Abstract

The study aims to analyze spatio-temporal variations in rainfall data over Indravati River basin, India. Three rainfall data sets, Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA), India Meteorological Department (IMD) grid, and IMD gauge were used. Data from 2001 to 2013 were analyzed for three time scales, namely, daily, monthly, and annual. Analysis showed good correlation between IMD gauge and TMPA grid rainfall at monthly time scale, poor correlation is observed at daily and annual time scales. Mann–Kendall (MK) trend test reveals a significant increasing trend of IMD gauge and IMD grid data, whereas TMPA rainfall is free from trends at the majority of stations for daily time scale. Nevertheless, both IMD grid as well as TMPA grid rainfall can be considered as a better representative of rainfall, since it is attained from observed rainfall data over the country. The Pettitt and standard normal homogeneity tests show that TMPA rainfall has a more non-homogeneous nature, whereas IMD grid rainfall and IMD gauge rainfall data are homogeneous. Overall, the trend and homogeneity analysis indicate that TMPA grid and IMD grid rainfall is in line with IMD gauge data, however IMD grid rainfall has the edge over TMPA grid data.

INTRODUCTION

Rainfall is one of the most important meteorological variables to understand climate change impact on water resources availability (Huang et al. 2015). Rainfall is a major driving force in hydrological modeling for quantifying water availability within river catchments, also changes in its trend at catchment scale need to be addressed while modeling (Taxak et al. 2014; Machiwal et al. 2016). Observed rainfall from observational gauging stations is considered to be most accurate in climate and hydrological studies because of its measurement at ground level. However, these observations are very poorly represented because of its limitation in the areal distribution of number of rain gauge stations (Xu et al. 2014). Recently, acquisition of satellite-based rainfall data and their application has been increasing in climatology, meteorology, hydrology, and hydrometeorology fields due to their large spatial coverage and high temporal resolutions (Kizza et al. 2012; Sun et al. 2016). Among satellite-based rainfall data set, Tropical Rainfall Measuring Mission (TRMM) rainfall data are being shown great interest among the research community due to their origin from multisatellites (Xu et al. 2014; Prakash et al. 2015). Further, merging of TRMM rainfall with in-situ rain gauge measurements has extended the application of these rainfall data sets in climate predictions and water resources studies (Kizza et al. 2012; Prakash et al. 2015). The recently released TRMM Multisatellite Precipitation Analysis (TMPA) rainfall product is an advanced version of TRMM 3B42 version 7 rainfall data set. TMPA rainfall benefits from both the TRMM version 7 and in-situ rainfall observations by merging techniques (Huffman & Bolvin 2013).

Validation of TMPA rainfall data with observed gauge rainfall data has been carried out by spatial averaging of TMPA rainfall at different time scales over the selected region (Prakash et al. 2015; Jiang et al. 2016). These studies showed good agreement between TMPA rainfall and gauge rainfall in terms of correlation coefficient because the variations of TMPA rainfall are minimized by spatial averaging over the region of interest. These studies provided the confidence to use those data in the field of climate change studies and hydrological studies at the macro-catchment scale. Further, application of this rainfall product to study climate change and water-related issues at watershed scale needs the evaluation of individual grid stations of TMPA rainfall against a regional watershed level standard rainfall or observed rainfall data set.

Climate change impact on runoff availability within a drainage area needs the trend and other analysis of rainfall data. Previous studies on rainfall around the globe have been carried out using traditional observed rain gauge data (Tomozeiu et al. 2005; Taxak et al. 2014; Machiwal et al. 2016). As TMPA rainfall data application is of increasing demand, analysis of rainfall data such as trend, homogeneity, etc., is required at individual grid level and is taken as the major aim of the present study. Statistical trend analysis of a rainfall can be carried out using both parametric and nonparametric tests. Parametric tests assume that the rainfall data used for trend analysis are distributed normally and independent. On the other hand, nonparametric tests have the advantage of working with rainfall data without passing normal distribution.

Mann–Kendall (MK) trend test is a widely used nonparametric method for computing trends of a given time series data (Mann 1945; Kendall 1975). The MK method has been successfully applied in different parts of the world for analyzing rainfall trends (Oliveira et al. 2014; Jaiswal et al. 2015). Rainfall trend detection using the MK method has been applied for rainfall in India also (Narayanan et al. 2013; Taxak et al. 2014; Joshi et al. 2016; Machiwal et al. 2016). Joshi et al. (2016) carried out trend analysis using MK test of drought variables in 30 subregions of India by considering monthly rainfall data from 1871 to 2011. The dominant area of interest considered in the present study research comes under subregion-26, reported by Joshi et al. (2016). Their study revealed that spatially averaged monthly rainfall trend in this region has an increasing trend during 1871 to 1940 and decreasing trend during 1941 to 2011, respectively.

Identifying the homogeneity/non-homogeneity in rainfall data is suggested before carrying out any modeling in climate change and water resouces studies (Yozgatligil & Yazici 2016). Non-homogeneity of rainfall data refers to abrupt discontinuities in original time series data subjected to changes in station surroundings. The presence of non-homogeneity in original hydrological time series data may affect the reliability of the model results. There exist several methods in the literature to find homogeneity in the rainfall time series data (Buishand 1982; Yozgatligil & Yazici 2016). Statistical homogeneity methods are categorized into the absolute method and relative method based on analysis of one station data with respect to surrounding stations. Absolute homogeneity methods are widely used by various researchers for analyzing rainfall time series data (Buishand 1982; Tomozeiu et al. 2005; Huss et al. 2009). Yozgatligil & Yazici (2016) suggested standard normal homogeneity (SNH) test for detecting non-homogeneity in rainfall data. Machiwal & Jha (2012) commented that one should consider more than one test for detecting non-homogeneity in rainfall time series data. Jaiswal et al. (2015) and Machiwal et al. (2016) applied the Pettitt homogeneity test developed by Pettitt (1979) for finding abrupt changes in rainfall data over the Indian region.

The present study compared the homogeneity in TMPA rainfall of all grid stations over Indravati River basin, India against the standard gauge-based gridded rainfall data and few observed gauge rainfall data provided by the India Meteorological Department (IMD). All these three rainfall series for three time scales, daily, monthly, and annual are analyzed. MK test is used to detect the trend in the rainfall analysis. Owing to the importance of the study area and research, the Pettitt test is used along with SNH test to detect homogeneity in the present study. Initially, the statistical analysis of individual stations is carried out along with spatial and temporal analysis. Nevertheless, the present study focused on trend analysis and homogeneity analysis of individual rainfall grid stations rather than using averaged rainfall over the basin.

