IO analysis studies feature a sub-stream dealing with disaster analysis. Okuyama (2007) provides a comprehensive review of the use of IO analysis for economic analysis of disasters. Quantitative disaster analysis is needed for understanding the impacts of a disaster, for driving effective disaster response, for informing disaster risk reduction and adaptation efforts and for pre-emptive planning and decision-making (Cannon, 1993; Lesk et al., 2016; Prideaux, 2004; Temmerman et al., 2013). It is intuitively clear that a disaster results in direct losses in the form of infrastructure damages and indirect higher-order effects in the form of subsequent losses in business activity (Rose, 2004). The ability of IO analysis to capture the upstream interconnected supply chains of an industry or region affected by a disaster makes it an ideal tool for assessing the full scope of impacts of a disaster event. In addition to IO analysis, computable general equilibrium (CGE) models, econometric models and social accounting matrices (SAM) are alternative modelling frameworks for estimating the indirect higher-order effects of a disaster (Cole, 1995; Guimaraes et al., 1993; Koks et al., 2016; Koks and Thissen, 2016; Okuyama, 2007; Okuyama and Santos, 2014; Rose and Guha, 2004; Rose and Liao, 2005; Tsuchiya et al., 2007). A discussion of these models is beyond the scope of this study and we focus on IO analysis, in particular the post-disaster consumption possibilities and possible spillovers (explained further below). IO modelling has been applied to many disasters such as earthquakes in Japan (Okuyama, 2014, 2004), floods in Germany (Schulte in den Bäumen et al., 2015) and London (Li et al., 2013), terrorism (Lian and Haimes, 2006; Rose, 2009; Santos and Haimes, 2004), hurricanes (Hallegatte, 2008) and blackouts (Anderson et al., 2007) in the USA, as well as diseases and epidemics (Santos et al., 2013, 2009), to name a few.

Prior research on disaster impact analysis, based on IO analysis, has sought ways of improving the standard IO model, for example by extending the standard framework to include temporal and spatial scales (Okuyama, 2007). For example, Donaghy et al. (2007) propose a flexible framework for incorporating short- and long-time frames using the regional econometric IO model (REIM), and Yamano et al. (2007) apply a regional disaggregation method to a MRIO model to estimate higher-order effects according to specific districts. Furthermore, a so-called “inoperability index” within the inoperability input–output model (IIOM) has been proposed as a way of assessing the effect of a disaster or initial perturbation on interconnected systems (Haimes et al., 2005). Both the static and the dynamic versions of IIOM have been applied to the case of terrorism for assessing the economic losses resulting from interdependent complex systems (Lian and Haimes, 2006; Santos and Haimes, 2004). Using the dynamic version of IIOM, it is possible to assess recovery times and also to identify and prioritise systems and sectors that are most economically critical for guiding the recovery process (Haimes et al., 2005).

One particular type of disaster IO analysis, proposed by Steenge and Bočkarjova (2007) aims to investigate post-disaster consumption possibilities as a consequence of production shortfalls resulting from a disaster. As this method uses Leontief's demand-driven model, it captures backward, upstream supply-chain impacts resulting from a disaster. Such an assessment has been applied, for example to widespread flooding in Germany (Schulte in den Bäumen et al., 2015) and electricity blackouts from possible severe space weather events (Schulte in den Bäumen et al., 2014). Here, we apply this method for the first time to undertake an estimation of post-disaster consumption possibilities, and subsequent losses in employment and economic value added resulting from the 2017 Tropical Cyclone Debbie in Australia. To this end, we use the Australian IELab to construct a customised subnational MRIO table for Australia with extensive detail on regions directly affected by the cyclone. In particular, and this is the novelty of our research, we examine detailed, disaggregated regional and sectoral spillovers including the consequences of this cyclone not only for directly affected regions and industry sectors, but also for the wider national economy.

