journal of computing::data mining and wireless sensor...

VOL. 3, NO. 6, June 2012 ISSN 2079-8407 Journal of Emerging Trends in Computing and Information Sciences

©2009-2012 CIS Journal. All rights reserved.

http://www.cisjournal.org

913

Data Mining and Wireless Sensor Network for Groundnut Pest Thrips Dynamics and Predictions

1 A. K. Tripathy, 2 J. Adinarayana, 3 S. N. Merchant, 4 U. B. Desai, 5 K. Vijayalakshmi, 6 D. RajiReddy, 7 S. Ninomiya, 8 M. Hirafuji, 9 T. Kiura

1, 2, 3 Indian Institute of Technology Bombay, Mumbai, India 4 Indian Institute of Technology Hyderabad, Andhra Pradesh, India

5, 6 A.N.G.R. Agricultural University, Hyderabad,India 7 The University of Tokyo, Nishi Tokyo, Japan

8,9 National Agricultural Research Center, Tsukuba, Japan 1 [email protected], 2 [email protected], 3 [email protected], 4 [email protected], 5 [email protected]

, 6 [email protected], 7 [email protected], 8 [email protected], 9 [email protected]

ABSTRACT

With the advent of data generation, collection and storage technologies, world is overwhelmed with data everywhere. Following this trend, more and more agricultural data are nowadays are virtually being harvested along with the crops and are being collected/stored in databases. As the volume of the data increases, the gap between the amount of the data stored and the amount of the data analyzed increases. Such data can be used in productive decision making if appropriate data mining techniques are applied. Data driven precision agriculture aspects, particularly the pest/disease management, require a dynamic crop-weather data. An experiment was conducted in a semi-arid region of India to understand the crop-weather-pest relations using wireless sensory and field-level surveillance data on the groundnut pest Thrips. Various data mining techniques were used to turn the data into useful information/knowledge/relations/trends and correlation of crop-weather-pest continuum. These dynamics obtained from the data mining techniques and trained through mathematical models were validated with corresponding surveillance data. Results obtained from 2009 & 2010 Kharifseasons (monsoon) and 2009-10 & 2010-11 Rabi seasons (post monsoon) data has been used to develop a prediction model. In this work an attempt has been made to develop a viable model for groundnut pest (Thrips) dynamics using the state of the art data mining techniques to understand the hidden correlations (crop-pest–meteorological continuum) and there by development of Multivariate Regression Models which led to development of forewarning system. Keywords: Data Mining, Knowledge Discovery, Wireless Sensor Network, Precision Farming and Pest/Disease Management 1. INTRODUCTION

Agriculture plays a significant role in India’s economy since a sizeable Indian population lives in the rural areas and earns a living either directly or indirectly. However with uncertain crop/weather conditions, farming community is encountering numerous problems in an attempt to maximize crop productivity. One of the chief reasons is that expert/scientific advice related to dynamic crop management aspects does not reach the farming community at appropriate time [20]. Crop losses, particularly oilseed, due to pests and diseases are quite considerable, particularly in the Indian semi-arid conditions [22. Among the oilseed crops, Groundnut (Arachishypogaea L.) extent its large space despite its prone to pests and diseases. Groundnut is an important crop both in subsistence and commercial agriculture in arid and semi-arid regions of the world [11]. Significant crop losses by these pest/diseases have been reported from Australia, India, South Africa, USA, China, etc. [25, 3, 11, 30]

The population dynamics of organisms responsible

for pest and disease outbreaks in crops is influenced by a number of factors such as weather conditions, quantum of food supply, availability of intermediary hosts or carriers,

presence or absence of natural enemies, etc. Weather and climate are very important in determining the precise epidemiology of outbreak of either pests or disease. Critical threshold of the meteorological elements for the incidence, spread and intensification of pests and disease determined in the laboratory condition have little relevance to the field condition. Therefore, they have to be determined and monitored under field conditions through simultaneous observation of micrometeorological parameters and the pertinent data [18].

A number of Thrips pest species are common to

both Asian Countries and the United States. These include four of the major global Thrips pests: Frankliniellaoccidentalis, Scirtothripsdorsalis, Thripspalmi, and Thripstabaci. These particular species have become global pests as they have been transported on agricultural produce [26]. Among these Thrips species, Frankliniellaoccidentalis, Scirtothripsdorsalis, Thripspalmiwere dominant in southern parts of India [21, 17, 35]. All the species of the Thrips cause extensive damage to the groundnut with significant yield loss. It is essential to have knowledge on the relative abundance and




914

their seasonality in relation to peanut crop to advocate appropriate and economic control measures. Besides causing direct damage to the crop, Thrips are known to cause more indirect damage by attacking as vector of viral disease like groundnut Bud Necrosis Virus, [34, 36], Peanut Yellow Spot Virus [9], etc. Among the viral diseases attacking groundnut crop, Bud Necrosis Virus (BNV) caused by tospovirus and is transmitted by Thripspalmiis considered one of the important diseases in India [34,36]. In India, the disease occurs with the incidence ranging from 0-98% [12, 35].

Weather plays an important role on the population

dynamics and distribution of pest and diseases. Temperature, humidity, rainfall, sunshine hours, leaf wetness, wind speed are the chief weather parameters influencing the pest and disease incidence. It is rather difficult to establish a direct cause and effect relationship between any single climatic factor and pest/disease activity, as the effect of these weather elements on the pest/disease is usually confounded [35, 37]. Weather based pest/disease forewarning model have been developed to certain extent [35]. However, a functionally viable model for pest/disease forecast considered to be one of the important components in the integrated pest management strategy. At the present scenario of pest resistance to pesticides and high cost of pesticides, it is mandatory requirement to develop early warning to provide caution to the farmers regarding the occurrence of pest, their peak activity and migration to develop effective and efficient management system. Hence the present investigation has been carried out to discover the hidden correlation and dynamics of pest/disease with weather parameters, which lead to developing a forewarning model.

