Indigenous food production in a carbon economy

Fig. 2A and Table 1 present the estimated edible weight of harvested food represented by the 2018 Inuvialuit Harvest Study (IHS) data. The reported 2018 harvest totals 122,117 ± 886 (mean, SD) kilograms of food. The most important food species by edible weight are caribou (21,430 ± 297 kg), broad whitefish (12,737 ± 316 kg), muskox (11,460 ± 233 kg), and inconnu (10,627 ± 470 kg). The total reported harvest corresponds to an average of 44.1 kg per resident Inuvialuit Beneficiary. Harvest estimates range from a high of 95.9 kg per Beneficiary in Ulukhaktok, to a low of 16.2 kg per Beneficiary in Inuvik. However, our estimates represent only a portion of the “true” total harvest, as survey participation rates suggest that reported harvests represent as little as 50% of the total harvest in some communities (SI Appendix).

Fig. 2B and Table 2 summarize the estimated replacement value of food reported in the 2018 IHS. Our estimate for the substitution value of the reported harvest is approximately 3.18 ± 0.02 million Canadian dollars. This corresponds to roughly $1,150 per Inuvialuit Beneficiary, ranging from $315 per Beneficiary in Inuvik to $2,802 per Beneficiary in Ulukhaktok. Again, these estimates are based only on the amount reported in the 2018 IHS and are thus a lower bound for the total harvest in the region. If we additionally account for food subsidies in the more remote communities, the amount of food reported in the IHS would result in an additional cost of approximately $394,546 (SI Appendix).

Fig. 2C and Table 3 report the carbon emissions that would be incurred through the production and importation of market substitutes for the harvests reported in the 2018 IHS. We report four estimates: “low” and “high” emissions projections based on the range of published emissions estimates for each transportation mode, for each of two different transport scenarios. The “barge” scenario represents a case in which foodstuffs are brought by road to Hay River and then by barge to Inuvialuit communities, while the “food mail” scenario represents a case in which foodstuffs are imported by road to Inuvik and Tuktoyaktuk and then flown into the smaller communities. Our analysis suggests that a quantity of food equivalent to the harvests reported in the 2018 IHS would produce over 1,000 tons of greenhouse gases per year, regardless of the mode of transport. The mean carbon emissions estimates for each transport scenario suggest that, per kilogram of country food consumed, 8.5 to 9.4 kg in gross CO₂-equivalent emissions are avoided, or between 375 to 417 kg CO₂-equivalent emissions per year per Inuvialuit Beneficiary (we consider carbon inputs to the traditional economy in the subsequent section). Once again, this is an estimate based on the reported, and not the true total, harvest.

Using parameters estimated in a regression analysis of the Tooniktoyok dataset, we estimate that 126,953 ± 15,303 L of gasoline would have been used in producing the harvest reported in the 2018 IHS. We further estimate an associated 781 ± 173 unsuccessful trips associated with the reported harvest, which would have consumed an additional 40,408 ± 15,668 L of gasoline. Thus, the total fuel consumption estimates for the harvest reported in the 2018 IHS is 167,362 ± 21,957 L. This corresponds to roughly 1.4 L of gasoline per edible kilogram harvested. Using $1.76 a liter as an approximation for the cost of fuel in 2018 (38), this translates to a total of $294,557 ± 38,644 in gasoline costs for the reported harvest, or roughly $2.41 per kilogram harvested. The Tooniktoyok data suggest that gasoline represents roughly 50% of the cost of supplies for harvesting trips (SI Appendix). Therefore, excluding major equipment costs, per trip expenses for harvesting are well below the replacement cost of market foods (which average $26.04/kg). Carbon emissions associated with the gasoline used in harvest production reported in the IHS range from 315 to 497 tons (low-emissions scenario 90% HPDI: 315 to 479 tons; high-emissions scenario 90% HPDI: 327 to 497 tons). Mean posterior estimates from these scenarios correspond to 3.3 kg (low) or 3.4 kg (high) of carbon emissions per kilogram of food produced and 145 kg (low) or 150 kg (high) of CO₂-equivalent emissions per Inuvialuit Beneficiary.

