Rabia ISST PhD Thesis - Drought in Pakistan (PDF)

1.1 Background Drought is an insidious natural hazard more influenced by climate change, which is now a globally recognized phenomenon (Handmer et al., 2012; Mokhtarzad et al., 2017). As a global challenge, climate change impacts the environment, traditions, human lifestyles, and health. In mo...

1.1 Background Drought is an insidious natural hazard more influenced by climate change, which is now a globally recognized phenomenon (Handmer et al., 2012; Mokhtarzad et al., 2017). As a global challenge, climate change impacts the environment, traditions, human lifestyles, and health. In most parts of the world, rising temperatures and changing rainfall patterns are the significant causes of drought severity and its frequent occurrence (Duffy et al., 2015; Gocic & Trajkovic, 2014; Zhao & Dai, 2015). Climate variables that influence drought characteristics assessment are critical for adapting to shifting drought patterns caused by climate change (Ahmed et al., 2018). Drought’s frequent occurrence and profound influence crucially limit economic and social development. Therefore, it is very crucial to keep track of and forecast dryness accurately for forewarning and hazard alleviation (Zhang et al., 2020). Droughts, along with other natural hazards, are getting more frequent and pose risks on regional, local, and global levels. Droughts occur in practically all environments around the world, causing adverse effects on plants, humans, and wildlife (Paulo et al., 2012). Droughts usually occur in those areas, which experience less rainfall than usual over an extended time. Droughts are grouped into four types: hydrologic, agricultural, meteorological, and socioeconomic (Mishra & Singh, 2010). The rainfall deficiency and a temperature rise (heat wave) cause Meteorological droughts (Agnew, 1990). Soil water deficiency is the main reason for Agricultural droughts occurrence (Agnew & Warren, 1996). Machine Learning (ML) is becoming common in every field of life; therefore these techniques are also beneficial for drought prediction. ML is an auspicious approach to knowing the complex nature of drought and its related factors. A short-term drought was predicted by researchers utilizing Deep Belief Networks (DBNs) and time information from several time-scale SPI (Chen et al., 2012; Agana & Homaifar, 2017a). Ali et al., (2018) employed ground-based products and remotely sensed data to predict upcoming drought episodes in Pakistan using the Comm-ELM model (Ali et al., 2018). Prodhan et al., (2021) assessed agricultural drought utilizing the Soil Moisture Deficit Index (SMDI) 2 and used rainfall, crop, and soil as input parameters for the deep forwarded neural network (DFNN). Using a single reanalysis dataset, Khan et al., (2020) demonstrated the feasibility of creating drought forecast models over Pakistan (NCEP/NCAR v1). ML techniques; SVM, ANN, and KNN are used and SVM found a better model as compared with others. (Khan et al., 2020). 1.2 Study Area Pakistan (latitudes 23°30ˈ N–37°30ˈ N and longitudes 61°E–78°E), is situated in South Asia with a covered area of 796,095 km2. Pakistan is a moderate zone, and its climate variations are similar to its topography—generally hot and humid at the coastline and the Indus River's lowland plains, and gradually cooler in the northern mountains and Himalayas (Khan, 2019). Because of its northern location, the country experiences four seasons that are differently characterized by temperature: the dry and cool winter (Dec- Feb), dry and hot spring (Mar-May), humid and hot summer (Jun-Aug), and dry autumn (Sep-Nov) (Khan, 2019). There are two wet monsoon seasons in Pakistan: the western disturbance (December-March) and the Indian monsoon (July-September)( Khan , 2019). India's monsoon (July–Sep) and Pakistan's western disturbance (Dec-Mar) are the two monsoon seasons. Except for the northern parts, where monsoons may bring as much as 200 mm of precipitation each month from July to September, almost the entire country observes minimal precipitation. Inter-annual precipitation varies substantially, often resulting in repeated occurrences of flooding and drought. El Niño significantly affects variations in Pakistan’s climate, leading to anomalies in temperature and flood frequency. Figure 1 depicts the Pakistan map with mean monthly measurements of surface air temperatures and precipitation during 1991-2022. A prominent pattern shows throughout the months, illustrating the seasonal fluctuations that reflect the area's environment. Temperatures rise gradually from the cooler months of January and February, increasing in July and August, before dropping gradually as autumn draws near. This pattern indicates the periodic structure of the climate in Pakistan. In June, the maximum monthly temperature peaked at 30.13⁰ C, whereas July