In the practices of agricultural information, the

agricultural intelligence monitoring system is an important and indispensable link [40]. The advanced sensor technology and the intelligence information processing technology are important method to guarantee correctly and quantitatively to gather agriculture information. With the advent of WSN information acquisition and the processing technology has seeped gradually into agricultural domain by the characteristics such as low power loss, low cost and redundant reliability [14, 41]. Sensor network technology (wired or wireless) is a potential system suitable for collecting the real time data on different parameters pertaining to weather, crop, soil and environment, which in turn helps in developing open solutions for majority of the agricultural processes. The wireless sensors are cheap enough for wide spread deployment in the form of a mesh network and also it offers robust communication through redundant propagation paths. Wireless Sensor Network (WSN) allow faster deployment and installation of various types of sensors as the network provides self-organizing, self-configuring and self-

diagnosing capabilities to the sensor nodes. It is a system comprised of radio frequency transceivers, sensors, microcontrollers and power sources. They are relatively low-cost, consumes low-power, small devices equipped with limited sensing, data processing and wireless communication capabilities, which perfectly suites for precision agriculture where decisions are made at micro-climatic level at right time/place/input [29]. In the present scenario, agricultural data virtually are being harvested along with the crops and are being collected/ stored in databases. As the volume of the data increases, the gap between the amount of the data stored and the amount of the data analyzed increases. Such data can be used in productive decision making if appropriate data mining (DM) techniques are applied. DM allows to extract the most important information from such a vast data and to uncover previously unknown patterns and hidden relationships within the data that may be relevant to current dynamic agricultural problems. With the ever-increasing amount of information about the farms, farmers are not only harvesting in terms of agriculture output but also a large amounts of data. These data should be used for optimization [8, 32]. Numerous advances in science and technology has made it quite essential that farming in the future would adopt techniques that aid better decision making during a crop cycle. Precision farming is an emerging methodology in today's context of agriculture and it definitely holds the key in the future. Researches on utility of macroclimatic data on precision agriculture has been carried out at [18, 33], however, very few research works available with sensory based microclimatic data. There have been a few studies concerning use theof WSN in pest/disease management. AgriSenswas used to test the feasibility of capturing and analyzing data and facilitated global data accessibility from multiple wireless sensor pods to study the efficient irrigation as well disease forecasting for grape vineyard (SPANN, 2009). Prabhakar et al.[18] discussed through a WSN, named COMMON-Sense Net, which monitors several environmental parameters and is deployed in an Indian semi-arid region. ‘U-Agri’ from Centre for Development of Advanced Computing (C-DAC), Hyderabad developed low cost sensor networks which encompass the farm environment and provide macro and micro climate information on groundnut crop for a Decision Support System to groundnut pest Leaf Miner and disease Leaf Spot [33]. WSN has been used for vineyard monitoring that uses image processing [13]. Matese et al.[15] designed and developed WSN with the aim of remote real-time monitoring and collecting of micro-meteorological parameters in a vineyard. Ronald et al.[23] developed crop management and decision support system with WSN. Wang et al.[38] have developed an intensive irrigated agriculture monitoring application, and presents a collection of




915

requirements, constrains and guidelines that serves as a basis for a general sensor network architecture for many such applications. It is essential that an efficient methodology should be capable of forecasting the pest & disease dynamics accurately. Thus, there is a need for development of a viable and functionally realistic model to correlate pest/disease with weather and surveillance data. In the present study, micro-level weather data (Temperature, Humidity and Leaf Wetness) were obtained through wireless Mote based AgriSens distributed sensing system, DM techniques and surveillance data have been used to understand and quantify hidden correlation between crop-pest/disease-weather parameters. Subsequently, one week as well as a cumulative prediction models for Thrips have been developed with which one can develop a Decision Support System (DSS) with multi-season data.

2. MATERIALS AND MEHTODS In order to study the crop-weather-pest/disease interactions, a test bed for WSN experiments was chosen at Agriculture Research Institute (ARI) of Acharya N G Ranga Agricultural University, Hyderabad falling in semi-arid tropic region. The test bed, where a long term experiments are being carried out on groundnut crop, will provide a platform for validation (including comparisons) of proposed model with existing pest/disease model. This work is a part of Indo-Japan initiative to develop a real time decision support system called GeoSense[30, 32] integrating Geo-ICT and WSN for Precision Agriculture. a. Standard Experimenatal Setup A standard field experiment design was laid out in the test bed (Figure 1 and 2). P1 stands for unprotected plot (normally situation like farmers plot) and P2 stands for weather based protection plots. Four different dates of sowing (D1, D2, D3 and D4) were taken in to consideration (Table -1). These different dates will determine the impact of pest and disease incidence in order to observe dynamics in pre (D1) and post normal weeks of sowing (D4). D2 and D3 are normal dates of sowing. Apart from this, to have uniform and unbiased observation, surveillance data has been collected from each plot in randomly selected one square meter area locations of the plot (S1D1, S1D2, S1D3, S1D4, S2D1, S2D2, S2D3, S2D4,………) through flowering to harvesting phenological stages (Figure 2) in three replicated plots (R1, R2 and R3). All these have been carried out for four seasons (two Kharif and two Rabi seasons) b. Surveillance Data Collection Thrips population dynamics (surveillance data) were obtained at every week from flowering to reproductive

pheonological stages, where majority of pest and disease incidences occur, at various locations in the experimental site. The surveillance data has been collected weekly instead of daily as there won’t be any significant visible changes in pest/disease incidences. A total of 48 observations (12 X 4) have been made with respect to different dates of sowing in each season. Along with this, Groundnut crop age (that is at which stage of the crop thepest/disease attack takes place and their dynamics trends) also recorded week- wise to understand the infection dynamics of Thrips. The weeks in a year are mapped into integer values by considering the first week of January as first standard week. Five sticks were placed in each one square meter and named 1, 2, 3, 4 and 5 (Figure 2).Subsequently data has been collectedmanually from the plants adjacent to eachstick for both Kharif (2009 & 2010) and Rabi seasons (2009-10 and 2010-11). The surveillance data like number of leaf-lets that Thrips have punctured/infected/visited, how many plants infected in the one square meter area, date of flowering, date of recording the surveillance data, etc.

Fig 1: Experimental Layout for Groundnut Crop

The plants have been numbered in the form of ID like R1P1D1-1, R1P1D1-2, R1P1D2-1, etc. so as to retrieve their respective data easily from data base. This is the one of




916

the standard design practice that has been observed in the long term experiment in the test bed.

c. Sensory Data collection Sensory data (Temperature, Humidity and Leaf Wetness) were collected from the field by using AgriSens and was transmitted through General Packet Radio Service

(GPRS) to the server (located remotely) for data storing, analysis and mining. The data collected through GPRS technology was stored in an OpenSource data base (PostgreSQL) for further analysis.

(a) (b)

Fig 2: Surveillance data collection from One square meter area (a) Flowering stage (b) Reproductive stage

The deployed WSN system consists of the battery-powered nodes equipped with sensors for continuously monitoring agricultural/weather parameters such as temperature, relative humidity, soil temperature and leaf wetness [29]. Figure 3 shows the schematics of wireless sensor network with agricultural / environment sensors deployed in the field. Each node was able to transmit/receive data packets to/from other nodes every 15 minute over a transmission range of 25 meter. Data collected by the sensors were wirelessly transferred in a multi-hop manner to a base station node (stargate) connected with embedded gateway for data logging. In a WSN, when the transmission range of a sensor node is not sufficient, it uses multi-hop communication to reach the destination node or sink node. This data forwarding mechanism continues till it reaches the sink node.

The base station has a GPRS connectivity through

which it routes data to the GeoSense server setup at Agro-Informatics Lab at CSRE, IIT Bombay and collect all the sensory information.The sensory data coming to the server through GPRS is raw and has been converted in to a usable format in real timethrough appropriate conversion formula(by using open-source server-side scripting language PHP) in the server end.Both raw data and the real-time data have been stored in different database for further analysis and mining.Other related weather data (sunshine hours SH, wind speed WS, rainfall RF and evapotransipiration ET) were obtained from the weather station with in the vicinity of the test bed.