A specific stream of IO analysis is disaster analysis (Okuyama, 2014, 2007), focused on IO databases employed to explore how an economy can be affected by a sudden slowdown or shutdown of individual industries. Since we are primarily interested in post-disaster consumption possibilities and ensuing employment and value-added loss, we utilise the approach by Steenge and Bočkarjova (2007). In essence, a disaster reduces total economic output x0 of industry sectors 1,…,N to levels x̃=I-Γx0, where Γ is a diagonal matrix of fractions describing sectoral production losses as a direct consequence of the disaster, and I is an identity matrix with the same dimensions as Γ. The entries of Γ are populated on the basis of primary data, in our case about cyclone Debbie (Sect. 2.4). Post-disaster consumption possibilities y1 are then the solution of the linear problem max⁡(1y1)s.t.(i)y1=(I-A)x1(ii)x1≤x̃,and(iii)y1≥0, where 1={1,1,…,1} is a summation operator, A=Tx1‾-1 is a matrix of input coefficients, T is the intermediate transaction matrix, the “^” (hat) symbol denotes vector diagonalisation, and x1 is the post-disaster total economic output. Constraint (i) in Eq. (2) is the standard fundamental IO accounting relationship stating that in every economy intermediate demand T and final demand y sum up to total output x. This can be seen by writing y1=(I-A)x1=x1-T1⇔T1+y1=x1. Constraint (ii) states that in the short term, post-disaster total output is limited by pre-disaster total output minus disaster-induced losses. Constraint (iii) ensures that final demand is strictly positive.

Condition (i) is different from the approach in Steenge and Bočkarjova, because we need to ensure the positivity of final demand y. Taking these authors' equation (Eq. 23) and recalculating for I-Γ=0.2000.8, we obtain negative post-disaster consumption possibilities -132.4. Our approach would yield the post-disaster situation 0.250.40.140.122037.5+030.2=2037.5, with non-negative post-disaster final demand and with post-disaster output x1=2037.5≤x̃=2040.

A disaster-induced transition to lower consumption levels y1=y0-Δy has implications for the state of regional economies, as it causes losses in value added and employment ΔQ=qΔx=qI-A-1Δy, where q holds value-added and employment coefficients. The sequence of these losses can be enumerated by carrying out a production layer decomposition, that is by unravelling the inverse in Eq. (3) into an infinite series (see Waugh, 1950) as ΔQ=qΔy+qAΔy+qA2Δy++qA3Δy+…=∑n=0∞qAnΔy, where the term qΔy represent the job and value-added losses borne by producers immediately affected by the reduction of consumption possibilities due to the cyclone, qAΔy describes first-order losses fielded by suppliers of cyclone-affected producers, qA2Δy 2nd-order losses for suppliers of suppliers and so on for subsequent upstream production layers. First- and higher-order upstream losses can in principle occur anywhere in Australia, depending on the reach of the supply-chain network of local northern Queensland producers.

In order to quantify indirect economic impacts of Cyclone Debbie, we first constructed a 19-region-by-34-sector IO model of Australia, with a particular regional detail for the regions close to disaster centres, that is, 10 subregions of Queensland as well as northern New South Wales (see also Fig. 1). The compilation of this table and underlying data are outlined in Sect. 2.5.

We used the Australian Industrial Ecology Virtual Laboratory (IELab; Lenzen et al., 2014) to construct a customised subnational MRIO table (including the input coefficients matrix A and initial total output x0) for Australia with extensive detail on regions directly affected by the cyclone. The IELab is a cloud-computing environment that allows for the construction of customised IO databases. IO tables document the flow of money between various industries in an economy – national IO tables present national data on intra- and inter-industry transactions between industries in a national economy, whereas MRIO tables harbour detailed data on trade between two different regions (Tukker and Dietzenbacher, 2013); see Leontief (1953) for an account of MRIO theory. MRIO tables can either be global or subnational. Global tables feature more than one country and provide detailed data on international trade between countries, whereas subnational MRIO tables provide detailed trade data for regions within one country. These tables have been extensively used for undertaking environmental, social and economic footprint assessments (Alsamawi et al., 2014; Hertwich and Peters, 2009; Lenzen et al., 2012; Oita et al., 2016; Simas et al., 2014; Wiedmann et al., 2013). Coupling of economic MRIO data with so-called physical accounts, as conceived by Nobel Prize winner Wassily Leontief in the 1970s, allows for the enumeration of direct as well as indirect supply-chain impacts (Leontief, 1970, 1966).

