ASSESSMENT AND MONITORING OF AGRICULTURAL DROUGHTS IN MAHARASHTRA USING METEOROLOGICAL AND REMOTE SENSING BASED INDICES

Drought is a recurring climatic event characterized by slow onset, a gradual increase in its intensity, and persistence for a long period depending upon the availability of water. Droughts, broadly classified into meteorological, hydrological and agricultural drought, which are interconnected to each other. India, being an agriculture based economy depends primarily on agriculture production for its economic development and stability. The occurrence of agriculture drought affects the agricultural yield, which affects the regional economy to a larger extent. In present study, agricultural and meteorological drought in Maharashtra state was monitored using traditional as well as remote sensing methods. The meteorological drought assessment and characterization is done using two standard meteorological drought indices viz. standard precipitation index (SPI) and effective drought index (EDI). The severity and persistency of meteorological drought were studied using SPI for the period 1901 to 2015. However, accuracy of SPI in detection of sub-monthly drought is limited. Therefore, sub-monthly drought is effectively monitored using EDI. The monthly and sub-monthly drought mapped using SPI and EDI, respectively were then compared and assessed. It was concluded that EDI serves as a better indicator to monitor sub-monthly droughts. The agricultural drought monitoring was carried out using the remote sensing based indices such as vegetation condition index (VCI), temperature condition index (TCI), vegetation health index (VHI), shortwave angle slope index (SASI) and the index which maps the agricultural drought in a better way was identified. The area under drought as calculated by various agricultural drought indices compared with that of the EDI, it was found that the results of SASI matched with results of EDI. SASI denotes different values for the dry and wet soil and for the healthy and sparse vegetation. SASI monitors the agricultural drought better as compared to other indices used in this study.


INTRODUCTION
Drought is complex natural hazard, which has a slow onset and is defined by the acute water shortage.It is described in many ways depending on its impact and varying characteristics.The prolonged low rainfall marks the onset of meteorological drought, which is followed by hydrological drought as a decrease in the surface water and ground water levels (Hisdal and Tallaksen, 2003), which has a direct effect on the agricultural crops resulting in agricultural drought (Mishra and Singh, 2010).* Thus depending on the impact, which drought has on various sectors, it is classified into Meteorological, Hydrological, and Agricultural drought.The effect of drought can be seen worldwide on the livelihood, environment, economy, and overall human well-being.Globally, around 11 million people died due to drought and 2 billion people were adversely affected by drought since 1900 (FAO, 2015).As per the reports from Food Security Information Network, around 108 million people from all around the globe were affected by food insecurity, caused by an El-Nino induced drought (FISN, 2016).India has a long history of drought events, and its impact on the livelihood of people.During * Corresponding author the period, 1871 to 2015, India witnessed around 25 major drought, the one in 1987 was a severe most, a similar drought event repeated in 2002, which affected a population of 285 million people and 59-60% crop area too.68% of the total cropped area in India is drought prone, 33%of which receives less than 750mm of mean annual rainfall (UNICEF, 2015-16).During the years, 2014-15 and 2015-16, major agricultural states were drought affected.The state of Maharashtra, the study area of this work, was severely drought affected in recent years (e.g.1996, 1997, 2001, 2002, 2003, 2004 and 2015).The drought, which occurred in 1996, affected around 266.75 lakh people, settled in 7 districts.Similarly, 17 districts were drought affected during year 1997.Around 50% of the drought prone area of the state is constituted by the Deccan Plateau, which receives an average annual rainfall of 600 to 750 mm (NIDM, 2016).Approximately 4.5 million hectares of crop was affected in the 2001 drought.The agriculture sector gets severely affected by these droughts impacting the livelihood of people depending on it.Taking into consideration, the changing climatic patterns and the irregular rainfalls, the risk of drought occurrence increases (Zhang et al., 2017).Studying the meteorological and agricultural droughts and how it affects the Maharashtra state helps in reduction of the impact of future droughts.Analysis and prediction of droughts is the necessity of the hour in the Maharashtra state.Thus, drought monitoring is of immense importance, which may be accomplished by using different remote sensing based and traditional indices.The monitoring of meteorological and agricultural drought can be done using meteorological and remote sensing based indices, respectively.