Sensory data was collected during the Kharif (monsoon) seasons of 2009 and 2010 as well as Rabi (post monsoon) season 2009-10 and 2010-11 with multiple sensor nodes, i.e. M1, M2, M3 and M4 for four different dates of sowing (D1, D2, D3 and D4) in four different test plots, respectively in groundnut field (Figure 4).

d. Data Mining and Statistical Models

Various data mining techniques and a few algorithms were used/developed to understand the pest dynamics and the general processing flow is depicted in Figure 5.Raw sensory data obtained from the experimental field is not uniform in its collection. Owing to the climatic conditions or non-function of field sensors or due to network errors, there have been a few breaks in collecting continuous sensory data that may lead to biased outcomes while developing the model.

Expectation–Maximization (EM) algorithm was used to deal with missing data [10]. Relative data from the nearest sensor node was used to fill the missing data with EM algorithm. The data set is provided in daily and weekly means wherever is required. Quality data is accomplished by performing satisfactory data pre-processing methodology such as data selection, data reduction and elimination of null values or other noise values. Though the real-time sensory data were collected at 15 minute interval,all such data were not used for the current experiment.

4 4 3

3

1 1

2

2

5 5




917

For example the temperature was

Table 1: Different Dates of Sowing groundnut crop in test bed

Date of Sowing D1 D2 D3 D4 Kharif 2009 07/07/2009 21/07/2009 07/08/2009 22/08/2009 Rabi 2009- 10 12/10/2009 26/10/2009 10/11/2009 27/11/2009 Kharif 2010 19/06/2010 03/07/2010 19/07/2010 04/08/2010 Rabi 2010-11 14/10/2010 29/10/2010 16/11/2010 04/12/2010

Fig 3: WSN architecture in the experimental site with Agri Sens

Fig 4: AgriSens sensor network deployed in the field

Fig 5: DM Processing Flow for Pest Dynamics




918

taken by computing maximum and minimum values of the respective day. Relative humidity was recorded as RH1 (7: 30 AM) and RH2 (3: 00 PM), which is a standard practice in Indian agriculture system (AgroMet-Cell, 2009). Leaf Wetness (LW) data has been used in the scale of 1 to 10 as Leaf Wetness Index, and the value above 5 has been taken as wet leaf and computed for wetness period [29], which is generally taken as the value that attract pest/disease. Data Mining (DM) techniques were used to understand Thrips dynamics as well as to convert with sensory and weather station based meteorological and other surveillance parameters in Groundnut crop. As Groundnut crop was infested with multiple pests and diseases, multi-level classification modules were developed in the model, which classifies crop pests & diseases based on the severity. Naive Bayes classification with Gaussian distribution [6, 16] was used in the experiment. Bayesian network principle was used to model uncertainty by combining experimental knowledge and observational evidences. Gaussian Naive Bayes (NB) classifier, which is a term in Bayesian statistics dealing with a simple probabilistic classifier based on applying Bayes' theorem with strong (naive) independence, was used to assume the presence (or absence) of a particular feature of a class unrelated to the presence (or absence) of any other feature [16]. Rapid association rule mining was used in association with above classification techniques to find out correlation of multiple weather parameters with respect to Thrips. This phenomenon is to identify signature patterns as well to discover their presence/dependency [1, 4] with other related pest and/or weather parameters. This algorithm helps in discovering effects of pest/disease over the other pest/disease with respect to weather and other related parameters. For example, how the Thrips incidence occurred due to presence/absence of Bund Necrosis Virus (BNV) or vice versa. The outcomes are in the form of correlation index values ranging from -1 to 1.

Regression Mining is a data mining (machine

learning) technique used for developing multivariate equation for the training dataset. It is a data mining function that predicts a number. For example, a regression model could be used to predict the value (infection index) of a pest/disease based on weather parameters, crop age, and other factors like carrier pest/disease. This technique provide basis to understand how the typical value of the dependent variable changes with varied independent variables, while the other independent variables are held fixed, which helps to estimate the conditional expectation of the dependent variable. Less commonly, the focus is on a quintile or other parameters of the conditional distribution of the dependent variable given the independent variables. In all cases, the estimation target is a function of the independent variables, called the regression function. In this analysis, it is also of interest to characterize the variation of the dependent

variable around the regression function, which can be described by a probability distribution.

When developing regression function for modeling experimental data with more than one independent argument, multivariate regression could be used. A multivariate regression is the use of more than one input variable for the pertinent of complex models. When multiple predictors are used, the regression line cannot be visualized in two-dimensional space. However, the line can be computed by expanding the equation for single-predictor linear regression to include the parameters for each of the predictors [31]. A sample equation is:

y = w1x1 + w2x2 + ..... wnxn-1 + e

(1)

Where, ‘y’ is dependent variable (e.g. Thrips) and ‘x’is independent variable (e.g. weather parameters and ‘e’ is the error functions. In a multivariate linear regression, the regression parameters are often referred to as coefficients (‘w1 … wn’ in the above equation). When building a multivariate linear regression model, the algorithm computes a coefficient for each of the predictors used by the model. The coefficient is a measure of the impact of the predictor ‘x’ (independent variable such as morning hours relative humidity RH1, maximum temperature Tmax, etc.) on the target ‘y’ (dependent variable such as Thrips).

A linear regression model can be used for

prediction of forecasting to fit a predictive model to an observed data set of‘x’ and ‘y’ values. Let D denote a data set that contains ‘n’ observations,

D = (xi, yi), where, i = 1 to n.

(2)

Each xi correspond to the set of attributes of the ‘ith’observation (can be called as independent variables) and yi correspond to the target (dependent variables). Mining the regression is the task of learning a target function ‘f’ that maps each attribute set ‘x’ into a continuous valued output ‘y’. The goal of Regression Mining in this model is to find a target function that can fit the input data with minimum error. The error function for a regression mining task can be expressed in terms of sum of absolute error (AE) or square error (SE):

AE = Absolute Error =

∑ −i

ii xfy |)(| (3)

SE = Squared Error =

∑ −i

ii xfy 2|))((| (4)




919

In the present study, the following equation can be computed with single dependent (target) /independent (predictor) variables to make a model with reduced error:

f(x) = w1x + w0 or f(y) = w1y1+w0

(5)

where, w0 and w1 are regression coefficients. This approach is an attempt to apply a method of least squares, which will help to find the parameters (w0, w1) that minimize the sum of the squared error (SSE) or residual sum of square errors.

SSE = ∑ −n

iii xfy 2)]([ = ∑ −−

n

iii wxwy 2

01 ][ =

0

(6)

The optimization problem can be solved by taking the partial derivatives of E (the error term) with respect to parameters w0 and w1, setting them to zero, and solving corresponding system of equations.