STUDY AREA

The study area, the Indravati River basin is a sub-basin of the Godavari River, India. The sub-basin lies within the global coordinates of 80° 0′ to 83° 6′ E and 18° 25′ to 20° 41′ N. The Indravati basin covers a geographical area of 38,963 km2. The highest elevation is 1,325 meters and the lowest elevation at the confluence with Godavari basin is 8 meters above MSL. There exists ten IMD observed stations within the basin, of which, only four stations were in fully operational condition with six stations having some missing data during the study period of 13 years from 2001 to 2013. The percentage of missing rainfall data at the six observed gauge stations is as follows: Dantewara – 14.65%, Jagdalpur – 1.95%, Kondagoan – 14.17%, Kosagudam – 11.6%, Narayanpur – 35.91%, and Nowrangpur – 2.57%. The missing data at the six rain gauge stations are filled by multiple imputation-based extreme learning machine (ELM) technique (Sovilj et al. 2016). The location map of Indravati basin along with TMPA, IMD grid, and IMD observed rain gauge stations is shown in Figure 1. It can be noted from Figure 1 that the locations of TMPA and IMD grid stations are the same, since the resolution of both the data sets are the same.

Figure 1

Location map of Indravati River basin with TMPA grid stations and IMD observational stations.

Figure 1

Location map of Indravati River basin with TMPA grid stations and IMD observational stations.

Indravati basin receives the majority of its rainfall during the south-west monsoon period (June–September). The average annual rainfall in this basin as per IMD, during the time period 1971 to 2013, is 1,288 mm, and this basin is situated in a moderate temperature region. In this study, the kriging interpolation method is used for geographical analysis of rainfall data. The details of data sets used in the study are discussed in the next section.

Data series and time scales studied

In the present study, three different rainfall data sets pertaining to Indravati River basin are studied. They are: (i) TMPA grid rainfall (79 grids), (ii) IMD grid rainfall (79 grids), and (iii) IMD observed rainfall (10 stations). TRMM rainfall data are widely used in climate change and water balance studies (Kizza et al. 2012; Prakash et al. 2015; Jiang et al. 2016). Microwave Imager (MI) and Precipitation Radar (PR) are the major instruments used in the development of TRMM rainfall (Huffman & Bolvin 2013). There exist several versions of TRMM rainfall data with different spatial and temporal resolutions covering the area from 50° N to 50° S (mirador.gsfc.nasa.gov/cgi-bin/mirador/presentNavigation.pl?tree=project&project=TRMM). Among all the available rainfall products of TRMM, merged rainfall product has a great attraction for many researchers because of its improvement in measurements by combining them with traditional rain gauge observations (Xu et al. 2014). The 3 hr TRMM version 7 merged rainfall product, namely, TMPA, developed by Huffman & Bolvin (2013) has been obtained from Goddard Earth Sciences Data and Information Services Center (GES DISC). The daily, monthly, and annual TMPA rainfall time series data sets at 0.25° resolution were obtained by accumulating 3 hr data sets corresponding to a particular time scale. TMPA data available for a period of 13 years from 2001 to 2013 were used in the present study.

The gridded daily rainfall data of 0.25° resolution was collected from IMD for a period of 13 years from 2001 to 2013. IMD daily grid rainfall was developed from more than 3,000 rain gauge stations across India (Prakash et al. 2015). Another advantage of using IMD grid rainfall to evaluate TMPA rainfall is that both the data sets have the same spatial resolution of 0.25°. There are around 79 gridded stations in Indravati basin. Daily rainfall data observed at IMD rain gauge stations within Indravati basin were also collected from the National Data Center, Office of Additional Director General of Meteorology (Research), IMD, Pune, India.

There exist ten IMD rain gauge stations within this basin, and their location is shown in Figure 1. Both IMD grid and IMD observed daily rainfall data sets were further converted into monthly and annual time scales by accumulating daily data and are used in this study. While analyzing the rainfall at all the grid stations, the statistics of TMPA and IMD grid stations close to ten IMD observed gauge stations were compared and verified. The temporal scales analyzed are daily, monthly, and annual time steps.

METHODOLOGY

The methodology applied in the study includes the preliminary analysis of IMD grid, and TMPA rainfall data sets along with IMD observed rainfall data available at ten stations, namely, Bhamragad (35), Dantewara (13), Etapalli (48), Jagdalpur (28), Kondagoan (52), Koraput (17), Kosagudam (29), Narayanpur (63), Nowrangpur (30), and Tentulikhuti (30), which are shown in Figure 1. The numbers in brackets indicate the grid that represent the stations. The preliminary evaluation includes the comparison of daily, monthly, and annual rainfall time series, a correlation analysis among the stations, and a number of rainy day analyses of all the three rainfall data sets under study. The trend and homogeneity analysis of IMD grid and TMPA rainfall is carried out and compared against the available IMD observed gauge stations. All the analysis of trend and homogeneity was carried out for daily, monthly, and annual time steps. As Indravati River basin receives its major contribution of rainfall during the monsoon period (June to September), trend and homogeneity analysis is carried out separately for this monsoon period also.

The MK test proposed by Mann (1945) and Kendall (1975) is a non-parametric test. The widely used method for trend detection analysis in rainfall time series, namely, the MK test is considerd in the present study (Narayanan et al. 2013; Joshi et al. 2016; Machiwal et al. 2016). Similarly, the homogeneous analysis is carried out using Pettitt and SNH tests. The Pettitt test by Pettitt (1979) is an absolute and non-parametric test used for finding non-homogeneities in the rainfall time series data. The advantage of the Pettitt test is that assumption of normality is not required, which is the major barrier for other homogeneity tests (Pettitt 1979; Jaiswal et al. 2015; Machiwal et al. 2016). The SNH test is a parametric method (Yozgatligil & Yazici 2016) and it needs to satisfy the condition that the input data must be free from dependency. This method uses standardized values obtained by comparing observations of a test station with the average of several other surrounding stations (Buishand 1982; Tomozeiu et al. 2005; Machiwal & Jha 2012; Yozgatligil & Yazici 2016).