This project was initiated by the Innovation, Inuvialuit Science, and Climate Change Division of the Inuvialuit Regional Corporation (IRC), whose mission is to coordinate research and develop policies for the ISR. The IRC is an Inuvialuit-led organization mandated to improve the economic, social, and cultural well-being of Inuvialuit through the implementation of the Inuvialuit Final Agreement, which was one of the first modern land claims agreement in Canada. E.R. was initially contracted by the IRC to produce a report estimating the size of the Inuvialuit Traditional Economy (40), as part of a series of reports focused on assessing the potential impacts of carbon tax policy on Inuvialuit Beneficiaries. After completion of the report, E.R. proposed publishing the results in a scientific journal, to disseminate the findings more widely. This process involved submitting a formal proposal to the IRC, who evaluated the potential costs and benefits of such a publication for Inuvialuit. The proposal and the subsequent manuscript were reviewed and approved by the IRC.

The data used in this study originate from two main sources, access to which were granted by the IRC and other relevant parties (Data, Materials, and Software Availability). The first data source, the IHS, is a long-term data collection initiative led by the IRC in collaboration with the Hunters and Trappers Committees (HTCs) in the six ISR communities. The second data source is the Tooniktoyok Project, a community-based research project conducted in Ulukhaktok in 2019 (41, 42).

We use 2,388 harvest reports from the 2018 IHS to estimate the quantity of food harvested in the ISR. The IHS is administered locally in each community in the Inuvialuit region by HTCs. Only members of the HTCs were included in the study, and participation in the study was voluntary. In principle, local interviewers contacted all HTC members each month to record all harvests and determine whether any harvesters were inactive or “out-of-sample” (e.g., away from town) for a given month. For a variety of reasons, the participation of HTC members in the 2018 survey was relatively low (close to 50% on average) and variable from month to month. For example, in some months, in some communities, no harvests were reported at all due to a lack of available interviewers. Due to these data-quality issues, extrapolating from the reported harvest to an estimate of the total harvest is not straightforward. In view of this, we emphasize that we focus on calculating the harvest reported in the IHS, which we take as a minimum estimate for the total harvest that year. In supplementary materials, we examine the data on participation rates in the 2018 IHS.

The harvest data in the IHS are based on participant recall. Because Inuit harvesters rarely count their catch, their estimates are approximate and strongly heaped (e.g., catches of several animals tended to be reported in round multiples of 5, 10, or 50). Below we detail the statistical procedure we use to account for this feature of the data. Also, due to similarities in variable names in the data collection platform, harvester IDs were sometimes recorded under the harvest number variable. To correct for these data-entry errors, we filtered the data to identify likely cases (n = 64) and flagged the harvest amount as missing. We then impute the missing harvest amounts using Bayesian methods during the modeling procedure.

Our modeling approach was also influenced by the fact that the IHS interviews focus only on harvests, and, as such, there is a lack of information about unsuccessful harvesting trips. Although unsuccessful trips are not required for estimating the total harvest, they are needed to estimate the cash inputs to and carbon outputs from harvesting. To deal with this issue, we use the Tooniktoyok dataset (described below) to generate estimates of the number of unsuccessful harvest trips and gas consumption associated with them, and then we adjust the cash and carbon estimates for the observed harvests using these rates.

Once harvest quantities are estimated, they need to be converted to edible weights. To obtain edible weight data, we reviewed three main sources: Usher’s calculation of edible weights for species harvested in the ISR (43); Ashley’s review of all available edible weight estimates for game in the NWT and Nunavut (44); and Brown et al.’s derivations of new estimates for several fish species caught in Alaska (45). Our list of edible weight values, including justifications for which values we adopted when multiple values were available, as well as descriptions of calculations for values not available in the published literature, are provided in SI Appendix.

Price estimates of the cost per kilogram of replacement foods (SI Appendix, Table S1) are based on average prices reported in each ISR community from an in-store pricing study (14), which established the lowest regular prices for preferred purchase volumes of a wide variety of foods. We group these products into three broad categories: mammals, poultry, and fish. For mammals, we use a mix of 50% pork and 50% beef (see SI Appendix for details and justification). For poultry, we use the price of an even mix of chicken legs and chicken breasts. For fish, we use the price of frozen fillets of sole, haddock, pollock, and halibut. The pricing study was conducted between 2014 and 2016, so we adjust for the change in the consumer price index of store food between 2016 and 2018 (0.989 (46); data for the NWT only available for Yellowknife).

