BTOPMC. Comparisons show that:(1) SRTM3 can reflect terrain information and soil moisture spatial pattern over the basin in most detail, followed by ASTER GDEM. Because the low resolution of HYDRO1 K overly smooths basin terrain, as a result of topographic index, soil moisture distribution tends to be uniform over the basin. In other words in the view of runoff generation, the basin is easily saturated. In hydrological meanings, the HYDRO1 K does not keep enough details of mountainous and hilly terrain in the scale like Hanjiang river basin.(2) When the horizontal resolution of DEM reaches details in a certain degree, its vertical accuracy, compared its horizontal resolution, more controls extracted terrain reality. As shown by comparing ASTER GDEM and SRTM3, although ASTER GDEM horizontal resolution is higher than SRTM3, but SRTM3 has better vertical accuracy and the extracted terrain information is also more accurate.(3) The terrain information extraction is also affected by pit-removal algorithm and flow direction assignment in DEM processing. BTOPMC scans every grid of DEM by row and row. When pits are found, they are filled using a small elevation increment in a loop way until they are removed but the terrain still follows local landscape. The elevation increment becomes small and small from the source to the outlet along the general drainage direction of basin. In this way, basin terrain information is kept well. In assigning flow direction for a grid using D8 method, when the flow follows along the grid neighboring steepest-slope but not the lowest, more grids tend to flow vertically or horizontally. Due to DCW blue lines were incorporated in the generation of HYDRO1 K DEM, the impact of flow direction assignment on topographic index is limited on these grids near the basin sources where no blue lines are located in DCW.(2) Based on the three DEMs of HYDRO1 K, SRTM3(version 4) and ASTER GDEM, drainage networks of Hanjiang river basin were extracted by using the terrain module of BTOPMC. Comparison shows that:(1) the accuracy of drainage networks extracted from SRTM3 is the highest, then these from ASTER GDEM. These from HYDRO1 K are not so good.(2) The vertical accuracy of DEM controls the accuracy of drainage network extraction. The accuracy of drainage networks extracted from ASTER GDEM is not so good as from SRTM3, both in whole basin and in sub-basins.(3) In large basin, drainage networks extracted from HYDRO1 K are reasonable but become bad in low-relief regions. HYDRO1 K is unsuitable for drainage network extraction in small-scale basins.(4) The accuracy of drainage network extraction from DEM is affected by local terrain slopes and the depression filling algorithm. Compared with Arc Hydro Tools, the terrain module of BTOPMC avoids most parallel channels. Also, the threshold contributing area used for channel sources should be identified by referring to the map scale.(3) Using AVHRR NDVI, IGBP land cover classification and observed data from weather stations, potential evapotranspiration(PET) over Hanjiang river basin in 2000-2006 was estimated by Shuttleworth- Wallace(S-W) model. It shows that PET is not only affected by climate, but also changes with vegetation type and its growth. Sensitivity of PET to climate and vegetation was analyzed. Results show that:(1) PET is very sensitive to vegetation type. The calculated PET of different vegetation in the same climate condition is quite different. The mean PET of evergreen needleleaf forests, croplands and woods savannas is 1136.6, 965.1 and 563.2 mm/year respectively. The maximum value is twice as the minimum value.(2) The sensitivity of PET to climate is quite different for different vegetation covers. The PET of evergreen needleleaf forests is the most sensitive to vapour pressure. The sensitivity of vapour pressure is much higher than that of air temperature and solar radiation. The sensitivity of wind speed can be ignored. The sensitivity of the PET of croplands to air temperature, solar radiation and vapour pressure is high, while the sensitivity of wind speed is low. The PET of croplands is the most sensitive to air temperature. The PET of woods savannas is also the most sensitive to air temperature. The sensitivity of vapour pressure, wind speed and solar radiation is also high and very close to each other.(3) The PET of all vegetation covers is sensitive to LAI. The sensitivity of LAI is lower than that of meteorological factors(except for wind speed). The sensitivity of PET to LAI is different for different vegetation covers. The PET of woods savannas is the most sensitive to LAI, secondly evergreen needleleaf forests, finally croplands.