Though in North Carolina, U.S. government officials only used Google Earth to verify code violation complaints, in places like India, New York, Argentina and Greece, Google Earth was used in the active reconnaissance of committed crimes. Marine researchers have also used analyses of Google Earth to evaluate the veracity of fish-catch reports made to the UN. Spain’s Green Party has reported illegal bottom trawling of beaches for fish using Google Earth images, as well. Google Earth has also been used to detect illegal dumping. For example, in Florida, a sheriff’s deputy used Google Earth to apprehend an individual who dumped a large boat; in Mississippi, a landowner identified a stolen and illegally dumped truck on his property using Google Earth; while in Bangalore, Google Earth was used to identify unauthorized and illegal waste dumping site. Illegal logging is also actively identified using Google Earth by such groups as local police departments in the Philippines, the Finnish Association for Nature Conservation and their associated NGOs in Russia, the Amazon Conservation Team and associated indigenous groups. Amateur Google Earth users have reported potential body-dumping based on the imagery available, as well. Some of the issues associated with Google Earth arise from the fact that its images are made available by a privately-owned corporation and are technology driven. Thus, as Sheppard and Cizek note, the visualizations of the Earth made available by this interface are more geared towards “efficiency, convenience…entertainment value, popular demand, flood tray and profit” than they are towards “truth, deeper understanding, improved civil discourse, safer and more informed decisions, and other ethical considerations”. As these and other authors point out, realism in landscape visualization is not the same as accuracy or validity.
Virtual globes, like Google Earth, may suffer from low data resolution, interfering with image clarity and accuracy, missing data or inaccurately displayed data. Further, it is often impossible to know the exact date of the imagery available on Google Earth and whether all images in a scene are from the same date . Thus, a potential crime sighted on Google Earth may be months or even years old or may be exaggerated by differing image dates. Finally, these data may be manipulated by the producers of these virtual globes for various privacy reasons; some areas are intentionally blurred or objects are not displayed. More significant than spatial and temporal accuracy is the consumption and use of these images by untrained or informal interpreters. These informal interpreters may not understand the temporal or spatial inaccuracies inherent in these data. Goodchild points out that users of Google Earth may be misled to think it is more accurate than it is in reality. Despite the fact that Google Earth images’ absolute positional accuracy is sufficient for assessing remote sensing products of moderate resolution, errors and positional inaccuracies are still a problem. Trained remote sensing analysts understand these limitations and may be able to account for them, whereas casual users may not. For example, in the case mentioned above, where amateur Google Earth users reported a dumped body, their interpretation of the image was flawed. In this case, the “dumped body” turned out to be a swimming dog. The dog’s watery trail on the cedar wood dock and the dog lying on that dock appeared to be a bloodied body rather than a picture of a sunny day at a lake. Un-validated identifications of “crimes” using Google Earth images by amateur analysts unfamiliar with the inaccuracies of these images or the nuances of image interpretation may be problematic for several reasons. First, they may cause law enforcement officers to seek places or things that are not where they are purported to be, are no longer present or never existed in the first place. This may result in a waste of funds, resources and personnel hours. Second, the misidentification of a site as a place where a crime is or has occurred opens that place and its residents up to potentially needless intrusion, intimidation, surveillance or violence.
Despite the increasing ease with which satellite images and other spatially explicit data flow to us, ethical and scientific rigor should not be laid aside. Finally, as Purdy and Leung note, Earth Observation data like those used in products like Google Earth may have their evidential weight in a court of law seriously reduced if un-validated, because the medium by which it was taken, the data management systems used or even the date the image was taken may be unknown. Given the potential for amateur misinterpretation or overconfidence in Google Earth images, it is obvious that crimes detected in this manner must be validated to ensure appropriate, timely and safe responses by government of law enforcement officers. While there have been a few cases where crimes detected using Google Earth were validated, either by fly-overs or personal ground validation missions , in the majority of cases, there is no discussion of accuracy assessment or validation. This dangerous trend toward trained and untrained analysts taking Google Earth images as “truth” with no validation may have broad reaching potential impacts on law enforcement efforts and personal security. Despite the fact that cutting-edge technologies are being used to remotely detect crime, the accuracy assessments of those analyses lag well behind current remote sensing standards. Indeed, as we have shown above, some studies that attempt to remotely sense crime do not perform accuracy assessments at all, depend on the opinions of “experts” or “surrogate ground truth data”, all of which are deemed to be substandard by today’s remote sensing community. Many of the studies noted above performed no accuracy assessment at all; they did not even use Google Earth or Digital Globes to validate their data. Particularly, in situations that may have life-and-death implications or serious environmental effects , law enforcement officers must strive to be as accurate as possible in their targeting of crimes and criminals. Although drones or unmanned aerial vehicles/systems may present excellent options for accuracy assessment, offering up quiet, real-time, high resolution imagery of remote or distant areas without threat to human life, they are not ideal solutions in every situation. The equipment, licensing, training and maintenance required to acquire and safely maintain a UAV may be well beyond the means of many local police departments or underfunded government agencies. In the United States, the Federal Aviation Administration has seriously restricted the use of unmanned aircraft in national airspace . Further, there are serious questions about the constitutionality of using UAVs for law enforcement. Critics of UAV use by law enforcement argue that these vehicles impede an individual’s reasonable expectation of privacy as protected by the fourth amendment Despite these concerns, law enforcement is increasingly using UAVs to detect crimes and facilitate law enforcement . In the following section, we propose some alternate or additional means of validating remotely sensed crime. We hope that this initial thought experiment may help spark a conversation about the methods and ethics of remote sensing in law enforcement. We define “first order” accuracy assessments as those described in the accepted remote sensing protocol , which include ground-based validation or the use of imagery of higher resolutions than the imagery to be validated.
