Source-linked AI summary
Forecasting Drought Using Multilayer Perceptron Artificial Neural Network Model
Zulifqar Ali, Ijaz Hussain, Muhammad Faisal, Hafiza Mamona Nazir, Tajammal Hussain, Muhammad Yousaf Shad, Alaa Mohamd Shoukry, Showkat Hussain Gani
TL;DR
Drought can affect water-resource planning, motivating evaluation of drought forecasting across Pakistan’s climatic zones. This paper applies MLPNN to monthly SPEI data and reports forecasting capability across time scales, with SPEI-1 correlations ranging from 0.887 to 0.987.
Problem
The study asks whether ANN can forecast multiscalar SPEI across Pakistan’s climatic zones, given drought’s implications for water-resource planning.
Source: §1 · pp. 1–2Method
The study uses MLPNN to forecast SPEI from monthly climate data collected at seventeen stations in Northern Area and KPK, Pakistan.
Source: §2.1 · p. 2, §4 · p. 6Results
MLPNN was reported to predict SPEI across multiple time scales; SPEI-1 correlations ranged from 0.887 to 0.987, while Balakot SPEI-12 correlation was 0.780.
Source: §3.2 · pp. 5–6Takeaways & Limitations
The authors conclude the model is convenient for operational water-resources management and may help planners take advance action where water scarcity is increasing.
Source: §4 · pp. 6–7
Abstract
from arXiv · showhide
These days human beings are facing many environmental challenges due to frequently occurring drought hazards. It may have an effect on the countrys environment, the community, and industries. Several adverse impacts of drought hazard are continued in Pakistan, including other hazards. However, early measurement and detection of drought can provide guidance to water resources management for employing drought mitigation policies. In this paper, we used a multilayer perceptron neural network (MLPNN) algorithm for drought forecasting. We applied and tested MLPNN algorithm on monthly time series data of Standardized Precipitation Evapotranspiration Index (SPEI) for seventeen climatological stations located in Northern Area and KPK (Pakistan). We found that MLPNN has potential capability for SPEI drought forecasting based on performance measures (i.e., Mean Average Error (MAE), the coefficient of correlation R, and Root Mean Square Error (RMSE). Water resources and management planner can take necessary action in advance (e.g., in water scarcity areas) by using MLPNN model as part of their decision making.
1. Introduction
The study evaluates ANN forecasting of multiscalar SPEI across climatic zones of Pakistan, motivated by drought’s effects on freshwater flows and water-resource planning.
1. Introduction
- Drought can reduce freshwater flows and substantially affect water-resource planning and management.Its occurrence is associated with factors including wind, relative humidity, temperature, and rainfall characteristics.Source: §1 · p. 1
- Drought indices, including SPI and SPEI, are widely used to assess drought and provide an alternative to decisions based on raw data.The introduction lists indices such as RAI, CMI, PDSI, BMI, SWSI, RDI, SPI, and SPEI.Source: §1 · p. 1
- Prior forecasting studies applied Markov chains, neural networks, and ANFIS to drought indices, while neural networks can learn complex patterns from data.The cited work includes SPI-based drought-class prediction three months ahead and ANN comparisons with linear stochastic models.Source: §1 · p. 2
- Prior ANN studies reported R^2 values of 0.66–0.79 for six-month SPI and EDI forecasts, better EDI than SPI results, and ANN advantages over KNN for nine- and twelve-month SPI forecasts.ANN-based synthetic NDVI forecasting was also reported for spring in the Ahar-Chay Basin.Source: §1 · p. 2
- The study evaluates ANN drought forecasting using multiscalar SPEI across different climatic zones of Pakistan.The paper also reviews drought indices to assess their appropriateness.Source: §1 · pp. 1–2
2. Material and Methods
The study calculates SPEI at four time scales from monthly rainfall and temperature records collected at meteorological stations in Northern Area and KPK, including the capital territory of Pakistan. Its methods estimate potential evapotranspiration, fit probability distributions, and standardize the precipitation–PET balance.
