Data-driven approaches to built environment flood resilience: A scientometric and critical review

Environmental hazards such as floods significantly frustrate the functionality of built assets. In addressing flood-induced challenges, data usage has become important. Despite existing vast flood-related research, no research has presented a comprehensive insight into global studies on data-driven built environment flood resilience. Hence, this study conducted a comprehensive review of data-driven approaches to flood resilience. Scientometric analysis revealed emerging countries, authorships, keywords, and research hotspots. The critical review revealed data-centric approaches such as Machine Learning (ML), Artificial Intelligence (AI), Flood Simulations, Bayesian Modelling, Building Information Modelling (BIM) and Geographic Information Systems (GIS). However, they were mainly deployed in hydraulic flood simulations for prediction, monitoring, risk, and damage assessments. Further, the potentials of computational methods in tackling built environment resilience challenges were identified. Deploying the approaches in the future requires a better understanding of the status quo. These methods include hybrid data-driven approaches, ontology-based knowledge representation, multiscale modelling, knowledge graphs, blockchain technology, convolutional neural networks, automated approaches integrated with social media data, data assimilation, BIM models linked with sensors and satellite imagery and ML and AI-based digital twin models. Nevertheless, reference to data-informed built-asset resilience decisions and clear-cut implications on built-asset resilience improvement remain indistinct in many studies. This suggests that more opportunities exist to contextualise data for built environment flood resilience. This study concluded with a conceptual map of flood context, methodologies, data types engaged, and future computational methods with directions for future research.


Introduction
Sustainable development targets are frustrated by catastrophes such as floods through the destruction of properties, disruption of water, power, and other utility supplies, alongside a general halt to normal community operations [75,130,155]. Records indicate that the number of flood incidents has risen dramatically during the past three decades [67] with devastating effects on built assets [66]. Globally, the economic losses caused by the flooding of built assets are estimated at over $1 trillion since 1980 [63]. The estimated economic losses from flood damages to built assets were reported as 49.8 billion dollars in the United States in 2022, over 333 million pounds in the United Kingdom, and 25.4 million US dollars in Shanghai [75]. These statistics highlight the challenge of flood damage to built assets and the consequential menace to the development of sustainable cities and communities [155]. Changing climates, growing populations, and extensive developments in urban areas threaten to increase these economic losses over the next few decades leaving sustainable city goals more vulnerable [158]. Hence, developing more resilient built assets is a topmost need [104,139].
Resilience is defined as the capacity of assets or the ability of a community to withstand a natural disaster and regain desirable performance after the occurrence of such disasters [61]. Built assets are critical to the normal functioning of communities, they enable access to essential resources and services and provide a platform for social and economic growth which are vital to societal sustainability. Built assets are broadly discussed with the surrounding built environment and, wide-ranging connections between different built environment components, including physical infrastructure (e.g., residential, and commercial buildings, bridges, roads, and utilities), and social infrastructure (e. g., community gathering places and emergency management systems).
It is important to ensure the stability and proper performance of built assets to enhance the social, economic, and environmental sustainability of cities and communities [20,156,155]. However, recurrent flood damages and losses to built assets have made achieving sustainable cities a challenge in terms of social, economic, and environmental developments [155]. Thus, decision-makers in society have highlighted the importance of built-asset flood resilience [13].
Existing resilience strategiesfor built assets are generally based on impact assessments conducted after event occurrence by human assessors or with post-disaster surveys. These conventional approaches require extensive labour and time to complete, which may delay the disaster response and recovery and cause extra losses to disaster-affected communities [75]. Thus, current solutions to increase the resilience of built assets are based on ad hoc procedures [60].
Recently, data-driven methods have been determined as an essential tool that can help enhance flood resilience through context-specific policymaking, administration, and research [135]. Data are a fundamental part of resilience, and data-driven approaches for decisionmaking have lately gained prominence in flood resilience [146]. A data-driven approach as a new paradigm of technology is based on analysing the data about a system, in particular, finding connections between the system state variables (input, internal and output variables) and it is widely regarded as an efficient method for making rapid and effective decisions [135]. The need of careful data pre-processing and feature engineering ability for the extraction of important information from raw data, issue of generalizing to unseen data, data overfitting, lack of causality, data bias, inteoperatability and need of high-quality and comprehensive data are some limitations of data-driven methods. However, data-driven methods have overcome traditional models' drawbacks and demonstrated them to be more accurate in modelling complex engineering and hydraulic needs [9]. Thus, highly integrated data-driven approaches are identified which play a significant role in flood resilience decision support systems through the effective and scientific use of various data sets [105,137]. Data-driven approaches provide a promising opportunity in reducing vulnerability and enhancing flood resilience for built assets, and it is helpful to ensure that problems are discovered before it becomes too late [122].
However, the existing data-driven flood resilience studies have perhaps focused on improving flood risk management through effective flood warnings and disaster relief systems while only a handful of studies have focused on enhancing the flood resilience of built assets ( [63,68]; Saker et al., 2019). These studies yet predominantly focused on property flood resilience measures based on onsite assessments and surveys. Li and Choo (2018) further stated that data that can play a critical role in flood resilience have been widely used for the coordination of people moving away from flood risks, and various data-driven approaches have been investigated so that people's evacuation and rescue can be optimized before and after flood events ( [68,104]; Saker et al., 2019; [120,75]). Despite the tremendous advantages, the use of data-driven methods to improve built asset flood resilience has received less attention [59,135]. The lack of comprehensive reviews on existing studies in this context impedes the available opportunities to identify research areas and potential future directions for further development. Mostly, the existing reviews merely discuss the disaster resilience for different building sectors [30,110], measurements of disaster resilience [21,149] and resilience assessments and governance [111,109,123,34], while there is a lack of reviews on data utilization towards built asset flood resilience. Though there are some other studies which have reviewed data-driven models for flooding resilience in terms of flood forecasting and the utilization of community-scale data for flood risk assessments To address this gap, a broader understanding of existing data-driven studies on flood resilience of the built environment is highly consequential. Based on this, this study has undertaken both scientometric and systematic literature reviews. The review aimed to provide a comprehensive understanding of past, current, and emerging research areas of data-driven research studies on flood resilience. The objectives of this study are to (1) conduct a scientometric analysis to reveal the status quo via details such as publication trend, authorship networks, research hotspots: co-occurrence keyword, document citation analysis, and country contribution analysis (2) to identify data-driven techniques deployed in flood resilience research (3) identify the data types and variables employed in data-driven flood resilience studies, and (4) establish gaps existing literature in the context of (2) and (3) together with the potential future opportunities.