and August experienced the greatest amounts of precipitation of approximately 50mm from 1990-2022. Precipitation values varied throughout the year, with spikes in the summer and lowest levels in the fall and the 3 beginning of winter. These monsoon patterns illustrate an association between climate variables and seasonal shifts, influencing regional ecosystems and human activities. In Pakistan, agricultural activity is one of the most important areas of the economy, accounting for over 70% of the population's income, whether directly or indirectly related to agriculture. Pakistan, with an estimated population of over 160 million and 32 percent residing below the poverty threshold, faces tremendous challenges (ESP, 2007). The climate of the country is continental, with substantial variation due to differences in topography and altitude (Khan et al., 2010). The population of Pakistan and its economic circumstances have fluctuated significantly during the past few decades. Being a hazard- prone country having a growing population residing in regions at risk, it is becoming more essential to deal with both man-made and natural hazards effectively. This involves minimizing the cumulative risks related to such hazards in multiple temporal and spatial domains as well. Figure 1: Study Area Map with monthly mean Temperature and Precipitation (1991-2022) 4 1.2.1 Drought in Pakistan According to Germanwatch, the publisher of the Climate Change Performance Index (CCPI), Pakistan is the seventh most susceptible nation to climate change, and due to insufficient resources for recovering from large-scale threats, it is particularly hard hit by the negative effects of climate change (Javid et al., 2019). Climate Change in Pakistan causes a temperature rise, a reduction in precipitation in the arid areas and increases in the monsoon regions, and lastly, accelerated glacial melt (Ahmed & Schmitz, 2011). Climate and topographic diversity are the primary reasons for past disasters, which mainly occur seasonally and impact the various regions of Pakistan. Droughts and floods are most common in the south of Punjab and Sindh, whereas Baluchistan is susceptible to droughts, earthquakes, and flash floods. Furthermore, Khyber Pakhtunkhwa is vulnerable to earthquakes, landslides, avalanches, and flooding. Javid et al., (2019) designed five new major classes, i.e., Aridity, Drought, Humidity, Cold Drought, and Wetlands, using remote sensing climate compound indices (RSCCI). The results prove that major areas of Sindh, Baluchistan, and southern Punjab are indicated as drought-prone areas. According to Ullah (2016) and Javid et al. (2019), such catastrophes will likely impact Pakistan more severely and frequently in the coming years (Ullah, 2016). Therefore, some steps need to be implemented to avoid the negative effects of forthcoming climatic problems such as droughts, flooding, heat waves, landslides, and cyclones (Javid et al., 2019). Pakistan is frequently experiencing an extreme surge in natural disasters causing human and wildlife loss, adverse effects on the social and psychological fiber of society, and property and infrastructure destruction. Moreover, drought is Pakistan’s most frequent natural hazard (Anjum et al., 2012; Pasha, Ali, & Waheed, 2015). Many researchers monitored drought using different indices and remotely sensed data in Pakistan and verified the possibility of drought occurrence in various areas of Pakistan. Tabassum et al., (2015) monitored drought in hot and arid areas of Pakistan using different drought indices extracted from remotely sensed data and observed severe drought in those regions. Adnan et al., (2015) and Ahmed et al., (2016) characterized drought using SPI. Akber et al., (2019) used statistical and meteorological indices whereas Adnan et al., (2018) used 15 indices for drought assessment and revealed that SPEI and RDI are the more suitable indices. 5 Ahmed et al., (2018) utilized SPEI to investigate the different characteristics of droughts caused by climate change during two significant cropping seasons. Ashraf and Routray (2015) investigated the temporal as well as spatial variability of drought in Balochistan and estimated the time series of drought index from monthly precipitation data using the Mann– Kendall test. I. Khan et al., (2016) used remote sensing MODIS-NDVI data and climate variables to assess drought in Pakistan and its surrounding areas. Pasha et al., (2015) study concluded that the major reason for drought in Sindh are continued lack of rainfall; however, the extraordinary destruction was due to the complete negligence of administrative authorities and the Sindh Government. To assess the negative impact of drought conditions in Thar, a deserted region of Sindh, Siddiqui and Safi, (2019) obtained the data by taking the views of local people. The Authors observed that drought is a natural phenomenon, but man-related activities, such as forest destruction and the using non- renewable resources for fuel, cannot be ignored. Drought has affected particularly poor people who live in countryside regions and rely significantly on agriculture because water deficiency directly influences agriculture. Adnan et al., (2015) identified the extreme historical drought years with the help of a drought hazard map of Sindh and found that 10 districts of Sindh were identified as highly susceptible to droughts. Drought has occurred in Pakistan at least once every three years in the past five decades and the bulk of its people continue to work in agriculture and agro-based industries (Khan et al., 2016). 