The IELab is capable of generating MRIO databases, wherein industry sectors can be distinguished for a number of Australian regions. Users are able to choose from a set of 2214 statistical areas (level 2; ABS, 2016f) to delineate MRIO regions with their specific research question. The regional and sectoral flexibility of the IELab (see Lenzen et al., 2017a) was exploited by generating a regional partition of Australia that is more detailed around the regions where the cyclone caused most of its damage (Queensland and northern New South Wales) and less detailed elsewhere (Fig. 1). As a sectoral breakdown we used the 34-sector industry classification from the Queensland regional IO database (OGS, 2004); see Supplement Sect. S1).

A number of national, state and region-specific data sources were used for constructing the MRIO database used in this work. These are the income, expenditure and product accounts (ABS, 2016c); the IO tables (ABS, 2016b, 2017b) for the national level; the accounts (ABS, 2016a) and the Queensland IO tables (OGS, 2002) for the state level; the household expenditure survey (ABS, 2011); Queensland regional IO tables (OGS, 2004); the business register (ABS, 2016d); the census (ABS, 2012) and the agricultural commodities survey (ABS, 2016g) for the regional level. Detailed regional employment data were taken from the labour force survey (ABS, 2016e).

Not surprisingly, Tropical Cyclone Debbie wreaked the most intense havoc where it made landfall, in the regions of Mackay (QLD-M), Fitzroy (QLD-F) and northern Queensland (QLD-N), and where heavy rains caused widespread flooding, around Brisbane (QLD-B) and in northern New South Wales (NSW-Rm & T; see Fig. 1). There is not a single region in the remainder of Australia that is unaffected by the cyclone. In the multi-region IO disaster model in Eq. (2), these spillovers come about because businesses experiencing production losses are unable to supply their clients and also cancel orders for their own inputs, thus leaving businesses elsewhere with reduced activity. Our results for indirect damage are obtained from a model and as such might only approximate the damage that really occurred in the regions. However, an application of the same model to a case study in which indirect effects were known (see Fig. 5 in Lenzen et al., 2017b) shows that measured outcomes were reproduced with reasonable accuracy.

Our results show that tropical Cyclone Debbie affected about 8487 jobs (Fig. 2) and caused a loss in value added of about AUD 2.2 billion (Fig. 2). Employment losses are expressed in terms of full-time equivalent (FTE) employment temporarily affected. Full-time equivalent means that part-time jobs are expressed as fractional full-time jobs, so that they are added into a total. The time span of a job disruption may range between a number of weeks, for example for coal mines that could be re-opened soon after the cyclone; (Ker, 2017; Robins, 2017) and 1 year (for example tree crops that will not yield until 1 year later).

The production layer decomposition defined in Eq. (4) indicates how the direct and spillover impacts of the cyclone unfolded regionally. In Fig. 2, production layers 1 & 2 indicate that the total value-added losses in all regions physically affected were about AUD 1.5 million. In addition, the cyclone caused another AUD 660 million of value-added lost across the supply-chain network of the directly affected businesses. These additional losses are shown in the production layers that follow.

As shown in Fig. 2, about 4800 jobs were directly affected (production layers 1 & 2) and an additional 3700 indirectly (from production layer 2 onward). The combined sectoral and regional spillovers are therefore significant.

While the coastal areas of northern Queensland, Mackay, Fitzroy, Brisbane (in South Queensland) and northern New South Wales (Richmond–Tweed area) were affected immediately by storm and flood damage, repercussions were subsequently felt in the rest of the affected regions and later on within the rest of Australia. Losses in value added and employment cascaded throughout inter-regional supply chains, as subsequent transactions were cancelled. Shortfalls were noticeable even by distant suppliers, removed from directly affected producers by four or more transaction nodes (Fig. 2).