0wE

∂∂

= -

2∑ =−−N

iii wxwy 0][ 01

1wE

∂∂

= -

2

∑ =−−N

iiii xwxwy 0][ 01

(7)

These equations can be summarized by the followingmatrix equations, which can be called normal equation:

∑∑∑

i ii i

i i

xxxN

2

1wwo =

∑∑

i ii

i i

yxy

(8)

Thus, the linear model that results in the minimum squared can be written as

][)( xxyxfxx

xy −+=σσ

(9)

Where ∑∑ ==i ii i nyynxx /,/ and

∑ −−=i

iixy yyxx ))((σ

∑ −=i

ixx xx 2)(σ and ∑ −=i

iyy yy 2)(σ

Where σ is Standard Deviation (measures of dispersions),

x and y is the arithmetic mean of ‘x’ and ‘y’ respectively. Multiple Correlations Coefficient (R2) is a

statistical measure to suggest how well a regression line approximates real data points. It is a descriptive measure between zero and one, indicating how good one term is at predicting another. R2 is used as an indicator of the reliability of a relationship identified by regression analysis. A regression task begins with a data set in which the target values are known. For example, a regression model that predicts the pest/disease infection index values can be developed based on observed data (sensory, meteorological data) over a period of time. The regression algorithm estimates the value of the target as a function of the predictors for each case of the test data. The relationships between predictors (e.g. Tmax, ET, RH1, etc.) and target (Thrips) is summarized in a multivariate regression equation, which could then be applied to a different data set in which the target values are unknown. The above technique may help in generating a prediction model.

Following is a multivariate regression equations developed by using XLminer[39] for Thrips.

Thrips- YTH = -4.84 + 1.23 *Tmax- 0.78

*Tmin-0.11 * RH1+ 0.25 * RH2 - 5.38 * RF - 2.05 * SH - 0.59 * WS - 0.28 * ET + 1.56E-02 * AC

(10)

Where, Tmax = Maximum Temperature in 0C Tmin = Minimum Temperature in 0C RH1 = Relative humidity in (%) recorded at 7: 30 am. RH2 = Relative humidity in (%)recorded at 3:00 pm. RF = Rainfall(mm/day) RD = Rainy days in a week SH = Sunshine Hours, WS = Wind Speed(km/hour) ET = Evapotranspiration (mm/day) AC = Age of crop. LW = Leaf Wetness in hours

This above multivariate prediction model can be

used by taking historical data and was used for one week




920

prediction. For example, one week prediction of Kharif 2010 season Thrips incidence level has been predicted for one week prediction by using Kharif 2009 data and similarly for Rabi 2010-11 seasons one week prediction was carried out with 2009-10 Rabi historical parameters.

In addition, Complex Polynomial Cumulative

model [2] was also adopted and modified by including various aspects viz. maximum pest population, time of first appearance, time of maximum pest population/disease Degree Days, Previous Year Record, Correlation with severity as well as Life Cycle, Season, Weather Parameter, Stage of crop, Incidence at flowering stage, Growing Other Pest/Disease, Previous Season Crop for pest/disease forecasting:

eZbZaaYj

jiijiij

p

iijji

p

i+++= ∑∑ ∑∑

== ≠=

1

0''

1

0 110

(11)

iw

n

nw

jiwij XrZ ∑

=

=2

1and

wiiw

n

nw

jwiijii XXrZ '

2

1'' ∑

=

=

Where, Zii’s and Zij’s are the independent variables which are functions of the basic weather variables like maximum temperature, relative humidity, leaf wetness, etc. Y: variable to forecast, Xiw = Value of ith weather variable in wth week, riw = Correlation coefficient between Y and ith weather variable in wth week, rii’w = Correlation coefficient between Y and product of Xi and Xii’ in wth week, n1= initial incidence, n2 = fist Peak population week. 3. RESULTS AND DISCUSSIONS

Thrips pest-crop-weather interactions were carried out with two Kharif seasons (2009 & 2010) and two Rabi seasons (2009-10 & 2010-11) data. Correlations studies (of Thrips –weather-crop infestation were carried out and their hidden correlation were discovered and quantified using DM techniques in the test bed.

The correlation values (both positive & negative) of predictor (e.g. Tmax, RH1, ET, etc.) versus target (e.g. Thrips infection index) were obtained from various datasets (sensory, weather-station and surveillance) during flowering to harvesting stages and are depicted in Figure 6 and Table 2 for all the four seasons. A correlation Index matrix was obtained by using Rapid Association Rule mining algorithm, in which correlation index of pest with weather parameters were quantified in the range of -1 to +1.

Fig 6: Correlation Index Values for Thrips with weather and crop age

Correlation indexing greater than 0.5 in the scale (-1 to 1) has been considered as strong +ve, whereas -0.5 and more for strong negative correlation. It was found that ET has strong positive correlation and AC has strong negative correlation with Thrips. In case of Thrips, ET, and AC found to be strongly correlated. Tmax, Tmin, RH1 and RH2, SH are also found to be positively correlated and others such as RD, RF, LW and WS are negatively correlated. Higher the wind speed, Thrips incidence will be less as high wind speed will blow away Thrips from the plot. Similarly, higher the rain fall Thrips infection will be less as due to the fact that Thrips will be washed away from the leaf or flower. Age of the crop found to be negartively correlated that means higher the crop age lower the pest incidence.

Table 2: Correlation Index Matrix from Kharif 2009 to Rabi 2010-11 for Thrips

Thrips Weather Kharif Rabi Kharif Rabi

Parameters 2009 2009-10 2010 2010-11 Tmax 0.206 0.186 0.316 0.298 Tmin 0.079 0.095 0.149 0.127 RH1 0.138 0.184 0.298 0.219 RH2 0.388 0.235 0.459 0.284 RF -0.106 -0.231 -0.136 -0.203 RD -0.098 -0.25 -0.158 -0.225 SH 0.165 0.107 0.278 0.157 WS -0.286 -0.239 -0.356 -0.273 ET 0.575 0.284 0.631 0.501 LW -0.178 -0.137 -0.359 -0.191 AC -0.596 -0.613 -0.651 -0.545

These (outcomes) reported are from the

experimental plot of P1D1, (unprotected plot with pre-normal date of sowing), as this situation is generally found in the farmers’ fields, which attracts pests/diseases more than the normal (D2& D3) and post normal season (D4) data. Table 3 shows overall interpretation of correlation (Thrips/Weather/Crop Age) values with negligible, moderate and strong level drawn from all four seasons. Based on concept of infection index with respect to pest/disease risk




921

model [24, 5] correlation index greater than 0.5 in the scale of -1 to +1 has been considered as strong +ve, whereas -0.5 to more for strong negative correlation.