1.3 Significance of the Study Drought represents one of Pakistan's most significant hazards, and its early prediction helps to establish the most effective mitigation strategy. The critical point highlighting the importance of this proposed research is that it presents deep knowledge about drought monitoring and prediction techniques that are extremely useful for water supply decision- makers and managers in managing catastrophes. The study results will be crucial for government agencies to recognize the forecasting capability of natural hazard occurrence and to develop sustainable management systems that provide safety from the catastrophic consequences of drought. Moreover, the study will provide valuable information about the best ML model for future drought prediction, which could be helpful for data scientists, researchers, disaster management agencies, engineers, and meteorologists. 6 1.4 Statements of Problems With global climate change, drought hazards occur frequently all over the world. Drought events monitoring, as well as drought prediction, are crucial contents for hazard management, planning, and water resource system management. An appropriate drought assessment helps limit the adverse effects on water resources, agriculture, and ecosystems (Mokhtar et al., 2021). There is a need for deep investigation into the possibility of future drought occurrence and monitoring past drought events to develop better management policies to cope with its catastrophic consequences. Moreover, ML is an emerging technique that provides better prediction than other statistical prediction techniques. To monitor and anticipate drought situations, historical data sets for climate characteristics must be analyzed. The various climate and satellite indicators like precipitation, snow cover, relative humidity, wind speed, sunshine hours, evapotranspiration, LST (day), LST (night), and soil moisture are efficient and helpful in predicting drought events by utilizing ML models. ML techniques present an efficient method for resolving drought management problems. (Harms et al., 2002) employed advanced statistics analysis as well as modeling techniques to identify drought-related predicting and descriptive patterns. Once a successful approach for abstracting and extracting suitable knowledge from these massive data sets is developed, earth scientists, agriculturists, and policymakers are going to be able to utilize data to improve drought mitigation and management. 1.5 Objectives The basic aim of this research is to model spatiotemporal drought over Pakistan’s agricultural land using ML techniques. Therefore, drought monitoring and prediction have been performed using remotely sensed and climate data. To obtain this objective, it is essential to fulfill the following sub-objectives: Drought monitoring is done by considering the different drought indices from remotely sensed data. Identify the drought-prone region based on different drought classifications i.e. Mild, Moderate, Extreme, and Severe drought. 7  Time series analysis is used to deeply monitor the drought condition over a certain period as well as its trend. Drought prediction using various ML Techniques and comparing the models to select the best model. 1.6 Research Questions The following research questions must be addressed to reach the objective: What are the spatio-temporal drought patterns in the area of interest? What is the relation between drought pattern function parameters and drought characteristics like onset, severity, persistence, and extent? What are the duration of past drought severity events that occurred in the past? What is the possibility that drought episodes will occur in the future? Which ML model is the most suitable to predict drought severity events? 