Available online at www.sciencedirect.

com

Procedia Computer Science 5 (2011) 106114

The 2nd International Conference on Ambient Systems, Networks and Technologies

Wireless Sensor Network Based Wildre Hazard Prediction System Modeling

Hakilo Sabita,, Adnan Al-Anbukya , Hamid GholamHosseinib
a Sensor

Networks and Smart Environment Research Centre (SeNSe) Auckland University of Technology (AUT), Private Bag 92006, Auckland 1142, New Zealand b Department of Electrical and Electronic Engineering, Auckland University of Technology (AUT), Private Bag 92006, Auckland 1142, New Zealand

Abstract This paper presents Wireless Sensor Network (WSN) based Wildre Hazard Prediction (WFHP) system. A systematic description of architectural details and requirements of WSN for WFHP applications is presented. A TrueTime model of WSN architecture is built in Matlab environment taking into account the requirements for in-network processing of WFHP for spatially explicit locations in forest layer. The model performance in terms of network latency, energy consumption, and scalability is analyzed through simulation. Verication of model sanity and performance are carried out taking real weather datasets and their corresponding wildre hazard outputs as benchmarks. Simulation results show the eciency and applicability of the model to real wildre hazard prediction system. Keywords: Wireless Sensor Network, Wildre Hazard Prediction, TrueTime model, In-network Distributed Computing

1. Introduction Wildres occur in natural environment such as forest or prairie and take hold burning uncontrollably [1]. Wildres are often very dicult to predict and control. They spread fast easily reaching and endangering homes, property and lives [2]. Wildres are usually not spotted until they spread to a large extent, by which time it is dicult to bring them under control. To prevent wildres and attack res before they can get out of control, WFHP systems are employed to predict where the most likely areas for re are at certain times, specially, during re seasons [3]. Many countries worldwide use Fire Weather Index system (FWI) [4] for wildre danger prediction. The FWI system generates national maps of re danger ratings which are used in trend forecasts. FWI system computations are based on once-daily observations from the re weather network [4] which are comprised of several metrology stations (weather stations) sparsely distributed throughout the country and satellite communication. These observations are reported to the central processing and repository centers (CPRC) where they are used each day at noon local time to generate re danger maps [5]. However, due to the nature of the communications used (satellite communication) and observation stations, the FWI system inherit several drawbacks; Limited measurement points due to the cost of metrology station set up and maintenance, A 24 hours time interval between re hazard reports due to the nature and
Corresponding

author Email addresses: hsabit@aut.ac.nz (Hakilo Sabit), adnan.anbuky@aut.ac.nz (Adnan Al-Anbuky), hgholamh@aut.ac.nz (Hamid GholamHosseini)

18770509 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Prof. Elhadi Shakshuki and Prof. Muhammad Younas. doi:10.1016/j.procs.2011.07.016