Drought Monitoring Traditional Indices
The impact of drought is severe as compared to other natural calamities.Drought can be characterized in terms of its severity, spatial extend and persistence.The severity of drought represents how intensely the drought has hit the area, the persistence defines how long the drought remained in the area (WMO, 2016 andGWP, 2016).The meteorological drought, which is marked by a substantial decrease in rainfall, is well monitored using many indices like Rainfall Anomaly Index (RAI), Standard Precipitation Index (SPI), Effective Drought Index (EDI), etc.These meteorological indices are calculated based on rainfall data, soil moisture, which are obtained for point locations.SPI and EDI uses the precipitation data for the analysis, with a minimum of 30 years datasets.The rainfall data collected are discrete and are sparsely located.The other parameters such as vegetation, temperature and soil moisture are not taken into account by SPI and EDI.These parameter govern the response of the agriculture system to the water shortage.Thus, remote sensing based indices are more preferred for agricultural drought monitoring.

Agricultural Drought Monitoring Using Remote Sensing Based Indices
Out of the total damage caused by natural hazards, around 22% affects the agriculture sector (FAO, 2015).Drought is the largest natural hazard for agricultural sector causing 84% of losses to the sector out of losses due to all natural hazards (FAO, 2014).

Datasets Used
The study uses meteorological, remote sensing and ancillary datasets.The meteorological datasets include the grid wise precipitation data obtained from India Meteorological Department (IMD), which has a spatial resolution of 0.25 degree.The data from 1901 to 2015 has been extracted grid wise for around 500 grids for whole state of Maharashtra.The ancillary datasets used includes the agro-ecological boundary map from the NBSS & LUP, the administrative boundary map from Survey of India (SOI).

Methodology
The methodology constitutes of two sections: The meteorological drought assessment and the remote sensing based drought assessment.The meteorological part involves the calculation of two meteorological drought indices namely Standard Precipitation Index (SPI) and the Effective Drought Index (EDI).The agricultural drought is monitored using the remote sensing based indices.The remote sensing (RS) indices used in the study are Vegetation Condition Index (VCI), Temperature Condition Index (TCI), Vegetation Health Index (VHI), and Shortwave Angle Slope Index (SASI).The comparison between different RS based indices was carried out to identify the best suitable index for the study area.

Meteorological drought monitoring:
Meteorological drought monitoring can be accomplished by using various meteorological indices like Rainfall Anomaly Index (RAI), Percent of Normal, SPI, EDI, Reconnaissance Drought Index (RDI), etc.In the present study, the meteorological drought monitoring has been carried out by using SPI and EDI due to their wide acceptance and intercomparability.The meteorological drought index, SPI, was calculated for 115 years from 1901 to 2015 using gridded rainfall data for over 499 grids covering entire state.Initially SPI computation involves the determination of probability density function, which explains the long-term time series of precipitation data.This is followed by the estimation of cumulative probability.The inverse normal function with mean zero and variance one is applied to cumulative probability function to obtain the required SPI values (Guttman, 1999).SPI computation involves fitting the rainfall values to the gamma distribution by using the following equations.
where α >0, α is a shape factor, β >0, β is the scale factor and x represents the precipitation data.
, A= ln x- where, Xi represents all non-zero precipitation values and N is the number of non-zero precipitation values.The value of A is obtained by subtracting the log of all precipitation data from the log of mean of all rainfall amount.The scale parameter and shape parameter can be calculated from the equation 3 and 4 (Shah et al., 2015).For zero values of precipitation, the gamma function is not defined, to incorporate those values.The cumulative probability is modified by the given formula.
In Equation 5, the probability of a zero rainfall is denoted by q.If the number of zero rainfall values is represented by m, then q can be estimated by m/n.