RESULTS AND DISCUSSION

The spatio-temporal rainfall analysis was carried out for each and every grid station, 79 TMPA, 79 IMD grid stations, along with ten IMD observed stations to see whether the characteristics of the gridded data are close to the observed ones.

Preliminary analysis

The daily, monthly, and annual rainfall data sets of the IMD grid and TMPA are analyzed and the results evaluated. The results at appropriate grid stations are compared with the results of rainfall data available at ten IMD observed gauge stations. The daily rainfall scatter plots between TMPA, IMD gridded, and nearest IMD observed stations are shown in Figure 2. From Figure 2, it can be observed that the TMPA daily rainfall has less magnitude compared to the IMD observed and IMD grid data sets. At the station Narayanpur, IMD grid rainfall was observed to be better compared to other stations, as shown in Figure 2(h). The monthly and annual scatter plots of three rainfall data sets are shown in Figures 3 and 4, respectively.

Figure 2

Daily rainfall scatter plots of IMD grid and TMPA with ten IMD observed rainfall stations.

Figure 2

Daily rainfall scatter plots of IMD grid and TMPA with ten IMD observed rainfall stations.

Figure 3

Monthly rainfall scatter plots of IMD grid and TMPA with ten IMD observed rainfall stations.

Figure 3

Monthly rainfall scatter plots of IMD grid and TMPA with ten IMD observed rainfall stations.

Figure 4

Annual rainfall scatter plots of IMD grid and TMPA with ten IMD observed rainfall stations.

Figure 4

Annual rainfall scatter plots of IMD grid and TMPA with ten IMD observed rainfall stations.

Figure 3(a)3(j) represent monthly IMD grid and TMPA rainfall with respect to ten IMD observed gauge stations, namely, Bhamragad, Dantewara, Etapalli, Jagdalpur, Kondagoan, Koraput, Kosagudam, Narayanpur, Nowrangpur, and Tentulikhuti, respectively. From Figure 3, it is clearly observed that monthly TMPA rainfall is correlated well with the IMD grid and IMD observed rainfall at all ten stations because temporal variations in daily data are added to get monthly. Nevertheless, the IMD grid rainfall is close to IMD observed rainfall indicating that IMD grid rainfall represents the reality better than TMPA rainfall. The smaller deviation in the IMD grid from that of IMD observed is because the grid data sets are derived by considering the rainfall stations throughout the country providing a spatial average over the region, reducing the variations in individual stations. Annual scatter plots of TMPA data show good agreement with the IMD grid and IMD observed rainfall at three observed stations, namely, Bhamragad, Etapalli, and Narayanpur, and poor correlation is observed at remaining stations, as shown in Figure 4. Among the grid data sets, IMD grid rainfall is better than TMPA at all observed stations because IMD grid data are derived from observed stations' data. Thus, it can be concluded that the IMD grid rainfall is very close to reality. TMPA monthly time scale data can also be considered.

A correlation study among the respective data sets with different time scales has been carried out and the coefficient of determination (R2) is presented in Table 1. The correlation between observed IMD gauge rainfall and gridded IMD rainfall is better than the correlation between observed IMD gauge rainfall and TMPA rainfall. The maximum correlation between daily rainfall of the IMD grid and IMD observed data is seen at Narayanpur (R2 = 0.33) and the minimum correlation is seen at Kondagoan (R2 = 0.07). On the other hand, TMPA has poor correlation at daily and annual scale (also seen in the scatter plots of Figures 2 and 4). The poor correlation with annual scale may be due to fewer data points, but TMPA shows a better correlation with monthly scale having R2 values ranging from 0.38 to 0.76 against IMD observed data, as shown in Table 1. In particular, from Table 1, it is very clear that TMPA rainfall shows no correlation with IMD observed gauge stations for the daily and annual time scale. The reason is that IMD observed no rainfall during the non-monsoon period but TMPA has, accounting for the variation in annual and daily. From the above discussion, it is clear that TMPA data are poorly mapped compared to IMD grid as well as IMD observed data for daily and annual time scales. Thus, TMPA rainfall, especially smaller time scale data, calls for a detailed bias correction before using it in any analysis and hydrological modeling.

Table 1

Coefficient of determination (R2) values of IMD observed rainfall against IMD grid and TMPA grid rainfall

S. No. IMD observed gauge station Location of station IMD grid rainfall
 
TMPA rainfall
 
Daily Monthly Annual Daily Monthly Annual 
Dantewara South 0.2 0.59 0.19 0.03 0.39 0.03 
Jagdalpur East 0.2 0.8 0.15 0.04 0.72 0.04 
Koraput East 0.19 0.7 0.5 0.05 0.6 0.02 
Kosagudam East 0.23 0.6 0.05 0.04 0.5 0.03 
Nowrangpur East 0.2 0.62 0.2 0.04 0.51 0.01 
Tentulikhuti East 0.22 0.73 0.35 0.05 0.65 0.07 
Kondagoan North 0.07 0.47 0.11 0.02 0.38 0.01 
Narayanpur North 0.33 0.63 0.02 0.07 0.57 0.37 
Bhamragad West 0.19 0.69 0.17 0.16 0.76 0.51 
10 Etapalli West 0.19 0.71 0.26 0.13 0.76 0.4 
S. No. IMD observed gauge station Location of station IMD grid rainfall
 
TMPA rainfall
 
Daily Monthly Annual Daily Monthly Annual 
Dantewara South 0.2 0.59 0.19 0.03 0.39 0.03 
Jagdalpur East 0.2 0.8 0.15 0.04 0.72 0.04 
Koraput East 0.19 0.7 0.5 0.05 0.6 0.02 
Kosagudam East 0.23 0.6 0.05 0.04 0.5 0.03 
Nowrangpur East 0.2 0.62 0.2 0.04 0.51 0.01 
Tentulikhuti East 0.22 0.73 0.35 0.05 0.65 0.07 
Kondagoan North 0.07 0.47 0.11 0.02 0.38 0.01 
Narayanpur North 0.33 0.63 0.02 0.07 0.57 0.37 
Bhamragad West 0.19 0.69 0.17 0.16 0.76 0.51 
10 Etapalli West 0.19 0.71 0.26 0.13 0.76 0.4 

The temporal statistics such as mean, standard deviation (SD), skewness, kurtosis, coefficient of variation, and maximum rainfall at daily time scale between the IMD grid and TMPA rainfall is shown in Figure 5. At the outset, all statistics at daily scale are close between TMPA and IMD grid stations. IMD grid data show maximum values of mean and SD over the eastern and western portion of the basin, whereas for TMPA it is in the north-west area (Figure 5(a)5(d)). This shows that all the grid rainfall has captured the temporal variation between mean and SD but with different magnitudes.

