The global climate is undergoing significant changes. Over time, there have been substantial in the weather. Climate change has caused rainfall to become unpredictable. The frequency of severe weather phenomena such as drought and flood has escalated due to climate change in Pakistan. Demanding more accurate and timely rainfall prediction. Rain forecasting is essential for strategic purposes such as agriculture, water resources management and agricultural design. The inherent non-stationary component in the rainfall time series hinders the effectiveness of models for hydrologists and assessors of drought risk. We propose a FBP model approach for predicting rainfall to tackle the problem of forecasting. Our research model involves forecasting the likelihood of rain the following analyzing data from the past dataset from 1960 to 2020. The FBP model is mostly used for time series data, and the result was a 99% confidence level. The standard deviation (94.34) and the RMSE and MAE values are around 45 and 50.6, respectively. Such a model result is quite good, and a recommendation for the Metrological department to use this for forecasting rainfall. The model’s performance will significantly enhance the accuracy of the rain forecast.

FBP: Facebook Prophet Model; ML: Machine Learning.

Rainfall is a highly studied part of hydrology and meteorology due to its significant short-and long-term impact on society. It is considered a vital factor in these fields. Within the hydrological cycle, precise rainfall forecasting is crucial for assessing risks, preventing catastrophic catastrophes, and managing water resources in everyday life. However, due to the dynamic and unpredictable nature of recorded meteorological data, forecasting rainfall is a complex task. Time series data in the fields of hydrology and meteorology can be predicted using datapowered technological systems and models. These systems are driven by a process, but they have their limitations. For instance, the former is not capable of expanding to incorporate new regions due to its requirement to complete complex computations involving large amounts of data. On the other hand, the latter may be quickly constructed and implemented due to its ability to efficiently extract enduring patterns of real events from data. Deep Learning (DL) and Machine Learning (ML) approaches, commonly used in sectors such as big data and the Internet of Things (IoT), are instrumental in developing data-driven models [1,2]. Data science, blockchain technology, computational genetics, and medical care [2].

Traditional mathematical frameworks are limited to effectively predicting linear or nearly linear time series and need help capturing nonlinear and irregular variables linked with the data. Using time series data of hydrology and meteorology, alternative approaches such as the Autoregressive Integrated Moving Average (ARIMA) and Exponential Smoothing models may yield less than optimal results. ML techniques have strong learning capabilities and are highly effective at predicting linear and nonlinear correlations in time-series data without understanding the underlying causes. Various machine learning (ML) and deep learning (DL) models, such as Neural Networks (NNs), Support Vector Regression (SVR), Linear Regression (LR), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM), have been employed to predict time series data in the domains of hydrology and meteorology. The anticipated results of unpredictable hydrological and weather patterns from the models above are favorable. While the accuracy of these individual strategies in predicting the future has increased, their shortcomings have also become apparent. For example, the SVR algorithm has limitations in accurately predicting highly irregular hydrological time series at different levels. The accuracy of the predictions is also affected by the amount of input data and the specific settings and kernel parameters used. The nonlinear solid projection abilities of neural networks (NNs) have significantly increased their application in predicting environmental information. 

However, as the data volume increases, the NN’s topology complexity also increases, resulting in a significant slowdown in computational speed. This can lead to convergence to regional minimums, reducing the accuracy of the forecasts [3]. ML approaches are commonly used to improve the dependability of time series forecasts, especially for rainfall data. The quantity Precipitation has been significantly affected by ongoing climate changes, more so than other environmental factors. In prior years, there have been significant temporal fluctuations and changes in the distribution of rainfall. This includes large downpours occurring during rainy droughts in dry periods and an overall decline in the total amount of rainfall [4]. These changes have become increasingly severe in the past few decades as a result of ongoing fluctuations in the environment. The rate of environmental change is expected to accelerate in the next years while it continues to undergo transformation. The accelerated rate of environmental change could adversely affect the anticipated precipitation, leading to significant effects. To predict future rainfall, experts should not rely solely on datadriven methodologies based solely on historical data [5].

Extreme precipitation is a significant natural hazard as it initiates degradation processes such as severe erosion, landslip triggering, and flash floods, which is problematic. Significant hazards to both human lives and physical assets. Recent research suggests that there have been alterations in the strength and occurrence of extreme events worldwide [6-9]. Assessing climate just by these methods may not be advisable, as it is also necessary to identify the distinctive features of extreme events within it. Means refers to the method or process by which something is accomplished or achieved. Therefore, if the climate is dynamic, the climatic extremes would also be dynamic. An extreme event is an occurrence that has an exceptionally high (or low) value and a very low likelihood of happening. The extremes of climatic variables are significant due to their potential for substantial human effect. Therefore, it is necessary. It is calculated using statistical methodologies. Monsoons can occasionally become a hazardous phenomenon and potentially cause significant destruction when they occur with full force [9-13].

