Ms Aditi Sharma, Dr.V.K.Dadhwal, Dr.C.Jeganathan, Dr.Valentyn Tolpekin
Drought is an insidious hazard of nature which is considered by many to be the most complex but least understood of all natural hazards. Large historical datasets are required in order to study drought which involves complex inter-relationship between the climatological and meteorological data. For the present study, daily rainfall datasets were acquired from Drought Monitoring Cell (DMC), Karnataka for the period of 1970-2004. Then, the Rainfall data from 1970-2004 was used to compute Standardized Precipitation Index (SPI). Ranking of various interpolation techniques to derive the spatial SPI distribution map is highlighted. SPI value -1.5 was used as threshold value and all the interpolated SPI maps were reclassified into two classes. i.e. SPI value from -3 to -1.5 as drought and SPI value from -1.5 to +3 as no drought category. SPI was then used to find the drought characteristics like onset, severity, and spatial extent.
To validate the findings from SPI, Govt. based drought assessment reports were used and correlation coefficient of 0.89 was achieved, which indicates strong positive correlation.
Drought has long been recognized as one of the most insidious causes of human misery and being the natural disaster that annually claims the most victims. Its ability to cause widespread misery is actually increasing (Perez et al.,1995). It is a temporary, recurring natural disaster, which originates from the lack of precipitation and brings significant economic losses. It creates an atmosphere of doom and despair. Water scarcity can be said as the cause and effect of drought. Drought occurs whenever and wherever the links in the hydrological cycle is broken or is destabilized. Drought is a slow poison, no one knows when it creeps in, it can last any number of days and its severity cannot be predicted. It has extensive spatial dimension and thus can have serious implications on the socio-economic stability of an entire region. It is not possible to avoid droughts. But drought preparedness can be developed and drought impacts can be managed. The success of both depends, amongst the others, on how well the droughts are defined and drought characteristics quantified (Smakhtin. et al.,2004).
Most often from literatures, operational definitions identify the beginning, end, spatial extent and severity of a drought. They are often region-specific and are based on scientific reasoning, which follows the analysis of certain amounts of hydrometeorological information. They are beneficial in developing drought policies, monitoring systems, mitigation strategies and preparedness plans. Operational definitions are formulated in terms of drought indices (Smakhtin. et al.,2004).Drought indices assimilate thousands of bits of data on rainfall, snowpack, stream flow, and other water supply indicators into a comprehensible big picture. A drought index value is typically a single number, far more useful than raw data for decision making(Hayes,2003).Drought indices can be derived from hydro-meteorological data and remote sensing data. This paper highlights the applicability of the SPI (Standard Precipitation Index) as a measure for Drought monitoring in the state of Karnataka, India. In addition, comparison of various interpolation techniques to interpolate SPI is discussed and analysed.
2. Standardized Precipitation Index (SPI)
SPI was developed in Colorado by McKee et al. (1993) and is based just on precipitation and, therefore, requires less input data and calculation effort than other indices like Palmer Drought Severity Index PDSI. A long-term precipitation record at the desired station is fitted to a probability distribution (e.g., gamma distribution), which is then transformed into a normal distribution so that the mean SPI is zero (Edwards and McKee 1997).
3. Study area:
The State of Karnataka is confined within 11.5 degree North to 18.5 degree North latitudes and 74 degree East and 78.5 degree East longitude. It accounts for 5.83 percent of the total area of the country and ranks eighth among major States of the country in terms of size.
Figure1: Map of India highlighting Karnataka State (left) & District Map of Karnataka(right)
Rainfall description/Variability of the Region:
The annual rainfall in the State varies roughly from 50 to 350 cm. In the districts of Bijapur, Raichur, Bellary and southern half of Gulbarga, the rainfall is lowest varying from 50 to 60 cm. The rainfall increases significantly in the western part of the State and reaches its maximum over the coastal belt. The south-west monsoon is the principal rainy season during which the State receives 80% of its rainfall. Rainfall in the winter season (January to February) is less than one per cent of the annual total, in the hot weather season (March to May) about 7% and in the post-monsoon season about 12%.
4. Data Requirement and Methodology
A. Rainfall dataset requirement:
Daily rainfall datasets were acquired from Drought Monitoring Cell (DMC), Karnataka for the period of 1970-2004. The India Meteorological Department (IMD) and State department have setup a rainfall monitoring station for each taluk.