Figure 5

Statistics of daily rainfall for IMD grid and TMPA rainfall data sets: (a), (b) mean rainfall, (c) and (d) standard deviation, (e) and (f) skewness, (g) and (h) kurtosis, (i) and (j) coefficient of variation, (k) and (l) maximum daily rainfall.

Figure 5

Statistics of daily rainfall for IMD grid and TMPA rainfall data sets: (a), (b) mean rainfall, (c) and (d) standard deviation, (e) and (f) skewness, (g) and (h) kurtosis, (i) and (j) coefficient of variation, (k) and (l) maximum daily rainfall.

The skewness of daily IMD grid and TMPA data shows similar variations with minimum values 4.65 and 5, maximum values 9.11 and 7.83, respectively (Figure 5(e) and 5(f)). This shows that both daily data sets are positively skewed. From Figure 5(g) and 5(h), the kurtosis of daily data set shows very high peakedness. The coefficient of variation and maximum rainfall also shows similar results at all grid stations in both IMD grid and TMPA rainfall data sets (Figure 5(i)5(l)). The variation in magnitudes at different scale may be linked to number of rainy days, because at a station it is an event of that particular station, whereas when gridded it may be affected by other stations' statistics.

Analysis of the number of rainy days in each year is also carried out for three rainfall data sets and is compared in Figure 6. The rainy day analysis of TMPA and IMD grid data follows a similar trend and both data sets show a high number of rainy days compared to IMD observed data in general and at station Dantewara is particular. Figure 6 indicates that IMD grid data can be considered for further evaluation of TMPA rainfall data at each individual grid station.

Figure 6

Scatter plots of rainy days' number in a year for IMD grid and TMPA with ten IMD observed rainfall stations.

Figure 6

Scatter plots of rainy days' number in a year for IMD grid and TMPA with ten IMD observed rainfall stations.

In addition to the above, the mean annual rainfall between IMD grid and TMPA data sets are carried out as part of the preliminary analysis. Figure 7 shows the spatial variation of mean annual rainfall of both IMD grid and TMPA rainfall data sets over the Indravati River basin. IMD grid data show more rainfall magnitude than TMPA data, as shown in Figure 7. Very high rainfall is shown in the eastern side by IMD grid data (2,616 mm), whereas the TMPA data show the maximum amount of 1,607 mm. This shows that, overall, TMPA data captured 1,000 mm less of the maximum rainfall compared to IMD grid data in this basin. Thus, it may be concluded that TMPA has more uncertainty on the eastern side of the Indravati basin. The spatial variations of TMPA and IMD grid data show a similar trend except at the eastern portion of the basin. The results of the present study show that TMPA data performed well in terms of spatial variations of rainfall, but less in magnitude of the rainfall when compared to IMD rainfall. It is also observed that when the areal extent becomes greater, both perform very well.

Figure 7

Mean annual rainfall in Indravati river basin: (a) IMD gridded and (b) TMPA rainfall.

Figure 7

Mean annual rainfall in Indravati river basin: (a) IMD gridded and (b) TMPA rainfall.

Prakash et al. (2015) showed a good correlation between the spatially averaged TMPA rainfall with IMD grid rainfall over the entire Indian sub-continent. Their good agreement between TMPA and IMD grid rainfall was due to spatial averaging of TMPA rainfall at several grids. However, when comparing individual grid stations, it is observed that TMPA rainfall is poorly correlated with IMD observed data at daily time scales.

To investigate further, the correlation between the IMD grid and TMPA rainfall is extended to each individual grid station. Figure 8 shows R values of each individual station obtained by correlating the monthly and annual rainfall data sets of the IMD grid and TMPA rainfall. From Figure 8(a), it is observed that monthly rainfall correlation between the IMD grid and TMPA data sets shows good agreement ranging from 0.58 at station 8 to 0.95 at station 79, whereas Figure 8(b) shows a poor correlation in terms of annual rainfall of IMD grid and TMPA (Figure 9). Especially, the eastern portion of the basin had very poor correlation. This may be due to the presence of the Eastern Ghats in this location and a lesser number of data points. Therefore, it reveals that TMPA data show uncertainty in rainfall at higher elevations. This is observed in magnitude of maximum rainfall also. The gridded data sets might have been developed and derived without giving weightage to the physiographic and hydrologic boundaries of a given river basin, which are very important in basin scale hydrologic models. If these gridded data were derived or corrected for bias by considering physiographic characteristics, then those data could be more useful and result in prediction of rainfall as well as estimating the rainfall. The overall statistical analysis shows that even though there is difference in magnitude of the process, the variation of the process remains almost similar among the gridded stations. To explore further about internal structure of the time series, the trend and homogeneity analysis of these three rainfall data sets are discussed below.

Figure 8

Correlation cefficient (R) values between IMD grid and TMPA rainfall for all stations: (a) monthly rainfall and (b) annual rainfall.

Figure 8

Correlation cefficient (R) values between IMD grid and TMPA rainfall for all stations: (a) monthly rainfall and (b) annual rainfall.

Figure 9

Scatter plot of TMPA rainfall versus IMD grid rainfall at all stations: (a) monthly and (b) annual.

Figure 9

Scatter plot of TMPA rainfall versus IMD grid rainfall at all stations: (a) monthly and (b) annual.

Trend analysis

Trend analysis of the observed and their respective grid-represented rainfall data set at all three time scales is carried out using the MK test and the results are shown in Table 2. For daily rainfall, IMD observed rainfall shows a significant increasing trend at stations Bhamragad, Jagdalpur, Kondagoan, and Kosagudam, whereas IMD grid rainfall shows no trend at these stations and saw an increasing trend at stations Nowrangpur and Tentulikhuti located in the eastern side of the basin, which is in conflict with IMD observed rainfall from Table 2. Daily TMPA rainfall shows increasing trend at stations Kondagoan and Kosagudam, as also observed for IMD observed rainfall, and it shows no trend at all other stations (Table 2). Similarly, both IMD grid and TMPA daily rainfall data sets show similar trend results except at the majority of stations located in the eastern portion of the basin, as observed from Table 2. The monthly rainfall of all three data sets shows no trend. In the case of annual rainfall, both IMD observed and IMD grid shows no trend at all stations. TMPA annual rainfall is also in line with IMD observed data except at station Etapalli.