1.7 The Framework of Dissertation This dissertation is organized into six chapters, each structured to address specific research objectives and methodologies, ensuring a coherent flow of information and analysis. The progression from the literature review to the application of methodologies and finally to the discussion of results and implications ensures a logical and coherent narrative throughout the dissertation. Chapter 1: Introduction: The introductory chapter provides a general overview of the research, outlining the background and context of drought conditions in Pakistan. It highlights the study's significance, presenting the problem statements, research objectives, and questions. This foundational chapter sets the stage for the subsequent chapters by establishing the need for effective drought monitoring and prediction using advanced techniques. 8 Chapter 2: Review of Literature: The literature review chapter offers a comprehensive overview of existing drought monitoring and prediction research. It covers various methodologies and tools, including ML models and remote sensing data. This chapter critically evaluates previous studies, identifies research gaps, and justifies the selection of methodologies employed in this study. It also highlights the use of ML for drought prediction in Pakistan. Chapter 3: Monitoring Spatial Drought Events and Time Series Analysis Using Google Earth Engine: This chapter introduces the Google Earth Engine (GEE) to assess spatial drought events and perform time series analysis of drought indices (VCI, TCI, VHI). It presents the data acquisition, preprocessing, and systematic processes involved and thoroughly describes the methodologies used to identify drought conditions over time. The GEE integration allows for the effective processing and study of large-scale remotely sensed data. Chapter 4: Machine Learning Modelling for Drought Prediction: This chapter discusses the various ML model’s development and assessment for drought forecast. It consists of the dataset collection, preprocessing stages, model training/testing, and evaluation metrics used to identify the best ML models(Random Forest(RF), Support Vector Machine(SVM), Extreme Gradient Boosting(XGBoost), and Multiple Linear Regression(MLR)). The outcomes of these models are compared, displaying their predictive efficiency and relevance to drought conditions in Pakistan. It integrates updated computational methods to build up the understanding of and predict drought events. Chapter 5: Discussion: This chapter presents the discussion on drought monitoring and prediction highlighting their implications for food security in Pakistan. The research’s findings are compared with current literature, verifying the outcomes and highlighting their significance for ensuring agricultural sustainability and resilience. Chapter 6: Conclusion: The conclusion chapter summarizes the study's main outcomes and contributions. It presents the results' implications for drought assessment and management, providing recommendations for upcoming research and practical applications. It also reflects on the research's overall significance, emphasizing how the 9 integration of remote sensing and ML techniques can improve drought forecast and management strategies. References are provided at the end. 10  Chapter 2 Review of Literature 11 Drought is a serious natural calamity that leads to serious socio-economic damage. Drought is a global risk to food and water security, and it is expected to get worse due to climate change (Nay et al., 2018). Monitoring and predicting regional droughts can assist farmers, water administrators, and others in mitigating the effects of drought. Several studies have been done on drought prediction and monitoring using remote sensing and meteorological data. 2.1 Drought Monitoring Drought is a widespread occurrence that generally happens in all types of geography, seriously harming both the environment and human life. Any region is more susceptible to extreme variations in temperature and precipitation, which can result in climate threats such as cyclones, heat waves, droughts, and floods (Handmer et al., 2012). Droughts are global, regional, and local issues that are becoming more frequent in addition to other natural calamities. Droughts affect practically every region on Earth, harming human life, wildlife, and plants (Paulo et al., 2012). In general, droughts happen in areas that receive less rainfall than normal over a longer duration. Droughts can be categorized into four classes: hydrological, agricultural, meteorological, and socioeconomic (Mishra & Singh, 2010). Meteorological droughts result from both an increase in temperature (hot wave) and a lack of precipitation (Agnew, 1990). Droughts in agriculture are caused by a shortage of soil water (Agnew & Warren, 1996). However, a lack of surface water in lakes, reservoirs, and streams causes hydrological droughts(Dracup et al., 1980), and socioeconomic droughts arise when variations in water availability due to weather lead to a rise in demand for products relative to their supply ( Adnan et al., 2018). The classification of drought types is given in Figure 2. Figure 2: Drought Classifications 12