Estimating the Benefits of Land Imagery in Environmental Applications: A Case Study in Nonpoint Source Pollution of Groundwater

Abstract

Moderate-resolution land imagery (MRLI) is crucial to a more complete assessment of the cumulative, landscape-level effect of agricultural land use and land cover on environmental quality. If this improved assessment yields a net social benefit, then that benefit reflects the value of information (VOI) from MRLI. Environmental quality and the capacity to provide ecosystem services evolve because of human actions, changing natural conditions, and their interaction with natural physical processes. The human actions, in turn, are constrained and redirected by many institutions and regulations such as agricultural, energy, and environmental policies. We present a general framework for bringing together sociologic, biologic, physical, hydrologic, and geologic processes at meaningful scales to interpret environmental implications of MRLI applications. We set out a specific application using MRLI observations to identify crop planting patterns and thus estimate surface management activities that influence groundwater resources over a regional landscape. We tailor the application to the characteristics of nonpoint source groundwater pollution hazards in Iowa to illustrate a general framework in a land use-hydrologic-economic system. In the example, MRLI VOI derives from reducing the risk of both losses to agricultural production and damage to human health and other consequences of contaminated groundwater.

Appendices

Appendix

Previous studies on the economics of information and uncertainty established that information becomes valuable when the information can be profitably employed in a decision making process by reducing the risk facing decision makers, especially when the consequences to the decision maker are uneven (Morgan and Henrion 1990). In this application where the consequences are uneven, the underlying behavior is the use of nitrate fertilizers on fields. Of critical interest to the regulator is when to intervene to control the rate of change in the groundwater quality to avoid the costly treatment of water wells. To intervene either too early or too late relative to a regulatory threshold is to affect individuals and firms, e.g., agricultural production, negatively or to contaminate either or both shallow and deep groundwater. On the other hand, to intervene too late allows a greater potential of contamination, i.e., the regulator’s risk. Bayesian Decision Analysis can be applied as a regulator’s decision problem involving uncertainty.

The loss function we face is asymmetric (Fig. 10.A.1) since groundwater resource damage is very costly to mitigate and thus the total economic value of diminished current and potential uses and existence of a pristine resource is large relative to the marginal loss of agricultural output. Edwards (1988) found the willingness to pay to prevent groundwater damage increased linearly with risk. The marginal risk to the groundwater increases as nitrate fertilizer application increases, we model the social loss of over application with an exponential functional form.

Fig. 10.A.1

Decision risk faced by regulators of societal cost of groundwater damage (red) vs. loss of agricultural production (blue). Asymmetric loss to regulator: too little NIT = crop income/profit loss; too much NIT = risk of aquifer loss

The modified linex loss function in Fig. 10.A.1 is adapted to the risk analysis in the hypothetical example (van Noortwijk and van Gelder):

$$ L\left( \Delta \right) = - {c_v}{\hbox{In}}\left( \alpha \right) \cdot \left( {{\lambda^*} - \lambda } \right) + \frac{d}{{1 - d}}{L_{i(j)}}\alpha \left[ {\exp \left\{ {{\hbox{In}}\left( \alpha \right) \cdot \frac{{{\lambda^*} - \lambda }}{\lambda }} \right\} - 1} \right] $$

(10.A.1)

where ${c_v}$ is the additional variable cost to reduce nitrogen use, ${\hbox{In}}\left( \alpha \right)$ is the q-quantile (level of risk tolerance) of the nitrate concentration distribution, and λ is the parameter of the nitrate loading distribution.

10. Commentary: Satellite Observations and Policy Improvements for Agriculture and the Environment

• Catherine Shelley Norman 

1. Department of Geography and Environmental Engineering and Department of Economics, The Johns Hopkins University, Baltimore, MD, USA

   Catherine Shelley Norman

Bernknopf, Forney, Raunikar and Mishra (BFRM 2010) consider the value of medium-resolution land imagery (MRLI) to a regulator focused on keeping water quality risks to an acceptable level. They present a general model of the biological, physical and economic processes at work, and then outline a specific example focusing on corn and soybean rotations and the fate and transport of associated nonpoint source pollution in a nitrogen-limited region of Iowa. MRLI allows the regulator to monitor changes in land use and optimize regulations to reflect nutrient burdens on the system, economic costs, and the value of mitigation plans.