Table 3: Overall Interpretation of Correlation index values Pest/Weather/Crop Age

Correlation

value (- 1 to +1)

Correlation Levels

(Such as Thrips with RH1)

Variables (Thrips )

0.0 to 0.1 1.0 (+ve)

Negligible or

No correlation

Tmin RD RF 0.0 to - 0.1

(-ve) > 0.1 to 0.5 Moderate

(+ve) Tmax,

RH2, SH < - 0.1 to - 0.5

Moderate (-ve)

WS RH1 LW

> 0.5 to 1 (+ve)

Strong (Good Correlation)

ET

< -0.5 to – 1 (-ve)

Strong (Good Correlation)

AC

Thrips dynamics in Kharif season 2009: Using

the developed mathematical model, with standard surveillance data in the ground level studies and weather parameters from sensory & weather station, pest/disease incidence analysis for Kharif 2009 has been carried out and presented both in tabular (Table 4) and graphical forms (Figure 7). In Kharif 2009 sensory data was collected from D1 sowing plot. A comparative study has been carried out and results obtained were depicted in table 4 and Figure 7.

Fig 7: Thrips incidence in Kharif 2009 for D1 sowing

From these, it has been observed that the multivariate regression (MVR) model has attained almost close to the ground level studies (GLS). A maximum of 6.43 % incidence has been observed during last week of August (i.e., on 28 August 2009). Thrips incidence found to be in rapid increase manner for the first four week staring from flowering stage and then gradually decreasing trends were observed.

Table 4: Thirps in D1 Sowing Kharif 2009

Date GLS MVR model 31/7/09 2.01 2.44 7/8/09 1.81 2.12 19/8/09 4.23 5.06 28/8/09 5.83 6.43 8/9/09 3.61 4.55 18/9/09 3.95 4.29 30/9/09 3.00 4.05 8/10/09 3.21 3.59

14/10/09 1.02 1.51 21/10/09 0.00 0.67

Thrips dynamics in Rabi season 2009-10: In Rabi 2009-10 season, sensory data was collected from D2, D3 and D4 sowing plot. By using Regression mining model, the outcomes found in Rabi 2009-10 are shown in graphical from in comparison with ground level surveillance data (Figure 8 to Figure 10) and the quantitative incidence were depicted in Table 5 to Table 7 for D2, D3 and D4 sowing dates, respectively.

It was found that a maximum of 22.54 % incidence

has been observed in the beginning (Flowering) stage of D3 sowing and also observed a decreasing trend subsequently. However, in case of D4 sowing, the Thrips incidence increases from the flowering stage (15.27%) to a peak of 18.07% and then decrease gradually with respect to increase of crop age. A very peculiar thing was notice in D2 sowing, i.e. Thrips incidence was decreased to 5.75 % in reproductive stage (starting from 15.17% at the flowering stage) and at later stage once again increased to 11.63% and gradually decreases. This could be an unexpected suitable weather conditions. Among the thee sowing dates in Rabi 2009-10 seasons D3 sowing date found to be more prone to due to dry weather which is favorable for Thrips attack.

Fig 8: Thrips incidence in Rabi 2009-10 for D2 sowing




922

Table 5: Thrips in D2 Sowing Rabi 2009-10

Date GLS MVR model 25/11/2009 14.39 15.17 4/12/2009 11.85 13.58

10/12/2009 9.66 10.87 18/12/2009 9.04 10.21 27/12/2009 5.03 6.75

7/1/2010 5.55 5.81 16/1/2010 8.39 7.54 11/1/2010 11.63 9.46 30/1/2010 6.20 7.13 3/2/2010 4.96 5.37

10/2/2010 4.44 4.77 18/2/2010 1.10 1.76

Fig 9: Thrips incidence in Rabi 2009-10 for D3 sowing


Date GLS MVR model 19/12/2009 22.54 21.21 27/12/2009 18.99 20.17

7/1/2010 17.00 18.53 12/1/2010 12.02 14.03 19/1/2010 9.87 10.54 27/1/2010 10.89 11.27

3/2/2010 6.53 7.54 10/2/2010 6.67 7.12 18/2/2010 7.94 5.37 26/2/2010 5.43 3.51

4/3/2010 2.23 1.59 9/3/2010 1.42 1.13

Fig 10: Thrips incidence in Rabi 2009-10, D4 sowing


Date GLS MVR model 7/1/2010 14.70 15.27

12/1/2010 15.54 17.29 19/1/2010 14.06 16.68 27/1/2010 15.88 18.07 3/2/2010 11.47 14.73

10/2/2010 9.63 10.85 18/2/2010 9.57 10.23 26/2/2010 6.28 7.11 4/3/2010 5.18 3.49 9/3/2010 0.36 1.71

20/3/2010 0.25 1.34 27/3/2010 0.19 0.83

Fig 11: Thrips incidence in Kharif 2010, D1 sowing

Table 8: Thrips in D1 Sowing Kharif 2010


14/7/2010 2.71 3.98 24/7/2010 11.02 14.21 30/7/2010 12.60 16.47 10/8/2010 16.09 17.01 18/8/2010 15.40 14.27 26/8/2010 6.27 8.37




923

2/9/2010 6.16 7.39 7/9/2010 10.82 8.88

16/9/2010 4.50 5.89 Thrips dynamics in Kharif 2010: Thrips

incidence in groundnut has been observed round the year with prevailing weather conditions. By using Regression mining model, following are the outcomes found in Kharif 2010 as shown in graphical from in comparison with ground level surveillance data (Figure 11 to 14). The presence of Thrips incidence is shown Figure 15.



Date GLS MVR Model 24/7/2010 3.67 2.45 30/7/2010 6.46 6.18 11/8/2010 6.39 8.78 17/8/2010 10.61 11.23 26/8/2010 10.56 12.01

4/9/2010 12.54 14.78 8/9/2010 15.48 18.29

16/9/2010 6.58 8.21 23/9/2010 1.78 5.35 30/9/2010 0.28 1.21



Date GLS MVR model

10/8/2010 8.92 10.23 17/8/2010 15.10 18.11 25/8/2010 8.67 14.84 31/8/2010 8.05 11.89 7/9/2010 11.84 11.29

13/9/2010 9.14 10.45 21/9/2010 4.99 6.73 30/9/2010 2.25 4.32 5/10/2010 1.90 2.02

13/10/2010 2.04 1.89



Date GLS MVR Model 25/8/2010 7.08 10.89 31/8/2010 10.81 15.48 6/9/2010 5.49 10.14

