Assessment of Three Spatial Interpolation Models to Obtain the Best One for Cumulative Rainfall Estimation (Case study: Ramsar District)

Assessment of Three Spatial Interpolation Models to Obtain the Best One for Cumulative Rainfall Estimation (Case study: Ramsar District)

Hasan Zabihi, Anuar Ahmad, Mohamad Nor Said

Department of Geoinformation, Faculty of Geoinformation and Real Estate,

Universiti Teknologi Malaysia, 81310 UTM, Johor Bahru, Johor, Malaysia

AbstractThere has been an increasing use of predictive spatial distribution of rainfall patterns for planning and regional management decisions. This study focused on three interpolation techniques including ordinary kriging module, linear regression method and inverse distance weighted (IDW) that were used to obtain the reliable spatial distribution of cumulative rainfall parameters in the study area. First, we selected twenty two meteorological stations in and around the study area. Second, rainfall data were analyzed in geographical information systems (GIS) to determine the accuracy of three models. Third, the distribution pattern was also validated by field investigations. Finally, the reliability map of rainfall was produced in GIS software. The results demonstrate that linear regression significantly performed better than inverse distance weighted (IDW) and ordinary kriging model.

Keywords Assessment; Cumulative rainfall; Spatial interpolation Models; RMS; GIS.

1. INTRODUCTION

   The effective of rainfall patterns are basically one the essential phenomenon in agricultural production as successful cultivation areas all over the world [1, 2]. Lack of meteorological station in regions is one of the main problems for future management and estimation of rainfall in given areas. However, using prediction methods to estimate amount of rainfall can help manager for future planning. On the other hand, the accuracy and reliability of these models also are very crucial for policy maker and managers. Geostatistical models are reported in numerous textbooks such as Kriging (plain geostatistics); environmental correlation (e.g. regression-based); and hybrid models (e.g. regression- kriging) [10, 11]. To present the accuracy of models, three methods including kriging, regressions and IDW were conducted to compare and benefits. Objective of this study focuses in comparison with three methods to obtain the best- performed one. The results show that regression method is well adopted with field real data set in study area and reliable one. In previous study, [7] used the geostatistical methods of kriging and cokriging to estimate the sodium adsorption ratio in an agricultural field. [8] produced a radon distribution map using the kriging and GIS techniques.

2. STUDY AREA

   The region is located in the northern part of Iran. Ramsar region is situated in the west of Mazandaran province, borders The Caspian Sea to the north and The Alborz Mountains range to the south. This region is one of the most important agricultural areas in Iran. The geographic coordinates of the study area are located between latitudes 36°3200 to 36°5911 N and longitudes 50°2030 to 50°4712 E. The total study area covers approximately 729.7 km2. The altitude of Ramsar County starts at a height of -20 meters near The Caspian Sea to 3620 meters above sea level. A map of the study area is shown in Fig. 1.

   Fig.1 The location of study area

3. METHODOLOGY

   Three conventional interpolation methods including linear regression, kriging and IDW are applied in this study to determine the best-performed one. Their principles are well explained elsewhere [3, 4]. Rainfall data was obtained from the Mazandaran Meteorological organization of 30 years (19802010). Rainfall data analysis was conducted by using curve expert software version 1.4 as a comprehensive curve fitting system. It employs a large number of regression models (both linear and nonlinear) as well as various interpolation schemes in the most precise and convenient way. In addition, the user may define any customized model desired for use in a regression analysis. The next step, elevation as independent imported in X axis or Y axis to get regression (linear fit) and coefficient data based on following equation (1):

   Y = a + b x (1)

   Moreover, kriging module and IDW are extracted in Geostatistical analyst into ArcGIS version 10.1. The coefficients give the values of the relevant parameters for the current model. These coefficients are always expressed as a and b. In other words, coefficients give the most vital information about the model in the graph. Twenty two meteorological stations were used that are given in Table 1. Rainfalls prediction, measure, errors and regression equation are shown in Table 2, 3, 4 and 5.