In offering a comprehensive, integrated approach to valuing the information provided by programs like Landsat, the authors’ focus is on valuing the decisions that hinge on the information provided by a given set of observations. Their model is unusual in that it takes multiple disciplinary perspectives—ecological processes, agricultural science, economics, and hydrology, among others—seriously and simultaneously. Regulation of land uses and management practices is well suited to medium-resolution land imagery. Monitors can use simultaneously updated, geographically complete information rather than relying solely on sampling programs that require regulators to select representative sites over time and across space. Space-based data can be coordinated with direct sampling to improve inferences from images and to support enforcement efforts.

Although the full data required to reach conclusions are not available to BFRM, this work is a strong and ambitious initial step toward a framework supporting improved program evaluation. I consider some clarifications, limitations, and possible extensions of this work, with a focus on producing credible estimates to inform decisions about the use of MRLI.

10.1.1 10.C.1. Major Contributions

Satellite imagery supports multiple overlapping objectives in the regulatory and political arena. BFRM focus on one agricultural regulatory application, working to quantify the benefits from that use alone. In theory, one could aggregate up to a total value for a space-based observation program by considering all users, though as we can see from this relatively straightforward application, the informational and technical demands associated with such an effort would be very high. Additionally, for many earth observations satellites, information on all uses is likely to be confidential for reasons of national security, and some part of the value will be strategic or political—and thus more challenging to quantify than ecosystem services. It is perhaps better to think of VOI in this framework as valuing data access or specific data uses rather than a program as a whole. If this kind of use is the primary driver of value for the program in question, it should still be emphasized that estimates produced in this way, no matter how comprehensive, will be lower bounds of value rather than estimates of total value. Even rough information about the scale of a given use relative to uses for the satellite information as a whole would provide useful context for decisionmaking.

Some VOI quantification of the sort proposed is critical to maintaining and instrumenting costly space-based information resources. Cost-benefit analysis is mandatory in many public decisionmaking processes and a common, well-understood framework in the remainder. If those of us who use this information cannot usefully answer questions about what it enables us to do, it is difficult to justify maintaining either the systems themselves or access to the information. BFRM present a model to capture the benefit to a policymaker focused on risk management (for avoiding threshold environmental effects) and thus assess the value, to regulators and the public good, of operating with the additional clarity and scope offered by space observation systems.

I attempt to lay out some of the most interesting questions this work posed below, focusing on the economics of regulation. Moving this model from the general to the specific requires the authors to confront political and public choices that are difficult to observe or theorize about in a way that yields usable numbers. It will be important in applications that the assumptions used to develop the regulator’s objectives and constraints are made transparent, and perhaps subjected to sensitivity analyses.

10.1.2 10.C.2. Social Preferences and Risk

In Sect. 10.2.3, the authors note that we can infer the optimal risk of failure (in a whole matrix of failure points reflecting reduced water quality due to pollution) of a given regulatory regime for the regulator and link this to social risk preferences. The regulator takes those social preferences as given from a ‘higher level of authority.’ In the illustrating example, the analyst must infer the political level of tolerance for a multitude of risks of reaching various pollution thresholds in various environmental media and locations. I would be interested in much more detail on how this matrix is populated; there is a small but longstanding literature on identifying the preferences of government bodies (McFadden 1975, 1976; Ross 1984 are identified in the chapter), but recent efforts (e.g., Ahlroth et al.2010; DeCanio and Norman 2005) are in much narrower applications, and even in such settings, inferences based on political choice are met with considerable skepticism.

Social decisions should yield some sort of information about the willingness of society to make trade-offs and incur costs, but methods for determining political willingness to pay (or similar) are by no means well established. BFRM view the regulator’s problem as one of reducing risk to an acceptable level without providing information on social risk aversion; individual and firm risk aversion are difficult to infer outside highly controlled settings (Chetty 2006 describes some of the complexities involved), and moving from individual choices to social choices involves a fairly fundamental determination about the relationship between a government and the citizenry: ought a government represent the median voter? Or should the state, with a (potentially) much longer lifespan and broader area of influence, worry more about the future than individuals?

It might be more feasible to go backward from costs incurred for specific environmental efforts rather than to look at political decisions directly: those probabilities can be translated into expected payouts to replace ecosystem services if they are reduced or eliminated by insufficiently stringent policy choices—and it may be that that’s what BFRM will do to quantify the requirements of the higher governmental authority—but there is insufficient information for the reader on the methods envisioned for general applications of this framework at present.

10.1.3 10.C.3. Regulator’s Objectives

In the BFRM example, the regulator’s objective is to maximize the total value of agricultural output, given the constraints established by the willingness of society to accept the ecological risks outlined above. The farmer’s objective is to maximize profit. Thus, a farmer would prefer a lower yield method if costs were sufficiently reduced to preserve profits, but the regulator cannot support this. In the agricultural sector it is often the case that polluting inputs can be replaced with less polluting alternatives or increased handwork; as regulatory environments and water and land quality evolve over time, this may be a profit-maximizing, output-reducing solution, which appears to be excluded in this analysis.