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

107

cost of satellite communication, no means of querying a particular domain of locations for re hazard ratings, low condence values of re ratings for locations further away from the metrology stations, and emergency deployment of the system at new regions is too slow. To address the drawbacks mentioned above a wireless sensor network based wildre hazard prediction (WSN-WFHP) system model has been developed in this research. WSN-WFHP system has several advantages such as high prediction certainty, timely hazard prediction, energy eciency, low-cost weather observation sensors, simple and rapidly deployable as well as being complimentary to the current FWI system. The characteristics and properties of the WSN-WFHP system model are described below. 2. Related work Wireless sensor networks have recently been used in a number of applications such as; wildre prediction, detection, and monitoring applications. The area of wildre prediction, detection and monitoring utilizing WSN has been investigated by several researchers [6, 2, 7, 8, 9, 10, 11, 12, 3, 13, 14, 15, 16]. Forest re detection using WSN equipped with only Light and Temperature sensors has been investigated [6]. Using light and temperature data as inputs to information fusion algorithm to detect forest res in the absence of other required sensors [6]. Sensor fusion technique to enhance performance of early forest re detection and re location estimation for Urban-Rural-Interface [2]. While [6] and [2] both detect forest res based on the inference from sensor fusion, they did not consider established forest re model such as FWI as in our work. The main challenge in WSN applications is managing energy consumption to impart long network lifetime [7, 8]. It is known that communication is the largest energy consumer in WSN system [9, 10, 11]. Towards this goal, [12] proposes a continuous sliding windows skyline (attribute space) computing algorithm for data suppression to minimize needless communication between sensor nodes for early forest re detection. However, the multidimensional attribute spaces (skylines) are merely an indication of re probability and lacks measure of possible re spread and other important components of re prediction found in FWI system. The requirements, considerations and experiences gained in the development and pilot implementation of the in-situ self-organizing network for forest temperature and wildre monitoring are presented in some details in [3]. Other dimensions of the problem such as utilizing short message serve through global system for mobile (GSM) communication technology for early bushre warning [13], detection and verifying rural and forest re using WSN and IP camera in a wireless network [14], and performance of WSN tracking re spread where motes are protected with a thermal insulation [15] are also attempted. The works of [3, 13, 14, 15] are all based on centralized data processing to detect forest res which is not robust to a single point of failures. The only work on application of WSN for early detection and modelling of forest res based on FWI system is the work by [16]. [16] model the early forest re detection problem as a k-coverage problem so as to estimate the dierent components of the FWI system. In their work, sensor cells compute FWI and FFMC (ne fuel moisture code) and periodically forward these values to a processing center. Their main focus is mainly on the degree of coverage (sensors density) to achieve a set FWI and FFMC accuracy for a give area. In contrast, this paper focuses on WSN architecture for distributed incremental in network processing of all the components of FWI system. 3. WSN-WFHP The proposed WSN-WFHP system consists of wireless sensor network organized in two-tiered cluster of weather sensor nodes as shown in Figure1. The weather sensor nodes are low-cost sensors capable of capturing weather data, processing weather observations and communicate using short range wireless communication. These weather sensor nodes are used as current sources similar to the metrology stations in FWI system. In addition, the weather sensor nodes compute partial re hazard prediction and runs wireless communication protocol stack to ensure network connectivity, organization and integrity. The rst tier of WSN-WFHP system consists of the links between weather sensor nodes and cluster head nodes. While the second tier consists of the link between cluster heads and sink (central coordinator node) links. The communication links here are low power industrial, scientic and medical (ISM) band wireless communication unlike satellite link in FWI system. 3.1. WSN-WFHP architecture The weather sensor nodes are typical wireless sensor nodes with limited energy and computational resources. The energy cost of data transmission in wireless sensor networks is known to be several orders larger than that of

108

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

Figure 1: Cluster organization of the proposed WSN-WFHP system.

computation [10]. This limitation usually calls for maximizing local computations and minimizing raw data transmission [11]. In WSN-WFHP system, computation of the complete re hazard prediction algorithm at individual sensor nodes is too resource intensive. On the other hand, transmission of raw weather observations long distance to the sink is too energy extensive. Therefore an architecture whereby individual weather sensor nodes accomplish partial re hazard prediction and transmit the partial prediction information to intermediate nodes (cluster heads) which in turn aggregates and computes further semi-re hazard prediction before transmitting to the sink is designed. In this work, in-network distributed WFHP processing is designed to achieve energy eciency and extended network lifetime. Based on two-tiered cluster architecture involving mobile nodes and backup nodes, this system presents unique and robust WSN architecture for wildre hazard modeling and prediction application. Architectural details of the system are described below with three main aspects; two-tiered cluster architecture, in-network distributed WFHP processing, and mobile relief nodes and backup nodes. 3.1.1. Two-tiered cluster architecture Three categories of functionally distinct wireless sensor node types namely Sensor nodes (SN), cluster head nodes(CHN), and central coordinator node (CCN) are deployed within target area to autonomously form a two-tiered cluster architecture. The SNs form the rst level of the two-tiered cluster hierarchy. The CHNs form the next level cluster hierarchy and the CCN node is at the top level of the hierarchy as shown in 1. The nodes deployment may be random. However, for this particular application the CH nodes are located at optimal locations such that every SN is within transmission range of at least one CHN and every CHN is within radio range of at least another CHN that can directly or indirectly reach the CCN node. The CCN forms a network on a given wireless channel and broadcasts a signal to recruit CHNs. The CHNs once join the network themselves broadcast a signal to recruit SNs. The SNs that receive the broadcast signals choose the closest CHN among the CHNs they can reach and join the cluster as member nodes. A CHN can reject SNs to join its cluster once it has admitted a given maximum number of member nodes so that balanced clustering can be achieved.