Table 2

Mann–Kendall test results for IMD observed, IMD grid, and TMPA grid rainfall

S. No. Station Location of station Daily
 
Annual
 
Monsoon daily
 
Monsoon total
 
IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid 
Dantewara South NT NT NT NT NT NT IT IT IT NT NT NT 
Jagdalpur East IT NT NT NT NT NT IT NT IT NT NT IT 
Koraput East NT NT NT NT NT NT IT NT IT NT NT IT 
Kosagudam East IT NT IT NT NT IT NT IT IT NT NT IT 
Nowrangpur East NT IT NT NT NT NT DT IT IT NT NT IT 
Tentulikhuti East NT IT NT NT NT NT NT IT IT NT NT IT 
Kondagoan North IT NT IT NT NT NT DT NT IT NT NT IT 
Narayanpur North NT NT NT NT NT NT DT NT NT NT NT NT 
Bhamragad West IT NT NT NT NT NT IT NT IT NT NT IT 
10 Etapalli West NT NT NT NT NT IT NT NT IT NT NT IT 
S. No. Station Location of station Daily
 
Annual
 
Monsoon daily
 
Monsoon total
 
IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid 
Dantewara South NT NT NT NT NT NT IT IT IT NT NT NT 
Jagdalpur East IT NT NT NT NT NT IT NT IT NT NT IT 
Koraput East NT NT NT NT NT NT IT NT IT NT NT IT 
Kosagudam East IT NT IT NT NT IT NT IT IT NT NT IT 
Nowrangpur East NT IT NT NT NT NT DT IT IT NT NT IT 
Tentulikhuti East NT IT NT NT NT NT NT IT IT NT NT IT 
Kondagoan North IT NT IT NT NT NT DT NT IT NT NT IT 
Narayanpur North NT NT NT NT NT NT DT NT NT NT NT NT 
Bhamragad West IT NT NT NT NT NT IT NT IT NT NT IT 
10 Etapalli West NT NT NT NT NT IT NT NT IT NT NT IT 

Note: DT means decreasing trend, IT means increasing trend, and NT means no trend.

As Indravati River basin receives its major contribution of rainfall during the monsoon period (June to September), trend analysis is carried out separately for this monsoon period also. In the case of monsoon daily rainfall, IMD observed rainfall shows increasing trend at stations Bhamragad, Dantewara, Jagdalpur, and Koraput. It also shows decreasing trend at stations Kondagoan, Narayanpur, and Nowrangpur (Table 2). The monsoon daily IMD grid rainfall shows different results from IMD observed rainfall except at station Dantewara (Table 2). Similarly, TMPA rainfall shows increasing trend at all stations except Narayanpur. It is also observed that both IMD grid and TMPA data sets never show decreasing trend in monsoon daily rainfall, which is observed in the case of IMD observed rainfall. This shows that gridded data give overall internal characteristics. From Table 2, it is observed that TMPA monsoon daily rainfall is in line with IMD observed rainfall at stations Dantewara, Jagdalpur, and Koraput. The monthly rainfall of the three data sets considered in monsoon season shows no trend at all stations. These results are similar to monthly rainfall considered during the full time series data. The IMD observed and IMD grid total monsoon rainfall shows no trend at all stations, whereas TMPA rainfall shows increasing trend at the majority of stations studied. From Table 2, it is observed that TMPA and IMD observed rainfalls mostly agree on the same trend at smaller time scales and differ at larger time scales, as seen from results of total monsoon rainfall (Table 2). It is also observed from Table 2 that IMD grid rainfall shows different trend results compared to IMD observed rainfall in smaller time scales in the eastern portion of the basin. This may be due to interpolation rainfall data to these grids from several nearby gauges. Further, trend analysis is extended to all grid stations of TMPA and compared with IMD grid data.

Trend analysis has been carried out individually at all stations of IMD grid and TMPA rainfall using the MK test for three time scales, and results are given in Table 3. Figure 10 shows the spatial distribution of Z value for both IMD grid and TMPA daily rainfall data sets along with their trends at individual grid stations. It is observed that daily IMD grid rainfall shows decreasing trend at stations 24 and 79, and increasing trend at grid stations located on the east side of the basin. Figure 10(b) also shows that no decreasing trend exists in TMPA rainfall at daily time scale throughout the basin, and increasing trend is observed at stations located in the south-east portion of the basin. Trend analysis of monthly and annual rainfall data sets of the IMD grid using MK test do not show any trend. Similarly, TMPA rainfall does not show any trend at a monthly time scale and 21 stations showed increasing trend at annual time scale.

Table 3

Number of stations showing increasing, decreasing, and no trend in Indravati basin at different time scales

Time step in time series IMD grid rainfall
 
TMPA rainfall
 
IT DT NT Total stations IT DT NT Total stations 
Daily 11 66 79 71 79 
Monthly 79 79 79 79 
Annual 79 79 21 58 79 
Monsoon daily 29 48 79 58 21 79 
Monsoon monthly 79 79 71 79 
Monsoon total 79 79 52 27 79 
Time step in time series IMD grid rainfall
 
TMPA rainfall
 
IT DT NT Total stations IT DT NT Total stations 
Daily 11 66 79 71 79 
Monthly 79 79 79 79 
Annual 79 79 21 58 79 
Monsoon daily 29 48 79 58 21 79 
Monsoon monthly 79 79 71 79 
Monsoon total 79 79 52 27 79 

Note: DT means decreasing trend, IT means increasing trend, and NT means no trend.

Figure 10

Spatial distribution maps of trend in daily rainfall of (a) IMD grid and (b) TMPA in Indravati River basin at 95% confidence level.

Figure 10

Spatial distribution maps of trend in daily rainfall of (a) IMD grid and (b) TMPA in Indravati River basin at 95% confidence level.

Surprisingly, the daily trend analysis of the IMD grid rainfall data set during monsoon season shows an increasing trend at 29 stations located centrally to the east side of the basin (Figure 11(a)). However, daily full year data show less stations with increasing trend. This may be due to more zeros in full year data. More in-depth research is needed to know how this longer length of zeros affects trend. The TMPA rainfall also shows increasing trend at 58 stations spread over the region of interest, as shown in Figure 11(b).

