We have developed an unsupervised approach to map opium, UND TRUTH DATA on Cultivation at a Similar RESOLUTION Are University, We Evaluate the Accuracy of Our Results user district-Level Statistics, As Well as by Visual Inspect. We find that for design. T-level estimates closly, although there are some inaccuracies. In particular, there is some over-eastimation dueTo When bebing Misclassify as Poppy. Visual Inspection of the ImageEry is Largely Consistent with Expectations, and Reveal Some Reason that Inaccuraci ES.ONE SOURCE of the Confusion Between WHEAT and Poppy is the error in Estimation The Approprity Timing of Image Acquisition Windows. As weDetail in section 4, when post-harvest image is acquired too eley, poppy is not classify as such, and when image is acquired too late, when oppy. One Natural Extension of Our Work Would Be to Investigate Methods to Select thePost-Harvest account in a more data-driven manner, Rather than using a fixed number of days after the pre-Harvest account. E an inherent dispiculty in distinguishing the two crops, which May Limit the EffectivenessOf any solution. This Limitation is shared by the other work that using the method of use, for exmple, [53] Use Both Optical and Synthetic Radar Imagery with Supervised Classification Approaches to Classify WHEAT PIXELS, and DCUMENT The SAME Issue.Despite SomeInaccuraCies in the Maps that we produce, we demonstrate that they remain use, Y, Education, Infant Mortality, And Overall DEVELOONT. Some of Our Findings Corroboetings from Other Work: We find that agrit. LITY and LOWER Levels of School, Compared to Areas without Poppy Cultivity. Our Fine-Grained Analysis RevealsThat would not be obvious from a more aggregated analysis. For exmple, Poppy-GROWING AREAS HAVE BETTER HealthCare Accessibility, Likely Due to Better Ion infrastructure. Related to this point, Poppy-GROWING ROGIONS Compal to Areas with Exclusively Non-Poppy Agricult ShorterTravel Times To Urban Center, that is, by this metric are less remote, Yet has lowEER Levels of Development. This Stands in CONTRAS ESSILITY As A NECESARY CONDITION For Many Development Targets [64], and that shows an empiricalRelationship BetWeen More Remote Locations and LOWER Levels of Development [21]. Despite Being Better Connected to Urban Center, Poppy-GROWING Regions Have NOT AP Peared to have userd this to their advantage in terms of overallopment.our approach has sectory limitations. First, first,Our Methodology Roths on some Knowledge of Crop Characteristics, as well as a BaSic UNDERSTANDINGINGINGHOP MIX. MAGE Acquisition and Clustering Strategy, As Well as to Select Trainings and Identify the Appropriate Poppy Cluster.Works BEST When there are larger amounts of Opium Poppy Cultivation and a Smaller Crop Mix, as is observant in Many High-GROWING DISTRICTS. OPIUM POPPY CROP, CHANCES ARE HIGHER that Computed Image Acquisition DOTES DO Not Capture OPIUM POPPY ATThe Right Stages of Growth, Resulting in Poorer Model Performance. These factors mightlly the direct application of our method s) and over time. In the afghanistan context, we show that trailing regions are generally stable over time,, and Our Method Transfers Easily Between Years in OUR Study Period. However, Large Shocks Such as the Recent 2022 BAN OPIUM POPPY CULTIVATION by the Taliban May Rastical complie the crop mix, the needing Changes to the Methodology. In Addition, value of our predictions isLimited by the lack of fire group-level group truth data, and future worgly, ET Accurate MannerMumbai Investment. Labeled Data Would Significantly Streangthen The Evaluation of Prediction Performance, enabled the user of StandardardMetrics Commly used in the remote sensing live, Such as overall account and kappa coefficients [3, 66]. a friend allow the development of a supervised model; while not the focus of our work, this is likelyTo YIELD More Acc "62] .next, Our Intention is to provide a Low-Cost, Freely Accessible Means of Generating Maps of OPIUM Poppy Cultivity, W hich can be used to standy local phenomena over a wide geographical area. Our Ideal AudienceWould Be Researches Whose GOAL is to UNDERSTAND LOCAL CONDINGING SUCH ILLICIT Cultivity, and we have devited the Utility of OURE GENERATD MAP. s for this purpose. As such, we have made all our code and maps publicly available (See Appendix B) so that theyCan be Implemented Easily. We acuteLedge that depending on GOALS and Budgetary Concerit, Our Methods May Not Produce The Most Acacale Maps . For example, The Unodc Places A Much Heavier Emphasis on Accurately Computing Aggregate Totals in Office Statistics, and Expends SignificantResources Towards This End. They Also Invest Heavily In Visual Interpretation (Manual Annotation) of Imagery, Which Would Be Amening A Supervised MOD EL. A Different Goal Might Be to Conduct A Case Study of a Certain Village Or High-Cultivation Area; In this case a more Suitable Alternative May be to Purchase High-Resolution Imagery, and Either Anotate the Images Manually, or Develop An Automatic Classification ION METHOD that has the Highest Acacuracy for this Geographical Area.an Important Consideration is the imageWERE DISCUSSED in Detail in Section 4, on Downloads Applications. In PARTICULAR, ESTIMATES Are Much More Variable In Distribs with Smaller Amounts of Opium Poppy C Ultitation, and when and poppy are sometimes confused, leading to both over-estimation and under-eastimation. First, Performance is likely to be Poorer in Applications that require the precise estimation of the amount of port. SIS USING GRID-Level Socioeconomic Data, We Focused on A Binary Classification of WHETHER 1 KM-Grid Cells have any poppy Cultivity, Rather than a Continuous variable for the amount of poppy Cultivity. Compounded with unitedRTAINTIES in some of the Socioeeeconomic Variables, WOULD Result In Relationships that