Rainfall prediction based on CNN-LSTM model under sliding window

In this paper, we have utilized ANN (artificial neural network) modeling for the prediction of monthly rainfall in Mashhad synoptic station which is located in Iran. To achieve this black-box model, we have used monthly rainfall data from 1953 to 2003 for this synoptic station. First, the Hurst rescaled range statistical (R/S) analysis is used to evaluate the predictability of the collected data. Then, to extract the rainfall dynamic of this station using ANN modeling, a three-layer feed-forward perceptron network with back propagation algorithm is utilized. Using this ANN structure as a black-box model, we have realized the complex dynamics of rainfall through the past information of the system. The approach employs the gradient decent algorithm to train the network. Trying different parameters, two structures, M531 and M741, have been selected which give the best estimation performance. The performance statistical analysis of the obtained models shows with the best tuning of the developed monthly prediction model the correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE) are 0.93, 0.99, and 6.02 mm, respectively, which confirms the effectiveness of the developed models.

Rainfall prediction plays an important role in understanding the pattern of rainfall movement for daily life. This research aims to predict monthly rainfall in Medan Belawan using the Tsukamoto fuzzy logic method. Tsukamoto fuzzy logic is used to predict monthly rainfall in Medan Belawan. This method uses several steps, starting from data input, fuzzification, inference, to defuzzification to help predict rainfall. This research aims to apply Tsukamoto fuzzy logic to monthly rainfall prediction. Fuzzy logic has tolerance for imprecise data. This rainfall prediction uses 4 (four) input variables, namely air temperature, air humidity, air pressure and wind speed, and 1 (one) output variable, namely rainfall as a crips value which is the output generated for rainfall prediction. The data used is 12 monthly data from January 2023 to December 2023. The results show the accuracy of the fuzzy method of 57.75% using the MAPE calculation and show that the calculation of the Tsukamoto fuzzy method is sufficient or feasible to do.

The importance of long-range prediction of rainfall pattern for devising and planning agriculture strategies cannot be overemphasized. Prediction of rainfall pattern remains a difficult problem and the desired level of accuracy has not been reached. In this paper, a committee of artificial neural networks (ANNs) based model with wavelet decomposition is proposed for prediction of monthly rainfall on accounts of the preceding events of rainfall data. Wavelet transform is used for extraction of approximate and detail coefficient of the rainfall data series. These coefficients are used along with a ANN for learning and knowledge extraction processes. The model has been tested on rainfall data for different geographical region of India and also for entire country. The proposed model is capable of forecasting monthly rainfall one month in advance

Monthly rainfall spatial interpolation is an important task in hydrological study to comprehensively observe the spatial distribution of the monthly rainfall variable in the study area. A number of spatial interpolation methods have been successfully applied to perform this task. However, those methods mainly aim at achieving satisfactory interpolation accuracy and they disregard the interpolation interpretability. Without interpretability, human analysts will not be able to gain insight of the model of the spatial data. This paper proposes an alternative approach to achieve both accuracy and interpretability of the monthly rainfall spatial interpolation solution. A combination of fuzzy clustering, fuzzy inference system, genetic algorithm and modular technique has been used. The accuracy of the proposed method has been compared to the most commonly-used methods in geographic information systems as well as previously proposed method. The experimental results showed that the proposed model provided satisfactory interpolation accuracy in comparison with other methods. Besides, the interpretability of the proposed model could be achieved in both global and local perspectives. Human analysts may therefore understand the model from the derived model’s parameters and fuzzy rules.

Modeling for rainfall forecasting is widely used in climate forecasting. A high volume of precipitation or scarcity brings serious problems for society. Thus, a methodology for monthly rainfall forecast is presented using Group Method of Data Handling (GMDH) and sea surface temperature (SST). The intelligent model gets the mean monthly SST, in predefined and temporally lagged areas, after a variable selection step. For model training, precipitation data from the Climate Prediction Center were used. The methodology was applied in a certain area of the municipality of Maraba, located in the southeastern region of the Para state. The results obtained with GMDH overcame those got by using a conventional multilayer Artificial Neural those got by using a conventional multilayer Artificial Neural Network (RNA), reaching values for regression coefficient (R) and root mean square error (RMSE) equal to 0.96363 and 30.8100 in the test stage, while the RNA-based model got 0.81712 and 68.2607, respectively. Results show the GMHDs effectiveness for the monthly rainfall prediction, constituting an important tool for the planning and assistance to decision makers.