Research methodology
This study conducted a comprehensive review and analysed the existing context of data utilization for flood resilience to identify the current research trends, research gaps and potential future research directions. To achieve this aim, this study was undertaken in two phases including a scientometric review using CiteSpace analysis results and a systematic review for an in-depth discussion. The scientometric review was conducted first by analysing the articles collected from the web of science core collection database, which comprises the most significant, impactful, and leading journal publications. For the preliminary collection of required publications, the retrieval code used in the Web of Science (WOS) core collection is TS = (Flood Risk Management* OR Flood Resilience* OR Built Asset Resilience* OR Flood* OR Flooding* AND Hard Data* OR Big data* OR Smart data* OR Data-driven* OR Data-centric*). Here, "*" denotes a fuzzy search and "TS" means an article subject. In this study, journal articles and conference papers were selected for analysis, while book reviews and editorials were excluded. In addition, publications irrelevant to floods such as biology, medicine, agriculture, other hazards etc. were also excluded. Finally, a total of 932 bibliographic records were collected from WOS from 2002 to 2022.
To analyse the collected bibliographic records the software package CiteSpace was used. CiteSpace is a tool that allows for the analysis and visualization of scientific literature in a particular field of knowledge. This field is broadly defined as a logically and cohesively organized body of knowledge. The use of domain analysis has been acknowledged as an effective method for uncovering important insights from a large amount of information and identifying areas of development [142]. CiteSpace is a strong software package for mapping knowledge domains by systematically creating various accessible graphs (Chen, 2016). Therefore, the latest version of CiteSpace 6.1.R1. was used to analyze the literature relating to data-driven flood resilience to built assets. Mainly, three types of bibliometric techniques were applied in this study: (i) co-author analysis which seeks author co-occurrences, country co-occurrences, and institution co-occurrences; (ii) co-word analysis which processes keywords or terms to analyse word co-occurrences; and (iii) co-citation analysis that identifies co-cited authors, co-cited references, and co-cited journals.
The trend of the number of annually published articles was evaluated first as an important indicator of the current state of research. These publications reflect the attention of researchers and the development of the research field from a macro perspective. Annually published papers collected from the WOS database from January 2002 to December 2022 were analysed and compiled into a line chart, to gain an overall view of data-driven studies for flood resilience. Fig. 1 shows the distribution of 932 bibliographic records from 2002to 2022. The number of published articles exhibited an upward trend from 2002 to 2009 and increased slowly from 2010 to 2017. Even though significant growth is not exhibited in the publication trends, an increase in published articles can be identified in 2019 but drops slightly from 2020.
The review methodology developed for the study is shown in Fig. 2. As mentioned in the review methodology, after the collection of relevant research publications, the scientometric review is conducted using the collected papers and subsequently, a systematic review was carried out with detailed analysis and discussions. The research publications identified from the scientometric review were studied for the systematic review to identify the context of data utilization for flood resilience of built assets. Mainly, data-driven methodologies, types of data and classifications necessary for flood resilience and future computational methods were identified and discussed. Finally, the existing research gaps are put forward, along with potential future research directions for data-driven flood resilience on built assets with conclusions.

Scientometric Analysis, Results, and Discussion
This section mainly discusses the results of the scientometric analysis of the study. The subsections of the analysis are co-author analysis, country analysis, institution analysis, co-word analysis, and co-citation analysis.

Co-Author analysis
The bibliographic records of the Web of Science database were used to extract the information of the article authors to identify the prominent researchers, countries, and institutions for data-driven studies of built asset flood resilience. Accordingly, a co-authorship network, a network of co-authors' countries and institutions was discussed.

Co-Authorship network
A co-authorship network was generated as shown in Fig. 3. Each node in the network represents an author and the links between the authors indicate the established collaboration through the co-authorship in the publications. The pathfinder function was used for network pruning which enables removing excessive and redundant links within the network which is recommended by Chen and Morris [39]. The coauthorship network was generated with 583 nodes and 1008 links. Modularity measures the overall clarity of a given decomposition of the network while the mean silhouette value measures the quality of a clustering configuration ranging from − 1 to 1. The highest value represents the highest quality of the network and thus, above modularity and mean silhouette values indicate the co-authorship network's proper dispersion and structural properties. The modularity of the coauthorship network was Q = 0.967 while the mean silhouette value (S) equals 1..
There are several research communities dispersed throughout the network which represent the researchers who have established cooperation in publications on data-driven flood resilience. Among them, the most highlighted research communities that have established strong collaborations in research publications were selected. In total, four major research communities were identified as highlighted in the figure.
In the first research community, Wang Y and Yang K were identified as central authors including Sung D G and Panahi Z M and Lee S were identified next as central authors including Zhang X Y, Rezaie F, and Pradhan B as the second research community. The third research community include Li Y, Meng L K, and Huang H as central authors including Chen Y, Liu R, Huan C, Xu T B, and Li L. Finally, the fourth research collaboration between Wu Y, Feng J, and Ding Y was identified.
The impact of the authors and collaborations for the research area was further analyzed by considering the citation bursts. The citation burst is calculated based on Kleinberg's algorithm [100] and it measures the sudden increase in citations within a short period. Seven authors were identified with citation bursts in the network. The most productive authors and the number of articles published by the authors were ranked and the burst strength of each author with the corresponding period was summarised in Table 1. The authors with strong burst strengths were identified where the value exceeded 1.8.

Network of countries and institutions
Based on the contributions of countries and institutions, two networks were generated to analyze the distribution of articles on datadriven studies on built asset flood resilience. The network of countries includes 759 nodes and 4539 links while the network of institutions includes 491 nodes and 513 links. The node size represents the total number of articles published from 2002to 2022. The modularity of the network is Q = 0.423 and the mean silhouette value is S = 0.795. As  shown in Fig. 4 two countries in the network have given greater contributions with more than 200 articles including the People's Republic of China (245 articles) and the USA (224 articles). In addition, Germany (60 articles), Japan (54 articles), England (53 articles), India (51 articles), Italy (48 articles), Australia (45 articles), South Korea (41 articles), and Canada (39 articles) have made major contributions to the articles on data-driven flood resilience. However, China and USA have been significant contributors to research publications on the study area.
Further, countries with high betweenness centrality scores were identified from the network. A node with a high betweenness centrality connects two or more large groups of nodes with the node itself inbetween, within the network. High betweenness centrality scores indicate the nodes which have a specific position and importance within the network. Accordingly, Countries such as the People's Republic of China (centrality = 0.63) and the USA (centrality = 0.44), Japan (centrality = 0.17), Australia (centrality = 0.16) and England (centrality = 0.16) have established key positions within the network reflecting the beneficial international influence of the research publications in the field of datadriven flood resilience. Moreover, citation bursts were identified which indicate the sudden increases in citations over a short period in countries such as Canada (burst strength     Table 2 shows the countries and institutions with a high frequency of articles identified from the networks.
Further, 10 keywords were identified to be citation bursts including "deep-learning" (burst strength = 6.29, 2021-2022), "calibration" The citation bursts can be identified for recent years from 2018 to 2022 for keywords such as "deep learning", "artificial neural network" and "machine learning". These values also indicate the popularity of research trends on modern data-driven techniques for flood-related research. Comparing the high frequencies and citation bursts of keywords, common keywords can be identified in both including "climate change", "system", "artificial neural network" and "machine learning". These keywords again reflect the recent development and concern toward data-driven systems and methods in floodrelated research. The summary of the results is shown in Table 3.