Figure 11

Spatial distribution maps of trend in monsoon daily rainfall of (a) IMD grid and (b) TMPA in Indravati River basin at 95% confidence level.

Figure 11

Spatial distribution maps of trend in monsoon daily rainfall of (a) IMD grid and (b) TMPA in Indravati River basin at 95% confidence level.

Monthly rainfall considered during the monsoon period shows increasing trend at eight grid stations for TMPA rainfall, whereas IMD grid rainfall was observed to have no trend at all grid stations, as shown in Table 3. Further total monsoon rainfall trends of the IMD grid do not show any trend in Indravati River basin (Figure 12(a)) and TMPA rainfall shows an increasing trend at 52 stations occupying an entire portion of the basin (Figure 12(b)). From Table 3, it is clear that only TMPA grid rainfall shows trends at annual and monsoon total time scales and IMD grid data are free from trends for both monthly and annual time scales. Overall, it is observed that the number of increasing trend stations becomes larger in TMPA rainfall when moving from daily time step to annual time step and from full time series data to monsoon alone data (Table 3). IMD grid rainfall is the exception to this. Therefore, IMD grid rainfall is observed to be free from trends when compared to TMPA rainfall. Thus, TMPA rainfall captures the smaller time step trend much better than IMD grid data.

Figure 12

Spatial distribution maps of trend in monsoon total rainfall of (a) IMD grid and (b) TMPA in Indravati River basin at 95% confidence level.

Figure 12

Spatial distribution maps of trend in monsoon total rainfall of (a) IMD grid and (b) TMPA in Indravati River basin at 95% confidence level.

Homogeneity analysis

Homogeneity analysis is carried out using the Pettitt test and SNH test for three rainfall data sets, namely, IMD observed, TMPA, and IMD grid. The test results of the Pettitt test and SNH test are shown in Table 4. As the Pettitt test is good in finding the change point in the middle of the time series data, three rainfall data sets available at ten observed stations are analyzed using this test for finding non-homogeneity. Similarly, the SNH test is also used to test homogeneity and is known to find the change point at the beginning and end of the time series data. The analysis is carried out using the above-mentioned homogeneous tests at daily, monthly, annual, monsoon daily, monsoon monthly, and monsoon total rainfall time scales of three rainfall data sets.

Table 4

Pettitt and SNH test results carried out between IMD observed, IMD gridded, and TMPA rainfall data sets for daily and monsoon daily time series

S. No. Station Location of station Daily
 
Annual
 
Monsoon daily
 
Monsoon total
 
IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid 
Pettitt test 
 1 Dantewara South NH NH NH H H H H NH NH H H H 
 2 Jagdalpur East NH NH NH H H H NH NH NH H H H 
 3 Koraput East NH NH NH H H H NH NH NH H H H 
 4 Kosagudam East NH NH NH H H H NH NH NH H H H 
 5 Nowrangpur East NH NH NH H H H NH NH NH H H H 
 6 Tentulikhuti East NH NH NH H H H NH NH NH H H H 
 7 Kondagoan North NH NH NH H H H NH NH NH H H H 
 8 Narayanpur North NH NH NH H H H NH NH H H H H 
 9 Bhamragad West NH NH H H H H NH H NH H H H 
 10 Etapalli West NH NH NH H H H NH H NH H H NH 
SNH test 
 1 Dantewara South NH H NH H H H NH H H H H H 
 2 Jagdalpur East NH H NH H H H NH NH H H H H 
 3 Koraput East NH NH NH H H H H NH H H H H 
 4 Kosagudam East H H NH H H H NH NH H H H H 
 5 Nowrangpur East H NH NH H H H H H H H H H 
 6 Tentulikhuti East H NH NH H H H NH H H H H H 
 7 Kondagoan North NH H NH H H H NH H H H H NH 
 8 Narayanpur North H NH NH H H H NH NH H H H NH 
 9 Bhamragad West NH H NH H H H NH H NH H H NH 
 10 Etapalli West H H NH H H H NH H H H H NH 
S. No. Station Location of station Daily
 
Annual
 
Monsoon daily
 
Monsoon total
 
IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid IMD obs. IMD grid TMPA grid 
Pettitt test 
 1 Dantewara South NH NH NH H H H H NH NH H H H 
 2 Jagdalpur East NH NH NH H H H NH NH NH H H H 
 3 Koraput East NH NH NH H H H NH NH NH H H H 
 4 Kosagudam East NH NH NH H H H NH NH NH H H H 
 5 Nowrangpur East NH NH NH H H H NH NH NH H H H 
 6 Tentulikhuti East NH NH NH H H H NH NH NH H H H 
 7 Kondagoan North NH NH NH H H H NH NH NH H H H 
 8 Narayanpur North NH NH NH H H H NH NH H H H H 
 9 Bhamragad West NH NH H H H H NH H NH H H H 
 10 Etapalli West NH NH NH H H H NH H NH H H NH 
SNH test 
 1 Dantewara South NH H NH H H H NH H H H H H 
 2 Jagdalpur East NH H NH H H H NH NH H H H H 
 3 Koraput East NH NH NH H H H H NH H H H H 
 4 Kosagudam East H H NH H H H NH NH H H H H 
 5 Nowrangpur East H NH NH H H H H H H H H H 
 6 Tentulikhuti East H NH NH H H H NH H H H H H 
 7 Kondagoan North NH H NH H H H NH H H H H NH 
 8 Narayanpur North H NH NH H H H NH NH H H H NH 
 9 Bhamragad West NH H NH H H H NH H NH H H NH 
 10 Etapalli West H H NH H H H NH H H H H NH 

Note: H and NH indicate homogeneous and non-homogeneous, respectively.

In general, IMD observed and IMD grid daily rainfall data sets are non-homogeneous at all stations, whereas TMPA shows a homogeneous nature only at station Bhamragad when tested with the Pettitt homogeneous test (Table 4). The monthly and annual rainfall of all three data sets is homogeneous in nature at all ten observed stations. Similarly, all rainfall data sets are non-homogeneous at the majority of observed stations during monsoon daily time scale, as shown in Table 4. The monsoon daily rainfall of IMD grid stations located in the southern and western portions of the basin shows conflict with the results of IMD observed rainfall when tested with the Pettitt test (Table 4). The results of TMPA daily monsoon rainfall is non-homogeneous at all observed gauge stations except at Narayanpur. Overall, the Pettitt test results of the three rainfall data sets at all time scales show similar results. From Table 4, it is also observed that SNH test shows mixed results for both IMD observed and IMD grid rainfall data sets during daily and monsoon daily time scales, whereas TMPA data are non-homogeneous at a daily time scale and homogeneous at monsoon daily time scale. The IMD grid stations located in the southern and western portions of the basin are homogeneous with the SNH test in daily rainfall which is different from the other two rainfall data sets. Similarly, it is also observed from Table 4 that SNH test results show an increase in number of homogeneous stations compared to Pettitt test at smaller time steps. The TMPA grid stations located in the north and south sides of the basin are non-homogeneous for monsoon total rainfall tested with SNH test, but IMD observed and IMD grid are homogeneous at this location. Thus, TMPA rainfall is more homogeneous in nature than IMD grid and IMD observed data during monsoon season at smaller time steps.