B. SPI Methodology:
Various interpolation methods were explored in detail to find the optimum technique for interpolating SPI from 1981-2003, which inturn helps to find the drought characteristics like onset, severity, and spatial extent. In addition, quantification of SPI pattern through Global Moran’s I method was performed in this study. Details workflow is given in Figure 2 below:
Figure2: Procedure for analyzing SPI and implementation of various Interpolation techniques
5. Analysis and Results:
A. Interpolating SPI using various interpolation techniques
In order to get spatial pattern of drought, interpolation of SPI was done. As the AVHRR-NDVI data is from 1981-2003, so SPI was interpolated for this time period for four months i.e. July, August, September and October. Latitude and longitude of all weather stations were taken and a point map was created. Then, each month was added to the point map and various interpolation techniques were used. Three important criteria’s i.e. elevation, distance from coast and aspect were considered while interpolating SPI. The results of each interpolation technique was compared on the basis of mean absolute error (MAE), mean squared error (MSE) and root mean squared error (RMSE). Equation of each is given below:
MAE=Σ(Predicted-Observed) / (Number of observations) (Equation 1)
MSE=Σ (Predicted-Observed)2 / (Number of observations) (Equation 2)
RMSE=Σ (MSE) (Equation 3)
B. Comparision of various interpolation techniques in order to interpolate SPI
All the interpolators used for SPI were compared on basis of mean absolute error (MAE), mean squared error (MSE) and root mean squared error (RMSE).
After calculating the errors for all the interpolation techniques i.e. Ordinary kriging, Co-kriging, Indicator kriging, Residual kriging, Thin plate spline, Completely regularized spline, Local Polynomial interpolation – using optimized weightage distance and Inverse Distance Weighting (IDW) using optimal power function; IDW proved to be best as it gave least error. Table 1 shows the ranking of all the interpolation techniques. Among the various criteria’s of spatial interpolation, extent of study area also plays an important role. The area of Karnataka State is very large 1,91,791 sq.km. Similar results were obtained in a study by Collins(2000), where he compared various interpolation techniques for different regions (region 1 had large spatial extent and region 2 was small). In his study also IDW using optimal power gave consistently better results than other techniques.
Table 1: Ranking various interpolation techniques
Figure 3: Interpolated SPI Maps for August 1984 (order as in table 1)
Evidence was found during the analysis of various interpolators that certain apriori data characteristics influence the choice of spatial interpolation technique. Distance, aspect, SPI correlation with elevation, spatial scale, relative spatial density and distribution of sampling stations, all influence the choice of interpolation technique.
C. SPI as Robust Index for drought analysis
SPI is a good drought monitoring tool. It also gives indication of the drought characteristics like onset, severity, and spatial extent.
C.1 Drought sensitive Station
Here, drought sensitive stations have been identified for the month of July, taken into account the whole time-period from 1970-2004. The onset of monsoon season start in the month of July, so the SPI value corresponding to this month for all the meteorological stations were analysed to determine the frequency of those stations whose SPI value falls below -1.5. 24 stations out of 175 stations were found to be more sensitive (Figure 4). It was observed that some stations like Sindgi recorded a SPI value less than -1.5, four times as compared to other stations where the frequency is less i.e.3, 2, 1 and 0, for the last 35 years. Hence, it could be concluded that those stations which records SPI values less than -1.5 more frequently for the last 35 years was identified and considered to be drought sensitive stations for further analysis. Moreover, it signifies the sensitivity of SPI values as a drought indicator index.
Figure4: Selection of Drought sensitive stations using SPI
C.2 Selection of Year
Figure 5: Selection of Drought sensitive year using SPI value recorded at Sindgi
Selection of Drought year using SPI was attempted and it could be clearly seen from the time-series graph from 1970-2004, that lowest SPI value was observed in the year 1972. Considering the satellite data acquisition from 1981-2003, the year corresponding to lowest SPI within this time –series was taken and considered as ‘drought year’ and its corresponding SPI value is -2, which is below -1.5, as shown in Figure 5. Similarly, year 2000 was chosen as a ‘normal year’ considering its observed SPI value which is 0.72.
C.3 Spatial Extent of Drought
Using SPI value -1.5 as threshold, all the interpolated SPI maps were reclassified into two classes. i.e. SPI value from -3 to -1.5 as drought and SPI value from -1.5 to +3 as no drought category. Results for the year 1994 and 2000 are shown in figure 6.
Figure 6: Interpolated SPI map for year 1994 (left) and 2000 (right).
