Drought Forecasting in Khash City by Using Neural Network Model
Hossein Negaresh
Associate Professor of Geography and Environmental PlanningFaculty, University of Sistan & Baluchestan
Mohsen Armesh
Holding Master Degree in climatology in Environmental Planning
Extended Abstract
1- Introduction
Drought is condition of lack of rainfall and increase in temperature occurring in any climatic condition. Since decrease of moisture discords vital systems that depend on aqua, drought may exterminate this systems. Khash city with a dry climate and arable land and gardens is always exposed to the climate risk and occurrence of droughts causing reduce in crop yields. The aim of this study is to forecast the drought in Khash city during one month, three months and twelve months which results in a model for forecasting drought to be used in the future planning to decrease the losses due to it.
2- Methodology
In this study monthly data, quarterly and twelve months of climatic elements such as precipitation, relative humidity, temperature and climatic indices affecting drought from 1961 to 2010 have been used to predict droughts. The data is standardized and then combined with 70 to 30 (70% for training data and 30% for testing models) and applied to all networks. To select the input on the drought affecting the city of Khash, stepwise regression and correlation methods were used. Standard precipitation index (SPI) of drought is employed as the outgoing model. Drought is in the time of Severe SPI -1 or less, when the drought index is positive and the event will end the drought. Start and end of the drought period is determined by negative numbers and values of the cumulative SPI and cumulative SPI shows the magnitude and severity of drought.
3- Artificial Neural Network
In this study, artificial neural network method has been used to predict droughts. The structure of networks of biological neural networks that comply with the regulation of weights is determined by the relationship between its components. After the network training, a particular input leads to a specific response. Elements of the neural network make up input vector, weight,. transfer functions and output. Neural network system composed of the community neurons and each neuron of it has been formed of three part bodies, dendrites and axon.
Figure (1) The non-linear neurons
Neuron output is calculated as follows:
In this statement (a) is output of neuron, f the transfer functions, w the weight vector, p the input neuron and b the bias value. Levenberg - Marquardt algorithm has been used for network training to comply with back propagation rule.
Neural network computing, the Feed Forward, is as follows:
In this relationship, apj is the value of the output before layer, Wij the weight of the layer and bi is the bias. This relationship is linear then the value F (Netpi) is calculated as an F is function of stimulus that nonlinearity in the hidden layer and output layer is linear. During the training process the same appropriate weight and bias are predicted to be a recursive algorithm for the training that we have here, the method of error back propagation (BP), and the calculations are as follows:
Where ΔWij amount Wij is added to the training and η, is rate training and fixed.
3- Discussion
3-1- Monthly forecasts of drought
After creation of the mass models, 8 models of network Radial Basis and 10 models of propagation network were selected as an appropriate model. Among the selected models Radial Basis, RBF6 model with five inputs and a hidden layer of 10 neurons was selected as the optimal model, this model has the highest R2 0.6816 and the lowest error made in the testing phase. Among Selected models of the back propagation network also model BP9 was selected as a choice model. This model will generate R2 0.6987 in the testing phase with less error. But at this stage BP8 model had slightly better results, but BP9 model results were very good in training phase of the model BP8.
Forecasting of the regression model is estimated by using Enter method. R2 regression model was estimated 0.6688, RMSE 0.6218 and the MAE 0.0463.
3-2- Quarter forecast of drought
In the three-month period of the selected models network Radial Basis, RBF6 model with six inputs and one hidden layer 20 neurons were known as the most appropriate models. RBF3 model results are very close to the chosen model, but on the whole better results were gained by RBF6 model. Among The back propagation network model, BP6 model with six inputs and two hidden layers with 15 neurons was the most appropriate model, this model with R2 0.8306 and the least error in the testing phase was chosen as the optimal model. In The regression model predicts were performed by using the Enter method that R2 was estimated 0.743, RMSE 0.8661 and the MAE 0.1118.
3-3- Twelve-month forecast of drought
Among the numerous models in the 12-month period, 7 models were chosen in Radial Basis Network and 7 models of the back propagation network. In Selected models of Radial Basis, RBF5 model with six inputs and one hidden layer of 30 neurons, were identified as the optimal model, R2 of the 0.7926 and a small error in the testing phase and the determination coefficient is very high and very low error in the training phase of the reasons is seen which is chosen as the most appropriate model. In the various models of propagation network, BP5 model with six inputs and four hidden layers and 15 neurons, were identified as the optimal model. Regression models also produced R2 0.64, RMSE 0.6342 and MAE 0.1638.
4- Conclusion