Co-Citation analysis
Co-citation analysis can be used to define the frequency with which two documents are cited together by other documents (Chen, 2016) and it is recognized as a proximity measure for documents. In this study, cocitation analysis has included journal co-citation analysis, author cocitation analysis, and document co-citation analysis.

Journal Co-Citation analysis
The network of journal co-citation was produced with 712 nodes and 3230 links to identify the most significant cited journals, as shown in       influential journals were identified with the frequency as stated in Table 4. These results indicated that the research articles published in these journals have received strong citations and importance in the research field.

Author Co-Citation analysis
Author co-citation analysis help determine the relationships between authors, whose publications are cited in the same articles, and analyses the evolution of research communities. The author co-citation network was generated with 719 nodes and 3263 links. The modularity of the network is Q = 0.625 and the mean silhouette value is S = 0.831. The node size of the network indicates the number of co-citations of each author while the links between authors correspond to indirect cooperative relationships based on co-citation frequency. The most highly cited authors were identified from the produced network as shown in the Authors with high betweenness centrality scores were also identified in the network. As resulted from the network, Li X (centrality = 0.13), Nash J (centrality = 0.12), Chen J (centrality = 0.12), Chen C (centrality = 0.11), Beven K (centrality = 0.10), Chen S (centrality = 0.10) and Das S (centrality = 0.09) were identified with centrality scores who established important positions connecting with different research communities. These authors can be identified as significant and influential contributors to data-driven flood-related research and help connect the various research communities. Summarily, the most highly cited authors with the frequency of citations and centrality scores demonstrate that studies on data-driven floods have been widely performed in China. However, Nash J from the UK has been one of the highly cited authors with high centrality score representing an important author in the research area.
In terms of citation bursts, several authors were identified with citation bursts, with rapid increases in citation frequency, including Intanagonwiwat C (burst strength

Document Co-Citation analysis
Document co-citation analysis can be used to analyze the underlying intellectual structures of a knowledge domain while demonstrating the number of references cited by publications. For document co-citation analysis, a network was generated with 759 nodes and 2550 links as shown in Fig. 9. The modularity of the network is Q = 0.829 while the mean silhouette value is S = 0.901. Each node represents a document labelled with the first author's name and the publication year and each link signifies the co-citation relationship between the two corresponding documents. The node size represents the co-citation frequency of the node document.
A total of 14 significant co-citation clusters were identified based on the keywords of the documents cited in each cluster, by the loglikelihood ratio (LLR) algorithm as the LLR method can select the best cluster labels in terms of uniqueness and coverage [44]. The top 14 highly cited articles were representative of clusters #0 to #19. The silhouette metric was included in co-citation clusters as it measures the average homogeneity of a cluster (Chen, 2016). For the clusters of similar sizes, a higher silhouette score represents more consistency of the cluster members [44]. The silhouette scores of the clusters ranged from 0.731 to 1.000, indicating that the members of each cluster were consistent enough. In addition, the representative document of each cluster was the document with the most co-citation frequency within a cluster. These representative documents can be identified as important and influential articles for the research area.
Totally, the top 10 cited documents were identified from the analysis as summarised in Table 5 The citation bursts were further identified from the document co-   . These findings indicated that the citations of these documents have been increased significantly over a short period in the corresponding years contributing to the development of research in datadriven flood resilience. Table 6 shows the 14 co-citation clusters produced in the network with the cluster-ID, size (number of articles within the cluster) silhouette scores of clusters cluster label, and the most cited article in each cluster with the corresponding journal and DOI.

Systematic review and discussion
This section of the paper includes a systematic review of the articles resulting from the co-documentation analysis of the scientometric analysis. The articles were subjected to full-text reading [65,124,140;125] to understand the current context of data utilization for flood resilience of built assets. The thematic focus of the review was laid on identifying the flood resilience focus of the studies, methodologies used, data categories and the variables utilized and future innovative computational methods that can be used to improve built asset resilience. The intention is to understand what exists and the opportunities available for data-driven approaches to built asset flood resilience. Existing research gaps were highlighted, and potential future research directions were subsequently suggested.

Methodologies deployed in the Data-Driven studies
The methodologies mainly deployed in the existing data-driven studies for flood resilience of the built environment were identified as machine learning techniques, artificial intelligence based modelling and hydraulic flood simulations. Significantly, tremendous attention and an increase in the use of machine learning techniques can be identified. Further discussion is as follows and a comprehensive summary is presented in Table 7.

Machine learning
Machine Learning can identify and learn patterns, discover trends and behaviours from large quantities of datasets and generate and predict new data to improve its performance. Due to the ability of machine learning models to deal with large datasets, the complex interdependent attributes and hidden behaviours of data can be identified [1]. With the proven capabilities and advantages of data handling and analysis of various machine learning techniques, a dramatic increase in flood resilience studies can be identified, they focused on flood prediction [1,7,15,17,19,43,24,86], flood monitoring [9] and flood damage assessments [74]. The different machine learning models developed and employed in these studies have significantly been imperative for meaningful identifications of flood hazards, different types of floods, direct flood damages and impacts on the long-term recovery of cost and time, through risk assessments and damage assessments. The approach provided the ability to categorize vulnerable communities, discover behaviours in extensive flood hazard datasets in different communities, and identify at-risk communities through spatial analysis [1]. However, limitations are further being argued in utilising machine learning models for more comprehensive resilience assessments for communities through robustness assessments, and rapidity evaluations on short-term  and long-term impacts. The issues mainly associated with the level of data, data training, data storage, and analytical errors in various machine learning techniques have been highlighted for thorough consideration in future studies [1].