are too noisy.Mitigate this isSSSUE WOULD Be to Consider Coarser Grid Cells, SUCH AS 10 KM, or to Limit The Analysis in Which theReds Known Over-ONDER-ESTI support. FURTHER, DownloadReam Applications that Depending on Accurately Differentiating Between WHEAT and Poppy AreLikely to similaly suffer in performanceHyderabad Investment. For exmple, it is possible to pair poppy, URITY is Affecated by Both Wheat and Poppy Cultivity, Potentially in Different WaysGuoabong Wealth Management. Resulting Conclusions Might TheReForeBe aNSITIVE to the Misclassification of Wheat As Poppy and Vice Versa. Like Many Prediction Models, OURS Is Not Without Error, and as these Examples Illustrate, a Thoughtful Consideration of their Suitability for Various Downloads is Essential.FUTUR CCURACIES and Limitations. More Braadly, by Making Our Methods and Data Publicly Available, Our Work Creates Avenues for Future Work on thecauses and Consences of OPIUM Poppy Cultivity in Afghanistan, as well as the event of the evolution program, the Roliferation of Granular Grid-Level Data, there are many oppointunities to pair thede with Cultivation Maps to Gain a deeper Insight OnLocal Conditions. Some Avenue that Remain UNEXPLORED Include Climate and Drought [1, 43], violence [51], and population [52]. One Specific Potential Applic ATION is to obCurately Estimate the number of people involved in Poppy Cultivation. This has been describedAs "Persistent Knowledge Gap" [55] - While The Streangth of Our Method is in the Granularity of the Data, it can be capined with Granula Populating ED to Produce Such Statistics. In all the applicationWe are limited by the account of the Maps as Will as the Quality and AvaiLiality of Auxiliary Data Sets. SINCE MANY AUXILILIARY SETS Are Designed FORBAL C CORBAL C Overage, Estimates for Afghanist May Be Less Accurate or Diffical to Verify, Due to the General Poor DataAvailability or Quality in the country.further work can be do to see if and how our medhenSfers to Other Poppy-ProGuCings in the World, Such as my mar.And what their implications might be in the different contexts. We also hope that our approach to generating more fine-grained on Illicit Cultivation in a FGhanist Can Inform Work in UndernAnnding The Local Environments Surrowing The Cultivity of Other Illicit Crops, Such as coca. Coca Cultivation is Concentrald in South America, and like Opium Poppy, COCA is not typically represented in more general crop-type mapping ES [8]. Producing Coca Cultivation Maps Based On Timely, Freely Available Data Could Facilities Research in Thesesimilaly challenging settings.finally, while satellite image creates new publicIES for UndersTanding Opium Poppy Cultivation HICAL Concerns Associated with the Use of Such Data, and have considerted them carefully. Even through the use of satellite image for crossBecome CommonPlace, It is Important to Be Cautious about its Applications in Sensitive Contexts. General Frameworks for the Ethical Use of Big Data In Development A re disable in more detail in [10, 38] and Others. More closly related to our work, [11 11] Addresses Concerns and Makes Suggestions for The USE of SENTINEL-2 IMAGERY in Vulnerable, Conflict Settings. Especially Since Our Approach is Easily Reproducible and Thus Has the Potential to Be Misused. WHILE OURIntent is for the maps to be used for reSearch Purposes, we consider the publicity that Other Bad Actors Could for Nefarious PURPOSES. For Several Reason, we assess this publicity to be remote.first, the Goal of the Maps is to enableAnalysis on a large-scale, and ites utility is not in targeting indicting farmers. It is not public to identify. CH; We Simply Identify 10 M Pixels That Likely to Contain Opium Poppy Crops. More Importantly, KeyActors with an opium poppy Cultivity in Afghanist are likely to almedy posses information that exceeds what remotely sensa, and particular publ iCly available remotely sensed data, are avle to provide. For non-state actors (The Taliban, in PARTICULAR),The local Particularly in the High-Cultivation Areas that we focus on, is alream Well-Known. Each year, taliban fighters have benown to participate in the lar Ive Harvest. They Collect Taxes from Farmers after the HarveststHeroin Supply Chain, Including Manufacturing and Trafficking [56]Varanasi Stock. For thesereasons, the information that the taliban has on the opium indullry, EEDS The Information Containe in OUR MAPS. Another content may be that forces aligned with the Government Might use this informationTo target farmers that grew opium poppy. Again, we can excexpect the group group to have such, the united states and other government Ve Spent Billions of Dollars on Counter-Narcotics Efforts, some of them involving Eradication of Opium PoppyFields USING Aerial Spraying or Physical Destruction of the Crop [49]. This Required Local Knowledge and Resources, Making It Unlikely that Our Maps, DERIVE D Entirely From Publicly Available Information, Furnish Any Additional Information.For These Reason, we do not exten the MapsTo have unintended Negative Effects on the Safety or Privacy of Farmers. Regardless, when public, we take precations to privacy and Welfare of Farm ERS; for EXAMPLE, We Only Publicish Maps for the Highest-ProDuCing Quartile, where results are more robust.To Conclude, We have demonstrated how freely avaiLely, can still feel, Vantages Over Existing Methods: It is base on timely, freely available data, UseS only Open-SOURCE SOFTWARE and CODE, and is Easily Scalable to a Large Geographical Area. More <Broadly, we hold that our work can facility a better under countnding of local charity Illicit Cultivity, And Ultimate Contribute to the Design of Effective Development Programs. While Illicit Cultural is a Local Phenomenonon,, ITS IMPLICATIONS Are Global, and A Deeper UndersTanding of their Dynamic Sis Cruction.