It is not clear to me that EPA or USDA, the primary agencies in this environment, value total output over total profitability in this context. Readers would benefit from efforts to explain the motivation behind this modeling choice and its implications for VOI calculations. In particular, this choice seems closely tied to the choice to value MRLI information as some fraction of total farm revenues.

The language of the chapter seems to suggest that there is a positive value for the information from the satellite system only if the information allows the regulator to relax restrictions on farmers. I don’t believe this is required in the model; if it is, the authors should explain their motivation in more detail or perhaps consider relaxing this assumption. Given the level of geophysical detail in the example, it seems plausible that relaxing nutrient loading restrictions in some places and tightening them in others might provide aggregate benefits. Additionally, there is value in improved information that leads to greater restrictions: if we’ve gotten the regulations wrong because a sampling regime or alternate set of instruments led us to underestimate aquifer risks, for example, we gain from the changes even if they reduce output or profits.

As an extension, a more dynamic perspective might allow adaptive management to be built in to policy choices. VOI would thus derive from a more flexible system as well as from a system that is better at a specific moment in time. Given the slow pace of the regulatory and legislative process, this would not entail changes in middle of a crop cycle; rather, a set of observations could mean that restrictions were automatically loosened or tightened according to a preset schedule, obviating the need for ongoing legislative action.

Also on a longer scale, it is worth constraining “optimal” policy choices in the model to those that are politically feasible and legally defensible. Detailed spatial data may reveal significantly different optimal restrictions on farmers in the same jurisdiction growing the same crop using the same technology. It would be disingenuous to suggest that the VOI is contingent on such policies’ being enacted.

That said, given a broader array of policy choices, MRLI data combined with hydrologic and other spatial data could be used to improve environmental quality (or reduce risks to environmental services) at minimum costs by treating neighboring parcels differently. We might use these data to identify land values that may be declining to a point where the parcels are appropriate for conservation easements, for example, and payments could be based on forgone income opportunities.

It is also worth noting that in the longer term, VOI is very sensitive to the national or international policy regime in place. In addition to the gains from improved management of water resources, monitoring of agricultural land uses and changes will provide credible baselines for measuring carbon sequestration and perhaps granting offsets. This is true globally, of course, and in an environment where offsets are valuable, the entire community of nations included in the carbon dioxide regime would gain from the sum total of land imagery available. Establishing a baseline now, assuming monitoring of changes will be needed in the future, creates an option value for farmers and regulators anticipating policy changes of this nature.

10.1.4 10.C.4. Monitoring and Enforcement

Some of the value associated with MRLI in this context will come from the broad applicability to monitoring and enforcement of existing law. Monitoring and enforcement are currently exogenous in this model, but repeated imagery of all farmers in a region will affect compliance behavior. Although 17-day satellite sweeps and missed observations due to weather mean that enforcement opportunities from satellite information cannot be perfect, MRLI offers more observation of practices and outcomes (with less awareness of the specifics of observation than an inspector arriving at the farm gate) than conventional practice. Over time, habitual offenders should be identifiable, and for some rules even a single pass can provide evidence of noncompliance with watershed protection law. Even limited enforcement actions taken based on this information could have significant spillovers in compliance behavior for a region (Shimshack and Ward 2005).

Agricultural nonpoint sources are enormously significant nutrient sources in strained watersheds; resource limitations and the difficulty of tracing a given pollutant load to a specific plot of land mean that enforcement is difficult and inspections relatively rare. Improved compliance as a result of farmers’ expectations about the use of satellite imagery (perhaps to direct site visits) will offer reduced uncertainty not only about watershed conditions but also about land use and growers’ behavior as regulations change.

10.1.5 10.C.5. Baselines

When fuller data are available to complete the analysis envisioned in BFRM, clarity about the specific baseline considered (and alternative baselines) will be essential. If we are considering a proposed Landsat 10 mission, for example, it will have a value relative to no use of satellite imagery in regulatory decisionmaking, a value relative to the other satellites whose data will be used if the mission does not take place, and plausibly a value relative to whatever other alternatives may be available. Each of these alternatives also has costs, either to build and run or to purchase data from. We can then estimate how much of the value of the management plan derives from the improved information associated with the MRLI source. We will never make decisions about regulating agriculture for water quality protection in an information vacuum, so the marginal value is always contingent on our expectations about the information that will be available without the mission (or without a specific instrument’s inclusion in the mission).