Our production layer assessment reveals the employment and value-added losses in different layers of production. Each of these layers are comprised of a range of industries. It is important to identify the industries affected in different layers of production. Our assessment shows that while only a selected number of industries and regions were directly affected by the storm and flooding (coal, tourism, sugar cane, road transport, vegetable growing; black stripes in Fig. 3), these direct losses resulted in many more indirect losses in the upstream supply chain. We further analysed the losses in different layers of production (Fig. 2) and identified the top 20 sectors that experienced the greatest total (direct and indirect) employment and value-added losses.

The top-ranking industries affecting employment directly and elsewhere are those connected to tourism (such as accommodation, restaurants, recreational services and retail trade (see Table 4 and Fig. 3). In the Richmond–Tweed area of New South Wales, 1132 jobs were affected directly in accommodation, cafes and restaurants, and about 466 indirectly in other industries and regions due to supply-chain effects (spillover). Similar effects are observed in Mackay and Brisbane in Queensland. The temporary coal mine shutdown in Mackay and Fitzroy affected as many jobs indirectly as directly. Damaged and closed roads affected road transport establishments and almost equally the industries that depended on them. Likewise, value-added losses are observed both directly and indirectly in the upstream supply chain (Table 5).

Analysis of the impacts of disasters, such as that undertaken in this paper, can have constructive uptake by informing disaster recovery plans as well as regional plans more generally. In August 2017, the government of the Australian state of Queensland released a management review of Cyclone Debbie and recommended improved Business Continuity Planning (BCP) as a way to build “… business and organisational resilience […] Enhanced BCP within state agencies, businesses and communities will help all to be more resilient to the impact of events. [… and..] should feature permanently in disaster management doctrine.” In addition, the report noted that “BCP needs to consider supply chains, and the numbers and skills of frontline staff required to ensure functioning of critical services”. (IGEM, 2017).

Consideration of the large indirect impacts identified in this article would help to improve future planning while recognising that only part of the impacts of the cyclone have been considered, and that wider analysis of positive and downstream impacts would be beneficial as suggested by Oosterhaven (2017). However, a step forward in consideration of negative upstream impacts could be achieved, for example, by considering the large number of employees indirectly affected by the disaster (as shown in Table 4), and the related services and products they provide. For instance, as shown in Table 4 for the indirect employment impacts for the “Accommodation, cafes and restaurants” sector, some 466 employees providing services were affected in the Richmond–Tweed area. However, this impact is currently not mentioned in disaster recovery planning documents.

In this work, we have focused on losses of employment and value added, because these are currently of immediate importance for governments, insurers and the media. Future work could investigate possibilities for restructuring the geography of production and supply-chain networks with the aim of finding more disaster-resilient configurations. In addition, there are variants of IO analytical methods that allow optimal recovery paths to be established (Koks et al., 2016), and these approaches could be integrated into the Australian Industrial Ecology Virtual Laboratory.

Future work could also consider the effects of cyclones beyond national borders. The disruptions of coal exports due to Tropical Cyclone Debbie, for example, caused bottlenecks in Indian and Chinese steel mills (The Barrel, 2017), and during the aftermath of the storm, steel producers were looking for alternative sources of coal such as Russia, Mongolia or Mozambique (Serapio, 2017). Such trade relationships can be taken into account using nested, multi-scale, global multi-region IO frameworks (Bachmann et al., 2015; Tukker and Dietzenbacher, 2013; Wang et al., 2015).

Our approach can be applied to other regions and ultimately extended to include impacts well beyond employment and value added, such as wider environmental or social consequences of disasters. The IELab already has many satellite accounts (and is being expanded) to assess broader environmental and social flow-on effects. The growing number of “virtual laboratories” for IO analysis (Geschke and Hadjikakou, 2017) for countries in disaster-prone zones (Indonesia, Taiwan, China) means that the work described in this paper can be readily applied to other geographical settings.