Advances in Applied Energy Heterogeneity of consumption-based carbon emissions and driving forces in Indian states

In India’s low carbon transition, policymakers should not only focus on a local mitigation policy in developed states, but on carbon leakage from the developing states, given the signiﬁcant heterogeneity in industrial distribution and population. Inter-state cooperation is recommended, with developed states subsidizing the mitigation in the developing states, which also entails a lower marginal cost to low carbon transition.


Introduction
India, responsible for 7% of global CO 2 emissions in 2018, is the third largest energy consumer and CO 2 emitter in the world [1] . Since 2010, India's emissions increased from 1750.56 Mt to 2621. 29 Mt in 2018, and were responsible for 21.5% of the global emission increment over the same period [2] . This sharp increase of CO 2 emissions in India is mainly driven by economic growth, and in particular by the rapid industrialization of some developed states [ 3 , 4 ]. Such performance suggests that India will be the next carbon giant. On the one hand, as home to 1.3 billion people, only 32% of India's population lives in urban areas, which means that it is likely to see the world's largest shift towards urbanization (with an estimate of a 50% urbanization level by 2050) [5] . On the other hand, India is the seventh largest country in terms of GDP, having an average annual growth rate of 7.0% GDP [6] but emits 1.9t CO 2 per capita a year, eight times lower than that of the United States strategies and actions at national and subnational level [12] . Due to its vast territory, diverse socioeconomic backgrounds and resource endowments, there is huge heterogeneity between Indian states. For example, less developed central and eastern states, such as former "BIMAROU " states, 1 including Bihar, Madhya Pradesh, Rajasthan, Odisha and Uttar Pradesh, are characterized by a weaker economy and lower socioeconomic status, but with sufficient natural mineral resources and cheap labor [13] . In order to comprehensively formulate regional mitigation policies, a pilot research project for feasibility and efficacy of interstate emission trading should be implemented, offering insights into the regional disparities discussed above [ 14 , 15 ].
Previous studies have already investigated subnational productionbased emissions and emphasized its policy implications for regional mitigation [16][17][18] . However, progress against emissions targets could also be evaluated through the use of consumption-based emission inventories that reallocate emissions from states where products are produced, to states where products are ultimately consumed [ 19 , 20 ]. Such consumption-based accounting of CO 2 emissions may better reflect the responsibility for paying the costs of emission mitigation [ 21 , 22 ]. Regarding research on consumption-based emissions for India, most studies focus on the national level, resulting in highly aggregated data and absence of regional variations [ 23 , 24 ]. For a large country like India, addressing spatial aggregation issues and considering regional heterogeneities, will help clarify root drivers and allocate emission reduction responsibility to regions. Previous studies have already proposed a multi-regional input-output model for regional emission studies and have explained the pattern of emission flows in other countries with vast areas of territory [ 25 , 26 ]. To the best of our knowledge, India's consumption-based emissions at a state level have not been quantified, leaving open the question of how to mitigate emissions from a regional and consumption-based perspective.
Here, our study focuses on the heterogeneity in consumption-based accounting of CO 2 (from combustion of fossil fuels) at state level in India and tracks CO 2 emission flows within India, by constructing a state-level multi-regional input output model (MRIO). Due to the lack of a MRIO table, the Flegg location quotient method and gravity model are applied to compile an Indian MRIO table in 2015, which includes 32 states and 50 sectors. The contribution of this paper is twofold: first, consumptionbased emission inventories at state level are estimated for the first time; and second, we track CO 2 emission flow among states in India and establish the pattern of outsourcing of CO 2 emissions. Our findings demonstrate that household consumption in affluent states dominate the total consumption-based emissions in India. The pattern of outsourcing of CO 2 emissions within India suggests that another contribution these affluent states could make is to introduce cleaner technologies into less developed states via some financial mechanism, such as an interregional clean development mechanism. Our findings and suggestions are also made for the improvement of regional energy efficiency and management of energy use, due to the connection between energy use and CO 2 emission.
The paper is organized as following: After the introduction section, we give a brief review of the existing literature. Then, we briefly introduce the accounting framework of consumption-based emissions in India and the underlying data. In the results section, we provide a descriptive overview on consumption-based emissions among 23 regions (including 3 union territories, 2 aggregated regions and 18 states) in India, illustrating ranking and composition of consumption-based emissions. We then rank emission flows within India and identify which states are key net emission exporters and importers. After the results section, we discuss the domestic and international policy implication of our results. Lastly, we conclude this study and state the uncertainties and limitations. 1 Note: Rajasthan and Odisha no longer belong to BIMAROU after 2018.