   Table 1: Top selection meteorological stations in and around study area (Climatic Atlas of Iran, I.R. of Iran Meteorological Organization, 2012)

   /tr>

   Name	Lat.	Long.	Station Class	Elev. (m)	Total rainfall
   Chaboksar	36.96	50.58	Raingage	-10	1012.1
   Mianlat	36.9	50.6	Raingage	350	633.5
   Sar limak	36.85	50.68	Raingage	200	659
   Azarak andokoh	36.85	50.7	Raingage	100	713.6
   Galesh mahaleh	36.81	50.73	Raingage	74	739.5
   Chapar sar	36.82	50.79	Raingage	5	862.6
   Soleiman abad	36.8	50.8	Raingage	25	616.5
   Tonekabon	36.81	50.87	Raingage	-15	747.2
   Golali abad	36.7	50.85	Raingage	50	801
   Lirasar	36.68	50.89	Raingage	300	688.5
   Balaoshtoj	36.64	50.76	Raingage	800	806.9
   Shahnetrash	36.64	50.72	Raingage	1550	668
   Javaherdeh	36.85	50.48	Raingage	2000	510
   Tomol	36.64	50.41	Raingage	2010	551.9
   Holoo-an	36.58	50.83	Raingage	900	732.2
   Tole lat	36.98	50.3	Raingage	40	902.5
   Malekot	36.89	50.11	Raingage	1428	471.2
   Zar abad	36.49	50.43	Raingage	1790	466.4
   parchkouh	36.63	50.17	Raingage	1600	554.3
   Ramsar	36.9	50.66	Synoptic	-21	1207.1
   Khorram abad	36.78	50.87	Climatology	50	1079
   Jannat roudbar	36.75	50.58	Raingage	1700	650.5

   To assess accuracy, we calculated the root mean square error (RMSE) is shown in following formula that is reported by [6] as in equation

   RMSE (2)

4. RESULTS AND DISCUSSION

   As a result of this analysis and to assess accuracy, we calculated the RMS of three methods and relationship between rainfall, altitude and stations. The root mean square error (RMS) indicates the spread of how far the computed values deviate from the observed. Based on Table 2, 3 and 4 higher RMS was obtained by linear regression and IDW. A lowest prediction error was achieved by kriging model. However, IDW can produce "bulls eyes" around data locations. Cumulative rainfall is classified into five classes. With regard to different interpolation analysis, it obvious that linear regression well performed for prediction of rainfall distributions in study area. The results of spatial interpolation into IDW method, ordinary kriging and linear regression are illustrated in the following Table 2, 3, 4 and 5. Moreover, Table 2 demonstrates cumulative rainfall into IDW interpolation method. Using IDW, the weight of any known point is set inversely proportional to its distance from the estimated point. It is calculated as basic formula follows:

   v = value to be estimated v i = known value

   di…, dn= distances from the n data points to the point estimated n.

   Root mean square value = 205.93.

   Table 2 Spatial prediction of rainfall using IDW method

   Regression function= 0.0296094924757434 * X + 701.926303235849

   The second method of spatial interpolation performed in this study was kriging. Kriging is used to estimate unknown values from data observed at known locations. Among the various existing Kriging techniques, both exact and inexact methods could be selected, depending on the measurement error model [5, 9]. A more detailed explanation of the kriging method is given by [12] and [13]. Kriging are based on statistical models that include autocorrelation. In reality, there are the statistical relationships among the measured points. Because of this, geostatistical techniques not only have the capability of producing a prediction surface but also provide some measure of the certainty or accuracy of the predictions [14, 15]. Table 3 shows ordinary kriging module using cumulative rainfall of twenty two meteorological stations.

   The general formula for both interpolators is formed as a weighted sum of the data:

   where:

   Z(si) = the measured value at the ith location

   i = an unknown weight for the measured value at the ith location

   s0 = the prediction location

   N = the number of measured values Root mean square value = 180.21.

   Table 3 Spatial prediction of rainfall using ordinary kriging method

   Regression function= 0.0371076672402633 * X + 717.536544660987

   The last method that was used in this study was regression approach. Regression of cumulative rainfall is addressed into Curve expert 1.4 and Excel that are shown in Table 4 and 5.

   Table 4 Regression trend between cumulative rainfall and elevation using Curve Expert software

   Y= -0.1691* X + 845.46, R2 = 0.40 and standard error = 226.91.