13/9/2010 4.51 7.56 20/9/2010 8.42 7.89 29/9/2010 6.63 5.78 5/10/2010 3.81 3.26

13/10/2010 2.88 2.57 21/10/2010 1.97 1.19

Fig 15:Thips incidence as seen in the test bed in Kharif 2010




924

Perusal of the Figures 11-14 indicate that the developed Regression model follows almost similar pattern in all the four dates of sowing for Thrips incidence with R2 0.897, 0.913, 0.937 and 0.941 for D1, D2, D3 and D4, respectively. The ET value was high in index (0.516) in previous week of 24 July 2010 (for D1 sowing) and its impact on Thrips has been seen in later for two or three weeks. Similar themes trends were also observer in the other dates of sowing as increase in ET has a significant impact in Thrips apart from other weather parameters. It has been found that initial few weeks from the growing stage to reproductive stage, Thrips damage has been obtained as high as 18.29 % in D2 date of sowing (Figure 12). Thrips incidence has gradually increased for few initial weeks and later decreases. However, in D3 and D4 sowing high incidences have been seen in second week of flowering stage and it may be due to the favorable weather condition for Thrips incidence. There is another peak seen in later stage of crop in D1 sowing. However, as the plants already leading to the harvesting stage, its affect may be very nominal. Thrips dynamics in Rabi 2010-11: Thrips incidence was seen more in Rabi 2010-11 for all dates and is as high as 23% in D3 sowing date. Figure 20 shows the supporting photograph/evidence for the presence of Thrips in the test bed during the season. As a general trend, Thrips incidence increases with increase of crop age and later from the reproductive stage onwards it decreases and as it damages the young leafs of the plant. This trend has been seen in D2, D3 and D4 sowing. However, in case of D1 sowing, the incidence levels were fluctuating. This is because the rain fall in between the peak and presence of wet weather the consequence had been seen in the next week or two weeks. The outcome of regression model matches almost with the ground level data in all dates of sowing except D1. There is a variation of 4% to 6% and need to be reinvestigated with many experiments. The R2 values forD1, D2, D3 and D4 were found to be 0.759, 0.897, 0.966 and 0.937, respectively. The initial incidence at flowering stage found to be more in D4 than D1, as it could be due favorable weather condition for Thrips.


Table 12: Thrips D1 Sowing Rabi 2010-11


12/11/2010 15.98 14.47 19/11/2010 6.67 10.23 26/11/2010 6.34 9.11 5/12/2010 7.32 12.45

12/12/2010 10.51 13.78 25/12/2010 8.22 10.31

2/1/2011 7.36 8.59 11/1/2011 10.04 11.78 22/1/2011 9.98 12.56 30/1/2011 3.85 5.14 9/2/2011 3.01 4.12

Thrips life cycle begins when adult females lay eggs in the leaves of host plants. These are preferably older plants.T. palmi ( during 150 C to 320 C) clutches average 50 eggs and those of F. occidentalis(during 150 C to 300 C) are about 200 eggs in size [3, 34]. The larvae feed in groups on the leaves (T. palmi) or flowers (F. occidentalis) of the host plant. They pass through four larval stages (instars) with the first two being most active. The third and fourth lead to pupation and the appearance of the adults. The adults feed on younger plants causing physical damage and also introduce viruses during their feeding. The length of the life cycle is very dependent upon temperatures.



Date GLS MVR model

19/11/2010 12.62 8.56 26/11/2010 7.01 7.89 5/12/2010 6.17 7.14

12/12/2010 10.44 11.47 24/12/2010 15.39 16.27

2/1/2011 17.09 18.19 11/1/2011 22.16 24.11 23/1/2011 19.21 17.51 30/1/2011 18.67 16.13 9/2/2011 13.69 12.45

17/2/2011 12.45 11.37 25/2/2011 11.27 10.21 4/3/2011 5.23 4.45




925



Date GLS MVR model 24/12/2010 8.95 10.45 13/01/2011 16.21 15.89 24/01/2011 23.02 22.12 30/01/2011 19.92 18.11 12/2/2011 21.11 19.87 20/2/2011 20.13 18.89 27/2/2011 16.81 15.11 5/3/2011 11.77 10.71

13/3/2011 6.35 6.28 20/3/2011 5.77 5.43

Fig 19: Observed Thrips incidence in Rabi 2010-11

for D4 sowing


Date GLS MVR model 13/01/2011 16.52 14.27 25/01/2011 11.81 13.59 30/01/2011 21.24 20.78 12/2/2011 22.87 23.44 20/2/2011 23.51 22.45 27/2/2011 18.24 16.58 5/3/2011 15.68 14.98

13/3/2011 11.64 11.23 21/3/2011 9.31 8.71

Fig 20: Thips incidence as seen in the test bed in Rabi 2010-11

PREDICTIONS

Pest/disease infestation in crops is highly influenced by meteorological factors. The weather based modeling for early warning of pest/disease infestation may provide appropriate tool for investigating and predicting pest/disease status. With correlation studies revealing the crop-weather-pest relationships/ interactions, there is a possibility of developing an early warning models (Cumulative and non-cumulative) on pest/disease infestations. Prediction computations have been carried out and presented for Kharif 2010 (Figure 21and Table 16) and Rabi 2010-11 (Figure 22 and Table 17) for Thrips for the plot D2 date of sowing which is a normal sowing date for farmers. Moreover to predict one week prediction model, previous year data of the same week, same date of sowing and same season has been used. These computations have been carried out for near to equal to the peak period only as rest of the prediction has a less pest/disease management significance if the crop is already infected during these peak stages. Two colour schemes were drawn as warnings (for Thrips management at that week) marked as yellow strip and near red as danger (leading to cause maximum damage) in all the Tables. The green broken line shown in the graph indicates the prediction outcome of one week prediction strategy and the pink broken line for the cumulative strategy.

Fig 21: Thrips Incidence Prediction for Kharif 2010




926

Table 16: Thrips incidence Prediction in Kharif 2010 with

D2 Sowing

Date GLS MVR model

Prediction 1wk

Prediction CWK

24/7/2010 3.67 2.45 2.45 2.45 30/7/2010 6.46 6.18 5.92 5.87 11/8/2010 6.39 8.78 8.39 7.78 17/8/2010 10.61 11.23 11.07 10.86 26/8/2010 10.56 12.01 12.59 11.89 4/9/2010 12.54 14.78 15.34 13.05 8/9/2010 15.48 18.29 19.22 17.51

16/9/2010 6.58 8.21 23/9/2010 1.78 5.35 30/9/2010 0.28 1.21

In figure 21 and 22, MVR model is the graph

obtained by empirical computation from the above mentioned weather parameters, whereas test bed data graph is the obtained from the ground level surveillance data. From the Figure 21 and 22 it has been found that MRV model appears to be close to the ground based observations. Thrips pest incidence increases with increase in the age of crop up to the reproductive stage and then decreases gradually with age of crop. Rabi season found to be more Thrips incidences (24.11%) in comparison with Kharif season (18.29%).

It has been observed (both in Kharif 2010 and Rabi

2010-11) that one week (1wk) prediction is very closer to the regression/empirical model (MVR), whereas the cumulative (CWK) method is probably a preferable prediction strategy as it is closer to the ground level data. These interactions indicate that if Thrips pest are controlled on 17/8/2010, there is a possibility to counter even the dependable BNV disease and the yield losses as far as D2 is concerned. Similarly, in case of Rabi 2010-11 season, the first control measure to be taken for Thrips should be on 12/12/2011 (marked in yellow colour in Table 17) failing which the incidence increases rapidly and after 2/1/2011 (two weeks marked red in colour in Table 17) applying control measure has no significance as the crop has already severely infected and will not save the crop.