It can be depicted from the figure 6 that year 1994 was a drought year whereas year 2000 is a normal one. Through SPI, we can compare two different stations in different climatic regions regardless of the fact that they may have different normal rainfalls; because the rainfall is already normalized and compares the current rainfall with the average. The rainfall of two areas with different rainfall characteristics can be compared in terms of how badly they are experiencing drought conditions since the comparison is in terms of their normal rainfall. Hence, SPI is more efficient than rainfall in spatial analysis of drought.
C.4 Onset of Drought
Figure 7 : Interpolated SPI map for the month July, August, September and October 1994
With the help of SPI it is also possible to detect the onset of drought. Classified drought map of the year 1994 is shown in the Figure 7.
It can be seen that in the year 1994, July month experienced only a small portion of drought in northern part, whereas in August small pockets of drought phenomenon can be seen observed in northern and southern part of Karnataka. On the contrary, in September drought can be seen concentrated mostly in northern part and some portion in the southern region. No indication of drought can be seen in the month of October.
Arrival of early drought can be seen at Sindgi taluk [refer : SPI-July 1994], whereas none of the place suffered from late drought in the year 1994.
C.5 Drought Severity
According to the standard classification of SPI values, a station can suffer extreme drought situation, if the SPI value is below -2. Drought severity graph for September 1994 is shown in Figure: 8.
Figure 8: SPI graph for September 1994 to identify drought severity.
Threshold line having SPI value -2 is drawn in order to know the stations which suffered from very severe drought. Out of 175 weather stations, 24 stations experienced very severe drought.
D. Validating SPI results with Govt. drought assessment report
Govt. drought assessment report on Taluk basis was used for validating the results. Area (in percentage) affected by drought was given in the report. Interpolated SPI map having a grid size of 1 km was overlaid with the taluk map of Karnataka.
Drought affected area in each taluk was extracted out by using the zonal statistics from the SPI map. This, inturn gives the percentage of drought affected area in each of the affected taluk. Thenafter, SPI area under drought was plotted against Govt. area under drought, as shown in figure 9 and the correlation coefficient achieved was 0.89(positive correlation).
Figure 9: Validation of SPI results with govt. reports Hence, the results obtained were validated.
From this case study, the following can be concluded :
- Rainfall varies spatially and temporally throughout the whole Karnataka. On analyzing the rainfall for all the 175 stations in the state from 1970-2004, it was found that there is a large variation in rainfall especially in the months corresponding to July- October in all the years. The minimum and maximum mean rainfall observed during this time period is about 28.10 mm to 1489.11 mm, where it indicates a large variation in distribution of rainfall in all the stations. Highest rainfall was found to be occurred in the coastal areas and least rainfall in this time-period was seen in the central and southern Karnataka Plateau.
- Occurrence of drought cannot be monitored by comparing the relative rainfall observed in various stations. To overcome these limitations, the use of SPI for drought monitoring was highlighted. SPI was computed at time-scale 1 and 2 for all stations. Further interpolation of SPI was carried out using various interpolation techniques in order to visualize it spatially. Among all the techniques i.e. Ordinary kriging, Co-kriging, Indicator kriging, Residual kriging, Thin plate spline, Completely regularized spline, Local Polynomial interpolation – using optimized weightage distance and Inverse Distance Weighting (IDW) using optimal power function; IDW proved to be best as it gave least error.
- Given the probabilities distribution of SPI, it was concluded that SPI is an excellent means that gives indication of the drought characteristics like onset, severity, and spatial extent. Through SPI, we can compare two different stations in different climatic regions regardless of the fact that they may have different normal rainfalls; because the rainfall is already normalized and compares the current rainfall with the average. The rainfall of two areas with different rainfall characteristics can be compared in terms of how badly they are experiencing drought conditions since the comparison is in terms of their normal rainfall. Hence, SPI is more efficient than rainfall in spatial analysis of drought.
- Collins, F.(2000).A comparison of Spatial Interpolation Techniques in Temperate Estimation www.ncgia.ucsb.edu/conf/SANTA_FE_CD_ROM/sf_papers/collins_fred/collins.html
- J.Hayes, M.(2003).Drought Indices
- Kogan., F. N. (2000). Contribution of Remote Sensing to Drought Early Warning. NOAA, NESDIS
- National Drought Mitigation Center.(2005).What is Drought
- Perez, E. and P. Thompson.(1995).Natural Hazards: Drought
- Smakhtin., V. U. and D.A.Hughes (2004). Review, Automated Estimation and Analyses of Drought Indices in South Asia. W. P. 83. Colombo, Sri Lanka, International Water Management Institute