2.1. Study Area
- Monthly rainfall and mean temperature were collected for 1975–2012 from 17 meteorological stations in Northern Area and KPK, including the capital territory of Pakistan.The selected locations represent precipitation regimes and catchments with minimal synthetic influences and good hydrometric performance.Source: §2.1 · p. 2
2.2. Standardized Precipitation Evapotranspiration Index (SPEI)
- SPEI combines precipitation and temperature through PET, estimated here with the Thornthwaite equation, then standardizes precipitation minus PET over specified time intervals.The study calculated SPEI at four time scales for each station.Source: §2.1 · pp. 2–3
- Gamma, Generalized Extreme Values, Log-Logistic, and Generalized Pareto distributions were fitted to d_i, with Kolmogorov-Smirnov and Anderson-Darling tests used in Easy-Fit before standardization.Source: §2.1 · p. 3
- For rainfall series containing zeros, zero and nonzero rainfall probabilities were handled separately, with zero-rain probability q estimated as m/n.Source: §2.1 · p. 3
- SPEI has mean 0 and standard deviation 1, enabling comparisons across time and space; zero corresponds to 50% cumulative probability under a Log-Logistic Distribution.Source: §2.1 · p. 4
3. Neural Network Forecasting
The section describes a feedforward multilayer perceptron for drought forecasting and reports validation of SPEI predictions across multiple time scales.
3.1. Multilayer Perception Architecture.
- With no universal network-structure rule, researchers often relied on trial and error; this study likewise used trial and error to select network parameters.Several structure-selection algorithms had been proposed, but none guaranteed optimal parameters for every forecasting problem.Source: §3.1 · pp. 4–5
- The selected MLPNN used 30 input neurons, 8 hidden neurons, and 1 output neuron, sigmoid activation, momentum 0.5, and 10,000 training epochs.Its input vector comprised the previous 30 values of each index; the model was validated on 20% of the data, and early stopping was used to address overfitting.Source: §3.1 · p. 5
3.2. Results and Discussion.
- Observed–predicted SPEI correlations ranged 0.887–0.987 (SPEI-1), 0.876–0.994 (SPEI-3), 0.876–0.994 (SPEI-6), and 0.780–0.970 (SPEI-12) across stations.Balakot showed apparent deviations for SPEI-12, with a correlation coefficient of 0.780; the authors also state forecasted drought indices could be used for multistep prediction.Source: §3.2 · pp. 5–6
4. Conclusion
The study concludes that MLPNN can forecast monthly drought indices from climate data across seventeen stations in Northern Area and KPK, with predicted SPEI variation low enough to support operational water-resource management. The model may help planners act in advance where insufficient rainfall is increasing water scarcity.
4. Conclusion
- MLPNN forecasts monthly SPEI from temperature and precipitation data recorded at seventeen stations in Northern Area and KPK from 1975 to 2012, with low variation between observed and predicted values.SPEI was estimated by fitting an appropriate probability distribution to the difference between precipitation and PET.Source: §4 · p. 6
- Validation-phase performance of multilayer perceptron models was reported for all study stations.Source: Table 3 · p. 7
- ANNs can capture variation in selected drought indices at a one-month time scale.Source: §4 · p. 7
- Planners may use the neural-network model to act in advance where insufficient rainfall is increasing water scarcity and may lead to drought.Source: §4 · p. 7
Ethical Approval
The manuscript states that it followed ethical standards for human experimentation and the 2008 version of the 1975 Declaration of Helsinki.
Ethical Approval
- The manuscript was prepared according to the responsible committee’s ethical standards for human experimentation and the 2008 version of the 1975 Helsinki Declaration.Source: Ethical Approval · p. 8
Disclosure
The manuscript uses secondary data.
Disclosure
- The manuscript was prepared using secondary data.Source: Disclosure · p. 8
Ecology
The supplied Ecology section contains no substantive ecological findings; its passages are publisher mastheads and journal-name fragments.
Ecology
- The passages consist mainly of repeated Hindawi Publishing Corporation Volume 2014 mastheads and unrelated journal-title fragments.No ecological methods, results, or claims are provided.Source: Ecology · p. 10