Remote sensing based drought monitoring
The remote sensing based drought monitoring has advantages like near real time monitoring at low cost with very less input of time and labor.The preprocessing of remote sensing datasets include the following Conversion of HDF file to TIFF formats Mosaicking ,clipping, and re-projecting Scaling-The scaling factor of LST, NDVI and Reflectance data are 0.02, 0.0001, 0.0001 respectively.
Masking-The MODIS LULC was used to map out the agricultural areas and cloud masking was also carried out.
The remote sensing indices VCI, TCI, VHI and SASI were computed.Agriculture drought is always marked by a decrease in amount of vegetation in fields (Tsiros et al., 2004).This can be easily identified using VCI as it depend completely on NDVI.The process of normalization is used to identify the drought and non-drought pixels (Quiring and Ganesh, 2010).The maximum and minimum NDVI values denoted by 100 and 0 are linearly scaled (Kogan, 1995).
The VCI values also ranges from 0 to 100 depending upon the absence or presence of vegetation in field.
VCI= 100*( In Equation 9, NDVImin, NDVImax represents the multiyear maximum and minimum values of NDVI. VCI has very low value in case of high cloud cover, thus wrongly depicting them as drought prone areas.To overcome such problems the temperature-based indices can be used, which uses the thermal band derived brightness values to compute TCI.Its values ranges from 0 to 100, a higher value suggest drought while the lower values indicate non drought areas (Kogan, 1995).
where BT, BTmax, BTmin are multiyear maximum and minimum of brightness temperature.An integrated approach of using both LST and NDVI, counters the limitation of individual indices.VHI can be computed using the formula given below.
VHI=a VCI + (1-a) TCI ( 11) where the value of 'a' denotes the weight constant for VCI, which shows the contribution of VCI in the calculation of VHI.The value of 'a' depends on the contributions of moisture and temperature during a vegetation cycle.It is generally assumed to have equal weightage, as it is difficult to determine it.As far as agricultural drought is concerned, the soil moisture is an important parameter.The above mentioned indices does not take care of soil moisture while identifying drought in an area.Palacios-Orueta et al. ( 2006) used an index SASI, which incorporates the effect of soil moisture in the drought measurement.SASI measures the angle formed between the bands NIR, SWIR1, SWIR2 of central wavelengths 858mm, 1240mm, 1640mm respectively.SASI has the capability of differentiating the dry and wet soil and the healthy and sparse vegetation (Das and Murthy, 2013).As compared to other indices, SASI depends on the inert band relationships and the angle formed between them.The inclusion of two SWIR bands makes SASI more sensitive to soil moisture.SASI values varies from positive to negative as the soil condition shifts from dry to wet.
SASI computation involves the following steps: where, a, b and c represents the Euclidian distances between the vertices NIR and SWIR1, SWIR1 and SWIR2, and NIR and SWIR2, respectively.The value of angle βSWIR1 varies from large to small as the soil moisture shifts dry to moist (Khanna et al., 2007).A high positive value of SASI is representative of a dry soil while a high negative SASI value indicates healthy vegetation (Das and Murthy, 2013).
The area under drought as estimated by different drought indices were identified and these results were compared with the area under drought as per EDI.The comparison was carried out to identify the best remote sensing index for assessment and monitoring of agricultural drought.

Meteorological Drought in Maharashtra
Two meteorological drought indices namely SPI and EDI were used to identify the meteorological drought in the Maharashtra state.The gridded rainfall data from 1901 to 2015 were used calculating SPI and EDI.The EDI values for the time period 2000-2015 were used further for the comparison with remote sensing based indices.The drought and non-drought years for this time period were also identified for the comparison.
The severity and persistence of drought were analyzed grid wise for different districts.The severity of drought represents the intensity of drought, while the persistence shows how long the drought prolonged.The grid wise plots of severity and persistence were created.The Figures 3 & 4 shows the number of drought that has occurred for the grid 164 (18.83N, 76.108E) and grid 166 (18.83N, 76.108E), both these grids represent are covered by drought prone Beed District.It is observed from Figure 3 that a maximum of four drought have occurred in a year (1951,1954,1960,1962) in the grid 164 and in grid 166 maximum of 5 drought have occurred in a year (1973).Figure 5 2002, 2007, 2009, 2011, 2014and 2015. Figure . Figure 9 shows the interpolated maps for selected month of the years 2001, 2002, 2009.

EDI based meteorological drought monitoring
EDI maps were composed for a time period of 16 years from 2000-2015.EDI were calculated on 16 daily basis and total 384 EDI maps were generated using IDW interpolation technique.It is observed that EDI has an advantage of identifying the sub-monthly drought.SPI was unable to identify the drought, which occurred for a time period of less than a month.As SPI uses the total monthly rainfall, if a rainfall deficit occurs in the first fortnight and a rainfall excess occurs in the second fortnight, SPI will not be able to account for the rainfall deficit, which has occurred in the first fortnight.When it comes to agricultural drought, the unnoticed drought in the first fortnight can lead to a decrease in production.This will drastically affect the yield.Therefore, sub-monthly scale of monitoring is very much crucial as far as agricultural drought is considered.Figures 10 (a

