To find out the variation of precipitation in the region (Assam) monthly
average rainfall data of 28 stations were taken for a period of 30 years
ranging from January 1988 to December 2018. Spatial variation of
rainfall of each month is evaluated using Geospatial analyst tool of
ArcGIS (version 10.2.2). Five interpolation techniques IDW, GPI, RBF,
LPI, Kriging were used to find out the spatial variation of rainfall
over the entire region of Assam. The results obtained by all the
techniques were cross validated by finding out the errors and efficiency
of all the models using the performance measures discussed above. Table
1 shows the performance of all the models. The rank column in the table
tells the overall performance of all the interpolation techniques on the
basis of performance measures. From the table we can see IDW at p=3 is
the most efficient model while 3rd order GPI is least
efficient out of the 20 models utilised. Second best model is again from
IDW for weight parameter p = 4. The other models which give reliable
results are Simple Kriging and RBF with basis function as spline with
tension. However, in some cases user is constrained to use a particular
technique only, for such cases also table 1 should be used as a guide to
choose the best parameter of a particular model. From table we find out
GPI gives its best result when the polynomial order is taken as one.
Performance of LPI is good when the kernel function is chosen as
Gaussian; similarly for Kriging the results are more favourable for
simple as compared to that of universal. Fig 2- Fig 6 shows how the
rainfall varies spatially in Assam during month of January using 5
models with the best parameters. Obviously as discussed above IDW will
represent the phenomenon more accurately compared to remaining four
models.
Fig 7 – Fig 9 shows how the errors are distributed spatially while
predicting the rainfall. It is observed that the regions where the
precipitation is more intense shows relatively higher error compared to
the region where precipitation is less intense. It is clearly visible
that the districts like Dhemaji, Tinsukia, Lakhimpur and Dibrugarh are
receiving higher amount of rainfall and the errors associated is also
higher in these regions.