   Table 5 Regression trend between cumulative rainfall and elevation using Excel 2007

5. CONCLUSION

Root mean squared (RMS) of regression method is higher than kriging and IDW methods in terms of rainfall patterns. In reality, the methods that have the lowest root-mean-square prediction error when performing cross-validation have been adopted. Indeed, the smaller the root-mean-square prediction error, the closer the predictions are to their true values and better are the interpolation method [5]. As a comparison of methods, however; an ordinary kriging method is the most accurate than linear regression and IDW methods in terms of lower error. But, linear regression method is reliable due to altitude. it is also considered as an essential spatial characteristic. In other words, based on Table 5 in Ramsar district, rainfall trend would be decreased by increasing altitude (e.g. R2 value < 0.50) across the study region. R2 describes the proportion of the variance in measured data explained by the model. R2 ranges from 0 to 1, with higher values indicating less error variance, and typically values greater than 0.5 are considered acceptable [16, 17]. Hence, this suggests that IDW and kriging perform very poorly in spatial interpolation, which may not be suitable for interpolation analysis to this purpose. Thereby, in spatial prediction models, regression performs accurately rainfall map than IDW and kriging interpolation performance in study area.

Fig. 2 Three different rainfall prediction maps using regression, kriging and IDW spatial interpolation

ACKNOWLEDGMENTS

We are thankful to Universiti Teknologi Malaysia for generously funding the research. The authors also wish to express sincere thanks to Iran Citrus Research Institute for kind support data during this study.

REFERENCES

1. S.S., Panda, G., Hoogenboom, J. Paz, Distinguishing blueberry bushes from mixed vegetation land-use using high resolution satellite imagery and geospatial techniques. Computers and Electronics in Agriculture. 2009, 67: 51-59.

2. W. Schlenker, and M. J. Roberts, Nonlinear effects of weather on corn yields. Applied Economic Perspectives and Policy. 2006, 28: 391 398.

3. C. D., Lloyd, Assessing the effect of integrating elevation data into the estimation of monthly precipitation in Great Britain. J. Hydrol. 2005, 308: 1 4.

4. P., Goovaerts, Geostatistical approaches for incorporating elevation into the spatial interpolation of rainfall. J. Hydrol. 2000, 228: 113 129.

5. ESRI. Guideline for User of ArcGIS 9.2. ESRI Company, 2009.

6. C. J., Willmott, On the evaluation of model performance in physical geography. In: Gaile, G. L. Willmott, C. eds., spatial statistics and models. D. Reidel Publishing Company, Dordrecht. 1984, PP. 443-460.

7. L., Pozdnyakova, and R. Zhang, Geostatistical analyses of soil salinity in a large field. Precision Agriculture, 1999, 1: 153165.

8. H. C., Zhu, J. M., Charlet, and, A. Poffijn, Radon risk mapping in Southern Belgium: An application of geostatistical and GIS techniques. The Science of the Total Environment, 2001, 272: 203 210.

9. ESRI, Using Arc GIS Geostatistical Analyst, Printed in the USA, 2003.

10. C.V., Deutsch, A. G., Journel, Geostatistical Software Library and Users Guide, 2nd ed. Oxford University Press, New York, USA, 1998, pp. 369.

11. T. Hengl, A Practical Guide to Geostatistical Mapping 2nd ed. University of Amsterdam, Amsterdam. 2009.

12. J. K., Yamamoto,. An alternative measure of the reliability of ordinary kriging estimates. Mathematical Geology, 2000, 32(4): 489509.

13. D., Mcgrath, and C., Zhang,. Spatial distribution of soil organic carbon concentrations in grassland of Ireland. Applied Geochemistry, 2003, 18: 16291639.

14. P. A.. Burrough, Principles of Geographical Information Systems for Land Resources Assessment. New York: Oxford University Press. 1986.

15. M. A. Oliver, Kriging: A Method of Interpolation for Geographical Information Systems. International Journal of Geographic Information Systems, 1990, 4: 313332.

16. C. J. G., Santhi, A. J. R., Williams, W. A., Dugas, R., Srinivasan and L. M., Huack Validation of the SWAT model on a large river basin with point and nonpoint sourcs. Journal of American Water Resources Association, 2001, 37 (5): 1169-1188.

17. M. W., Van Liew, T. L., Veith, D. D., Bosch, J. G., Arnold, Suitability of SWAT for conservation effects assessment project: A comparison on USDA-ARS experimental watersheds. Journal of Hydrologic Engineering, 2007, 12(2): 173-189.