In Pakistan, climate change has led to increased unpredictability in rainfall patterns, exacerbating the frequency and intensity of droughts and floods. Existing forecasting methods must help capture the non-stationary nature of the rainfall time series data, hindering their accuracy. This limitation poses challenges for critical sectors like agriculture and water resource management. We use the FBP model to forecast the result, which is quite good for the time series data. The rest of the paper has an introduction, methodology, result, discussion, and conclusion.

FBP is an efficient and fully automated prediction method that eliminates the need for manual chores. Due to its resilience to data shortages and modifications in the pattern, this method successfully handles irregularities in time series data. The FBP model establishes upper and lower limits that organize the information into the most suitable framework for anticipated rainfall patterns. The FBP platform needs to have a local perspective and provide a means for individuals to access it. This limitation hinders the ability to broaden understanding and is crucial for predicting the near future. NP, a modified version of Prophet, uses deep learning frameworks, such as autoregressive neural networks, for time series prediction. FBP model maintains the architectural approach provides the same essential system components. Autoregressive methods seek to forecast the future results of a parameter by analyzing its previous measurements.

The authors utilized the FBP model to forecast rainfall. The results indicated that the FBP technique was used exclusively for estimating values based on observable data. The results showed that the FBP model technique had outstanding effectiveness, especially in periods of little rainfall, due to its multiplicative periodicity feature. It has been found that dry season reconstruction can be done more effectively utilizing FBP compared to rainy events. This may be due to an enhanced understanding of the multiplicative periodicity characteristic of FBP during dry seasons resulting from improved trainingrelated knowledge [14]. The authors utilized the FBP model to forecast rainfall. The findings revealed that the Facebook model accurately forecasted rainfall, with Root Mean Square Errors (RMSEs) varying between 1.24 and 7.31. They utilized the FBP model to predict the low-frequency components [15]. FBP model is utilized not just for forecasting rainfall but also for estimating wheat yield in many applications [16], Heart disease [17], air quality [18].

STUDY AREA

Pakistan is situated within the latitudinal range of 24°C to 37°C north and the longitudinal range of 61°C to 75°C east. Pakistan experiences a subtropical, semi-arid climate. Pakistan has substantial yearly variations in precipitation, with the southern plains receiving 125 mm and the sub-mountainous regions and northern plains receiving 500 to 900 mm. The majority of Pakistan’s annual rainfall, accounting for almost 70% of the total, occurs from July to September. In contrast, the remaining 30% of Pakistan’s total precipitation occurs during the winter season. The Pakistan Agricultural Research Council (PARC) has categorized the country into ten specific agro-ecological zones, namely the Indus delta, southern irrigated plain, sandy desert, northern irrigated plain, barani (rainfall) land, wet mountains, northern dry mountains, western dry mountains, dry western plateau, and Sulaiman piedmont. Multiple criteria, including geography, climate, agricultural land utilization, and water accessibility, determine these classifications [19].

FBP model is a forecasting method that uses an additive model to predict time series data. It incorporates non-linear trends and includes yearly, monthly, and daily seasonality and holiday effects. It is most effective when used in time series data exhibiting pronounced seasonal patterns and having multiple seasons of historical data available. The Prophet model is resilient to missing data and trend variations, and it handles outliers effectively. If many targets, ranging from tens to hundreds or even thousands, require completion, FBP’s forecasting speed is relatively slow when forecasted simultaneously. Impoverished. If there are significant components in addition to holidays, when there are changes in demand, FBP is not adequate. Not following a Gaussian distribution FBP prohibits the occurrence of noisy dispersion. Regarding remaining food, FBP does not consider heterogeneity. FBP does not infer heterogeneous patterns. FBP stands for Efficient and effective time-series modelling and analysis. The prediction methodology was developed by Central Data, which Zuckerberg owns. The science department in 2017. The main objective of this framework is to manage non-uniform time-series datasets. Incorporate components such as patterns, variations in demand, and Effects of festivals. Significant fluctuations in the seasons within the Datasets enhance the model’s efficiency. Furthermore, it is also a robust model capable of handling abnormalities and inaccuracies and data or facts that provide knowledge or details about a particular subject.