Since these first order assessments can be limited by security, funding and terrain issues and drone use presents funding and legal issues, we propose a “second order” level of accuracy assessment. This second order accuracy assessment analyzes the larger geographical and social context in which remotely sensed crimes are detected by remote sensors. Such assessments could utilize crowd sourcing, big data mining, landscape-scale ecological data and anonymous surveys to determine whether and how crimes are occurring and where remote sensing analysts think they are. Second order accuracy assessments may allow remote sensors and law enforcement officers to confirm that crimes are taking place where analysts say they are without facing rugged terrain, 4×8 grow tray insecure conditions or using costly overflight methods. Further, second order validation may enable analysts to gain better contextual understandings of those crimes, allowing for more ethical and proportionate responses by law enforcement. While these second order validation techniques may not be as reliable as first order techniques, they are better than no validation at all. Alternatively, these second order techniques could be incorporated into interdisciplinary crime detection techniques that may increase detection accuracy. Urban areas are well suited to second order accuracy assessments because of the amount of available social data produced and available at any given moment. For example, Oakland’s Domain Awareness Center plans to link public and private cameras and sensors within the city limits into a single hub for law enforcement use . While highly controversial, these centers present numerous opportunities to validate remotely sensed crimes with closed-circuit television , as well as readily available on-the-ground policing. Rural or more remote areas present more of a challenge, however. These places typically lack surveillance cameras and mounted sensors. It is also in these places that large-scale drug production, human and drug smuggling frequently occur. Thus, here, we use illicit cannabis production as a case study to think through three potential second order accuracy assessment techniques in non-urban zones. Though we acknowledge that each of these methods would require further development and thought and that methods may exist beyond those we propose here, it is our hope that this will be the first effort in a larger conversation as to second order validation techniques in the remote sensing of crime. Social media: Location-based social network analysis may be helpful in validating crimes remotely sensed in other ways through geolocated self-reporting or observations by others. LBSN has been shown to provide reliable spatio-temporal information about incidents occurring in a broad landscape. For example, researchers from the Institute of Environment and Sustainability in Italy used a Twitter application programming interface to retrieve tweets and related metadata for a specific topic, the 2009 Marseille forest fire. These tweets were then organized into meaningful summary statistics using data mining and web crawling scripts. These researchers found that the LBSN data collected were temporally synchronized with actual events and provided some geographically accurate reporting. They note that Twitter “could offer promising seeds for crawlers to collect event-related data where time and location matter”. Some products already exist to facilitate such second order validation of crimes. Products like SensePlace2, Twitter-based event detection and analysis system, DataSift, Gnip, SABESS, and others, enable those interested in crime or emergency detection to gather and aggregate publicly-available, geo-located, time-stamped information in real time about where and when an incident may have occurred, who was involved and how serious it was. Because these data are publicly available, issues that other forms of remote sensing bring up in terms of the invasion of privacy are avoided. Further, because reports are on the ground and produced by humans, they may offer information on the context of crimes and their perpetrators and an interpretation of the events that took place rather than leaving this work up to far-removed remote sensing analysts. While connectivity in rural areas is more limited than in urban spaces, the Pew Research Group has found that as of January 2014, 88% of rural Americans have a cellphone and 43% of rural Americans have smartphones, making such data gathering feasible in these areas. Landscape-scale ecological data: Remote sensing of large-scale cannabis production can be validated using landscape-scale ecological data, as well. Down-stream water quality is one way remote sensing of these grow sites can be validated, for example. Large-scale outdoor cannabis production can threaten water quality through water diversion, erosion and sediment deposition due to grading, terracing, road construction, deforestation and clearing; as well as the inputs of harmful chemicals or other pollutants, such as rodenticides, fungicides, herbicides, fertilizers, trash, human waste, gasoline, oil and insecticides, into local water sources. Using stream water quality analysis that picks up the chemical signatures of such pollutants may be one way to affirm that remote sensing analysts were correct in their characterization of given drug production sites. Though no studies using this approach to detect upstream drug growth exist to date, similar methods have been used in the early detection of sudden oak death. Stream monitoring efforts are able to detect Phytophthora ramorum even before signs of infection are even visible from over-flights.