Comparison of SPI and EDI
The area under drought as obtained from EDI and SPI were compared with each other for different ecological zones.The comparison of SPI and EDI can give three kinds of observations.Firstly, when both SPI and EDI shows drought, the observaions of these kind are plotted in the Figure 11.As it can be seen that in April 2000, October 2000, May 2010, May 2011 and May 2012, the area under drought as per SPI and EDI is same.Secondly, when it is a drought as per SPI while either of the EDI values disagree or vice-versa.The Figure 12 shows this case, where those areas that are marked as drought by SPI are not under drought in one of the fortnights.These differences have occurred in those areas, which are brightened areas in the XOR map shown in Figure 13.It was found that for grid 412 (20.84 N, 78.826 E), which belongs to the brightened area, first fortnightly rainfall is 0 mm and for the second fortnight, the rainfall is 135.58 mm.Thus, the sub-monthly drought could not be mapped by SPI while EDI could effectively identify that.As it can be seen from these graphs that the area under drought obtained by using EDI matches well with that of the area obtained using SASI.This shows that SASI has an advantage over other indices.SASI discriminates between the dry and wet soil as well as the dry and moist vegetation too.VCI and TCI on the other hand over predicts the area under drought.The comparison was done for drought year, 2002 and non-drought year 2010.For the year 2002, 41% of the state area was under drought as per TCI, while as per SASI and EDI, it was around 22.9 % and 27.8 % respectively for the October first fortnight.For the nondrought year 2010, VCI predicted 22.64 % drought area in the state, while as per EDI and SASI around 3% and 4.9% area was under drought.Figures15 (a) and (b) show the comparison of area under drought as per various remote sensing indices and EDI for the year 2002 and 2010.This shows that the results of EDI and SASI are comparable.
The remote sensing based datasets where used for the computation of remote sensing based indices.The Moderate Resolution Imaging Spectroradiometer (MODIS) based Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), Enhanced Vegetation Index (EVI), Reflectance data and Land use land cover (LULC) datasets were obtained from U.S Land Processes Distributed Active Archive Center (LPDAAC) (http://lpdaac.usgs.gov).

Figure 5 .
Figure 5.Graph showing persistence of drought in grid 164 (18.83N, 76.108E) ) & (b) show the interpolated EDI maps.As it can be seen for April 2008, the first fortnight has very less area under drought, but the second fortnight shows more area under drought.The use of EDI helps in identifying the submonthly drought which has occurred in the second fortnight.

Figure 12 .Figure 15 .
Figure 12.SPI map of October 2000 was compared with EDI maps of October 2000 first second fortnight Figure 13.XOR Map Thisproves that majority of the agricultural is rainfed.The scarce and inadequate rainfall causes these areas to be severely drought hit, thus around 24% of country's drought prone area lays in the state.
like Godavari and Krishna.The entire plateau is composed of rocks from diverse origin, which has undergone substantial amount of metamorphism.These ancient rocks mainly comprise of the Deccan traps, which covers up to 80% area of the state.The Deccan Plateau has black soil, which is rich in iron and moisture retentiveness but has less nitrogen and organic matter content.National Bureau of Soil Survey and Landuse Planning (NBSS&LUP) has broadly classified the soil of the state into four groups, which are Soil of Konkan coast, Soil of Western Ghats, Soil of the Upper Maharashtra, Soil of Lower Maharashtra.The climate of Maharashtra varies between heavy monsoon showers to hot summers.Physiography and location govern the type of climate prevailing in an area, the coastal region receives heavy monsoon while it decreases as we proceed towards the east.The annual rainfall reaches a maximum of around 6000mm in Ghats and decreases to the east to reach a level of below 500mm.It further increases in the east to reach a second peak of 1500mm.The temperature in the state varies between a maximum in the range of 27°C and 40°C to a minimum temperature in range 14°C and 27°C.Based on vegetation, soil and rainfall, the state has been divided into 9 agro climatic zones which are South Konkan Coastal Zone, North Konkan Coastal Zone, Western Ghats, Transitions Zone-1, Transition Zone-2, Scarcity Zone, Assured Rainfall Zone, Moderate Rainfall Zone, and Eastern Vidharbha Zone.Agriculture is of immense importance for the state of Maharashtra.Around 61% of the total population of the state depends on agriculture directly or indirectly.Jowar, Pulses, Rice, Sugarcane, Bajra, Turmeric, Cotton, oil seeds like Sunflower, Groundnut and Soyabean are the major crops cultivated.Grapes, Oranges, Mangos, and Bananas account for the fruits cultivated in the state.Out of the total 226.1 lakh ha area under cultivation, 33.5 lakh ha area is under irrigation.
(Shah et al., 2015)ability H(x) is then converted to SPI values by converting them into standard normal random variable (Z).Based on the Z values, the area can be classified as wet and dry, as positive SPI indicate wetness and negative SPI values indicate dryness(Shah et al., 2015)