(4) Based on the observed data from some ground meteorological stations, spatial distributed meteorological data was achieved by Kriging interpolation method. PET over Hanjiang river basin was estimated by S-W model. It was compared with PET calculated by CRU data. Results show that the accuracy of weather data affects PET estimation results. Input data with low resolution may homogenize PET to a certain extent in the same coverage. Spatial weather data is created by interpolating observed data from meteorological stations, so density of the stations will affect PET estimation results. The density of Kriging spatial interpolation weather data is much larger than that of CRU data.Space distribution and seasonal change of PET calculated by Kriging spatial interpolation weather data and CRU data was compared. It shows that PET calculated by Kriging spatial interpolation weather data is more accurate than that of CRU data.(5) Three IGBP land cover classification system based datasets, including global data IGBP DISCover and MODIS, and recent China data MICLCover, were compared and evaluated over Hanjiang river basin. Their spatial distributions were compared in the basin and accuracies were assessed by identifying Google satellite images in sample sites.(1) For IGBP land cover, land cover types of IGBP DISCover are fewer and its spatial variability is smaller than that of MODIS or MICLCover. But for big classes, their spatial distributions match well.(2) The components of woodlands classified by the three kinds of land cover data are quite different. Only MICLCover corresponds to the reality of the basin. The area of croplands classified by MODIS or MICLCover corresponds to the reality of the basin, while the croplands area of IGBP DISCover is much larger. The areas of urban, barren, wetlands and water bodies classified by the three kinds of land cover data are smaller than the reality of the basin, especially for IGBP DISCover, because resolution of 1km is not high enough to describe the land cover types which are small in area or distribute dispersedly.(3) Accuracies assessment by identifying Google satellite images in sample sites shows that classification accuracy of MICLCover is much higher than that of IGBP DISCover or MODIS. In IGBP DISCover, the main error is that many other land cover types are classified as croplands, while many are classified as grass in MODIS.(6) Three kinds remote sensing NDVI data from AVHRR, SPOT-VGT and MODIS were analysed to compare their similarities and differences on different vegetation types over Hanjiang river basin for 2001-2006 and their correlation was analysed using linear regression method.(1) In space, although they behave in a generally similar distribution pattern, and MODIS and SPOT-VGT NDVI match well, the distribution of AVHRR NDVI is somewhat different. Among them, MODIS can recognize objects more clearly on earth surface due to its spectrally narrow sensors with high spatial resolution. In MODIS NDVI, values vary in a wide range. It means that more vegetation types are detected and more scattered spatially.(2) In time, three kind NDVI changes seasonally similar with an almost same amplitude. In specifics, MODIS NDVI reflects vegetations seasonally change more accurate. In AVHRR NDVI, some vegetation types change not really.(3) Although different vegetation types seasonally change in a same way in three kind NDVI, in specifics, they change in a more temporally same pace in MODIS NDVI and SPOT-VGT NDVI than in AVHRR NDVI. In MODIS NDVI, distinguish seasonal change of croplands, closed shrublands and savannas, very different from other vegetation types, was detected more clearly than in SPOT-VGT NDVI and in AVHRR NDVI. It shows that MODIS NDVI has a high resolution on reflecting vegetation types and their development.(4) A linear relationship between different NDVI can be found for whole basin or at different vegetation types, and the linear relationship is most strong between MODIS NDVI and SPOT-VGT NDVI. Based on the linear regression of MODIS NDVI and AVHRR NDVI, MODIS NDVI is estimated accurately using AVHRR NDVI over Hanjiang river basin in 2000. In this way, MODIS NDVI time series can be extrapolated to the past years.(7) The spatial and temporal consistency of MODIS, CYCLOPES and GLASS LAI datasets was analyzed over Hanjiang river basin. Comparisons show that:(1) CYCLOPES LAI is observed to contain a large number of missing pixels, while MODIS and GLASS LAI products are more spatially and temporally complete. MODIS LAI contains many invalid pixels, whose LAI becomes much smaller abruptly comparing with the LAI values just before or after this time.(2) The spatial distributions of MODIS, CYCLOPES and GLASS LAI are mainly consistent with the vegetation types of the basin. The spatial distributions of MODIS and GLASS LAI are more consistent than that of CYCLOPES LAI. MODIS LAI is larger than GLASS LAI in forest pixels, while it is contrary in other pixels. CYCLOPES LAI is much smaller than MODIS and GLASS LAI in forest pixels.(3) MODIS, CYCLOPES and GLASS LAI products generally depict similar temporal trajectories with differences in magnitude. GLASS LAI has the smoothest and completest trajectories, while the trajectories of MODIS LAI contain a large number of erratic fluctuations. All of these three LAI products depict similar seasonal change for different vegetation types. Comparing with CYCLOPES LAI, a good agreement is achieved between MODIS and GLASS LAI values.