Weather-state models have been shown to be effective in downscaling the synoptic atmospheric information to local daily precipitation patterns. We explore the ability of non-homogeneous hidden Markov models (NHMM) to downscale regional seasonal climate data to daily rainfall at a collection of gauging sites. The predictors used are: ensemble means of seasonal rainfall as forecast by the DEMETER and ECHAM models, and the preceding seasonal outgoing long-wave radiation (OLR). As the downscaling of seasonal GCM-based predictions lacks the ability to capture the intra-seasonal variability, we augment the seasonal GCM-driven inputs with statistically-driven predictions of the monthly rainfall amounts. The pooling effect of combining seasonal and monthly estimates of the regional rainfall enhances the capacity of the NHMM to simulate the stochastic characteristics of rainfall fields. The monthly rainfall prediction is derived from a wide range of climate precursors such as the El Niño-Southern Oscillation, local sea-level pressure, and sea-surface temperature. Application of the methodology to data from the Everglades National Park region in South Florida, USA is presented for the seasons May–July and August–September using a 22-year sequence of seasonal data from eight rainfall stations. The model skill in capturing the seasonal and intra-seasonal rainfall attributes at each station is demonstrated graphically and using simple statistical measures of efficiency. The hidden states derived from NHMM are qualitatively analysed and shown to correspond to the dominant synoptic-scale features of rainfall generating mechanisms, which reinforces the argument that physical processes are appropriately captured. Citation Khalil, A. F., Kwon, H.-H., Lall, U. & Kaheil, Y. H. (2010) Predictive downscaling based on non-homogeneous hidden Markov models. Hydrol. Sci. J. 55(3), 333–350.

This study aimed to improve rainfall prediction accuracy by integrating spatio-temporal Generalized Autoregressive Integrated Moving Average (GSTARIMA) with Deep Neural Network (DNN) techniques. Accurate rainfall forecasting is vital for regions like West Java, which face risks of flooding and landslides during heavy rains in December, January and February. The research focused on enhancing monthly rainfall prediction using GSTARIMA model residuals, transformed from daily data, with a spatial resolution of 0.5° × 0.625°. The GSTARIMA-DNN model was developed to reduce prediction errors and closely align with actual rainfall values. The DNN architecture utilized a Multilayer Perceptron (MLP) with eight input nodes for spatio-temporal residuals, two hidden layers with Leaky ReLU (LReLU) activation functions and a linear output layer for regression. The integration of GSTARIMA and DNN significantly improved prediction accuracy compared to traditional methods. Specifically, the GSTARIMA(0,1,1)-DNN model achieved a 7.28% Mean Absolute Percentage Error (MAPE) at four key locations (Tasikmalaya, Cirebon, Majalengka and Garut) among ten locations with the same lag order. This hybrid model effectively managed complex temporal and spatial data, underscoring the benefits of combining spatio-temporal models with DNN for enhanced rainfall forecasting. This approach offers valuable insights for advancing precise spatio-temporal models in various research fields.

Rainfall is one of the most critical parameters in a hydrological model.A few models have been created to investigate and predict the rainfall conjecture. in recent years, soft computing models like Artificial Neural Network (ANN) have been widely used to model a complex hydrological processes.In this paper an attempt has been made to find an alternative feed forward backpropagation neural network (FFBNN) architecture for rainfall prediction.The FFBNN with 12-10-5-1 architecture have been attempted and demonstrated with a case study on Tukad Mati watershed.The developed model on rainfall prediction pattern have been calibrated and validated with appropriate statistical methods.For model performance criteria used mean square error (MSE) and coefisient correlation (R) as an indicator.It was found that the FFBNN are capable in modelling monthly rainfall prediction in the Tukad Mati watershed with 0,011913 for average of MSE and 0,85769 for average of R testing value from three rain gauge station.The FFBNN approach could provide a very effective to solve problems in monthly rainfall prediction.

<p>The development of information system technology is needed in facing current and future agricultural challenges. An integrated agricultural information system is needed based on climate information system technology in answering this challenge. The Indonesian Agency for Agricultural Research and Development Ministry of Agriculture has developed a web-based Indonesia Integrated Cropping Calendar Information System (ICCIS). The ICCIS provides information for users in preparing planting plans. Likewise, Jaxa's Satellite-based Monitoring Network system (JASMIN-JAXA) web application that uses a satellite database covers all provinces in South Asia, Southeast Asia, and East Asia. This paper aims to review the use of web-based climate information systems in the ICCIS and the JASMIN-JAXA web-based application. Integrated climate information system in ICCIS, among others: flood estimation, drought and crop pests, prediction dry season and rainy season, district-level monthly rainfall prediction map. While JASMIN is a map (rainfall, drought index, soil moisture, groundwater index, solar radiation, surface temperature, and vegetation index) with time series and data sources. Both are able to monitor climate developments for agricultural purposes, dynamic, and growing. The importance of updating and verifying data in accordance with field conditions and equipment maintenance in the system is to be continuously improved to get benefits and accuracy of results in accordance with the interests of agriculture.<strong></strong></p>

