STATISTICAL SPATIAL DOWNSCALING TECHNIQUE OF GLOBAL PRECIPITATION MEASUREMENT (GPM) PRECIPITATION USING SATELLITE DERIVED VEGETATION AND TOPOGRAPHIC DATA

摘要

Global Precipitation Measurements (GPM) is the latest satellite mission operated to map the atmospheric precipitation over the globe by virtue of 0.1° spatial resolution and 3-hour temporal resolution. The recent GPM precipitation data improves the hydrological modelling quality by giving reliable now casting hydrological parameters from its raster-based precipitation pixels. Such advantages are best to be applied to map precipitation at regular period and higher diversity tropical rainfall particularly in Kelantan. Malaysia. Spatial resolution on pixels has greatly improved the modelling results by increasing the reliability and spatial accuracy of point measurement by gauge. Yet, the spatial variation propagates into GPM precipitation and has poorer correlation with gauges. This becomes more difficult if the gauge is sparsely located and having frequent void and unreliable readings. Therefore, localized precipitation map is needed and this may lead to spatial downscaling of the GPM precipitation image. Spatial downscaling approach determines the best scale at which the relationship between precipitation and other environmental parameters was established in the algorithm. Vegetation and local topography in the context of Normalized Difference Vegetation Index (NDVI) and Digital Elevation Model (DEM) respectively are common parameters being tested. Relationship of NDVI and DEM with GPM precipitation has never been tested in the Malaysia tropical rainfall and thus no intention was made for downscaling the GPM precipitation pixels. This study focuses to downscale the GPM precipitation by means of the statistical spatial downscaling method from the autocorrelation with the NDVI in Kelantan river basins. Online data repository of the GPM IMERG precipitation. Moderate Resolution Imaging Spectroradiometer (MODIS) derived NDVI and DEM from the Shuttle Radar Topography Mission (SRTM) were used. Temporal integration derives a monthly averaged 3-hour GPM precipitation and 16-day MODIS NDVI products taken from Nov 2014 to Dec 2014. DEMs STRM at specific locations were selected. Polynomial and exponential model between NDVI and GPM were used to predict the monthly precipitation. Monthly precipitation at 1 km2 was generated by superposition of the regressed precipitation estimate and interpolated the residual. The root mean square error (RMSE) and correlation coefficient (R2) were evaluated by comparing with gauge data. NDVI shows good correlation with GPM precipitation particularly in pixels where the intense rainfall is available. Exponential regression gives more representation of relation between NDVI and GPM as the bias was reducing. Spatial downscaling on GPM using NDVI is straightforward and can be used to determine large-scale precipitation map in Kelantan river basin area.