Artificial intelligence
While machine learning techniques have been undertaken in such studies being a part of artificial intelligence, some researchers have mainly focused on artificial intelligence-based modelling and algorithms such as neural networks which are similarly applied with the benefit of modelling and analysing large data sets. Amininia and Saghebian [17] and Araghinejad et al. [19] have conducted studies on river flow modelling using artificial intelligence algorithms to enhance flood predictions. A recent study has also been undertaken by Dai and Cai [53] using artificial intelligence for flood prediction. Thus, machine learning and artificial intelligence can be identified as emerging data-driven approaches in the research field, especially for the prediction of floods. Artificial intelligence as a standalone technology has extensively been utilized with specific algorithms and models due to its inherent benefits in the prediction and detection of flood parameters (ex: flood River streamflow data Flood prediction depth) along with real-time data processing, and estimations of past flood events to estimate future risks. The key advantage is the consistency of many interrelated processing features working uniformly to resolve problems. However, the major limitations mentioned in the previous studies are the extensive time needed to train data networks and the computational expensiveness for more sophisticated tasks. A further challenge of this method is the intermediate process which is hidden, allowing the user to feed and modify the end results without having real access to the major decision-making process (Saravi et al., 2019).

Hydraulic simulations
Over the past years, credible attention has also been given to hydraulic flood simulations and studies have widely utilised flood simulations in terms of 2D flood inundation modelling and 3D flood inundation modelling. Flood simulation models have significantly been utilized in flood monitoring [4], flood prediction [7]; Chen, Chen & Shen, 2019; [49], flood risk assessments [8,48] and flood damage assessments ( [119]; Altinaka, 2011; [108]). Recently, advanced flood simulation software developments such as Citygml and OpenFloods can also be identified in existing studies [4,7,8,120]. These studies mainly drew attention to flood risk and prediction at the urban and community level. Hydraulic simulations enable prediction-related and management-related decision-making through estimating and assessing risks and potential damages under various climate conditions, hydraulic components, and land use changes. In contrast, limitations have also been discussed in the existing studies, such as not considering all flood parameters and oversimplification of real-time conditions reducing the significance of simulated results and indices (Parra et al., 2018).

Building information modelling (BIM) and geographic information systems (GIS)
Building Information Modelling (BIM), geographic information systems (GIS) and integrations of BIM and GIS with flood simulations in this regard were identified as viable data-driven approaches for flood risks and damage assessments [18,56,120,119]. Some of the existing datadriven studies focused on flood resilience in individual properties. The studies have mainly aimed at facilitating comprehensive flood damage assessments and risk assessments for buildings, they are applicable for visualising vulnerability and property flood loss potentials. Due to the advanced ability of 3D visualisation, data visibility, and generating digital descriptions of all aspects of buildings providing a virtual environment, BIM and GIS have gained popular attention in previous studies. However, the coordination of these methods is yet to be examined due to the limitations of interoperability, lack of standards and issues of information exchange [55]. Hence, BIM and GIS integrated research on flood resilience so far, have merely been focused on the interrelationships between individual buildings than complex environmental and urban settlements.
Few studies have employed Bayesian Network Modelling [139,134] and the Extreme Learning Machine (ELM) method [45] aiming at flood recovery processes and flood predictions. Bayesian network modelling is also gaining attention in current disaster-related studies however, limited research is available, especially to built asset flood resilience. This review identified a study using bayesian network modelling for flood recovery processes and only one of the articles in the reviewed sample is based on the extreme learning machine method, it was used for flood prediction. Table 7 shows the overall article summaries in terms of methodologies and data types under each focused flood resilience aspect.

Classification of data types
As data-driven methods utilize large quantities and varieties of data types and attributes, it is logical to classify these data into similar categories [120], this would give more insight into what exists and stir thoughts on what more data could be considered. Based on the review, the main categories of data were classified as locational/geographical data, locational flood data and building or built asset-specific data. Data relating to built asset characteristics, and construction were classified as built-asset-specific data, data directly relating to flood characteristics corresponding to flooded locations were classified under locational flood data, and data representing topographic or geographic attributes of locations were grouped as locational/geographic data.

Built Asset-Specific data
The data variables found in this category in the data-driven studies include building geometry [18], topology, type of buildings, materials, building use [120,139], exterior and interior components and characteristics, building value, number of stories, type of foundations, elevations [119], damages and cost data [18,8,120,119,1,121,139] and, building size, type, quality and economic value [16]. The focus of these papers is available in Table 7, this data type is largely used for vulnerability and damage assessment, other limitations and opportunities are discussed in the further research section of this study.

Locational flood data
Finally, locational flood data includes flood type [1], flood duration and frequencies [1]; Altinaka, (2011), flood depth [22,16], flood volume, flood velocity [16], rate of rising, rainfall intensity, flood duration [16], and patterns [45] (a); [43,9,53,121,90], water level and flow pathways, water channel structure, watershed morphometry data [9;15], frequency inundations, flood probabilities [74,35,139], water level and flow direction, water channel structure, rivers' length and slope [45]  A deeper look into the technical connections and potential integrations of data-driven models suggests that the various models developed in the existing studies can complement each other and lay the foundation for the attainment of more comprehensive approaches to built asset flood resilience. Existing models are based on specific aspects of flood resilience, such as flood prediction, flood hazard and risk assessments, structural design, damage assessments and emergency planning. Flood prediction models for example, are often coped with hydrological and meteorological data and these data provide valuable insights for early warning signals, extent, timing and intensity of potential flood events, while structural design models measure the vulnerability of built assets to flooding. By integrating these models to new models, decision-makers can prioritize investments and mitigation strategies in flood-resistant design features. This combination will help optimize resource allocations for strengthening the most vulnerable built assets and will ensures that the structural design is informed by accurate flood predictions, leading to more resilient built assets.
Existing models based on GIS and remote sensing techniques can also be integrated with other various flood prediction models to generate more comprehensive flood maps, gain a more comprehensive understanding on flood risks in areas and flood impacts to built assets. This will be a valuable integration focusing on the alignment of flood risk and flood impact potential. As these models use data from satellite imagery, aerial surveys, and LiDAR for flood mapping and monitoring, combination of these models can be used to upgrade structural measures, risk assessment models to quantify the potential consequences of floods on built assets. Existing BIM models on the other hand, capture more detailed data on physical characteristics of built assets and integration of these models with existing risk assessments and prediction models will permit the simulation of the behaviour and interactions of communities within a system. This will enable decision-makers to assess and evaluate the resilience of built assets at a more granular level, appraising different flood scenarios, potential failure points and optimizing flood mitigation strategies. Existing data-driven models can further be enhanced by combining historical and real-time data collection and monitoring methods. For instance, sensors installed in buildings to collect real-time data such as water levels, structural integrity, and environmental conditions can be fed into other flood prediction models, hazard and risk assessment models and existing structural design models empowering timely updates and adaptive decision-making. Such combination ensures that resilience strategies to built assets are based on the most updated and accurate data and information available. These combinative models demonstrate the potential for further enhancements that can complement and bolster the existing models, for more comprehensive and robust resilience measures for built assets resilience. It further helps leveraging capabilities and strengths of existing multiple models and addressing their respective limitations for future data-driven models. This iterative process of integration and model refinements not merely fills the existing technical gaps but also inspires future academics and researchers to explore novel approaches to advance and drive innovation for built assets flood resilience.