Literature review
As the Indian climate issue has become an increasing concern for 1.5 °C global mitigation target, a considerable body of research has investigated the potential trajectories to achieving low carbon development and emission mitigation policies [27] . In the existing literature, most research has focused on production-based emission management. Some scholars focus on specific industries, such as the iron and steel industry [28] , the electricity generation industry [29] , the cement industry [ 30 , 31 ]. William et al. have initialed a forward projection of output from iron and cement industries and estimated the energy and CO 2 emissions savings, based on a bottom-up conservation supply curve methodology [31] . Some researchers also focus on production-based driven factors of CO 2 emissions. Rapid economic growth in India has been the dominating driving force contributing to the increase in carbon emissions, while energy intensity has been the main factor reducing carbon emissions [ 7 , 32 ]. Meanwhile, research on demand-side management has gradually emerged. Some papers have quantified national indirect emissions and complete emissions [ 24 , 33 ], and have identified dynamic or global drivers from a consumption-based perspective [34][35][36] . For example, Wang et al. have found that since 2008, final demand has been the major driving force behind the increase in carbon emissions transfers from capital formation to household consumption [34] . Few studies have investigated regional analysis to illustrate the specific policy implications. Jemyung et al. have quantified regional carbon footprints of households, and highlight the need to differentiate individual responsibilities for climate management [37] . However, previous studies have not clarified the root drivers, taken into account regional heterogeneities in the industry chain, which is the problem addressed in this paper.

Accounting scope
A clear definition of the scope of emission accounting could avoid double-counting in emission estimates. The International Council of Local Environmental Initiatives initialed 3 scopes of emission boundaries [38] . Scope 1 includes direct emissions from fossil fuels combustion within territory boundaries. Scope 2 includes emissions from consumption of imported electricity from upstream power plants. Scope 3 includes the indirect emission from upstream productions, that is, emissions embodied in trade [39] .
In this study, carbon emissions were calculated using two approaches: production-based emissions (Scope 1, Sectoral Approach) and consumption-based emissions (Scope 3, Regional Input-output Model) [40] . Production-based carbon emissions in this study only refer to energy-related emissions from the combustion of fossil fuels, excluding emissions from industry processes or agriculture production. In addition, this study only considered CO 2 emissions without other non-CO 2 greenhouse gasses (CH 4 , NO 2 ), because of lack of regional emission factors. Based on the production-based emission inventories, we applied the MRIO model to calculate consumption-based emissions.

Flegg location quotient method to compile the MRIO
An input-output table (IOT) based on extensive surveying by statistical institutions is considered better than a non-survey-based IO table [ 41 , 42 ]. Many practitioners are also critical of applying some nonsurvey methods for simulating IO tables [ 43 , 44 ]. Some hybrid methods are favoured, requiring fewer primary data but retaining significant accuracy. However, hybrid approaches are still costly in some developing countries because of the imperfection of statistical systems [45] . Therefore, given the limited data resources in sub-national IO tables in India, we used the Location Quotient (LQ) method, a non-survey method, to compile the MRIO table for India's 32 states.
In the standard LQ method, regional input coefficients are specified as a piecewise function of national input coefficients and LQ. Mathematically, it is expressed as: Where is intraregional input coefficients; is national input coefficients; is the proportion of regional requirements of input purchased from within region .
can be obtained directly from the national IO table. can be derived from the steps in supporting information (SI). Further Obviously, the residual is non-negative because the intraregional input coefficients are bounded by national input coefficients. The residual then needs to be distributed among the interregional transactions with . Here, refers to a proxy (or an initial estimate) of interregional transactions or an initial estimate.
indicates the distance between region and region .
Then, consistent estimations are constructed with the help of a key parameter of , scaling the , so that they sum to unity: Therefore, the interregional intermediate transaction estimates can be simply proportional to the size of the sending sector and the receiving sector.
Initial estimates of final demand, imports and export of each region are shown in supporting information (SI). These estimates based on sectoral value added, output, sample survey datasets, and etc., do not meet the "double sum constraints ", in which the row and column totals match the known values in national IOT [46] . The minimum entropy approach is used to adjust the initial estimates to ensure agreement with the summed constraints. The procedure of minimizing the squared distances tends to preserve the structure of the initial estimates as much as possible with a minimum number of necessary changes to restore the row and column sums to the known values.