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

109

Figure 2: Weather sensor model

3.1.2. In-network distributed WFHP processing Once the clustering is achieved and stable two-tiered architecture is formed, the steady state operation of the network proceeds. At this stage, the CHNs sample and instantly broadcast wind speed and rainfall values to their cluster members and wait for their responses. The SNs on receiving the wind speed and rainfall values immediately start sampling ambient temperature and relative humidity. The SNs then compute partial WFHP indices using the weather data (wind speed, rainfall, temperature, and relative humidity). Every SN then transmits the partial WFHP indices to its CHN. The CHNs, if received WFHP information from their member nodes within a given time period, proceed computing the semi-WFHP and transmit these information to the CCN node. The CCN node transmits the semi-WFHP information to the central processing and repository center (CPARC). The CPARC computes the WFHP for each SN location. The CPARC then stores the WFHP of each SN to the weather database which feeds the PC application for modelling, analysis and system reconguration. This cycle will be repeated for the duration of the steady state stage until the result of the PC application suggests system reconguration. 3.1.3. Mobile relief nodes and backup nodes The mobile nodes in this architecture are guest nodes that are allowed to enter and leave the network as desired. They are typically tted to the re ghter gadgets, personnel or vehicles. They play the role of monitoring re ghter safety through vital signs sensors and relieving energy constrained CHNs from long range transmissions. Mobiles nodes eectively come into the network during high re hazards as the re ghters arrive at the location to bring the situation under control. During these situations as the re ghters roam the hazard area, the mobile nodes allocated to the re ghters or their equipments assist in forwarding information to CCN on behalf of the closest CHN. The other aspect of this architecture is the backup cluster head nodes role and activities. Since the CHNs are local aggregators of member SNs, failure at the cluster head results in large information loss. Therefore this architecture supports the CHNs with idle back up cluster head nodes. The backup cluster head nodes remain in idle listening mode till they discover that the CHN they support is critically low on energy or malfunctioned. In such situations the backup cluster head nodes immediately assume the role of the CHNs. The demoted CHNs in turn assume the role of backup cluster head nodes and cease the opportunity to recharge their battery through the solar panel onboard. 4. WSN-WFHP models Weather Sensor Node model: The weather sensor node model consists of a kernel, several sensors, a wireless radio transceiver, and energy source. The kernel provides modelling of Microcontroller unit with memory, signal conditioning circuitry and peripherals to interface sensors. The node model settings are based on TIs MSP430F2274 Microcontroller, CC2530 IEEE802.15.4 radio model, powered from two AAA size batteries of each 1.5 Volts and 1200 mAh. A TrueTime model of weather sensor node is shown in Figure 2. Wireless Sensor network Model: The WSN-WFHP model consists of the weather sensor models linked through TrueTime ZigBee wireless network. The weather sensor nodes cluster organization, data transmit period for each node, WFHP processing algorithms and processing time delays are congured through Matlab scripts which are

110

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

The WSN-WFHP Model Parameters Model Parameter Value Network Type ZigBee Network Number 1 Number of Nodes 19 Data rate (bits/s) 25000 Minimum frame size (bits) 16 Transmit power (dBm) 0

Model Parameters Receiver signal threshold (dBm) Path-loss exponent ACK timeout (sec) Retry limit Error coding threshold