            FBPT = GT + ST + HT + ET

The input dataset, FBPT, consists of a single-variable time series. The function GT reflects the trend connected with the data. ST represents the seasonal character present in the data. HT refers to the repercussions of holidays that occur on one or more days with possibly irregular schedules. The error term ET is typically modelled as a distribution with standard features. It accounts for any unexplained variation in the data not captured by the predictive algorithm.

FIG: Theoretical Framework.

This study uses secondary data from 1901-2020 from different geographical stations in Pakistan. The data was provided by the Pakistan meteorological Department.

The FBP model and its settings and features are used for rainfall prediction. After completing the data preparation process, the revised dataset is considered. The model’s final component must generate a specific amount of rainfall. Initially, time-series rainfall data was gathered from the Pakistan Metrological Department. After collecting the data, an exploratory data analysis was conducted. The data was divided between training and testing sets at 80 to 20. Subsequently, the dataset was subjected to individual application of the FBP Model to forecast annual rainfall in Andhra, Pakistan. The findings were interpreted and analyzed to identify any inaccuracies, and the performance was evaluated using the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics.

The predictive accuracy of the developed model for forecasting monsoonal rainfall in Pakistan is high. For example, the estimated and actual rainfall shows a firm agreement with a correlation coefficient of 0.76, which is statistically significant at a 99% confidence level. In addition, the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) have values of around 45.0 and 50.6, respectively. Given that RMSE and MAE values greater than zero indicate imperfect precision, we consider these measures in conjunction with other descriptive measurements like SD (standard deviation). The RMSE and MAE values for the validation period are lower than the measured rainfall’s standard deviation (SD) (94.34). This indicates that mistakes do not have a significant impact on the accuracy of the forecast.

Figure 1: A village underwater in Sindh province Pakistan. Over 4.5 million residents of this province alone have been displaced due to the flooding. (Photo: Emmanuel Guddo/Concern Worldwide).

Figure 2 displays the anticipated rainfall variability using the FBP model. The FBP model demonstrates reduced intra and intervariability compared to the observational data, which is evident due to its ability to estimate. The mean values are calculated without considering the variability. The highest amount of rainfall recorded for 15 years did not surpass 150 mm; however, the observational data exceeded 140 mm on two occasions. The highest recorded rainfall in 2006 occurred twice and ranged between 100 and 150 mm. In all other years, the maximum recorded rainfall was below 150 mm. A higher quantity of rainfall is recorded throughout the summer each year (June-JulyAugust). However, in specific years, September also experiences significant rainfall. Furthermore, it is observed that the highest amount of precipitation occurs in the late winter and early spring (February-March) each year, following the summer season. Examining the lower panel of Figure 3 shows that the projected rainfall closely aligns with the observed rainfall pattern. Both datasets exhibit the highest levels of rainfall in July and August. Based upon the results the results are align with previous study the study [20]. The inter-annual rainfall variability reached its peak in the year 2010, which experienced severe flooding in Pakistan as a result of intense monsoonal rainfall. The level of variation. The data indicates that the values were also elevated in the years 2001, 2006, 2007, 2013, 2014, and 2015. The majority of these years are characterized by flooding [21].

Figure 2:  Annual cycle of the Rainfall from 1960 to 2020.

Figure 3: Line chart of the Rainfall.

In this step, we applied Meta’s time-series forecasting engine, FBP model, to our dataset and examined the projected rainfall ranges for the upcoming years. As from the data we observed that the rise in the rainfall trend shown in Figure 5. In Figure 6 show the trend of the rainfall as we observed that the rainfall was rising and show as the yearly and monthly trend of the rainfall. Due to climate change the rainfall was rising due to melting of the glaciers and global temperature rising. Figure 5 show the scatter plot show the data from 1960 t0 2020. The result was align with the previous research [22]. Global warming directly affects the variations and variability of precipitation, which challenges developing countries’ regional food and water security. The current moment in time.

Fig: Time series forecasting.

Figure 4: Scatter plot for the Rainfall Data.

Figure 5: Actual and predicted Rainfall.

Figure 6: Monthly and yearly trend line show in the Rainfall data.