Overall, IMD observed, IMD grid, and TMPA rainfall data sets show similar results at the majority of stations by the Pettitt test at all time scales. The SNH test shows different characteristics compared to Pettitt test for all rainfall data sets and also saw an increase in homogeneous stations' number at smaller time step compared to the Pettitt test. Therefore, both IMD grid and TMPA rainfall data sets were further compared to test homogeneity characteristics at all grid stations over the Indravati basin.

The number of homogeneous and non-homogeneous stations at different temporal durations estimated using Pettitt and SNH tests is given in Table 5. All grid stations, except station 23 in Indravati basin, show a change in mean rainfall for daily IMD grid rainfall data set with the Pettitt test which is shown in Figure 13(a). While the daily TMPA rainfall also shows the change in mean rainfall at 64 stations, listed in Table 5, these non-homogeneous stations are spread throughout the entire basin, as shown in Figure 13(b) when tested with the Pettitt homogeneous test. It is also observed from homogeneity analysis that all non-homogenous stations of IMD grid and TMPA daily rainfall data sets show a jump in the middle of time series data when tested with the Pettitt test.

Table 5

Number of stations showing homogeneous and non-homogeneous characteristics in Indravati basin at different time scales

Time step in time series IMD grid rainfall
 
TMPA rainfall
 
Pettitt test
 
SNH test
 
Pettitt test
 
SNH test
 
H NH H NH H NH H NH 
Daily 78 41 38 15 64 79 
Monthly 64 15 79 48 31 79 
Annual 78 78 77 65 14 
Monsoon daily 24 55 47 32 70 72 
Monsoon monthly 78 76 79 79 
Monsoon total 79 78 70 52 27 
Time step in time series IMD grid rainfall
 
TMPA rainfall
 
Pettitt test
 
SNH test
 
Pettitt test
 
SNH test
 
H NH H NH H NH H NH 
Daily 78 41 38 15 64 79 
Monthly 64 15 79 48 31 79 
Annual 78 78 77 65 14 
Monsoon daily 24 55 47 32 70 72 
Monsoon monthly 78 76 79 79 
Monsoon total 79 78 70 52 27 

Note: H means homogeneous and NH means non-homogeneous.

Figure 13

Spatial distribution of homogeneity test results for daily rainfall: (a) Pettitt test for IMD grid rainfall, (b) Pettitt test for TMPA rainfall, (c) SNH test for IMD grid rainfall, (d) SNH test for TMPA rainfall.

Figure 13

Spatial distribution of homogeneity test results for daily rainfall: (a) Pettitt test for IMD grid rainfall, (b) Pettitt test for TMPA rainfall, (c) SNH test for IMD grid rainfall, (d) SNH test for TMPA rainfall.

There are 38 stations displaying non-homogeneity in daily IMD grid rainfall when tested with the SNH test and these stations are located on the north-western to east side of the basin, as shown in Figure 13(c). The daily TMPA rainfall shows non-homogeneity at all grid stations in Indravati river basin with the SNH test, as shown in Figure 13(d) which is contrasted to the characteristics shown by daily IMD grid rainfall when tested with the SNH test. The number of homogenous stations is increased in IMD grid data and decreased in TMPA data when tested with the SNH test compared to Pettitt test, which can be observed from Table 5. Similarly, from Table 5, monthly and annual rainfall data sets of IMD grid data display homogeneity at the majority of grid stations within the basin when tested using both Pettitt and SNH tests. TMPA rainfall data show 48 stations as homogeneous and 31 stations as non-homogeneous in monthly rainfall, but only two stations as non-homogeneous in annual rainfall with the Pettitt test, as shown in Table 5.

From Table 5, the Pettitt test shows 24 stations located in the western side of the basin as homogeneous in IMD grid monsoon daily rainfall data, as shown in Figure 14(a), and the same test showed nine stations are homogeneous in TMPA monsoon daily rainfall, shown in Figure 14(b). SNH test is non-homogeneous at 32 stations for IMD grid monsoon daily rainfall data and only seven stations for TMPA monsoon daily rainfall data, as shown in Figure 14(c) and 14(d). The change in mean rainfall is also observed at the ends of time series data for non-homogeneous stations when tested with SNH test. Overall, from Table 5, it is observed that TMPA rainfall is more non-homogenous in nature compared to IMD grid rainfall data set. The monsoon monthly rainfall shows all stations as homogenous in TMPA data sets when analyzed using Pettitt and SNH tests, whereas IMD grid monthly monsoon rainfall is non-homogenous at one station for the Pettitt test and three stations for the SNH test, which is also shown in Table 5 and Figure 15. It is also observed from Table 5 that the number of homogeneous stations becomes greater in both IMD grid and TMPA rainfall when moving from daily time step to annual time step and from full time series data to monsoon alone data.

Figure 14

Spatial distribution of homogeneity test results for monsoon daily rainfall: (a) Pettitt test for IMD grid rainfall, (b) Pettitt test for TMPA rainfall, (c) SNH test for IMD grid rainfall, (d) SNH test for TMPA rainfall.

Figure 14

Spatial distribution of homogeneity test results for monsoon daily rainfall: (a) Pettitt test for IMD grid rainfall, (b) Pettitt test for TMPA rainfall, (c) SNH test for IMD grid rainfall, (d) SNH test for TMPA rainfall.

Figure 15

Spatial distribution of homogeneity test results for total monsoon rainfall: (a) Pettitt test for IMD grid rainfall, (b) Pettitt test for TMPA rainfall, (c) SNH test for IMD grid rainfall, (d) SNH test for TMPA rainfall.