Value -85 3.5 .0004 5 0.03

Table 1: The WSN-WFHP model parameters

attached to the model through an interface dialog. The model takes the path-loss of the radio signals into account taking x and y inputs to specify the true location of the weather sensor nodes. The WSN-WFHP set model parameters are shown in Table 1. 5. WSN-WFHP simulation The WSN-WFHP system model described in section 4 is set up for simulation. The initial simulation conguration has consisted of 16 weather sensor nodes self-organizing themselves autonomously into two clusters. Once steady wireless network state is achieved, the weather sensor nodes acquire weather data from a specic input le and send partial re hazard indices to their respective cluster heads every 30 Seconds and varied when required. The cluster head nodes upon receiving indices compute semi-re hazard indices and transmit the same to the sink for further processing. The initial simulation conguration only consisted of 16 weather sensor nodes for the purpose of re hazard prediction model sanity check. However the rest of the simulation has consisted of up to 200 weather sensor nodes and 25 cluster head nodes within an area of 1.72 Sqauare Killometers to mimic a naturally dense WSN. The minimum number of sensor nodes used in simulation setups to investigate the eect of number of sensor nodes on re prediction results has been set to 16. This is because simulations showed that weather network of less than 16 nodes only captured mimimal network dynamics such as data collision and packet latency which is not the case in real weather network scenario. 5.1. Model sanity The WSN-WFHP model is simulated for sanity check feeding inputs from real weather data sets with known wildre hazard prediction results. The real dataset consists of 24 records of wind speed, rainfall, temperature, relative humidity measured hourly over the last 24 hours for each 100 adjacent pixels of a given area, and one record of FWI system calculated re danger prediction rating, and ease of re ignition rating for each pixel over the same time period (24 hours). 100 WSN-WFHP sensor node models are each given wind speed, rainfall, temperature, relative humidity from a single pixel every hour. The WSN-WFHP sensor node models each produced re danger prediction rating, and ease of re ignition rating every hour. The average over 24 hours of model produced re danger prediction rating, and ease of re ignition rating and the FWI produces re danger prediction rating, and ease of re ignition rating are ploted in Figure 3 and Figure 4 respectively. This experiment has been repeated for dierent number of sensor nodes (16 to 200). From the wildre hazard prediction point of view the WSN-WFHP model average error is in order of about 15 % in all cases. 5.2. End-to-end delay The average end-to-end delay of the WSN-WFHP model is shown in Figure 5. This is the sum of transmission, propagation, WFHP processing and queuing delays. In this simulation the sensor nodes transmit their partial prediction information every 60 seconds. The cluster heads process semi-prediction and queue packets for transmission as soon as the channel is available. 18 weather sensor nodes and two cluster head nodes computing for the wireless channel, the maximum delay observed is 10 seconds, with an average delay of 5.0747 seconds. This has been repeated for 54 weather sensor nodes and six cluster heads node and similar result has been observed as shown in Figure 6.

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

111

40 FWI WSNWFHP 35

30

Wildfire hazard prediction rating

25

20

15

10

10

20

30

40

50

60

70

80

90

100

Dataset number

Figure 3: Comparison of WSN-WFHP and FWI

100 WSNWFHP FWI 90

80

Ease of fire ignition rating

70

60

50

40

30

20

10

10

20

30

40

50

60

70

80

90

100

Dataset number

Figure 4: Comparison of WSN-WFHP index and FWI index

112

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

12 Endtoend delay linear y = 0.049*x + 4 10

Delay time [seconds]

10

15

20

25

30

35

40

45

Number of packets transmitted

Figure 5: Average end-to-end delay

12 Endtoend delay linear y = 0.014*x + 4.6 10

Delay time [seconds]

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

Number of packets transmiitted

Figure 6: Average end-to-end delay

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

113

35 16 sensor nodes 54 sensor nodes 90 sensor nodes 30

25

No. of packet loss

20

15

10

15

30

45

60 75 90 105 120 Data transmit cycle interval [seconds]

135

150

165

180

Figure 7: Packet loss performance

5.3. Packet loss The WSN-WFHP model is evaluated for the amount of data packets loss under dierent sensors data transmit period and whether the amount of packets dropped have signicant impact on the hazard prediction results. The simulation scenario involves sensor nodes transmitting data packets at varying data transmit periods. The results obtained as shown in Figure 7 indicates that the number of packet loss decreases exponentially as sensors transmit data packets less frequently for a network of 16, 54 and 90 weather sensor nodes. The prediction results obtained under this scenario indicates that data collusion is not an issue at such network scales as long as the number of packet losses are not signicantly high for a given cluster of sensor nodes. This also shows the fault tolerance of the WSN-WFHP under these conditions. 5.4. Energy consumption The models energy consumption is analysed based on the TrueTime battery model. The battery model computes energy consumption due to kernel data processing, packet transmission/reception, and idle waiting consumption. The battery performance of the model is shown in Figure 8. The energy source is two AAA batteries of each 1200mAh (2 x 1200 mAh). The Figure shows a simulation of WSN-WFHP application for continues run till the battery remaining power is below 300 mAh for both weather sensor nodes and cluster head nodes. 6. Conclusion This paper presents a wireless sensor network based wildre hazard prediction system through TrueTime modelling and simulation software. The WSN-WFHP model sanity has been veried through real wildre datasets. Distributed in-network processing of wildre hazard prediction based on WSN has several advantages while producing similar results to satellite communication based FWI system. The end-to-end delay, packet loss and energy consumption performance of WSN model have been observed through simulations. The simulation results indicate that for two-tiered WSN architecture, the inuence of end-to-end delay, energy consumption and packet loss on the wildre hazard prediction results is tolerable. This system provides high spatial and temporal resolution wildre hazard prediction system which is cost-eective, energy ecient, easily deployable for emergency situations and provides means for interactions.