Consumption-based accounting of CO 2 emissions
The multiregional input output model is the accepted tool employed to account for consumption-based emissions, which illustrates the interregion and inter-industry relationships along supply chains and interregional trades [47] . Specifically, consumption-based CO 2 emissions can comprehensively present emissions directly and indirectly embodied in the final consumption [48] . In this study, consumption-based emissions in certain regions includes emissions from local production and emissions embodied in domestic imports, excluding emissions embodied in international imports.
In the MRIO model, the total output of each industry in each region, x , stimulated by a vector of final demand, y , is given by Where ( − ) −1 is the Leontief inverse (matrix of total requirements), with indicating the total output of industry in region that is stimulated by a unit of demand of in region ; where is the direct requirement matrix, with indicating the amount of direct inputs from industry in region required to generate one unit output of in region ; where is the final demand matrix with indicating the final demand for goods of industry in region from region ; where is the total output matrix with indicating the total output of industry in region r. Using familiar matrix notation and dropping the subscripts, we have the following: To calculate the embodied emissions in the goods and services, we extend the MRIO model with environmental extension by defining direct emission intensity, which indicates CO 2 emissions per unit of economic output for all industries and regions. Mathematically, it is expressed as: = ∕ , where indicates the CO 2 emission emitted from industry in region . Therefore, total consumption-based emissions embodied in goods and services used for final demand can be estimated by: where is a vector of emission intensity for all industries in all regions; is the final demand including four categories for region from all regions.
In turn, the total embodied emissions in exports from region to region can be estimated by: where represents the direct emission intensity for region but zero for all other regions; similarly, indicates the final demand including four categories for region from all regions. ̂ is the diagonal matrix of .

Data sources
The production-based carbon emissions from fossil fuels (e.g. coal, petroleum and gas) for regions in India came from the GHG Platform of India (GHGPI), which is the most comprehensive GHG (greenhouse gas) emission datasets at a state level in India (see Table 1 and Table 2 ). These emission data included 35 regions and 22 sectors ( SI Table S2 ). The Multi-Regional Input-Output Table (MRIOT) used in this paper was compiled based on 2015 Indian Supply-Use Table (SUT) at a national level. The MRIOT contains 32 regions and 50 economic industries for each region. In addition, imports from other countries appeared as a single row vector on the input side in MRIOT, and the intermediate transaction matrix and final use only included goods and services produced domestically. Hence, we mainly discuss the consumption-based emissions supported by domestic production, excluding imported emissions from other countries. To facilitate discussion, several states with lower population density and GDP (see SI Table S1 ), were clustered into geographic regions (i.e. Himachal Pradesh, Jammu and Kashmir. Uttarakhand are aggregated into North region, and Assam, Sikkim and Tripura are aggregated into Northeast region). Basic data for compilation of the MRIOT came from CEIC datasets, Annual Survey of Industry (ASI) datasets and Fig. 1. a shows the ranking of state-level consumption-based emissions. b shows the ranking of state-level consumption-based emissions intensity per GDP. c shows the ranking of state-level consumption-based emissions intensity per capita. For all bar chart, length of bar indicates its size according to the x-coordinate, and color of bars corresponds to state GDP per capita from the least developed states in yellow to the most developed states in blue.
the Ministry of Statistics and Programme Implementation (MoSPI). Detailed information about how we compiled the MRIOT can be found in the Supporting Information (SI).