Further capabilities of existing data-driven methodologies and new insights towards built environment flood resilience
This section of the review discusses further capabilities of datadriven methodologies from existing literature, future innovations and new insights which can be applied to built asset flood resilience. Though there is a rapid increase in advanced data-driven methods which are currently being used for built asset flood resilience, the engineering challenges of such methods should further be recognized beyond the mere identification and general utilization, to address the limitations associated with such methods. Evaluating the existing works on identifying and remedying such engineering challenges are immensely important to increase and facilitate the encouragement and utilization of novel data-driven methods to flood resilience. In light of this, further advanced computational methods embedded with such data-driven methods are vital and need to be discussed.
One of the limitations of identified data-driven methodologies is data processing for flood resilience assessments. Hu and Gong [81] mentioned that how to process the data into high-quality information is an issue which often creates the so-called "data-rich-but-informationpoor" situation. To address this issue the authors suggested a solution by introducing an information salience framework based on Data Envelopment Analysis (DEA) to arrange the order of information processing tasks. The proposed model combines the DEA efficiency score with a group decision-making process using language, and the findings demonstrate that the framework is capable of arranging geospatial data processing tasks in a systematic way and speeding up the process of extracting information from geospatial data sets associated with disasters. The main three challenges addressed in the study are formulating the value of information in disaster scenarios, modelling the information articulation process and dynamically balancing the conflicts in information processing needs, mainly considering the scenarios of disaster response. Taking the implications of this study, the challenge of data processing can be further investigated and addressed at other flood resilience aspects at built asset level where time-critical information is needed such as flood risk assessments, damage assessments, flood prediction and recovery along with different datasets, to improve the efficiency and effectiveness of data processing for data-driven models.
Problems of spatial mapping have also been identified as one of the challenges in data extraction and processing of GPS data, GIS data and high-resolution aerial images captured from drones for flood risk assessments. In large-scale applications of disaster response and mitigation a key limiting problem is contextualising collected data from disasteraffected objects like buildings and flooded areas to decision-making due to the problem of spatial mapping. To address this, Nath, Cheng, and Behzadan [117] created a solution by devising an approach that enables the quick scanning and mapping of objects of interest within unstructured visual data that is shared through crowdsourcing, even if GPS information or camera metadata is absent. The outcome of the study is a map that includes crucial spatial information about disaster-affected objects such as buildings. This map can aid in decision-making in different stages of disaster management, including response, damage assessment, and recovery. The potential applications of the developed approach were further suggested in estimating flooded areas and accurate counts of damaged buildings and infrastructures in flooded areas as it is a critical piece of information for local governments, insurance firms, and private owners who will have to prepare and report damage estimates to secure necessary funds and/or other federal and state resources for rebuild and recovery efforts.
Alizadeh et al. [12] presented a technique for determining floodwater depth that enhances flood gauge data with pictures contributed by users of flooded streets. The methodology enables a dependable estimation of floodwater depth, which can be used to optimize routes and assess risk. The study developed an improved map of flood depth that accurately portrays the movement of floodwaters in nearby areas, particularly in residential neighbourhoods. The objective of this mapping is to enhance the quality of decision-making during floods, such as evacuating people or transporting goods and services while avoiding flooded areas. The outcomes provide a significant contribution to overcoming the problems of spatial mapping associated with traditional flood mapping and simulations. The developed depth estimation methodology can be used as a forward step in assessing damages specific to built assets in future studies.
Pi, Nath, and Behzadan [129] employed convolutional neural networks, a type of deep learning technique, to detect objects in aerial imagery for disaster response and recovery. The study proposed a solution to the challenge of processing large amounts of data to identify and map objects of interest on the ground in real-time by introducing a series of Convolutional Neural Network (CNN) models. These models can recognize important ground assets such as buildings (both damaged and undamaged), vehicles, vegetation, debris, and flooded areas. The study outcomes enable the detection and quantification of natural disaster damage in aerial imagery using CNN models trained on footage from past disasters.
In another study by Bui et al. (2020), a new approach for flash flood susceptibility mapping was developed using the Deep Learning Neural Network (DLNN) algorithm. A hybrid GIS and deep learning model was created to predict different levels of susceptibility to flash floods. The model was developed using variables such as elevation, slope, curvature, aspect, stream density, soil type, lithology, and rainfall of hazard areas. This model can be used as an alternative to aid government authorities and other stakeholders in developing effective flash flood mitigation strategies and land-use planning. The method introduced in the study is a promising opportunity for studies in the future to integrate with built asset variables to conduct comprehensive flood risk assessments, damages and mitigation strategies for the built environment.
Using ensemble learning Doycheva et al. [58] developed a methodology towards improving flood forecasts by weighting ensemble members based on supervised machine learning. The methodology addresses overcoming the challenges of uncertainties of numerical precipitation forecasts and these precipitation forecasts form the input into hydraulic models, creating uncertainties in flood forecasts. The outcomes are useful to improve flood simulations and flood risk estimations and future studies can extend the use of ensemble models combined with built asset features. To address the challenges of the accuracy of data-driven models, Hosseini et al. [80] developed a hybrid approach using ensemble models and the bayesian method for flash flood hazard assessments. The findings discussed its potential application to flood hazard areas with flood-affected variables and future studies can further extend the implications using built assets variables for in-depth flood hazard assessments in relation to the built environments. For example, ensemble learning could be used to combine the predictions of multiple machine learning models that are trained on different types of data such as flood maps, building plans, and hydraulic models. By combining the predictions of these models it may be possible to generate more accurate assessments of flood or risk or building vulnerability than would be possible using any single model alone.
Virtual and augmented reality is gaining prominence in recent research and the potentials of the methods have widely been discussed in relation to the architecture, and engineering construction industry (Delagda et al., 2020; [133]. Lai et al. [102] developed a 3D virtual environment to address the challenges of excess engineering details, lack of engineering accuracy, integrated communications between decisionmakers and flood flow visualizations [76]. However, limited research is still available on flood resilience. Virtual and augmented reality can be used to create immersive simulations of flood events, allowing decisionmakers to visualize the potential impact of floods on buildings and infrastructure. This can help to inform flood risk management strategies and improve the resilience of communities to floods.
Xu, Huang, and Fang [154] employed cloud computing, another emerging technology, to enhance urban flood control by developing a cloud-based platform capable of sensing the real-time status of physical assets of a city. This cloud asset is designed to adapt to varied working scenarios, be controlled remotely, and be shared among agencies. The physical assets monitored by the system include dams, sewers, pipes, pumps, hoisters, and sandbags, which can pose a threat to the success of urban flood control. Managing these physical assets is challenging due to the complexity of urban flood control, which involves a large and diverse range of physical assets that are widely distributed over large areas and located in extreme environments. The proposed method can further be investigated in future studies taking the built assets into consideration for flood control.
Another challenge associated with data-driven methodologies is the availability of data. In some cases, there may not be sufficient data available to develop accurate flood resilience assessments. Transfer learning is a machine learning technique that allows a pre-existing model to be used as a foundation for a new task, which can be useful in overcoming limited data availability. By using a pre-trained model as a starting point, the model can be fine-tuned on a new dataset, thereby reducing the amount of data required for training. Zhao et al. [160] introduced the first transfer learning-based machine learning method for enhancing urban flood susceptibility mapping in catchments with both data-rich and data-sparse scenarios. The authors concluded that their proposed approach is a promising method for improving machine learning-based assessments of natural hazards, especially in areas with limited data. However, in the field of research application of transfer learning for built asset flood resilience is yet less straightforward.
Data augmentation involves generating new training data by applying various transformations to the existing data. This can help to increase the diversity of the training data and reduce overfitting. Data augmentation can be particularly useful in flood resilience assessments where the available data is limited [99]. For example, data augmentation techniques could potentially be used to generate additional training data for machine learning models [25] that are used to classify images of flood damage or assess the effectiveness of different flood mitigation strategies. Further, Explainable AI (XAI) involves developing machine learning and deep learning models that can provide explanations for their predictions. This can help to improve the transparency and accountability of flood resilience assessments and reduce errors that may be caused by opaque models. Jiang et al. [86] conducted a study exploring flooding mechanisms and their trends in Europe through explainable AI. Along with this, XAI can be a useful methodology for built asset flood resilience, particularly in contexts where it is important to understand the factors that contribute to flood risk or building vulnerability. XAI helps identify the specific features or characteristics of buildings or infrastructure that contribute to their vulnerability to flooding. By providing explanations of how a particular model arrives at its predictions or decisions, XAI can help to identify areas of weakness or areas where improvements could be made to mitigate flood risks. Moreover, XAI can be used to address the analytical problems of machine learning and deep learning techniques. By providing an explanation of how a model arrives at a decision, XAI can help to increase the trust and reliability of machine learning models [23].
To overcome the challenges associated with the level of data, data training, data storage, data processing and analytical problems of machine learning and deep learning techniques, data pre-processing techniques, model selection and hyperparameter tuning, cloud computing and distributed computing methods can be used as highlighted in the existing literature [151]. Data pre-processing is a crucial step in the machine learning pipeline. Choosing the appropriate machine learning model and hyperparameters can have a significant impact on the performance of the model. These techniques can also be used to avoid overfitting and underfitting problems. Cloud computing and distributed computing can be used to overcome data storage and processing limitations and on-demand access to computing resources and storage capacity. These resources can be used to store and process large amounts of data, train machine-learning models, and optimize hyperparameters [151].