This study aims to investigate the patterns and geographical variations of precipitation in Pakistan by analyzing historical data from 1960 to 2020. Pakistan experiences an annual precipitation of 500 millimeters, with a small area receiving over 1,000 millimeters. The monsoon precipitation in the country varies from 50 and more than 450 millimeters. The overall precipitation in Pakistan comprises around 90% of the annual precipitation. This consists of 60% from the Indian monsoon during the summer and 30% from the western disturbances during the winter. The findings indicate that the summer monsoon is the primary contributor to the overall annual precipitation in Pakistan, The winter season follows in terms of its contribution [23-25]. As previously mentioned, the country experienced significant droughts following the 1990s. From 1989 to 2016, there has been a decrease in annual precipitation at many elevations, except for areas below 250 meters and between 1,500 and 2,000 meters. There is a noticeable and modest increase in precipitation with elevation for the stations located in the monsoon region of the country. The rise in precipitation could be attributed to the amplification of the temperature difference between land and sea, leading to an enhanced moisture movement from the ocean to the land [26]. Temperature rises enhance the capacity of the atmosphere to store moisture, leading to increased precipitation. This is caused by a decrease in the number of wet days and an increase in the frequency of precipitation episodes [27,28].

To achieve sustainable growth in Pakistan, it is necessary to forecast monsoon rainfall accurately every year. In this study, we utilize an FBP statistical forecast model to predict the inter- and intra-annual variability of monsoonal rainfall in Pakistan and identify potential drivers of this variability. Prompt and accurate rainfall forecasting enables improved supervision of hurricanes, agriculture, water supply, power generation, and infrastructure planning and development. However, because of its exceptional instability in different times and locations, predicting this essential meteorological element has become a highly challenging task in terms of accuracy. Over the past few decades, there has been a rise in the seasonal and regional variability of rainfall due to ongoing climate variations. In recent years, the global community of climate experts has become increasingly preoccupied with predicting rainfall. The extreme precipitation results in floods and disasters. This work focuses on utilizing FBP Model-based techniques for time-series forecasting. The Model was evaluated to predict the rainfall the Model’s performance is assessed in terms of the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). After evaluation, the FBP Model provided accurate predictions as the anticipated values were closer to the actual values.

1. Desanamukula VS, Priyadarshini MA, Srilatha D, Rao KV, Kumari RL, Vivek K. “A Comprehensive Analysis of Machine Learning and Deep Learning Approaches towards IoT Security”. International Conference on Electronics and Sustainable Communication Systems (ICESC). 2023; 1165-1168.

2. Vivek K, Rao MS, Kiran P, Nambi PA, Kumar CK. “Artificial Neural Network Based Effective Detection of Breast Cancer By Using Mammogram Data”. International Conference on Innovations in High Speed Communication and Signal Processing (IHCSP). 2023; 437- 440.

3. Basha SJ, Prasad GLV, Vivek K, Kumar ES, Ammannamma T. “Leveraging Ensemble Time-series Forecasting Model to Predict the amount of Rainfall in Andhra Pradesh”. International Conference on Artificial Intelligence and Signal Processing (AISP). 2022; 1-7.

4. SJ Basha, T Ammannamma, K Vivek, VS Veesam. “Comparative analysis of time series forecasting models to predict amount of rainfall in Telangana”. International Conference on Advanced Computing and Communication Systems (ICACCS). 2022; 1: 1918-1922.

5. Prasad GLV, Teja BR, Govathoti S, Dhanikonda S. “Leveraging ARMA and ARMAX Time-Series Forecasting Models for Rainfall Prediction”. International Conference on Advanced Computing and Communication Systems (ICACCS). 2023; 1: 353-357.

6. Colombo AF, Etkin D, Karney BW. “Climate variability and the frequency of extreme temperature events for nine sites across Canada: implications for power usage”. Journal of Climate. 1999; 12: 2490-2502.

7. Parry ML. Climate change 2007-impacts, adaptation and vulnerability: Working group II contribution to the fourth assessment report of the IPCC. Cambridge University Press. 2007.

8. Leary N, Dokken D, White K, Berger A. “Climate change 2001: impacts, adaptation, and vulnerability”. Intergovernmental panel on climate change. 2001; 301-305.

9. Sivanantham S, Kumar PH, Vardhan SN, Kumar SC, Kumar TA, Pradeep T. “Rainfall Prediction using Machine Learning Techniques–A Comparative Approach”. Third International Conference on Artificial Intelligence and Smart Energy (ICAIS). 2023; 1472-1480.

10. Rasul G, Chaudhry QUZ, Sixiong Z, Qingcun Z. “A diagnostic study of record heavy rain in twin cities Isl?m?bad-R?walpindi”. Advances in atmospheric sciences. 2004; 21: 976-988.