Good quality of rainfall data is required for the hydrological studies, water resources planning and sustainable environmental management. Consequently, the assessment of the homogeneity of rainfall data at different region is becoming increasing popular in the past few decades. In this study, the homogeneity analysis of rainfall data was carried out in Kelantan River Basin, Malaysia. The methods, namely standard normal homogeneity test (SHNT), Buishand range test, Pettitt test and von Neumann ratio test were applied to the monthly, yearly and seasonal data. The historical rainfall data from 10 rainfall stations covering the study period from 28 to 60 years were selected. The four tests were applied to 120 monthly series, 10 yearly series and 40 seasonal series. ‘Useful’, ‘doubtful’ and ‘suspect’ were used to classify the results of the four tests. The results showed that 94.17% of the monthly rainfall series, 70% of yearly rainfall series and 97.5% of seasonal rainfall series are labelled ‘useful’. There is 5% of monthly rainfall series, 30% of yearly rainfall series and 1% of seasonal rainfall series are classified as ‘doubtful’. Meanwhile, there is only 0.83% of monthly rainfall series and no yearly rainfall series and seasonal rainfall series detected in the class ‘suspect’. Overall, the percentage of inhomogeneity detected in the monthly, yearly and seasonal rainfall data series of Kelantan River Basin is very small, thus most of the data is suitable to be used for further hydrological and variability analysis.

In this study, a new hybrid model integrated adaptive neuro fuzzy inference system with Firefly Optimization algorithm (ANFIS-FFA), is proposed for forecasting monthly rainfall with one-month lead time. The proposed ANFIS-FFA model is compared with standard ANFIS model, achieved using predictor-predictand data from the Pahang river catchment located in the Malaysian Peninsular. To develop the predictive models, a total of fifteen years of data were selected, split into nine years for training and six years for testing the accuracy of the proposed ANFIS-FFA model. To attain optimal models, several input combinations of antecedents’ rainfall data were used as predictor variables with sixteen different model combination considered for rainfall prediction. The performances of ANFIS-FFA models were evaluated using five statistical indices: the coefficient of determination (R 2 ), Nash-Sutcliffe efficiency (NSE), Willmott’s Index (WI), root mean square error (RMSE) and mean absolute error (MAE). The results attained show that, the ANFIS-FFA model performed better than the standard ANFIS model, with high values of R 2 , NSE and WI and low values of RMSE and MAE. In test phase, the monthly rainfall predictions using ANFIS-FFA yielded R 2 , NSE and WI of about 0.999, 0.998 and 0.999, respectively, while the RMSE and MAE values were found to be about 0.272 mm and 0.133 mm, respectively. It was also evident that the performances of the ANFIS-FFA and ANFIS models were very much governed by the input data size where the ANFIS-FFA model resulted in an increase in the value of R 2 , NSE and WI from 0.463, 0.207 and 0.548, using only one antecedent month of data as an input (t-1), to almost 0.999, 0.998 and 0.999, respectively, using five antecedent months of predictor data (t-1, t-2, t-3, t-6, t-12, t-24). We ascertain that the ANFIS-FFA is a prudent modelling approach that could be adopted for the simulation of monthly rainfall in the present study region.

Monthly-to-seasonal precipitation forecasts are important in water resource management. Hidden Markov models (HMM) are widely applied in precipitation simulation due to its simplicity and advancements in associated computing techniques. HMMs, however, lack the flexibility to accommodate external factors in the dynamics. In this study, we consider a novel semiparametric Bayesian non-homogeneous hidden Markov model (BNHMM) that can explicitly incorporate the effects of observed covariates, e.g., climate index, on the state transition probabilities as well as the observation emission distributions. The proposed approach is tested for three rainfall stations in the service area of Tampa Bay Water, a regional water supply agency in the Southeastern United States. The BNHMM is first examined to simulate historical monthly rainfall data from the period 1979 to 2016. It is then evaluated in a retrospective mode to generate 3-month-ahead precipitation forecasts. Results indicate that the proposed BNHMM can capture historical rainfall properties well, and it is a promising alternative in providing operational rainfall forecasts. Although large-scale atmospheric and oceanic teleconnections, e.g., ENSO, have a strong influence in regional rainfall at the seasonal time scale, especially for the winter and early spring; its influence on monthly scale rainfall is limited. Potential improvement in forecasting performance is also discussed.