Modern hybrid Data-Driven approaches to improve the flood resilience of the built environment
The impact of flooding on built assets has become a growing concern in recent years due to the increasing frequency and severity of extreme weather events. In response, researchers and practitioners have been exploring modern hybrid data-driven approaches to improve flood resilience in built assets. These approaches combine various data sources, such as sensor data, satellite imagery, and weather forecasts, with advanced analytical techniques, including machine learning and artificial intelligence, to enhance flood risk assessment, early warning systems, and mitigation measures. By leveraging the power of data and advanced analytics, these approaches hold the promise of improving the resilience of built assets against flooding and reducing the potential for damage and disruption.
Costache [52] conducted a study comparing four hybrid models for accurate flash-flood assessments for the improvement of flash-flood forecasts and warnings. Hong et al., [79] developed a hybrid model of ANFIS-DE (adaptive neuro-fuzzy inference system) in flood susceptibility mapping. Bui et al., [32] also developed a novel hybrid approach based on a swarm intelligence-optimized extreme learning machine for flash flood susceptibility mapping. The research introduced a novel soft computing technique, called PSO-ELM, for predicting the spatial occurrence of flash floods. PSO-ELM is a combination of two methods, namely extreme learning machine (ELM) and particle swarm optimization (PSO). Similarly, Chen et al., [43] developed a hybrid approach combining machine learning models and GIS for flood susceptibility assessments. While the direct focus of these studies is less on the built environment future computational hybrid approaches can be suggested to improve the flood resilience of the built environment using different methods as indicated in Table 8. These hybrid approaches can be used to improve the accuracy and effectiveness of flood resilience measures in the built environment. By combining different modelling techniques, these approaches can provide a more comprehensive and nuanced understanding of flood risks and can help to prioritize flood resilience measures to built assets that are tailored to specific locations and conditions.
These data-driven approaches can complement existing methods by providing valuable insights to improve flood resilience to the built environment. For example, when combining traditional flood models with machine learning and artificial intelligence tehcniques can complement existing hydrological and hydraulic models. While machine learning algorithms can analyse larger datasets and define complex patterns and relationships in flood data, artificial intelligence methods, such as neural networks, can define non-linear relationships and capture the flood dynamics improving the understanding and prediction of flood behavior. Hydraulic simulations enrich existing flood models by featuring up-to-date information for example, water levels, and flow rates allowing more precise flood hazard mapping, damage and risk assessments. Integration of hydraulic models with data-driven approaches, enhance the accuracy and reliability of flood predictions enabling more effective flood resilience strategies. Hybrid data-driven approaches, could combine the strengths of physics-based models such as hydraulic models with data-driven techniques. Incorporation of flood models with physics-based models, more accurate flood predictions can be taken especially in complex or poorly understood systems. Incorporation of machine learning and artificial intelligence, into agent-based models, help understand the behavior and decision-making of communities in response to flood events. This helps better understanding on developing more realistic and effective strategies for built assets resilience considering community factors and social dynamics. Therefore, such integrations enable more comprehensive understanding of flood dynamics and improved decision making in flood resilience of the built environment.

Future computational methods to improve the flood resilience of the built environment
The field of flood resilience is constantly evolving, and computational methods can be used to reinforce engineering knowledge in flood resilience studies by providing new insights, tools, and techniques. These methods can provide new insights to optimize flood resilience measures, reduce the impact of floods on communities and infrastructure, and enhance the sustainability of flood resilience strategies.