Figure 15

Spatial distribution of homogeneity test results for total monsoon rainfall: (a) Pettitt test for IMD grid rainfall, (b) Pettitt test for TMPA rainfall, (c) SNH test for IMD grid rainfall, (d) SNH test for TMPA rainfall.

Overall, the statistical characteristics, as well as internal characteristics such as trend and homogeneity analysis, show that TMPA grid rainfall is in line with IMD observed data. The study also reveals that TMPA rainfall is free from trends, but more non-homogeneous in nature compared to IMD grid rainfall data in Indravati river basin.

CONCLUSIONS

The preliminary statistical comparison of three rainfall data sets, namely, TMPA, IMD grid, and IMD observed gauge data sets at a river basin has been carried out and results reveal the monthly rainfall of TMPA data as well as IMD grid data are correlated well with IMD observed data, whereas a poor correlation is observed for daily and annual time scales. From the spatial maps of mean annual rainfall estimated for TMPA and IMD grid data sets, around 1,000 mm difference in maximum rainfall is observed between the two data sets at the west side of the basin. Therefore, differences in TMPA rainfall magnitudes are observed at a higher elevation when compared to observed data. Furthermore, the number of rainy days analysis shows very good agreement between TMPA rainfall and IMD observed rainfall. IMD grid data also show good results in terms of all preliminary statistics when compared with IMD observed data, since the data set itself originated from IMD observed rainfall.

The trend analysis was carried out using MK trend test and homogeneous analysis using Pettitt and SNH tests at various time scales. IMD grid rainfall showed a significant increasing trend at stations located on the east side of the Indravati river basin for daily time scale (for both full time series as well as monsoon alone). TMPA rainfall is observed to be free from trends in the majority of locations for daily rainfall data. This places more emphasis on using TMPA rainfall data in climate- and water-related studies. The results of homogeneity analysis showed that daily TMPA rainfall displays non-homogeneity at the majority of stations spread over the area of interest, whereas IMD grid rainfall displays non-homogeneity at stations located on the southern side of the basin. Overall, the trend and homogeneity analysis shows that the IMD grid is close to the observed data and TMPA has uniformly no trend with non-homogenous characterized rainfall. Thus, it is concluded that TMPA and IMD grid data are a good representative of rainfall for climate- and water-related studies at large catchment scales especially in data-scarce regions.

ACKNOWLEDGEMENTS

The authors would like to thank the India Meteorological Department (IMD) for providing gridded rainfall data as well as the observed rain gauge stations data. The authors would like to extend their thanks to Goddard Earth Sciences Data and Information Services Center (GES DISC) for providing free access to download TMPA precipitation data.

REFERENCES

REFERENCES
Huang
,
S.
,
Chang
,
J.
,
Huang
,
Q.
&
Chen
,
Y.
2015
Identification of abrupt changes of the relationship between rainfall and runoff in the Wei River Basin, China
.
Theor. Appl. Climatol.
120
(
1–2
),
299
310
.
Huffman
,
G. J.
&
Bolvin
,
D. T.
2013
TRMM and Other Data Precipitation Data set Documentation
.
NASA Greenbelt USA 1–40. https://pmm.nasa.gov/sites/default/files/document_files/3B42_3B43_doc_V7.pdf (accessed 21 August 2017)
.
Huss
,
M.
,
Bauder
,
A.
&
Funk
,
M.
2009
Homogenization of long-term mass-balance time series
.
Ann. Glaciol.
50
,
198
206
.
Jaiswal
,
R. K.
,
Lohani
,
A. K.
&
Tiwari
,
H. L.
2015
Statistical analysis for change detection and trend assessment in climatological parameters
.
Environ Process.
2
(
4
),
729
749
.
Jiang
,
S. H.
,
Zhou
,
M.
,
Ren
,
L. L.
,
Cheng
,
X. R.
&
Zhang
,
P. J.
2016
Evaluation of latest TMPA and CMORPH satellite precipitation products over Yellow River Basin
.
Water Sci. Eng.
9
(
2
),
87
96
.
Joshi
,
N.
,
Gupta
,
D.
,
Suryavanshi
,
S.
,
Adamowski
,
J.
&
Madramootoo
,
C. A.
2016
Analysis of trends and dominant periodicities in drought variables in India: a wavelet transform based approach
.
Atmos. Res.
182
,
200
220
.
Kendall
,
M. G.
1975
Rank Correlation Methods
.
Charles Griffin and Co. Ltd.
,
London
,
UK
.
Kizza
,
M.
,
Westerberg
,
I.
,
Rodhe
,
A.
&
Ntale
,
H. K.
2012
Estimating areal rainfall over Lake Victoria and its basin using ground-based and satellite data
.
J. Hydrol.
464
,
401
411
.
Machiwal
,
D.
&
Jha
,
M. K.
2012
Hydrologic Time Series Analysis: Theory and Practice
.
Springer
,
Netherlands
.
DOI:10.1007/978-94-007-1861-6
.
Mann
,
H. B.
1945
Nonparametric tests against trend
.
Econometrica
13
,
245
259
.
Narayanan
,
P.
,
Basistha
,
A.
,
Sarkar
,
S.
&
Kamna
,
S.
2013
Trend analysis and ARIMA modelling of pre-monsoon rainfall data for western India
.
C. R. Geosci.
345
(
1
),
22
27
.
Oliveira
,
P. T. S.
,
Nearing
,
M. A.
,
Moran
,
M. S.
,
Goodrich
,
D. C.
,
Wendland
,
E.
&
Gupta
,
H. V.
2014
Trends in water balance components across the Brazilian Cerrado
.
Water Resour. Res.
50
(
9
),
7100
7114
.
Sovilj
,
D.
,
Eirola
,
E.
,
Miche
,
Y.
,
Björk
,
K. M.
,
Nian
,
R.
,
Akusok
,
A.
&
Lendasse
,
A.
2016
Extreme learning machine for missing data using multiple imputations
.
Neurocomputing
174
,
220
231
.
Tomozeiu
,
R.
,
Stefan
,
S.
&
Busuioc
,
A.
2005
Winter precipitation variability and large-scale circulation patterns in Romania
.
Theor. Appl. Climatol.
81
(
3–4
),
193
201
.
Xu
,
X.
,
Li
,
J.
&
Tolson
,
B. A.
2014
Progress in integrating remote sensing data and hydrologic modeling
.
Prog. Phys. Geogr.
38
(
4
),
464
498
.