Geographically weighted regression model for spatial downscaling of global precipitation measurements data in Kelantan
Rainfall is one of the prominent parameters in the hydro-climatic process, it is typically measured by using gauge which is limited to station samples and inherent systematic error with the requirement for regular instrument calibration. To overcome such limitations, the Tropical Rainfall Measuring...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/99806/1/NoorEmiFadzilahMFABU2022.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Rainfall is one of the prominent parameters in the hydro-climatic process, it is typically measured by using gauge which is limited to station samples and inherent systematic error with the requirement for regular instrument calibration. To overcome such limitations, the Tropical Rainfall Measuring Mission (TRMM) and the Global Precipitation Measurement (GPM) which are remote sensing precipitation satellite missions, have been used at a regional scale to provide reliable precipitation estimates over large spatial extent within the spatial resolution of 0.25o and 0.1o respectively. However, it is difficult to spatially match with the point-based gauge data at an acceptable local scale and thus, gives a poor empirical relationship. Previously, spatial downscaling algorithms using simple statistical models were devised by spatially correlating them with normalized difference vegetation index (NDVI), and digital elevation model (DEM) data at the higher spatial resolution, but the outcomes were unsatisfactory due to goodness of fitting dependent and spatial non-stationary influence. As such the aim of this research was to apply the Geographically Weighted Regression (GWR) method which put forward local regression with spatial non-stationary modelling to downscale both satellite precipitation data by showing the cross-correlation between NDVI and DEM data at high spatial resolution. The objectives were to develop a local downscaling method using multi and single variables to estimate rainfall at 1km spatial resolutions by using GWR modelling based on non-linear regression method; to assess the impact of spatial variability on local downscaled rainfall algorithm in the different model considering the different spatial resolutions; and to evaluate the quality of downscaled rainfall data with rain gauge measurements in differentiating the light and heavy rainfall. GPM, TRMM, Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI, Shuttle Radar Topography Mission (SRTM) DEM and ground gauge data were applied over Kelantan area for three consecutive periods from October 2013 to December 2016. Polynomial Regression (PR) and GWR were employed to downscale annual and monthly satellite precipitation from 25km and 10km to 1km spatial resolutions. Ground gauge data were used to validate the accuracy of light and heavy rainfall at below and above 200 mm, respectively. The GWR model improved the precipitation accuracy obtained by GPM as compared to TRMM by about 40% due to better spatial resolution pixels. PR models were limited for higher spatial non-stationary exhibited by homogeneous vegetated areas at low elevation and heterogeneous elevation. GWR had the least impact of the spatial non-stationary with 30% reduction of Root Mean Square Error (RMSE) similarly obtained by PR. Light rainfall was evident along the coastal line and the heavy rainfall was concentrated in the vigorous vegetated areas around the Kelantan area. This study proves that the GWR downscaling approach is suitable for tropical rainfall types in Kelantan and cross-correlating it with other rainfall related geo-parameters such as vegetation index and elevation. |
|---|