If, as the authors suggest, the next-best information is from the Advanced Wide Field Sensor (AWiFS), which is under the charge of the Indian Space Research Organization, the VOI in the proposed Landsat mission is likely to be increased by the greater control U.S. agencies would have over Landsat. This is harder to quantify than the ecosystem services approach outlined in BFRM, but a simplified approach might rely on estimates of the probability of having to move from AWiFS to another next-best solution in a given year.

10.1.6 10.C.6. VOI and Time-Series Continuity

In the example given by BFRM, we are concerned with the value of information associated with Landsat imaging. One important area that is not yet extensively developed is the value of continuity in the time series. On a short-term basis, the value of one satellite mission relative to another may be roughly comparable, but most researchers would prefer a mission that offers greater continuity with past and future missions over one that does not. In particular, scientists and policymakers interested in climate change and in carbon sequestration rely on comparability of current and older imagery. Direct comparisons are especially useful for looking at trends over large scales when medium- and low-resolution imagery does not allow detailed observation of the phenomenon of interest: we may not be able to count the trees on a plot of land, but seeing how the same instrument records changing levels of things associated with trees (height of ground cover, albedo, etc.) allows us to make inferences about changing ecology, land use, and productivity over time.

Valuing this is more complex than valuing a given piece of data over its entire use cycle, which is itself not straightforward. The switch from SIC codes to NAICS codes for the collection and provision of industrial data by the U.S. government prompted some discussion of the value of continuous time-series information and of the processes used to connect different time series; the Census Bureau’s Economic Classification Policy Committee (1993) report presents the core concerns of data collectors and users. It details millions spent to construct “bridge” data to help users get some of the benefits of the long series even after classification schemes has been changed to reflect new patterns of production and provide improved comparability across nations.

What value there is in continuity will not automatically accrue to ongoing Landsat missions. Each mission has differing instrumentation and data characteristics. The additional value of information that is part of a 40-year stream of information of the same sort is thus contingent on efforts to merge the varying Landsat series (or to merge Landsat data usefully with alternative space-based observations). Information about those costs or about the number and type of users who would benefit from more determined efforts in this direction is not readily available. It does, however, seem likely that efforts to connect data from various Landsat missions will be easier, cheaper, and more likely to be pursued by a single entity responsible for all the data than if Landsat observations were replaced by AWiFS, as in the illustrative case.

10.1.7 10.C.7. Additional Considerations

This work covers a very large area, and space precludes consideration of everything that is important or interesting in this context. An extended analysis might take into account some or all of the considerations in this section.

In BFRM, financials are allowed to fluctuate weekly in the model. This makes sense for inputs to production but is less appropriate for revenues, since farmers can typically store corn and soybeans for a time if prices are not favorable. This storage decision could be added to the model, or annual output price figures could be used to approximate outcomes of storage decisions and costs.

Consideration of farm- and community-level spatial interactions would be interesting. Do we allow the production decisions of farmers to affect those of other producers spatially near them? If one farmer changes her crop rotation, do the neighbors respond? There may be thin local markets in processing or storing facilities or farm labor, or limited transportation or storage infrastructure that vitiate incentives toward more homogenized crop patterns in a region. This could operate through increased input costs or input cost volatility, affecting expected profit and income volatility.

Analysts would also need fuller details on the actual instrumentation considered for a given mission and how it could be used. For those unfamiliar with the details of the program, information about what spectra will be measured at what level of detail, and which will be used for the regulatory purposes envisioned, would clarify the application of the methodology to the illustration. Is the regulator hiring staff to visually inspect images of farms? Are heat and albedo sensors informing estimates of growth rates or nitrogen concentrations in specific locations? VOI will be very sensitive to the specific instrumentation and uses envisioned.

Lastly, broader consideration of ecosystem services, including carbon sequestration, biodiversity support, and erosion outcomes, would provide more opportunities for enhanced regulatory decisionmaking, thus increasing the VOI to environmental management based on MRLI.

10.1.8 10.C.8. Conclusion

The authors integrate a complex set of hydrogeologic, biophysical, social, and economic models to provide an estimate of the state of water quality and agricultural products in a given region with regulators with identical preferences but two possible data sources. The value to society of better information in this setting is estimated by examining the effects of changes in regulations associated with higher-quality MRLI. Although applying this approach will be data and time intensive, and explaining it to decisionmakers will need to done very carefully, this is an important step toward estimating the value of satellite information in a broad array of uses.