The heterogeneous distribution of consumption-based emissions for India in 2015
In 2015, India emitted 1766 Mt CO 2 from the burning of fossil fuels, accounting for 6% of global emissions. 83% of emissions were consumed by domestic final demand and the remaining 17%, driven by export to other countries. Fig. 1 ranks consumption-based emissions in 2015 for all 23 regions in India. Some states where services and light industries are substantially developed, display high consumption-based emissions. For example, the highest consumption-based CO 2 emissions of 195.1 Mt was in Maharashtra, (accounting for 13.5% of the total), which is mainly due to the state having the largest GDP in India (accounting for 14% of national GDP). More than 70% of its GDP is attributed to services and light industries. Similarly, 96.7 Mt emission in Karnataka and 128.7 Mt emission in Tamil Nadu also are associated with high levels of total GDP. By contrast, some states such as Madhya Pradesh, Chhattisgarh and Odisha, have lower consumption-based emissions, which is associated with being less populous and having lower per capita GDP. The economy in these states is heavily dependent on coal use and energy intensive activities such as energy industry, heavy industry or materials manufacturing, with the result that consumption-based emissions are significantly lower than production-based ones. However, when standardized by socioeconomic factors (GDP and population), the ranking changes drastically (see Fig. 1 ). Uttar Pradesh, as the second largest consumption-based emissions consumer, has a middle level of emission intensity by GDP and a lower level of emission intensity per capita. This is because Uttar Pradesh is the most populous state and has lower GDP per capita. In other words, the consumption-based carbon intensity (emissions per unit of GDP) is greatest in less developed states in central and east areas. By contrast, the high GDP and more developed states in the west, south and union territories, tend to be the least carbon intensive. For example, Odisha, whose value-added iron and steel accounted for 18% of its GDP, had the highest consumptionbased emission intensity, 27.04 g CO 2 per Rs, which is more than double the consumption-based emission intensity of Maharashtra. Conversely, per capita consumption-based carbon emissions in developed states and union territories such as Goa, Delhi, Haryana and Puducherry, are more than three times that of Odisha, Madhya Pradesh, and Jharkhand. Affluent states with a GDP per capita of around 200,000 Rs or more such as Goa and Delhi, have consumption-based emissions per capita of around 3 tonnes. In contrast, Jharkhand, an eastern state with a GDP per capita of just around 60,000 Rs, has consumption-based emissions per capita of around 1 tonne.

The heterogeneous drivers of consumption-based emissions in India
Consumption-based emissions can be supported by domestic productions and imported goods and services (shown in Fig. 2 a), which explains where the emissions come from. Consumption-based emissions can also be triggered by different drivers, such as four categories of final demand (shown in Fig. 2 b) and different sectoral final demand (shown in Fig. 2 c), which demonstrate the heterogeneity of drivers in Indian states. In 23 states or regions of India, with the exception of Odisha and Chhattisgarh, more than 50% of consumption-based emissions were attributed to imports from external regions, which indicates that in 2015, consumption relied heavily on foreign products as opposed to local products. The above results are consistent with previous studies which state that approximately 40-80% of emissions in Japan were attributed to imports [49] . For example, Delhi (94%), the top state in terms of shares of imported emissions, corresponded with its overwhelming consumptionbased emissions that were three-fold its production-based emissions. Chhattisgarh and Odisha, by contrast, accounted for the minimum percentages of imported emissions, reaching a low level of 40-50%, because they are typical production-based emission producers. From the perspective of categories of final demand, household consumption induced the largest share of consumption-based emission, followed by fixed capital formation. The results are consistent with previous researches on consumption-based emissions in India at the nation-level [33] . More than 60% of consumption-based emissions in each state were triggered by households, especially in Chhattisgarh (78%) and in Odisha (75%). Emissions induced by capital formation accounted for a larger proportion of total consumption-based emission in more affluent states, such as the top three consumers of Tamil Nadu, with 33%, Maharashtra and Uttar Pradesh, with 27%, than less developed states of Odisha and Chhattisgarh, with 18%, and Madhya Pradesh, with 22%. An underlying explanation is that industrial sectors in less developed states with high emission intensity, consumed more but invested less in new high-tech growth and urbanization. In contrast, development models in more developed states, including rapid economic growth associated with rapid urbanization, push investment in transportation infrastructure and construction, which accounted for 70% of total capital formation in India. From the perspective of the consumption-based emissions of different sectors, services, transport, energy industry and construction are the main contributors. Fig. 2 c shows regional disparities between  sectoral consumption-based emissions. Energy industry accounted for a large share of consumption-based emissions in heavy-industry intensive states, while services and equipment industry are the main drivers of consumption-based emissions in developed states. For instance, energy industry accounted for 53% and 17% of total emissions in Chhattisgarh and Maharashtra, respectively, while services and equipment industry accounted for 15% and 37% of total emissions in these two states. Table 3 A, shows the largest gross emission flows within India, reflecting the patterns of CO 2 outsourcing. Maharashtra and Tamil Nadu are the most populous and affluent regions in India, accounting for nine of the top 10 emission inflows (seven inflows to Maharashtra, two inflows to Tamil). Former "BIMAROU " states with lower socioeconomic status, accounted for six of the top 10 emission outflows. For example, 40% of consumption-based emission in Maharashtra are imported from Gujarat (13%), Madhya Pradesh (7%), Chhattisgarh (6%), Odisha (7%) and Uttar Pradesh (7%). Table 3 B and D show emission flows induced by fixed capital formation and household consumption, respectively. Similar to gross emission inflows, Maharashtra is also the major destination of the largest emission inflows triggered by household consumption. People living in locations with the largest inflows have much higher per capita household consumption than people living in other states that export emission intensive goods to more affluent regions. This indicates that the higher levels of household consumption in developed states are supported by carbon intensive production in less developed, heavy industry-oriented regions. In keeping with rapid growth, more developed regions, such as Maharashtra and Tamil Nadu, invest a larger share of GDP (27-33%) to promote urbanization than less developed states (only 18-22%) in east and central areas of India, resulting in a higher level of per capita fixed capital formation. However, the largest emission inflows embodied in capital formation normally originate from less developed states, such as Odisha and Chhattisgarh, to those developed states. This is mainly caused by the difference in industrial structure and endowments of natural resources. For example, the central and eastern regions have the highest reserves of metal and non-metal minerals and a lower position in the supply chains.

