Geostatistical approach to Estimating Rainfall over Mauritius

1
Geostatistical approach to Estimating Rainfall
over Mauritius

• Mphil/PhD Student
• Mr.Dhurmea K. Ram
• Supervisors
• Prof. SDDV Rughooputh
• Dr. R Boojhawon

2
Objectives of Research

• Analyse qualitatively and quantitatively the
  spatial distribution of rainfall over Mauritius.
• Develop models for predicting values of rainfall
  at other locations from available measured
  rainfall values using geostatistics coupled with
  regression models.
• Generate digital results that can eventually be
  used in Geographical Information Systems (GIS).

3
Importance of modelling

• Precipitation measurements are unevenly
  distributed over a region and even less in a
  mountainous areas.
• One of the fundamental problem in hydrology is
  to estimate precipitation at unmonitored site
  using data from available surrounding
  precipitation stations.
• Hydrological models require precipitation fields
  on grid systems and in digital form.

4
Background work

• Study particularly aim at modelling the
  long-term variations in the distribution where
  the monthly mean values of precipitation for the
  period 1971-2000 are analysed for each month,
  i.e. for January till December using
  geostatistics.

5
Geostatistics

• Rainfall is also affected by elevation of an
  area and also by prevailing atmospheric
  conditions
• Rainfall is such a variable whose structure
  depends on direction (anisotropic).
• Need for another method whereby this anisotropic
  behavior can be modeled.
• Most common technique is geostatistics
• Special branch of statistics developed by George
  Matheron (1963) (Centre de Morphologie
  Mathematique).

6
VARIOGRAM

• It is the principle tool of geostatistics
• The variogram is a measure of how quickly things
  change on the average.
• The variogram gives an insight of the geometry
  and continuity of the variable
• The variogram is a function of direction, i.e. it
  characterizes the dependence that exist between
  variables at different points in space
• The underlying principle on the average, two
  observations closer together are more similar
  than two observations farther apart.

7
Exploratory data analysis

• Analyze data for trend, outliers and skewness.
• Outliers observed for most of the months.
• No significant trend over short interval but
  trend exist over long period.
• Data found to be skewed. Square root and log
  transformation applied to normalize data.
• Modelling is carried out using raw data, outlier
  removed data and detrended and transformed data

8
Model Approaches

• Basic geostatisitics model
• Raw data
• Outlier removed
• Transformed data
• Detrended data model
• Raw data
• Outlier removed
• Geostatistical- topographical model

9
Results (1)

• More accurate estimates obtained from detrended
  data model.
• Better estimates for winter months.
• High error values limited to less than 25 of
  the stations. Models yield very good estimates
  for 75 of stations.
• No improvements in estimates when removing
  outliers.

10
Results (2)

• Comparing with the rainfall volume calculated at
  the Mauritius Meteorological Services shows that
  actually the long term mean monthly rainfall, for
  1971-2000, has been underestimated for most of
  the months using the isohyetal analysis.

11
Challenges

• Optimize variograms particularly for summer
  months to improve accuracy of models
• Derive values of humidity, wind speed and
  direction, pressure and temperature on a regular
  grid from few available stations.

12
Limitations

• Very few stations on mountainous areas.
• Availability of finer resolution Digital
  Elevation Model .
• Effective estimation of missing rainfall data.

13
Future Works (1)

• To develop individual model templates for
  different weather systems.
• Include atmospheric variables such as pressure,
  temperature, humidity and wind speed and
  directions as additional parameters in the
  models.
• Besides using Ordinary Kriging and regression
  models, use cokriging to incorporate the
  secondary variables.

14
Future Works (2)

• Model using a smaller number of stations. This
  aims at removing redundant, rain gauges and their
  relocation.
• Implement precipitation forecasting models using
  artificial neural networks techniques.

15
Conclusions

• Geostatistics turned is a very good approach for
  estimating rainfall.
• Accuracy of model estimates depends on degree of
  optimisation of variograms.

16
Acknowledgements

• University of Mauritius for providing logistic
  support.
• Mauritius Research Council for funding this
  study.
• Mauritius Meteorological Services for providing
  required data.