Time series analysis and forecasting has become a major tool in different applications in meteorological phenomena, such as rainfall, humidity, temperature, draught and environmental management fields. It has two goals, perception or modeling random mechanism and prediction of future series quantities according to the past. In this research, ARIMA (Auto Regressive Integrated Moving Average) model has been used to carry out short term predictions of monthly rainfall in Sylhet and Moulvibazar district (north-eastern region) for years 2012 to 2014. Based on the inspection of the ACF, PACF autocorrelation plots, the most appropriate orders of the ARIMA models are determined and evaluated using the AIC-criterion. For the monthly rainfall in Sylhet district at Tajpur and Kanairghat station ARIMA (1,1,1) (0,1,1) is obtained, whereas the respective models in Moulvibazar district at Chandbagh, Sreemangal and Manu railway bridge are ARIMA (0,1,1) (1,1,1), ARIMA (1,1,1) (1,1,1) and ARIMA (1,1,0) (0,1,1). Among five rainfall stations PBIAS is the least (-1.07%), NSE (88%) and Index of agreement (87%) are the highest at Kanairghat station. Negative Mann-Kendall test statistics of monthly rainfall series (for the period between 1980 and 2011) indicates that monthly rainfall is Original Research Article British Journal of Applied Science & Technology, 4(11): 1644-1656, 2014 1645 decreasing with time except Kanairghat station (0.075). Mean rainfall with their standard deviation indicates rainfall is fluctuating with time. The outcomes from this study will assist water engineers and hydrologists to establish strategies, priorities and proper use of water resources in Sylhet and Moulvibazar districts.

Indian summer monsoon rainfall (ISMR) is a crucial factor affecting the economy of the country. Prediction of ISMR is crucial for formulating strategies of agriculture and its planning and for better management of water resources of the country. Thus, in the present study, significant lagged circulation indices affecting rainfall of months of ISMR are determined by application of linear correlation, removal of multi-collinearity present among significant lagged circulation indices, and formulation of monthly composite index (MCI). Four models are formed for assessment of hydro-climatic teleconnections between rainfall of months of ISMR and different lagged circulation indices, each having model development phase and testing phase. Four models are having model development phase periods 1950–1999, 1950–1994, 1950–1989, and 1950–1984. The MCIs derived for the different months of ISMR for all four models are used for prediction of rainfall for the period of 2000–2014, which is a common testing period for all the four models. From the study, it is observed that significant lagged circulation indices used in development of MCIs for rainfall of each month of ISMR are changing with respect to time periods, having some common indices and some uncommon indices. Also, in the present study, circulation indices in addition to El Nino-Southern Oscillation (ENSO) and Equatorial Indian Ocean Oscillation (Equinoo) which are affecting rainfall of each month of ISMR significantly are determined for better prediction of ISMR.

Although the variability and prediction of rainfall is an essential issue of the Santhal Pargana Division of the Jharkhand State but the issue is still far from its’ conclusive statement till date. Therefore, this study aimed to simulate the monthly rainfall from 1901 to 2020 using an eight-step procedure. After downloading the monthly rainfall for the Santhal Pargana Division from 1901 to 2020, the TBATS and Naive models were used to simulate the rainfall. The accuracy assessment of each model was done by using the MASE, MAE, RMSE, ME, and R. For the Naïve model, the Godda station was noticed with a comparatively high combined error. The lowest combined error was found for the Pakur station in case of Naïve models. Similar result was also obtained for the TBATS model. The TBATS was found with comparatively higher accuracy, as the combined error was less for the TBATS. The spatial assessment for the standardized rainfall varied from 84.419 mm. to 149.225 mm. For the Naïve predicted model, the rainfall was marked in between 8.133 mm. to 67.059 mm. For the TBATS fitted model, the rainfall fluctuated from the 37.127 mm. to 62.993 mm. Dumka station was noticed with comparatively low rainfall (i.e.,37.127 mm.). Deoghar and Jamtara stations were marked with a moderate rainfall. Remaining stations were marked with higher amount of rainfall for the TBATS fitted model. The Wilcoxon test proved that each model was significant at 95% confidence interval. The result produced in this research is fruitful enough to be utilized for agricultural planning in the Santhal Pargana Division of the Jharkhand state, India.
 Keywords : TBATS model; Naive model; simulation; accuracy
 
 Copyright (c) 2022 Geosfera Indonesia and Department of Geography Education, University of Jember
 This work is licensed under a Creative Commons Attribution-Share A like 4.0 International License