Fig 22: Thrips Incidence Prediction for Rabi 2010-11

For Kharif 2010 season, the CWK value has been

observed to be in the range of 2 to 5% increased value with respect to ground level data. However, in case of 1wk prediction approach it has been found 5 to 10% increase in prediction value as compared to the ground level data. Thus, the CWK model found to be more accurate prediction method. Similar trends were also observed during 2010-11 Rabi season and care should be taken after the appearance of first peak incidence of Thrips.

Table 17: Thrips incidence Prediction in Kharif 2010 with

D2 Sowing

Date GLS MVR model

Prediction 1wk

Prediction CWK

19/11/2010 12.62 8.56 8.56 8.56 26/11/2010 7.01 7.89 7.51 7.39 5/12/2010 6.17 7.14 6.89 6.55

12/12/2010 10.44 11.47 11.13 10.88 24/12/2010 15.39 16.27 16.11 15.89

2/1/2011 17.09 18.19 17.91 17.53 11/1/2011 22.16 24.11 23.87 23.12 23/1/2011 19.21 17.51 30/1/2011 18.67 16.13 9/2/2011 13.69 12.45

17/2/2011 12.45 11.37 25/2/2011 11.27 10.21 4/3/2011 5.23 4.45

It has been observed that there were minimum

deviations between actual and predicted (Cumulative prediction) values of Thrips population during certain months, indicating the feasibility of predicting population occurrence using the prevailing weather parameters.

Table 18: Thrips incidence in the year 2010-11 Sowing Date Kharif 2010 Rabi 2010-11 D1 16.09 15.98 D2 15.48 19.21 D3 15.11 21.11 D4 10.81 23.51

Fig 22: Thrips Incidence in 2010-11




927

Also, It has been observed that, during 2010-11, the data obtained from the unprotected groundnut field with different dates of sowing during Kharifand Rabi seasons indicated that highest mean incidence of Thrips was observed during Rabi season in D2, D3 and D4 date of sowing compared to Kharif season, however in Kharif 2010 with D1 date of sowing had highest Thrips incidence (Figure 23 and Table 18). Among all D4 date of sowing in Rabi has been seen more prone to Thrips incidence and there is increased trend of incidence with respect to D1, D2, D3 and D4. D4 date of sowing in Kharif 2010 season has been seen as less prone to infection. 4. CONCLUSIONS An attempt has been made to understand the hidden relationships between most prevailed pest (Thrips) and weather parameters in Groundnut crop. WSN was established in the semi-arid tropic test bed to obtain real-time weather parameters (Temperature, Humidity and Leaf wetness) at micro-climatic level and a few related weather parameters were taken from the weather station in close proximity to the test bed. The crop-weather-pest dynamics and hidden relations were obtained and quantified using different DM techniques. The statistical approach together with regression mining based correlations helped in developing multivariate regression model that has been used to develop an empirical prediction model (non-cumulative) to issue the forecast for population buildup, initiation & severity of Thrips. Apart from this, a cumulative prediction model has also been developed (which found to be more accurate than the non-cumulative one) and tested using two seasons data. This will help to take strategic ameliorative decisions so as to save the crop from Thrips affects and improve the crop yields. REFERENCES [1] Agrawal, R., Imielinski, T., Swami, A.N.(1993).

Mining Association Rules between Sets of Items in Large Databases. SIGMOD. June 1993, 22(2): pp. 207-16.

[2] Agrawal, R. and Mehta S.C.(2007). Weather Based Forecasting of Crop Yields, Pests and Diseases - IASRI Models, Journal of Indian Society of Agricultural Statistics, 61(2): pp. 255-263.

[3] AgroMet-Cell: Entomology Work 2008-09 AgroMet-Cell, Agriculture Research Institute, ANGRAU, Hyderabad, Annual Report, pp. 57-69.

[4] Das, A., Ng , W. K. and Woon Y. K. (2001). Rapid association rule mining, CIKM '01, Proceedings of the tenth international conference on Information and

knowledge management (CKIM'01), November 5-10, 2001, Atlanta, GA, USA, pp. 474-481.

[5] Das, I, Naveen, C.P.R.G., Yadav S S., Kodilkar, A A., Shah, N.G., Merchant, S.N. and Desai, U.B. (2009) WSN Monitoring of Weather and Crop Parameters for Possible Disease Risk Evaluation for Grape Farms - Sula Vineyards, A Case Study, Proceedings of Conference on Geospatial Technologies and Applications, Geomatrix- 2009, IIT Bombay, Mumbai, 27-29, 2009.

[6] George, H. J. and Pat, L. (1995). Estimating Continuous Distributions in Bayesian Classifiers. Proceedings of the Eleventh Conference on Uncertainty in Artificial Intelligence, Morgan Kaufmann, San Mateo. Pp.338-345.

[7] George, N. A. (2005). Plant Pathology. Elsevier Academic Press. 5th ed., ISBN 978-0120445653.

[8] Georg, R., Rudolf, K., Martin, S., Peter, W. (2008).Data Mining with Neural Networks for Wheat Yield Prediction. P. Perner (ed.), Springer LNAI, Vol. 5077, pp. 47-56.

[9] Gopal, K., Reddy K.M., ReddyD.V.R. and MuniyappaV.(2010). Transmission of peanut yellow spot virus (PYSV) by Thrips, Scirtothripsdorsalis Hood in groundnut, Archives of Phytopathology and Plant Protection, Taylor & Francis, Vol. 43, No. 5, 20 March 2010, 421–429.

[10] Gupta, M. R. and Chen, Y. (2010). Theory and Use of the EM Algorithm, Foundations and Trends in Signal Processing, Vol. 4, No. 3 (2010) 223–296.

[11] Izge, A.U., Mohammed, Z.H. and Goni, A. (2007). Levels of variability in groundnut (Arachishypogaea L.) to cercospora leaf spot disease – implication for selection, African Journal of Agricultural Research Vol. 2 (4), pp. 182-186, April 2007.

[12] Kendre, M.S., Patange, N.R., Neharkar, P.S., and Telang, S.M. (2010). Occurrence of Thrips plami, a vector of bud necrosis disease of groundnut in Marathwada region, Journal of Soils and Crops, 10:2, pp. 226-230.

[13] Lloret, J., Bosch, I., Sendra, S., Serrano A.(2011). A wireless sensor network for vineyard monitoring that uses image processing. Sensors (Basel). 2011; 11(6):6165-96. Epub 2011 Jun 7.

[14] Liu, H., Liao, G., Yang, F. (2008). Application of wireless sensor network in agriculture producing.




928

Journal of Agricultural Network Information, 2008, (11):16 – 18.

[15] Matese, A., Gennaro, S.F. Di, Zaldei, A., Genesio, L., Vaccari, F.P.(2009). A wireless sensor network for precision viticulture: The NAV system, Computers and Electronics in Agriculture, Volume 69, Issue 1, November 2009, Pages 51-58.