Spillover effects from inter-state emission flows across India
In addition, our results show that states with higher emission intensity of production tend to be net emission exporters. Table 3 C , shows that the largest net inflows originate from high intensity states to low intensity states (comparisons of sectoral intensity are shown in SI Table S4 ). Per GDP emission intensity consists of energy intensity (energy   consumption per GDP) and carbon intensity (CO 2 emission per energy consumption). In the case of energy intensity, developed states or union territories, such as Maharashtra and Delhi, entail a lower level of energy intensity than those of less developed states, due to advanced technology for production and greater shares of low-energy but high-value-added products in more developed regions. In another case of carbon intensity, states located in north or northeast India tend to consume a greater share of low-carbon energy sources (i.e. solar power, wind power and nuclear power, see SI Table S3 ) than states located in east and central parts of India, but rely heavily on heavy industry. Fig. 3 shows, at the sectoral level, the imported emissions and exported emissions of each state or region. For some service-oriented states, the largest share of embodied emissions in imports can be attributed to the service sector. In addition, there is a relatively large amount of embodied emissions in the imports of light industry in these states (e.g. Maharashtra, Tamil Nadu, Delhi and Karnataka), followed by construction, transport and food-related industries. For some heavy industry-oriented states (e.g. Chhattisgarh, Odisha and Jharkhand), the largest share of embodied emission in exports can be attributed to construction, followed by light industry.

The heterogeneous drivers of emission flows across India
In terms of net emission flows between inflows (emissions embodied in imports) and outflows (emissions embodied in exports) of each region, the net emission exporters are highly centralized, with the top four exporters accounting for 80% of total net outflow in India, while the net emission importers are more decentralized. The largest net emission importers are dominated by the affluent states and highly urbanized union territories such as Delhi, Maharashtra and Tamil Nadu. Some less developed regions, such as those in the north (including Himachal Pradesh, Jammu and Kashmir, Uttarakhand) and northeast (including Assam, Sikkim and Tripura) also import large embodied emissions, produced elsewhere in India, with respective net imported emissions of 26.4 and 11.0 Mt. The underlying explanation is that there is a cleaner energy mix in those regions, where a large share of the electricity supply is borne by solar and wind power. In addition, lower levels of industrialization in those regions prompt more imports of high-tech industrial goods from external regions [50] . Normalizing net emission flows per unit of capita and per GDP, further highlights the imbalance of emission flows from less developed eastern regions to the developed south and west regions or highly-urbanized union territories. For example, per unit GDP carbon intensity of net emission exporters is greatest in Chhattisgarh (34.1 g CO 2 per Rs), Odisha (25.6 g per Rs), and Jharkhand (10.3 g per Rs), due to the prevalence of emission intensive products (i.e. iron and steel, non-metal minerals, electricity mainly supported by fossil fuel) exported from these states. Meanwhile, per unit capita carbon intensity of net emission importers is greatest in Delhi (1.94 Tonne CO 2 per capita), Chandigarh (1.50 Tonne per capita) and Goa (1.50 Tonne per capita) due to the higher per capita consumption in affluent regions.