11. Rasul G, CHAUDHRY Q, Zeng Q, Zhao S, Shouting G. “Interaction of a mesoscale low and diffused tropical depression during south Asian summer monsoon”. Mausam. 2008; 59: 453-460.

12. Babar ZA, Zhi X, Zhang L, Ge F. “Determination of precipitation return values in connection to severe rainfall event of 2010 in Pakistan”. International Conference on Advanced Computational Intelligence (ICACI). 2012; 840-844.

13. Muneer MS, Mustafa SMN, Zehra SS, Maqsood H. “Rain Predictive Model using Machine learning Techniques”. International Multidisciplinary Conference in Emerging Research Trends (IMCERT). 2023; 1: 1-6.

14. Hossain MM, Anwar AF, Garg N, Prakash M, Bari M. “Monthly rainfall prediction at catchment level with the Facebook Prophet model using observed and CMIP5 decadal data”. Hydrology. 2022; 9: 111.

15. Li D, Jinfeng M, Kaifeng Rao, Xiaoyan Wang, Ruonan Li, Yanzheng Yang et al. “Prediction of rainfall time series using the hybrid DWTSVR-Prophet model”. Water. 2023; 15: 1935.

16. Mittal Desai, Amisha Shingala. “Time Series Prediction of Wheat Crop based on FB Prophet Forecast Framework”. ITM Web of Conferences. 2023; 53: 02014.

17. Yi Lu. “Heart Disease Prediction Model based on Prophet”. Highlights in Science, Engineering and Technology. 2023; 39: 1035-1040.

18. Fangjie Cai. “The Prediction of the Air Quality based on the Prophet Algorithm”. Highlights in Science, Engineering and Technology. 2023; 39: 1056-1060.

19. Khan Imran, Hongdou Lei, Khan Ahmad, Muhammad Ihsan, Tehseen Javeed, Khan Asif, et al. “Yield gap analysis of major food crops in Pakistan: prospects for food security”. Environmental Science and Pollution Research. 2021; 28: 7994-8011.

20. Muhammad Adnan, Firdos Khan, Nadia Rehman, Shaukat Ali, Sher Shah Hassan, Muhammad Mubashar Dogar, et al. “Variability and predictability of summer monsoon rainfall over Pakistan”. Asia-Pacific Journal of Atmospheric Sciences. 2021; 57: 89-97.

21. Muhammad Adnan, Nadia Rehman, Shaukat Ali, Shahbaz Mehmood, Kaleem Anwar Mir, Aftab Ahmad Khan, et al. “Prediction of summer rainfall in Pakistan from global sea?surface temperature and sea- level pressure”. Weather. 2017; 72: 76-84.

22. S Tewari, A Kaur, NN Das, M Bhatia. “Predict the Temperature  Rise using Facebook’s Prophet Model”. International Conference  on Advances in Computation. Communication and Information  Technology (ICAICCIT). 2023; 1319-1323.

23. Safi Ullah, Qinglong You, Waheed Ullah, Amjad Ali. “Observed changes in precipitation in China-Pakistan economic corridor during 1980- 2016”. Atmospheric Research. 2018; 210: 1-14.

24. Waheed Ullah, Guojie Wang, Dan Lou, Safi Ullah, Asher Samuel Bhatti, Sami Ullah, et al. “Large-scale atmospheric circulation patterns associated with extreme monsoon precipitation in Pakistan during 1981–2018”. Atmospheric Research. 2021; 253: 105489.

25. Kamal Ahmed, Shamsuddin Shahid, Eun-Sung Chung, Tarmizi Ismail, Xiao-Jun Wang. “Spatial distribution of secular trends in annual and seasonal precipitation over Pakistan”. Climate Research. 2017; 74: 95-107.

26. F Ikram, M Afzaal, S Bukhari, B Ahmed. “Past and future trends in frequency of heavy rainfall events over Pakistan”. Pakistan Journal of Meteorology. 2016; 12.

27. Somayeh Salehi, Majid Dehghani, Sayed M. Mortazavi, Vijay P. Singh. “Trend analysis and change point detection of seasonal and annual precipitation in Iran”. International Journal of Climatology. 2020; 40: 308-323.

28. Azfar Hussain, Jianhua Cao, Shaukat Ali, Sher Muhammad, Waheed Ullah, Ishtiaq Hussain, et al. “Observed trends and variability of seasonal and annual precipitation in Pakistan during 1960–2016”. International Journal of Climatology. 2022; 42: 8313-8332.