Review of recent advances in climate change detection and attribution studies: a large-scale hydroclimatological perspective

The rapid changes in global average surface temperature have unfathomed in ﬂ uences on human society, environment, ecosystem, availability of food and fresh water. Multiple lines of evidence indicate that warming of the climate system is unequivocal, and human-induced effects are playing an enhanced role in climate change. It is of utmost importance to ascertain the hydroclimatological changes in order to ascertain the characteristics of detection and attribution (D&A) of human-induced anthropogenic in ﬂ uences on recent warming. Climate change D&A are interrelated. Their study enhances our understanding about the rudimentary causes leading to climate changes and hence, considered as a decisive element in all Intergovernmental Panel on Climate Change Assessment Reports. An extensive discussion of the concerned scienti ﬁ c literature on climate change D&A is indispensably needed for the scienti ﬁ c community to assess climate change threats in clear terms. This study has reviewed various processes and advances in climate change D&A analyses at global/regional scales during the past few decades. Regression-based optimal ﬁ ngerprint approach is majorly employed in climate change D&A studies. The accumulation of inferences presented in this study from numerous studies could be extremely helpful for the scienti ﬁ c community and policymakers as they deal with climate change adaptation and mitigation challenges.


INTRODUCTION
The 'warming of the climate system is unequivocal' because of the consistent overall warming trend since the mid 20th century, which can be attributed extremely likely to humaninduced anthropogenic influence (Stocker et al. ). The cardinal aim of the Paris agreement is to confine the global warming rate 'well below' 1.5 C to 2 C. Since the pre-industrial era the concentration of CO 2 in the atmosphere has increased by about 40% globally. Observed changes in ocean properties such as sea level, ocean heat content, acidification and salinity are consistent with the changes in the atmosphere due to human-induced anthropogenic effects (Bindoff et al. ). Extreme event attribution is a recent growing research field which deals with extremes such as heatwaves, droughts, floods and wildfires which vary greatly in different parts of the world. Attributions of these events never conclude concrete inferences, as these compare the probabilities of occurrence of a particular event in the world under the presence/absence of global warming. These help in better analysis of the processes involved, and the inferences can be potential for future policy interaction. Integrated knowledge from various streams such as climatology, hydrology and sociology can be useful for event attribution science in analysing the effects of extreme events. However, to date, the science of extreme event attribution is in a nascent stage in most important parts of the globe. Heatwave changes are more rapid under anthropogenic climate change and have calamitous effects on human health (morbidity and mortality rates) and biosphere (Perkins-Kirkpatrick et al. ). Climate change has affected oceans in different ways, such as ocean atmosphere circulation, ocean acidification and upper ocean warming, which has led to global sea level rise since 1970.
There is a reduction in snow cover both at continental and regional scales due to anthropogenic influence. The global water cycle has significantly changed since 1960, which is attributed to human-influenced combined changes in ocean and atmosphere. Global, continental and regional scale intensification of climate extremes have been in evidence since the middle of the 20th century.
There is a need of climate change detection and attribution (D&A) studies as they yield comprehensive knowledge about climate science and help in assessing the causes of recent changes in climate. D&A studies improve our knowledge in assessing the impact of human activities on climate change and help in ascertaining the risks and impacts associated with climate change comprehensively. 'Detection' and 'attribution' are interlinked processes, and challenging because of the associated complex spatio-temporal variations in the atmospheric system and interactions between the natural internal and external (natural and anthropogenic) drivers.
A significant gap exists, although this research has been in progress for more than a quarter of a century.
With recent progress in observation, sophisticated climate model simulation and developed methodology, climate change D&A studies have enriched the evidence on human-induced anthropogenic impacts. This paper reviews the recent advances in climate change D&A studies.
It also reviews the role of climate models in D&A analysis. Widely adopted climate change D&A methodologies are discussed thoroughly and their suitability for arriving at reliable attribution statements in different spatio-temporal scales are highlighted. The study reviews the evidence depicting human-induced or naturally driven significant changes in different hydroclimatological variables at regional as well as large spatial scales, namely, global, continental and sub-continental. The effect of humaninduced anthropogenic influences, natural internal and external variability on changes in the cryosphere, climate extremes, circulations and oceanic changes are discussed briefly. Extreme events and associated mechanisms with growing interest on event attribution worldwide are focused on specifically. It is difficult to diagnose regional forcings and their responses in the observational record. Hence, there is a high chance of misattributions at regional scales. Hence, special attention is accorded to analyse the regional effects which can have heterogeneous effects across the globe. The IPCC has published five comprehensive assessment reports (ARs), respectively, in the years 1990, 1995, 2001, 2007 and 2013. The sixth AR is expected to be completed by the year 2022. Each AR consists of three volumes based on three working groups (WG).
As per IPCC AR5, the process of establishing climate change in a defined statistical sense, without assigning any specific reason is known as detection and the process which assesses the relative contributions of multiple potential causal factors for the detected changes is defined as attribution. These two processes are essential components of all IPCC ARs and substantial progress has been accomplished over the years in different IPCC ARs from 1990 to 2013 (Liu & Xia ). Over the years, the confidence level on attribution results has been reported in firmer statistical footing represented as 'likely', 'very likely' and 'extremely likely', respectively, in third, fourth and fifth ARs of the forcing estimates and advancement in D&A approaches summarized in Table S1 in the Supplementary Information (Knutson ). Accumulation of evidence indicates that human influence has considerably enhanced the probability of occurrence of heatwave events across the globe. Precise prediction of future warming trends at regional scale is difficult compared to that at higher spatial scales. Few existing studies are directed to detect (i.e., distinguish from expected natural internal variability) and attribute (i.e., ascribe a cause to) the observed changes in climate at a regional scale.
It is claimed that global warming has stopped or slowed down. However, studies claim these short-term warming/