Ontology-based knowledge representation
Ontologies are formal and explicit descriptions of concepts and their relationships, which can be used to represent the knowledge of experts in a domain. Ontologies can help to capture the semantics of engineering concepts and provide a framework for integrating and sharing engineering knowledge. By providing a structured way of representing and organizing knowledge about flood risk and resilience measures an ontology can be created to represent the various aspects of flood risk and resilience, including information about the built environment, flood hazard maps, climate change projections, and flood protection measures [11,115].

Multiscale modelling
Multiscale modelling can be used for built asset flood resilience by providing a more comprehensive understanding of how floods affect the built environment at different scales. For example, multiscale modelling can be used to simulate the impact of floods on buildings and infrastructure at the micro and macro scales. It can help to identify the most critical components of a system and to optimize the design of flood resilience measures (Kawaike et al., 2021).

Knowledge graphs
Knowledge graphs can be used to represent the relationships between different concepts and entities in a domain. It helps to capture the Iqbal et al. [84]  complex interdependencies between different aspects of flood resilience, such as building codes, flood insurance, and infrastructure investments. It can also be used to support reasoning and decision-making in flood resilience studies (Johnson et al., 2021).

Integration of real-time data and sensor networks
The use of real-time data and sensor networks can help decisionmakers to monitor flood conditions and respond quickly to changing risks. As sensor technology continues to improve, it may become possible to develop more accurate flood forecasting and early warning systems, as well as more effective flood risk management strategies [96].

Use of high-resolution mapping and remote sensing technologies
High-resolution mapping and remote sensing technologies, such as LiDAR and satellite imagery, can provide detailed information on the built environment and its vulnerability to flooding risks. As these technologies continue to improve, they may become even more powerful tools for developing effective flood resilience strategies [113,93].

Integration of blockchain technology
Blockchain technology can be used to develop secure, decentralized databases that can be used to track flood risk and damage, as well as to store and share data on flood resilience measures. As this technology continues to mature, it may become a valuable tool for enhancing collaboration and coordination among stakeholders in the flood resilience ecosystem [150,153].

Convolutional neural networks (CNNs)
CNNs are already being used to detect damaged built assets due to floods, however, in the future, it may be possible to improve the accuracy of these models by integrating other types of data, such as sound or vibration data, which could provide additional information on the extent of damage to built assets. Also, CNN trained to detect damage to buildings after an earthquake could potentially be adapted to detect damage from floods as well, by fine-tuning the model on a smaller set of flood-specific data [46]. CNNs can be trained on large datasets of historical data, but they can also be improved by integrating real-time data from sensors and other sources [92], which could be critical for decisionmaking in the context of flood resilience [157].

Integration of social media data
Social media data can provide real-time information on flood conditions and impacts in affected areas. In the future, automated approaches could be developed to integrate this data into flood forecasting and prediction models to improve their accuracy [95].

Cloud-based computing
Cloud-based computing also has the potential to revolutionize flood resilience by enabling real-time monitoring and analysis of flood events. It can enable the collection and analysis of real-time data from a variety of sensors, such as remote sensing platforms, ground-based sensors, and crowd-sourced data. This data can be used to monitor water levels, flow rates, and other parameters during a flood event and can be used to create predictive models that can forecast the potential impact of a flood event on built assets. These models can be updated in real-time as new data becomes available, allowing decision-makers to make more informed decisions for built assets during a flood event [70].

Bayesian models
Bayesian-based machine learning models have the potential to improve flood risk assessments by allowing for more accurate and precise predictions of flood risk. These models can be used to fuse data from different sources, such as remote sensing data, ground-based sensors, and historical flood records. This can help to improve the accuracy and coverage of the data used in the flood risk assessment. The models can also be used for real-time updating and multi-hazard modelling to model multiple hazards, such as flooding and storm surges, simultaneously. This can help decision-makers to understand the interactions between different hazards and make more informed decisions [71,3].

Improved data assimilation
Data assimilation is the process of combining observations with model simulations to improve the accuracy of predictions. Future hydraulic flood simulations could include more sophisticated data assimilation techniques, such as the use of machine learning to learn the relationships between observations and simulated variables [54].

Integrated building information modelling
In the future, BIM software could be developed to integrate flood risk data, such as flood depth and flow velocity, into the modelling process. This would allow for more accurate predictions of the impact of flooding on built assets. BIM models could be linked to real-time flood monitoring systems, such as sensors and satellite imagery, to provide up-to-date information on the extent and severity of flooding in the surrounding area. This helps building owners and operators quickly identify areas of risk and take appropriate action [55]. BIM could be further used to support the development of flood-resistant building design, including the use of materials and technologies that can better withstand flood damage and to create dynamic flood simulations, allowing building owners and operators to model the impact of flooding on their assets and evaluate the effectiveness of flood protection measures.

Digital twin integration
Digital twin models can be integrated with machine learning and artificial intelligence to automatically detect and respond to flood risks. This can include the deployment of sensors and other devices to alert authorities to potential flooding events and the use of automated responses to mitigate the impact of floods. Digital twin models can be enhanced to better simulate urban water systems, including drainage, sewage, and water supply networks. These models can provide a more accurate understanding of flood risk and inform the development of more effective flood prevention and response strategies [69].

Summary of the systematic review
The use of data-driven techniques has been largely limited to flood predictions, flood monitoring, flood risk assessments and damage assessments (Table 7). There are still significant opportunities for further utilization [77,147], including applications in structural and nonstructural measures planning, preparation, asset performance evaluation and mitigation, alternatives assessment, investment allocation and prioritization, flood response, comprehensive insurance analysis, potential recovery procedures and processes, built asset recovery resource planning, considerations for building back better, and overall built asset safety. These further utilisations can be through the integration or expansion of existing models as well as the development of new models. Further, the limitations of the data-driven methods as well as the difficulties triggered by built environment system complexity highlight huge potentials for novel computational methods. Thus, this review discussed the significant studies on data-driven and evolving computational methods and application to future flood resilience assessments, planning and recovery of built assets. Fig. 10 is a diagrammatic representation of the systematic review findings.