In this study drought in the city of Khash was predicted in three periods, monthly, three months and twelve months by the network Radial Basis and Back propagation and regression model which result in network performance and compared lists. Modeling results showed that the predicted three-month and twelve-month period of the drought is far better than monthly drought anticipated. In monthly and quarterly forecasting drought, propagation network gave better results. All predicted three-month drought with back propagation network with two hidden layers, 15 neurons, and input maximum relative humidity three-month, relative humidity in the third quarter of the delay, the southwest monsoon rain with four seasons delay, the monsoon rainfall southwest with two season delay, average relative humidity three-month and minimum relative humidity three-month, most appropriate model for predicting drought was declared in the Khash city.
Key Words:
Drought, Artificial Neural Network, climatic indices, Khash, forecasting.
5- References
- Alami M. T, Y. Hosseinzadeh and M. Kamasi (2009), Evaluation and Comparison of drought indices and predict it’s on SPI index from dabelvariable modeling with artificial neural network method, the eighth International Seminar on River Engineering.
- Alizadeh Amin (2002), Principles of Applied Hydrology, press 14, sanctum court Imam Reza, Mashhad.
- Afkhami, H., M. T. Dasturati, H. Maleki Nejad and H. Mobin (2010), investigating the effect of climate on the higher accuracy of artificial neural network in predicting the drought Yazd area, Science and Technology of Agriculture and Natural Resources, Volume 14, Issue 51, pp. 157-170.
- Barua Shishutosh(2010), Drought Assessment and Forecasting Using a Nonlinear Aggregated Drought Index, M.Sc. (Water), Engineering and Science Victoria University, Australia.
- Barua, Shishutosh and Perera, B. J. C, A. W. M. Ng and Tran, D. H (2010), Drought forecasting using an aggregated drought index and artificial neural networks, Journal of water and climate change, 1 (3).pp.193-206.
- Cutore P, Di Mauro G, Cancelliere A (2009), Forecasting Palmer Index Using Neural Network and Climatic Indexes, Journal of Hydrologic Engineering/ vol 14/ Issue 6, pp. 588-595.
- Eivazi M., A. Mosaedi and A. A. Dehghani (2009), Comparison of different approaches for predicting SPI, J. of Water and Soil Conservation, Vol. 16(2), pp. 145-167.
- Farajzadeh M (2010), Climatology Techniques, press 3, SAMT, Tehran.
- Farokhnia A, S. Morid and H. Ghaemi (2010), data mining on large-scale climate signals to predict the long drought, the third conference on water resources management.
- Hayes Michael, J. Svoboda, Mark. D. Wilhite, Donald A. Vanyarkho, Olga V. (1999), Monitoring the 1996 Drought Using the Standardized Precipitation Index, National Drought Mitigation Center, Vol. 80, No. 3, pp. 429-438.
- Juan Camilo, Ochoa-Rivera (2008), Prospecting droughts with stochastic artificial neural networks, Journal of Hydrology, Vol 352, Issues 1-2, pp. 174-180.
- Kavyani M. R, Alijani B (2000), The Foundations of Climatology, press 6, SAMT, Tehran.
- Keya Mostafa (2009), NEURAL NETWORKS, press. Kian Rayaneh, Tehran.
- Khoshhal Dastjerdi J. and S. M. Hosseini (2010), Application of ANN in the simulation of climatic and predictable cycles of drought (Case Study: Isfahan Province), Geography and planning environment, age 21, 39, No. 3, pp. 107-120.
- Menhaj M.B (2009), Fundamentals of Neural Networks, vol. 1, press. 5, Amir Kabir University, Tehran.
- Mishra A.K, Desai V.R Singh P (2006), Drought forecasting using feed-forward recursive neural network, ecological modeling, vol.198, pp.127–138.
- Mishra A.K, Desai V.R, Singh P(2007), Drought Forecasting Using a Hybrid Stochastic and Neural Network Model, Journal of Hydrologic Engineering, Vol. 12, No. 6, pp. 626-638.
- Morid S , Smakhtin V, Bagherzadeh K(2007), Drought forecasting using artificial neural networks and time series of drought indices, International Journal of Climatology, Vol. 27, pp. 2103-2111.
- Nasiri Maryam (2011), Forecasting Drought by using Artificial neural Networks Case Study Shiraz Synoptic station, Dissertation of M.Sc Climatology, University of Sistan and Baluchestan, Zahedan.
- National Organization Oceanic Observation and Atmospheric of USA: http://www.cdc.noaa.gov/ClimateIndices/List/,(2011)
- Niknam. H, M. Azhdari Moghadam and M. Khosravi (2010), Drought Forecasting Using Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Teleconnections Patterns, Case Study Zahedan, 4 th International Congress of the Islamic World Geographers (ICIWG).
- Akhtari R, M. H. Mahdian and S. Morid(2007), Assessment of Spatial Analysis of SPI and EDI Drought Indices in Tehran Province, Water Resources Research, Volume 2, No. 3, pp. 27-38.
- Salajegheh A, A. Fath Abadi and M. Najafi Haji Ver (2008), Comparison of neural network and time series forecasting drought (case study: Khorasan Razavi province), Science and Engineering Watershed Iran, the second year, No. 4, pp. 74-77.
- Santos, Celso Augusto G Morais, Bruno S Silva, Gustavo B L(2009), Drought forecast using an artificial neural network for three hydrological zones in San Francisco River basin, Brazil, Proceedings of Symposium HS.2 at the Joint Convention of The International Association of Hydrological Sciences (IAHS) and The International Association of Hydrogeologists (IAH), No.333, Hyderabad, India, pp. 302-312.
- Sedaghat Kerdar A And E. Fattahi (2008), indicators of drought pre-knowledge in Iran, Journal of Geography and Development, No. 11, pp. 59-76, Zahedan.

Type of Study: Research |

Received: 05/Nov/12 | Published: 15/Jan/12

Received: 05/Nov/12 | Published: 15/Jan/12

Rights and permissions | |

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |