Winter storms in Texas, floods in Europe, heatwaves and wildfires in California, hurricanes in Louisiana… This year, a severe natural disaster makes the front page every few weeks. Many of us have begun to panic about whether our homelands will remain hospitable in fifty years and whether our children will grow up with enough food. On the bright side, we learn and improve every time we suffer from a natural disaster. As Dr. Matthew Kahn, an environmental economist at USC, argues, technology has advanced by leaps and bounds in recent decades, so a lot of unknowns about climate change have become “known unknowns.” This new information enables us to rely on an efficient free market to adapt to environmental crises. 

One of the crucial developments is big data, which significantly increases the accuracy of disaster forecasting. In July, severe floods inundated Western Europe and caused hundreds of deaths. However, this tragedy could have been alleviated if officials had acted on predictions from early warning systems. These systems use real-time spatial data captured by sensors to estimate when and where floods will occur. In Germany, officials were warned three days before the actual flood, but they failed to take action. In contrast, Dutch officials responded quickly to the warning and organized an evacuation in advance. As a result, the Netherlands suffered no casualties.

Because of its use in environmental crisis prediction, real-time environmental data is the key to building an efficient market for climate adaptation products (e.g. masks, air purifiers, air conditioners, etc.). Inventory allocation and dynamic demand fulfillment are two crucial problems for e-commerce companies like Amazon. These companies usually have a network of distribution centers. To ensure high-quality customer service, they need to build a neural network to calculate the optimal inventory in each center and determine which center should fulfill each new order. If firms can predict demand, they can avoid keeping excess supplies or running out too soon. Currently, they estimate demand using web traffic and recent purchase data and restock more frequently to regions where demand is rising. This works under normal circumstances, but not when there is a sudden natural disaster. Research has documented that the demand for masks surges by 20% on the first day of pollution when compared to the week before. If companies make inventory decisions solely based on pre-crisis demand, they are unlikely to prepare sufficient inventories.

With environmental spatial data, e-commerce companies can use machine learning algorithms to capture the subtle changes in environmental indicators and identify the regions at risk of an environmental disaster. The algorithms can then predict how the demand for each product would change (by learning from how demand responded to a similar crisis in the past) given the predicted type and intensity of environmental shocks. This enables e-commerce companies to improve the accuracy of their demand functions for each region or even each user. Granted, some environmental shocks have become more frequent only recently, and some extreme crises are occurring for the first time in history. It would be hard to predict the demand under these circumstances because there are few historical data points to train the algorithms. Yet these models would still be more reliable than those based only on past demand. As climate crises become more frequent, firms should integrate real-time sensor data into their demand models for more accurate predictions.   

Nevertheless, even if firms forecast a rise in demand and take action beforehand, a temporary shortage may still occur because it takes time to manufacture and restock goods. The e-commerce companies thus face spatial allocation trade-offs. One strategy is to redistribute adaptation products from lower-risk regions to regions severely affected by crises. This way, supply would match demand in the more severely affected regions, preventing a surge in price so that more people can afford protection. Meanwhile, companies could raise the price of adaptation products in other regions to signal a scarcity of inventories and incentivize consumers to hold off their purchases.

Although allocating more products to the affected regions could maximize social welfare, it may not maximize the firm’s profits. It depends on which portion of the demand curve they are operating on. When demand is inelastic, quantity is less responsive to price changes. A price decrease leads to a smaller increase in quantity, and firms lose revenue. If e-commerce companies sell on the inelastic portion of the demand curve of the affected regions, they will be reluctant to keep down the price and redistribute sufficient products to these regions. Vice versa, firms lose when reducing inventories and increasing prices in unaffected regions with elastic demand. This is likely the reality. People in affected regions need immediate protection regardless of the price, while people in unaffected regions are more price sensitive. Admittedly, these are the scenarios when demand in each region follows the classical supply-demand model (i.e. no price discrimination). In practice, companies could predict individual demand and charge each consumer a different price. Such specific price discrimination makes it possible to maximize profits and public welfare simultaneously, though not guaranteed. Other factors can influence firms’ decisions as well. For example, firms may be discouraged from delivering products to harshly-affected populations due to disruptions in transportation by a natural disaster. The discrepancy between private and public interests thus remains a relevant concern.

If e-commerce companies do not locate sufficient adaptation products to severely affected regions, low-income people will be harmed disproportionally. Wealthier consumers are less price-sensitive because adaptation products constitute a relatively small proportion of their income. They can afford them at higher prices. However, poorer people are more price-sensitive as they need to spend most of their income on necessities like food. They have to give up purchasing adaptation products due to the price spike. Additionally, they often live in neighborhoods with less resilient infrastructure and have less flexible jobs (i.e. unable to work remotely). This means they would benefit more from self-protection (compared to people in richer neighborhoods), yet their inflexible employment renders them physically and financially incapable of protecting themselves. 

To ensure private firms do not maximize profits at the expense of low-income people’s well-being, the government should partner with e-commerce companies to ensure the fair distribution of adaptation products. The government has a comparative advantage in installing loT sensors (useful in measuring humidity, PM2.5 levels, etc.) in urban neighborhoods and collecting real-time data. By sharing these detailed data with e-commerce firms, the government can in return ask them to sacrifice some profits and allocate more products to the affected regions when an environmental shock hits. Why would private firms accept such a deal? Because the real-time data would bring them significantly more profit over time than their one-time loss in an environmental crisis. By factoring in spatial environmental indicators, firms could take advantage of users’ location data to build highly accurate, real-time demand models. They could then identify each user’s consumer surplus and charge more for those able and willing to pay more. 

Despite the immense potential of big data to facilitate climate crisis adaptation, we should take a step back and consider a more fundamental question: Are people from all income and racial groups equally likely to be represented in data? Nothing I discuss would work if marginalized groups are omitted in the data collection process! The Federal Communications Commission (FCC) reported in 2020 that about 18 million Americans still lack access to stable and affordable internet (some other reports suggest that the actual number is even higher). These people are less likely to shop online and would be underrepresented in e-commerce companies’ datasets (built based on past purchases). Consequently, e-commerce companies might underestimate the demand from low-income communities during an environmental crisis. This would drive up the price, rendering products unaffordable. 

Going forward, the government could subsidize internet services for low-income neighborhoods. To meet the rising demand, internet companies should expand their networks to previously inaccessible regions. This would enable more low-income people to buy adaptation products online. E-commerce firms could then gain more data from them and better model their demand. These models could help firms take low-income individuals’ needs into account (and even give greater weight to their needs) when allocating inventories. 

Aside from the inequality in internet access, another source of underrepresentation arises from low-income neighborhoods’  dilapidated infrastructure. Poorer neighborhoods usually have trouble attracting investments because businesses in low-income areas face more restrictions on getting loans. There is also a lack of public investment because the government gains less tax revenue from these neighborhoods (due to the low average income and the lack of business activities), and governors might face negative political consequences if they subsidized public projects in poor communities using taxes from other neighborhoods. It is thus challenging to install sensors to capture real-time data since nobody would pay for them. The missing environmental data would cause an underestimate of demand and lead to shortages during a natural disaster. 

Today, big data infrastructures (i.e. infrastructures used to collect and process big data) are as important as conventional infrastructures.  Biden administration’s new $1 trillion infrastructure package includes a $65 billion broadband package, $42.45 billion of which would be directly distributed to local governors. These funds are technologically neutral, so governors can decide what technologies to invest in based on the needs of communities. Most governors would seek to expand broadband internet coverage, the foundation of online (virtual) data collection. This is a promising trend, but the importance of physical data (e.g. weather and pollution data) may seem less obvious to governors. Since online data would not give us a complete picture of consumers’ demands, it is crucial to install sensors and data hubs (for physical data collection and processing) simultaneously with internet infrastructure. This is an important issue for policymakers to consider as they allocate funding from the new bill.