114

Hakilo Sabit et al. / Procedia Computer Science 5 (2011) 106114

2500 Cluster head Weather sensor

2000

Battery current [mA]

1500

1000

500

20

40

60

80

100

120

140

160

Time [Hours]

Figure 8: Energy consumption performance

7. References
[1] B. S. Lee, M. E. Alexander, B. C. Hawkes, T. J. Lynham, B. J. Stocks, P. Engleeld, Information systems in support of wildland re management decision making in canada, Computers and Electronics in Agriculture 37 (1-3) (2002) 185198, doi: DOI: 10.1016/S01681699(02)00120-5. [2] E. Zervas, O. Sekkas, S. Hadjieftymiades, C. Anagnostopoulos, Fire detection in the urban rural interface through fusion techniques, in: Mobile Adhoc and Sensor Systems, 2007. MASS 2007. IEEE Internatonal Conference on, 2007, pp. 16. [3] V. Vescoukis, T. Olma, N. Markatos, Experience from a pilot implementation of an in-situ forest temperature measurement network, in: Personal, Indoor and Mobile Radio Communications, 2007. PIMRC 2007. IEEE 18th International Symposium on, 2007, pp. 15. [4] N. R. Canada, Canadian forest re weather index (fwi) system (2009). URL http://cwfis.cfs.nrcan.gc.ca/background/summary/fwi [5] A. J. Dowdy, G. A. Mills, K. Finkele, W. d. Groot, Australian re weather as represented by the mcarthur forest re danger index and the canadian forest re weather index, Cawcr technical report no. 10 (2009). [6] S. d. P. Osman, E. F. Nakamura, Fusing light and temperature data for re detection, in: F. N. Eduardo (Ed.), The IEEE symposium on Computers and Communications, Vol. 0, Riccione, Italy, June 22-25 2010, pp. 107112. [7] A. S. John, Research challenges for wireless sensor networks, SIGBED Rev. 1 (2) (2004) 912, 1121780. [8] N. Srivastava, Challenges of next-generation wireless sensor networks and its impact on society, Journal of Telecommunications 1 (1). [9] I. Kurtis Kredo, M. Prasant, Medium access control in wireless sensor networks, Comput. Netw. 51 (4) (2007) 961994, 1223795. [10] D. Ganesan, A. Cerpa, W. Ye, Y. Yu, J. Zhao, D. Estrin, Networking issues in wireless sensor networks, Journal of Parallel and Distributed Computing 64 (7) (2004) 799814, doi: 10.1016/j.jpdc.2004.03.016. [11] H. Sabit, A. Al-Anbuky, H. Gholam-Hosseini, Distributed wsn data stream mining based on fuzzy clustering, in: Ubiquitous, Autonomic and Trusted Computing, 2009. UIC-ATC 09. Symposia and Workshops on, 2009, pp. 395400. [12] Kre, P. imir, i, B. Hrvoje, V. Marin, Early forest re detection with sensor networks: Sliding window skylines approach, 1430151 725-732 (2008). [13] L. Liu, R. Sun, Y. Sun, S. Al-Sarawi, A smart bushre monitoring and detection system using gsm technology, Int. J. Computer Aided Engineering and Technology 2 (2/3) (2010) 218233. [14] J. Lloret, M. Garcia, D. Bri, S. Sendra, A wireless sensor network deployment for rural and forest re detection and verication, Sensors 9 (11) (2009) 87228747. [15] T. Antoine-Santoni, J.-F. Santucci, E. d. Gentili, X. Silvani, F. Morandini, Performance of a protected wireless sensor network in a re. analysis of re spread and data transmission, SENSORS 9 (8) (2009) 58785893. [16] M. Hefeeda, M. Bagheri, Wireless sensor networks for early detection of forest res, in: Mobile Adhoc and Sensor Systems, 2007. MASS 2007. IEEE Internatonal Conference on, 2007, pp. 16.