Discussion
As potential giant emitter, India's climate actions to address climate change are critical to the 1.5 °C global mitigation target [51] . The largest challenge to seeking a cleaner pathway to industrialization or urbanization in India is that there is no successful previous model to follow, among either developed countries or industrialized emerging countries like China. Mitigation in a vast country like India cannot be achieved without considering the regional heterogeneity in industries and socioeconomic stages. Like most countries with vast territory, regional emission disparities in India also show that the most affluent regions dominate the total consumption-based emissions. For example, consumptionbased CO 2 emissions in Maharashtra with the largest GDP in India, accounted for 13.5% of the national total. The overall economic growth rate of the state is expected to be at a compound annual growth rate of 7% until 2051, and is likely to remain the highest contributor, accounting for about 12.8% of national GDP by 2051 [12] , which may correspond to its position as largest contributor to total emissions. High economic growth in the affluent states would undoubtedly increase their consumption-based emissions, especially carbon flows embodied in the supply chains from other under-developed states with higher carbon intensity. The spillover effects could undermine mitigation efforts, observed in other developing countries, and require the policymakers to shift the scale of mitigation more broadly with the participation of multistates, especially for those inextricably connected in terms of supply chains.
Exploring the green growth way requires cooperation to balance economic growth and inequality reduction. In a carbon-constrained future, developed countries' efforts to become carbon neutral (e.g. capturing carbon and storing) [52] , will save the carbon quota for industrialization for developing countries, such as India. Similarly, the carbon quota should be also allocated at the sub-national level between affluent and developed states such as Delhi, Goa and Maharashtra, and less developing states such as Odisha and Telangana. Therefore, these affluent states should also set ambitious targets of mitigation (e.g. carbon neutrality) to provide room for less developed states. Our study found that household consumption was the largest contributor to consumption-based emissions at state level. More than 60% of total emissions in 2015 were triggered by embodied emissions in households in each state, which are highly associated with transport, food processing and energy. A lowcarbon lifestyle of wealthier households is critical. Considering the comparative advantages in wealthier areas, planting trees or carbon capture and storage (CCS) in wealthier areas is not a reasonable approach to achieving carbon neutrality. Connecting poor regions and rich regions through a carbon credit mechanism will provide opportunity for a carbon sink [53] .
By contrast, capital formation, as the largest contributor to consumption-based emissions in most developing countries [54] , only accounted for 20-30% of total regional consumption-based emissions in India. This reverse was mainly caused by India's large population but lower level of urban infrastructure expansion. However, with rapid industrialization since the financial crisis in 2008, India as the second largest populated country, also has the world's largest potential for urbanization, which will make capital formation dominant in India again. We also found that more urbanized states are characterized by a higher proportion of emissions embodied in capital formation, for example 33% in Tamil Nadu, 27% in Maharashtra but only 18% in Odisha and Chhattisgarh. In terms of percentage, in less developed states, the urban population is significantly lower than the national average. Therefore, theses states are in a unique position to chart out an urbanization path that learns from the mistakes or experiences of other more developed states. Given the climate change dimension, these states should go further by defining a climate-friendly urbanization path. For example, developing an energy efficient building and transport infrastructure in the primary stages of urbanization, could be a low-cost way to substantially reduce future energy use caused by the potential expansion of urbanization.
Interregional trade in India, triggered by comparative advantages and regional heterogeneity, brings economic benefits both for developed states (lower costs) and developing states (GDP growth). Meanwhile, the high carbon flows embodied in inter-state trade indicate the considerable spillover effects. To be more specific, household consumption in developed states such as Maharashtra and Tamil Nadu, were supported by outsourcing emissions to surrounding regions, while capital formation was mainly supported by emissions occurring in less developed states that are heavily reliant on heavy industry. The pattern of outsourcing of CO 2 shows the trade between developed states and developing states tend to be carbon-intensive and involve low value-added goods, such as metal and non-metal industry. In addition, we found that the net emission flows tended to originate from high intensity states with to low intensity states. The major mechanism for interregional trade to drive CO 2 emissions rests on differences of production-based emission intensity between importers and exporters [55] . Carbon-intensive manufacturing in less developed states, such as former "BIMAROU " states, entails drastically more emissions than making the same products in the most afflu-ent states, using advanced technology; thus, interstate trade increases India's CO 2 emissions. We also found that net emission exporters are highly centralized and mainly located in the east. Restructuring the industrial sector among Indian states goes against existing comparative advantage. Therefore, there is a potential opportunity to decarbonize, by improving emission intensity in the main net exporters in India, in which more energy-efficient technologies can be installed. Such action will upgrade the industry chain and enhance the value added of products in developed states. Given that consumption in developed states is supported by carbon intensive production in less developed heavy industry-oriented regions, such low marginal cost approaches can be supported by the affluent regions' efforts to introduce accessible, hightechnologies into less developed states in India. For example, an interregional clean development mechanism (CDM) will incentivise developed net importers with cleaner production processes to buy carbon emission permits [56] , by investing in less developed net exporters to use advanced technologies and upgraded productions.