Conclusion
This research conducted a scientometric and systematic review for gaining a comprehensive understanding of the current context of data utilization for flood resilience of built assets for societal sustainability. This study established an inclusive scientometric analysis that covers coauthorship analysis, country and institution analysis, co-word analysis, author co-citation analysis, journal co-citation analysis and document co-citation analysis. The purpose of the scientometric analysis was to understand the past, current, and emerging research areas of data-driven flood resilience. The key findings from the scientometric review are as follows: (1) The co-authorship analysis showed four major research communities in the research area, the country analysis showed the top five countries: the people's republic of China, the USA, Germany, Japan, and England and in the institution analysis, the Chinese Academy of Science and Wuhan university were identified as the top two institutes in the study area. (2) The most frequent keywords identified from the co-word analysis were model, big data, flood prediction, machine learning, artificial neural network, climate change (and flood), flood impact, flood risk, and algorithm. In terms of journals, the 10 most influential journals were identified from journal co-citation analysis and the Journal of Hydrology and Water Resource Research journal were the top journal which had more than 200 articles in the research field. This describes the hotspots and research frontiers respectively of data-driven flood-related studies.
(3) Based on the author co-citation analysis, the top 7 most cited authors were identified; 6 from China and 1 from the UK. Finally, 10 highly cited articles were identified resulting from document co-citation analysis, their focus is flood forecasting.
The systematic review was conducted to obtain further insight into previous studies on research area, methodologies deployed, data types and variables. The key findings from the systematic review are as follows: (1) The existing data-driven studies have largely limited the focus of flood resilience to flood prediction, flood monitoring, flood risk assessments and flood damage assessments. Areas such as built asset and infrastructure planning, flood insurance, potential recovery processes with improved integration of built asset data attributes, response, and safety in the context of built asset resilience have not benefited from data-driven approaches.  buildings. Hypothetical data raise questions about the reliability of the outcomes. (3) Seven major data-driven techniques identified are machine learning, artificial neural network, BIM, GIS, hydraulic flood simulations, bayesian network modelling and extreme learning machine method. Each approach has peculiar limitations, and they are all faced with the general challenge of data acquisition (such as lack of data and difficulty in accessing data) and governance issues. (4) Data types identified are mainly locational flood data, locational/ geographical data, and building-specific data. Studies that describe the intersection between flood hazard characteristics and buildings exposure data are limited, and the actual dynamics between flood attributes and built asset characteristics remain underexplored. (5) The future computational methods identified for built asset flood resilience include various hybrid data-driven approaches, ontology-based knowledge representation, multiscale modelling, knowledge graphs, sensor technologies integrated with real-time data, high-resolution mapping and remote sensing technologies, blockchain technology, convolutional neural networks, automated approaches integrated with social media data, cloud-based computing, Bayesian-based machine learning models, data assimilation, BIM models linked with sensors and satellite imagery and ML and AI-based digital twin models.

Research gaps and directions for future research
The following research gaps and future research directions in the context of data-driven techniques and data types have been identified.
(1) In terms of the built environment, most existing data-driven flood resilience studies have merely focused on the overall perspective of flood risk management at the urban level, while a more focused but well-integrated attention can be placed on the built asset level and issues relating to built asset management. Hence, future research studies can be undertaken using the modern data-driven methods for a broader view of flood resilience aspects giving more consideration to built asset resilience and connections with asset management practices. (2) Future studies can be extended to include the collection of relevant data from existing built assets, potentially in real-time and in a more structured manner. This approach would provide more reliable assessments compared to hypothetical scenarios used in some studies, ensuring greater accuracy and applicability to the original context. [57,89,94,116]; Pistrika & Jonkman, 2010). Data creation procedure, classification and ownership, and access facilitation needs further attention. (3) By incorporating more comprehensive variables within the datasets and giving more details, future research studies can focus on utilizing machine learning and AI-based methods to enhance the flood resilience of built assets at the community level rather than individual property level [1]. (4) More comprehensive solutions can be achieved by the careful complementary usage of data-driven models across the different aspects of resilience such as flood prediction, with built assets structural measure selection at asset unit, community and larger scale. (5) A unique perspective will be the exploration of the influence of interconnected or neighbouring assets on individual and collective asset resilience, using data-driven methods. This can influence decisions on the planning, operation, and management of built assets. This will strengthen the understanding of resilience of built environment that encompasses all the built assets including individual buildings, sites, infrastructures and critical facilities.
(6) As data-driven methods need large quantities of data sets, lack of data availability and uncertainty have become challenging and hindered the application of such methods for built asset flood resilience. Especially, since machine learning and AI-based techniques heavily rely on long-term records of floods and built assets, accessing those data sometimes in mixed data formats (numerical, categorical, and qualitative formats) is a challenge [1]. Hence, to address this issue further research efforts are needed in the processing and extraction of numerous flood resilience-related data for the built environment [43,139,15]. (7) More sophisticated tools are still needed to broadly analyse and improve flood resilience at the built asset level determining the intersection between flood hazard characteristics and buildings exposure data [72]. (8) Since data-driven models and data processing methods are highly data sensitive, future studies should be conducted by applying more ML and AI techniques using wider data ranges for built asset flood resilience. This will be beneficial in expanding insight, understanding, and strengthening capabilities in various aspects of flood resilience in the built environment [31]. (9) It is also important to conduct more research studies to identify different resilience-related data sets and ranges needed to enhance the flood resilience of built assets as current studies have been limited to specific data variable types. More insight is needed into the specific contribution of each data type to flood resilience. (10) The awareness towards novel computational methods should also be increased to address the challenges of the level of data usage, data training, data storage, data analytical errors, sophisticated data training, computational expensiveness and difficulties, and incorporation of parameters into data models and platforms. The future computational methods suggested in the review hold great promise for enhancing the efficiency and transparency of built environment flood resilience efforts by offering powerful toolkits to help better understand flood risks, develop more effective flood resilience measures, and improve decision-making related to flood response and recovery efforts. Future studies can explore the use of these methods for built asset flood resilience at both individual and community level resilience. (11) Future studies can also integrate these novel computational methods for different types of built assets such as buildings, transportation infrastructure, drainage, water and wastewater systems and explore their unique challenges related to flood resilience. Researchers can further explore the development of new computational tools which are user-friendly, accurate, and reliable that can help in assessing flood risk and designing flood protection measures promoting data-driven flood resilience. (12) As flood resilience is not merely about physical infrastructure and systems, future research studies can explore the incorporation of socio-economic data such as population density, income levels, and land-use patterns in computational models with respect to built environment flood resilience. (13) Case studies can be undertaken in future studies to provide insights into the effectiveness of computational methods for flood resilience in different contexts. Researchers can conduct case studies in different regions to evaluate the applicability of computational methods in various settings so that contextspecific data utilization can be improved. As it is essential to validate and verify the accuracy of computational models used in flood resilience studies future researchers can also undertake validation studies to compare the performance of different computational models and improve their accuracy.
Summarily, much more can be done by future researchers to expand data utilization for flood resilience of built assets in respective individual, communities, and sectors [5,46;14].

Implications of the study
This study provides insights to researchers, built environment professionals, environmental experts, and flood experts on the opportunities available to generate more sophisticated and reliable data-driven solutions to enhance the flood resilience of built assets. It sets the scene for contextualising data-driven flood resilience researches, especially in the area of built environment, built asset and infrastructure resilience. Addressing the research gaps will contribute to the attainment of resilience and sustainability goals of built environment, societies, and communities.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
No data was used for the research described in the article.