In assessing the emissions resulting from substituting traditional foods with market substitutes, it is our view that a comprehensive life-cycle approach that includes indirect emissions incurred in the production of food is most appropriate (SI Appendix). As above, we create three general categories of food products: beef and pork, poultry, and fish. For beef, pork, and poultry, we take the most recent “to-the-farmgate” emissions estimates from western Canada (47–49) (SI Appendix, Table S2). For the purpose of estimating carbon emissions of market substitutes for local fish, we use a mix of 70% common market whitefish (cod, pollock, haddock) and 30% salmon/trout, based on the rough proportions of the main fish species harvested in the ISR in 2018. We draw median emissions estimates for each of these groups of fish from a recent review (50). We convert the published estimates (per kilogram live or carcass weight) to bone-free meat weights and add average carbon costs per kilogram for processing, packing, and transport to retail distribution centers.

Finally, stores in the Inuvialuit region are exceptionally far away from major distribution centers. Food may travel to the Inuvialuit region through several different routes and modes of transport, including road, barge, and air freight. Carbon emissions of these different modes of transport depend on a wide range of conditions, including the size of the vehicles, engine type, river/sea conditions (in the case of barges), and so on. Given this variance in conditions, we adopt a simple approach in which we use high- and low-end estimates of carbon emissions (per kilogram per kilometer) for each mode of transport from the 2014 IPCC report on transport (51) (SI Appendix). At this stage, we consider direct carbon emissions only (i.e., fuel burned), not indirect emissions (e.g., produced in vehicle manufacturing). This is a conservative approach as it assumes that no additional vehicles would be built to transport additional imported food.

Finally, we need a means to infer, from the available harvest data, cash and carbon inputs to harvest production. To do this, we use data on 132 harvesting trips recorded in the Tooniktoyok study, conducted with 10 harvesters in Ulukhaktok in 2019 (41, 42). This dataset contains estimates of fuel usage and harvested quantities. Usher’s edible weight estimates were used in the Tooniktoyok data, and so the edible weight data should be highly comparable to the IHS data. We note that in the Tooniktoyok interviews, interviewees generally reported their individual harvesting expenses, while harvests were often reported for groups of harvesters (e.g., if one caribou was caught). Consequently, to avoid overestimating the returns from harvesting when using these data, we take the estimates for individual harvests where provided, and where they were not, if it was a snowmobile or ATV trip, we divide the harvest by the number of participants on the trip. This is a conservative approach that is equivalent to assuming that each participant had their own machine. In contrast, for trips using boats, we include the total harvest, which assumes that only one boat (with multiple people in it) was involved in the reported catch.

To calculate the edible harvest represented in the 2018 IHS, and to estimate its market value and carbon cost, we adopt a Bayesian approach that allows us to account for measurement error in the harvest reports and to account for unobserved processes in data generation (i.e., the absence of reports on failed harvesting trips in the IHS). The model has four main components: 1) we account for measurement error in the harvest data; 2) we run a regression model using the Tooniktoyok data that estimates i) fuel use per kilogram of edible weight, and ii) the proportion of trips that did not yield any harvest; 3) we use the posterior distribution of harvest sizes from step (1) to calculate the edible weight represented in the IHS and its associated market replacement costs and carbon emissions; and 4) we use the posterior distributions from steps (1) and (2) to calculate fuel usage represented in the IHS data and its associated gasoline consumption. We detail each of these steps below.

The first component of the model deals with measurement error in the IHS data. This involves accounting for the problem of “heaping” and, to a much lesser extent, imputation of missing harvest values. To account for the error in harvest estimates induced by heaping, we model each reported harvest as being taken from a lognormal distribution with a median equal to a latent variable representing the “true harvest” and a scaling factor of 0.15. The scaling factor was chosen on the basis of the relationship between the median and SD in lognormal distributions with different scaling factors. A scaling factor of 0.15 implies that a reported harvest of 10 items would have a SD of about 1.5 items, and a reported harvest of 500 items would have a SD of about 50 items (SI Appendix). This distribution accounts for the scaling of error with the size of the estimate (i.e., larger estimates are more heaped than smaller ones). We then model the true harvest as being taken from a probability distribution of harvest sizes which is unique for every “species type” within the dataset. We use the term species type because, for this procedure, we grouped multiple species into groups that are ecologically similar and harvested using similar techniques, and therefore are expected to have similar distributions of harvest sizes. For instance, we grouped several species of geese into a single group “geese,” and treated all subspecies of caribou as “caribou.” We model the log of the true harvest for each report in the IHS as drawn from a normal distribution with mean equal to the mean logged observed harvest size for that species type and with SD equal to the SD of the logged observed harvests for that species type. For missing data, we draw from this distribution to generate a predicted harvest value. Modeling true harvests from this species type distribution is a conservative approach that treats large harvest size estimates for a given species with greater skepticism. In SI Appendix, we consider an alternative approach that does not fit the true harvest estimates to a species-typical distribution, instead using them only to impute missing data.