ROLE OF CLIMATE MODELS IN DETECTION AND ATTRIBUTION ANALYSIS
In order to comprehensively describe the observed warming, combined contributions from natural and anthropogenic forcings are required. Quantification of uncertainty due to modelling and forcing is vital, but uncertainty varies across different forcings, such as small for well-mixed GHGs forcing and large for aerosol and land-use change forcings. It can be complicated further because of feedback processes (Forster et al. ). Aerosols may alter cloud microphysical properties and reduce the amount of solar energy reaching the surface, but at present, our knowledge in this aspect is limited. Any knowledge about volcanic forcing prior to the 20th century is limited unlike the recent history of volcanic activities, which leads to greater uncertainty (Crowley et al. ). Similarly, prior to the pre-satellite era, solar forcing influences on climate were not evaluated clearly (Gray et al. ).

Progress in climate model simulation
As climate model simulations are associated with different uncertainties (as discussed above), it is essential to evaluate them before using further in statistical analysis. This process reduces the chances of spurious detection. The ability of the climate model for simulating the observed changes across a wide range of climate indicators has infused confidence in D&A analysis by reducing levels of uncertainties.
Over the period, across the globe, the evaluation process has greatly expanded with the addition of a range of various performance metrics and performing the process over different hydro-climatic variables ( Johnson et   Since AR4 there has been important improvement by the widespread usage of Earth system models (ESMs), which have the capability of using time-evolving emissions of constituents from which concentrations can be computed interactively. ESMs are the current state-of-the-art models, which are improved from the standard models, i.e., atmosphere-ocean general circulation models (AOGCM).
Interactive representation of the carbon cycle, aerosol and anthropogenic sulphur dioxide emissions are included in ESMs.
Time-varying ozone (stratospheric) is included in the latest suite of models. Hence, CMIP5 climate models (climate and Earth system models) are able to simulate many significant aspects of observed climate (Flato et al. ). Hence, these crucial improvements in CMIP5 promoted our confidence in the model's suitability for application in D&A analysis and for quantitative future projection. CMIP5 includes more comprehensive models with higher-spatial resolution and a wider set of experiments which can address a broader variety of scientific questions.

CMIP5 experiments
Majorly, CMIP5 includes two types of experiments: longterm (century time scale) and near-term integrations, namely, a decadal prediction experiment which is entirely a new addition. These decadal predictions explore the predictive skill of each variable. Long-term simulation is the core simulation and includes atmospheric model intercomparison project (AMIP) run, a coupled control run and historical run (reflecting both anthropogenic and natural sources Pre-industrial control simulations are based on non-evolving pre-industrial conditions which serve for the estimation of unforced variability and provide the initial conditions for historical simulations. The model-derived pre-industrial control simulations obtained from the 'piControl' experiment are available over many centuries, incorporating no change in external climate drivers such as GHG level and solar irradiance, and hence such control simulations do not exhibit the observed warming. These long pre-industrial control simulations were procured from climate models as these were difficult to obtain from observed data which are not free from the effects of external influences. The equivalent natural internal variability (which is essential for D&A analysis) estimation using too short an instrumental record would not be reliable.  which is shown in Figure 1, could be directly employed for selecting appropriate models at regional scale. Improvements in AR5 over AR4 include covering global to regional perspectives with a comprehensive focus on spatial pattern across the globe instead of global mean change. The science of attribution depends on climate model simulation, hence, improvement is needed. It should be borne in mind that good quality, unbiased observed and model data sets are crucial to obtain positive attribution results by minimizing the uncertainty associated with attribution analysis. In the conventional frequentist approach, the fingerprints maximize the ratio of the observed climate change

Regression-based fingerprint approach
One key approach for D&A is the regression-based finger- Optimal fingerprint approach The optimal fingerprint approach is a classical approach which is most frequently used for climate change D&A analysis (Hasselmann ; Allen & Tett ). This approach has been refined over the years (Huntingford Signal-to-noise (S/N) ratio is generally low in the case of variables other than temperature and at regional spatial scale. A thorough description of optimal fingerprint can be found in Hasselmann (). In order to improve the S/N ratio, model-simulated responses and observations are normalized by internal variability. There is a need for inverse covariance matrix estimation using the pre-industrial control simulations of climate model or by considering the variations within an initial-condition ensemble. Several difficulties arise in estimating full covariance as it is obtained from control simulations, which are too short for this to as temporal optimal detection approach. It is different from the classical optimal fingerprint approach as it allows to infer the spatial distribution of the detected signal without providing any spatial guess pattern. They applied this approach to data sets of temperatures and precipitation over France. This approach is well suited to regional scale as spatial properties of the internal climate variability (which is very challenging to estimate at regional scale) are not needed. Hannart () proposed a methodological advancement in the classical optimal fingerprint approach.
Several issues may arise with the compartmentalized treatment involved in the classical optimal fingerprint approach. Hence, the proposed approach presents all available data (i.e., observation, model responses and control simulations) in a high-dimensional spatio-temporal format, i.e., represented in a single statistical model.