[16] Mitchell, T.M.(2010). Generatvie and Discruninative Classifier: Naive Bayes and Logistic Regression, In Machine Learning, McGraw Hill (Draft of January 19, 2010), pp. 1-17.

[17] Nandagopal, V. and Vasantha, S. (1991). Influence of plant growth regulators on the population of ThripsCaliothripsindicus and damage in groundnut. Entomon 16: 87-89

[18] Prabhakar, T.V., Jamadagni, H. S., Sahua A, Prasada R.V. (2010), "Lessons from the Sparse Sensor Network Deployment in Rural India", 11th International Conference on Distributed Computing and Networking (ICDCN’10).Springer LNCS, Vol- 5935, pp. 104-115.

[19] Prasad, R.D.V.J. (2009).Thrips Transmitted Viral Diseases in Groundnut and Management, Proceeding of workshop Changing Insect Pest Scenario and Management strategies in Different Crops, Hydearbad, pp. 89-97.

[20] Ratnam, B.V., Reddy, Krishna, P., Reddy, G.S. (2006).eSagu-An IT based personalized agricultural extension system prototype. International Journal of Education and Development using Information and Communication Technology, Vol. 2, Issue 1, pp. 79-94 (2006).

[21] RangaRao, G.V. and Wightman, J.A. (1993). Insect pest problems in groundnut and their management. In Proceedings of the National Seminar on Changing Scenario in Pests and Pest Management in India, organized by the Plant Protection Association of India, Jan 31-1 Feb 1992 at CPPTI, Rajendranagar, Hyderabad, India. Pp. 70-77.

[22] Reddy, D.D.R and Ravindra, B. (2008). Groundnut Insect Pests, Diseases, Nutritional Disorders, Acharya N.G. Ranga Agricultural University, Information Bulletin No.4, pp. 20-36. (2008).

[23] Rolando, A., Cárdenas, T. M. G., Lugo, I. and Antonio, G. M.J. (2010). Better Crop Management with Decision Support Systems Based on Wireless Sensor Networks, 7th International Conference on Electrical Engineering, Computing Science and Automatic

Control (CCE 2010), Tuxtla Gutiérrez, Chiapas, México. September 8-10, 2010.

[24] Royle, D.J., Burr, D.J. (2008). The Place of Multiple Regression Analysis in Modern Approaches to Disease Control, EPPO Bulletin, 9(3), 155-163.

[25] Singh, F. and Oswalt, D.L.(1992). Major Diseases of Groundnut, ICRISAT-skill development series No.6, pp. 50-36.

[26] Stuart, R. R., Yu-lin, G., Zhong-ren L. (2011).Thrips: Pests of Concern to China and the United States, Sciences in China, Volume 10, Issue 6, June 2011, Pages 867-892, ISSN 1671-2927.

[27] Shekh, A.M., Patel, H.R. and Pandey, V.(1999). Predicting the outbreak and spread of pests and diseases using weather parameters. Dept. of Agri. Meteorology, Gujarat Agriculture University Anand, Annual Research Report, Vol.2, pp. 249-255.

[28] Shining, L., Jin, C.,Zhigang, L. (2011). "Wireless Sensor Network for Precise Agriculture Monitoring," Intelligent Computation Technology and Automation (ICICTA), 2011 International Conference on , vol.1, no., pp.307-310, 28-29 March 2011.

[29] SPANN Lab (2009). SPANN Lab. WSN project report, Electrical Engineering Department, IIT Bombay, Available at,(Accessed on 23 March 2009) Http://www.ee.iitb.ac.in/spann/agrisens/.

[30] Sudharsan, D., Adinarayana, J., Tripathy, A. K., Desai, U. B., Merchant, S. N., Ninomiya S.,Hirafuji,M., Kiura, T. and Tanaka, K.(2011). GeoSense: Geo-ICT and Wireless Sensor Network based Dynamic Real-time System for Precision Agriculture (2011); Proceeding of CIGR, Conference CIGR2011, September 19-23, 2011, Tokyo, Japan.

[31] Tan, P.N., Steinbach, M., Kumar, V.(2006). Data Mining and Regression, Pearson Addison Wesley (edn.), pp. 729-733 (2006).

[32] Tripathy, A.K., Adinarayana, J., Sudharsan, D., Merchant, S.N., Desai, U.B., Vijayalakshmi, K., Reddy, D.R., Sreenivas, G., Ninomiya, S., Hirafuji, M., Kiura, T., Tanaka, K.; Data Mining & Wireless Sensor Network for Agriculture Pest/Disease Prediction (2011); 2011 World Congress on Information and Communication Technologies (WICT 2011), 11-14 December 2011, Mumbai, India. IEEE Catalog: CFP1168R-CDR, ISBN: 978-1-4673-0125-1, pp. 1233-1238.




929

[33] U-Agri (2008), Ubiquitous Agriculture, CDAC, Hyderabad (Project Report accessed on 21 February 2009) http://www.ubicomp.in/uagri

[34] Vijayalakshmi, K. (1994). Transmission and ecology of ThripspalmiKarny, the vector of peanut bud necrosis virus. Ph.D. thesis, Andhra Pradesh Agriculture University, Rajendranagar, Hyderabad, India. pp: 99.

[35] Vijayalakshmi, K., Reddy,Raji D., Varma, N.R.G. and Pranuthi, G. (2009).Weather based pest and disease forwarning models in groundnut in the context of climate change, Workshop Proceedings: Impact of Climate Change on Agriculture, ISPRS Archives XXXVIII-8/W3: pp. 48-50.

[36] Whitfield, A.E., Ullman, D.E., German, T.L. (2005).TospovirusThrips interactions. Annual Review of Phytopathology, 43, 459-489.

[37] Waiganjo, M.M., Gitonga, L.M. and Mueke, J.M. (2008). Effects of weather on onion Thrips population

dynamics and its Implication on the Thrips pest management Afr. J. Hort. Sci. (2008) 1:82-90.

[38] Wang, Y., Huang, L., Wu, J. and Xu, H. (2007). "Wireless Sensor Networks for Intensive Irrigated Agriculture", 4th IEEE Consumer Communications and Networking Conference, 2007. CCNC 2007, pp.197-201, Jan. 2007.

[39] XLminer- a Cytel Software, http://www.resample.com/xlminer/index.shtml (Accessed on 17 September 2009). [40] Yuanguai, L., Haixia, X., Wei, Y. and Xingwu, Z.

(2009). Application of Zigbee wireless sensor network in Precision agriculture [J]. Journal of Qiongzhou University, 2009, (5):32-34, 37.

[41] Yuanguai L. (2010). An Intelligent Monitoring System for Agriculture Based on Zigbee Wireless Sensor Network. 2010 The International Conference on Electrical Engineering and Automatic Control (ICEEAC 2010), 2010, vol.7: pp. 424-428

journal of computing::data mining and wireless sensor...

Documents