The second step of the modeling procedure estimates gas consumption as a function of edible weight harvested and the frequency of unsuccessful trips, using the Tooniktoyok dataset. First, we model the probability that a trip was successful or not using a Bernoulli distribution with success probability θ, which is estimated from the data. For successful trips, we then model log fuel consumption as a linear function (intercept and slope) of log edible weight harvested. For unsuccessful trips, which returned no edible weight, we estimate the mean and variance in log fuel consumed using a normal distribution.

The third component of the model involves simulating true harvest amounts from the posterior distributions and multiplying each estimated harvest by the edible weight for the type of animal concerned (not pooled by species type). It is then straightforward to multiply these estimates by the per-kilogram cash and carbon costs of market replacements that we generated (above). We assign each reported harvest to the poultry, fish, or beef/pork category, and calculate sums of the estimates for each category. These operations are undertaken such that uncertainties at each step are fully propagated through to the final estimates. We do not account for any error or variation in edible weight; but our deheaping procedure results in fractional harvest counts, which accounts to some extent for variability in animal size and/or the portion used. Given the absence of available data, we do not account for measurement error in the market cost or carbon emissions estimates, although we do calculate estimates for different plausible scenarios for carbon emissions (SI Appendix).

Finally, the last component of the model uses the edible weight estimates from the previous step along with the parameters from the Tooniktoyok regression (step 2) to estimate the fuel used for each harvest reported in the IHS. A limitation here is that we could not reconstruct individual harvest trips from the IHS data, so we treat every harvest as a distinct trip. As such, our fuel use estimates may be somewhat high. We use θ, the probability of a failed trip, and the estimated mean and SD of log fuel use on failed trips from the Tooniktoyok model to infer the number of unobserved unsuccessful trips that are likely to be associated with the successful harvests observed in the IHS data and the fuel usage for these trips. To estimate the carbon emissions incurred from the shipping and burning of gasoline used in harvesting, we calculated high and low emissions scenarios for shipping gasoline to the ISR via rail to Hay River followed by barge to each community, converted kilograms to liters using a density of 0.749 kg/L, and then added emissions per liter shipping to the carbon emissions produced from burning a liter of gasoline (2.319 kg/L (52)).

The computer code for the analysis is included in SI Appendix. We used the R statistical computing environment (4.3.2) (53), Stan (2.32.2) (54), RStan (v2.32.5) (55), and the rethinking package (2.40) (56) for analysis. Our priors for all estimated parameters were mildly regulating, and are defined in the accompanying R code. The model was fit using Hamiltonian MCMC in the Stan engine, on three MCMC chains with 4,000 iterations each. Convergence was diagnosed by the Gelman–Rubin R-hat statistic, and manual inspection of traceplots to confirm that the chains were well mixed. Fig. 2 was produced using rphylopic (57).

The research reported here used only previously collected, deidentified data. A proposal for the article and the resulting manuscript were reviewed and approved by the IRC. The IRC, in collaboration with the Inuvialuit HTCs, designed and implemented the IHS. The Tooniktoyok Project was collaboratively designed and implemented by Angus Naylor and the project participants, and overseen by a volunteer Inuit Oversight Committee. The Tooniktoyok study protocols were approved by the research ethics boards of the University of Guelph (REB 17-12-012) and the University of Leeds (AREA 18-117), and the project was licensed by the Aurora Research Institute (No. 16533).

E.R., C.T.R., B.B., and J.P. designed research; E.R., C.T.R., and B.B. performed research; E.R., C.T.R., and B.B. analyzed data; and E.R., C.T.R., B.B., and J.P. wrote the paper.

This article was developed from an impact report completed by E.R. under contract for the Inuvialuit Regional Corporation.