Non-optimal fingerprint approach
Qualitatively, non-optimal fingerprint approach can assess the consistency of observed changes with model-simulated changes with respect to different forcings. Thus, non-optimal fingerprint approaches were widely adopted in various

Other approaches
Various approaches have been employed other than  anthropogenic impact or the combination of both (Sonali & Nagesh Kumar , b). This approach is suitable for sub-regional scale and is subjected to uncertainty due to observation, model simulation, climate forcings, model response and simulated internal climate variability. Among all these approaches, risk-based approach was widely used as it can assess the possible anthropogenic influence on an extreme event. Details about the riskbased approach can be found in the section 'Weather and climate extreme events attribution'). Progress in climate change D&A approaches over the last few decades is shown in Table 3.

Hydro-climatic variables
In addition to temperature analysis, scientific attribution of observed hydro-climatic changes, climate-related risks and hazards to human influence can extend to many other aspects such as changing patterns in different variables like precipitation, streamflow, humidity and ocean heat content and will help better to cope with the adverse conditions  Table S3 in the Supplementary Information.

DETECTION AND ATTRIBUTION: DIFFERENT PERSPECTIVES AND IMPLICATIONS
D&A of climate change at continental and regional scales is more challenging compared to global scale (Zwiers & Zhang ; Stott et al. ). Several concerns exist in regional D&A analysis, such as estimation of proper contribution of natural internal variability at a regional scale, utility of excluded important regional forcing in the global climate model simulations and non-guarantee of accurate model simulations at regional scale. Along with large scales, evidence from regional studies reflect a growing interest in ascertaining causes and effects of climate change, which can vary significantly across the globe. Policymakers are more concerned about regional inferences obtained from D&A analysis. Various studies have established the significant contribution of anthropogenic influence in changing climate at global and regional scales (Stott et al. ). Extreme events pose various challenges to society, such as health hazards and crop damage.
Regional D&A analysis is more difficult than at global scale. The contribution of internal variability is amplified at regional scale. Climate model simulations (such as preindustrial control simulation and historical natural, GHG and miscellaneous forcings) are less dependable at regional scale compared to global scale and it is difficult to apportion responses to different forcings at regional scale.
The crucial challenges in climate change D&A studies at regional scale are due to dominant natural internal variability, uncertainties in the climate model outputs and uncertainties in observational data sets, uncertainty in the regional forcing such as land use change and impact of aerosols. Various studies have been conducted at continental, sub-continental and regional scales, and manifested the This could be accomplished by suggesting the combinations of forcings which produce the highest accuracy in estimating β GHG (GHG-induced warming scaling factor), which was used in the regression-based statistical model suggested by Allen & Stott (). They suggested the optimal strategy (Combination of all þ Aerosol only þ Natural only forcings) which can be adopted by many modelling centres in the upcoming CMIP6 D&A exercise. They mentioned the allocation of large ensemble size to the weaker forcing. The new phase CMIP6, which was enacted for the betterment of model simulation in every possible way, could be helpful in D&A analysis. There is always important scope to investigate the influence of upgraded data sets, sampling and model uncertainties on the existing conclusions. Regional scale attribution remains challenging because signal separation is limited by lower S/N ratio due to the dominant impact of internal variability. Important regional forcings (such as land use change and short-lived forcings) should be considered along with improved spatial resolution of global climate models for successful regional D&A analysis.
Periods with prolonged abnormal hot weather known as heatwaves are increasingly becoming common and have a disastrous impact on human health and the environment.
Increase in frequency of heatwave is seen in many parts of the world, but it is more predominant over tropical regions is no unanimity about the best methodology to be adopted for event attribution.

CONCLUSIONS
The discernible human influence on global climate is a major cause of concern and profound focus should be devoted to it. The objective of this paper was to review the advances made in assessing different climatic changes across the globe due to human influence and natural variations. Successful adaption strategies necessitate primal understanding to be obtained from an extensive review on climate change detection and attribution (D&A) analyses.
This review majorly discussed the ongoing research on climate change D&A at global, continental and regional scales considering various hydro-climatic variables, role of climate models in D&A analysis, the associated uncertain- Considering future challenges for the science of D&A, to better analyse the present pace of change and to understand the physical processes driving the regional-scale changes, a refined understanding of the effects of external forcings and internal variability is highly essential.