Conclusion
In the study, we used the Flegg location quotient method and gravity model to construct a multi-regional input-output table for India in 2015. Consumption-based CO 2 emissions at state level for India in 2015 were first measured to establish regional disparities and supplement regional mitigation policies. We found that the developed western region dominated India's carbon footprint. Household consumption dominated consumption-based emissions in all states, while investment-led emissions were relatively higher in developed regions than in the developing regions. Consumption-based emissions were significantly redistributed among Indian states via interstate supply chains. We found that states not only emitted CO 2 emissions within territory boundaries but also imposed emissions on other regions through interregional trade. Especially, carbon-intensive productions in eastern region are significantly triggered by demand in developed western region.
Mitigation policies should focus on the connection between developed and developing states. A low-carbon lifestyle of wealthier households in developed states and planting trees or carbon capture and storage in developed states, via a carbon credit mechanism (payment from developed states) will save the carbon quota for developing states. Seeking a climate-friendly urbanization path in developed states will not only reduce local carbon emissions, but also future energy use and carbon emissions caused by the potential expansion of urbanization in developing states. Introducing accessible, high-technologies from developed states via an interregional clean development mechanism will improve emission intensity in developing states and also bring about high valueadded industries. Future work can link our multi-regional input-output table with global input-output table (for example Global Trade Analysis Project) to the relationship between Indian states and other countries. A time-series research, such as Structural decomposition analysis, is also proposed to explore dynamic drivers in different regions.

Limitation and uncertainty
This study faced a number of limitations and assumptions in the process of conducting the research. First, we only considered consumptionbased emissions from domestic production in our study. The data of imported goods at a state level in MRIO were not officially available during the study period. However, India, as a "manufacturer " in global supply chains, was characterized by higher exports than imports. Consumption in India was mainly supported by domestic goods. Therefore, the uncertainty about a possible disconnect between India and the rest of the world, was limited. Second, our analysis focused on energy-related CO 2 emissions excluding other greenhouse gasses (for example, CH 4 , NO 2 , emissions from agriculture production, and emissions from industrial processes), such that we may undervalue the implication of agriculture on mitigation. Third, the interregional trade estimated in this study is based on the FLQ model and gravity model due to data availability, which inevitably leads to some uncertainty regarding interregional trade.

Author contributions
H.Z. and J.M. designed the study. Q.H. performed the analysis and prepared the manuscript. Q.H., H.Z., J.M. and interpreted the data. D.G. and J.L coordinated the project. All authors participated in writing the manuscript.

Declaration of Competing Interest
The authors declare no competing financial or nonfinancial interests.