The Data of Coca
The registry of its cultivation and destruction
Data creates narratives, policies and laws. Data clarifies, but also obstructs and obscures, and in turn, becomes a source of violence. Data shapes the earth. In this case, we examined the role of data in the violence of the forced eradication of coca in the Colombian Amazon. Images of coca are captured by satellites, photographs, and maps. Together they create an informational complex that constructs the plant as a clandestine being that must be eliminated. From the moment that coca was outlawed its presence and destruction began to be quantified and documented. But most of these records are still classified or withheld from the public.
This project interrogates data from the United Nations Office on Drugs and Crime, the US Department of State, the US Drug Enforcement Agency, and The Colombian Ministry of Defense. We work with coca-growing farmers, use artificial neural networks to process satellite images, and compare our analysis with official results. We demonstrate the structural lack of transparency and critical flaws in the methods of capturing and quantifying coca since their inception and the effect of these failures in the escalation of violence in the war on drugs. A violent past which is crucial in order to understand the geopolitics of the present.
Map of Colombia showing the locations of this investigation: the departments of Guaviare, Putumayo, Meta and Caqueta with the military base of Larandia, used as an operations base for aerial fumigation in the country’s southern departments.
(Plano Negativo, 2025)
Guaviare: Coca Eradication, Deforestation & Cattle
El Píxel y la Parcela, Plano Negativo, 2025
The Pixel and the Plot, 2025, 17 min. (Plano Negativo, 2025)
Aerial fumigation with herbicides in Latin America dates back to the 1970s and was imposed by the United States on different countries, in line with Cold War strategy. In Colombia, while the first account of the use of herbicides in eradication is from 1978, the decades of documentation of this practice as it became the prime method of combating marijuana, coca, and poppy, has been concealed from the public.
In this project we analyzed the small fraction of these datasets that have been declassified, trying to verify them with agricultural communities that have been subjected to eradication since the early 1980s, and working with remote sensing scientists to understand the continuum of eradication, deforestation and cattle-ranching in the department of Guaviare, one of the most intensely fumigated regions of the country.
A zoomed in view in the project’s study region in the department of Guaviare in the Colombian Amazon. This map is a result of our remote sensing analysis which combines fumigation centroids (white), coca (green), cattle (red) and deforestation (grey). The spectrum of these colors begins in 2000 and ends in 2024.
(Plano Negativo, 2025)
Archive I: Records of the US Department of State
The documents presented here were released following a 2018 Freedom of Information Act (FOIA) request to the US Department of State, the entity that holds the records of aerial fumigation conducted in Colombia. The request was made for all records of aircraft parameters recorded during missions to eradicate so-called illicit crops in Colombia for all aerial spraying operations conducted in the country from 1994 to 2015, when aerial spraying was discontinued in the country. This request encompassed maps of the flight paths, aerial photographs and video footage taken from the planes during the missions, and data log recording the date and GPS coordinates of each flight.
The records released in response to this request document what appear to be a single fumigation event, a fumigation experiment over a military base in the department of Caqueta called Larandia, which was used as a center of eradication operations for the southern departments. No explanatory information was given along with these documents, they are presented here with their original file names.
24 selected files from the DynCorp Papers are shown here for examination. Full Repository of DynCorp Papers available here.
The documents also evidence the evolution of tactics of eradication and the visual technologies that were used as a mode of verification. Verifying the act of eradication through quantifying the survival and destruction of coca became a point of controversy between the US and the Colombian governments throughout the 1990s. In 1995, the writers detailed a disagreement between the two governments over the number of hectares of coca quantified by each. Debate surfaces as to the correct methodology to use to measure and quantify the coca presence and eradication of the plant. The US calculated coca using aerial imagery and high resolution satellites, while the Colombian government subtracted the total amount of hectares sprayed with herbicides to calculate the yearly cultivation. This led to more moderate estimates of the plant killed by the US, which angered the Colombian government. This provoked the signing of a “Verification Protocol" between the two countries in 1996, which laid out the terms of how the plant, both its cultivation and destruction, should be quantified. By 1997, the Colombian government was still not satisfied by the numbers calculated by the US, and called for an independent assessment to be conducted. These documents demonstrate how the methods of detecting coca were a source of conflict even before the United Nations Office on Drugs and Crime began to make their studies (see Part II for this discussion). What is also clear is the level of importance of detailed data that each fumigation flight collected on the eradication missions, beyond the specific flight path, speed, and volume of the herbicide applied, these missions were equipped with cameras, sometimes also videocameras, that were meant to record the flights, and document the effects of the herbicides on the ground.
Given the lack of data, it is difficult to determine the specific effects of a fumigation event, or a series of fumigation events in a single area. One of the questions we sought to answer was what could be understood about the effects of this practice, comparing this very partial dataset with satellite imagery. Since it is not known when the first case of eradication occurred in a given centroid, a part of our task was to determine the first date in which coca can be detected in the GPS point. In the majority of the 3,012 points, the first occurrence of coca could be detected between 2000 and 2015, the time period of Plan Colombia. In 974 centroids we could not detect any coca.
Remote Sensing Research
Film Still: a coca plant in Putumayo as seen in infrared.
(Plano Negativo, 2025)
A time series of detection layers and fumigation centroids over Landsat 5, Landsat 7, Landsat 8 and PlanetScope satellite images using infrared, red and green bands.
(Plano Negativo, 2025)
We then used open source satellite data to classify the land into forest, coca or cattle between 1985 and 2024. We analysed these transformations in relation to the centroids provided by the Ministry of Defense. Coca cultivation is often blamed as a main driver of deforestation. In this analysis of tracking land transformation around each point, we demonstrate how this is a misconception, one that leads to the demonization of the plant. We show how cattle ranching is the main force driving deforestation in the region, and how this deforestation is a part of the continuum of destruction produced by forced eradication.
Clip from Pixel and the Plot
(Plano Negativo, 2025)
Map 5: The combined dynamics of coca, cattle, and deforestation from 2000-2024 overlaid with the centroids of fumigation.
(Plano Negativo, 2025)
F-2019-00271 - Final Response+ar.pdf
(US Department of State, 2024)
This series of maps separates each dynamic (coca, cattle, and deforestation) to allow the viewer to see the transformations over the territory of Guaviare. Looking at the maps of coca and cattle for example, the difference of scale of transformation between the two is apparent. Cattle expansion is so massive, that it disappears the coca plots when they are placed together in Map 5 below.
Beyond the fumigation centroids, we can see the progression from coca to cattle that prevails on much of the deforestation frontier throughout Guaviare. Map 7 depicts coca plots that have transitioned to cattle ranching between 2000-2024.
(Plano Negativo, 2025)
Map 8 shows the land classification of Coca, Cattle, and Forest in 2024 on every pixel under the fumigation centroids as detected by our dataset. 8% of the fumigation centroids now have coca, 58% of them are now cattle, and 34% are now forest.
(Plano Negativo, 2025)
Archive II: The DynCorp Papers
Archive III: DIRAN fumigation centroids
In 2020, as a part of the investigation into the effects of the war on the Nukak people in the department of Guavaire, the Colombian Truth Commission made a petition to the Colombian Ministry of Defense for all records of aerial fumigation in the region. What the Ministry gave to the commission was a PDF with what they describe as “centroids of fumigation lines”, which are GPS coordinates without a specific date tied to them. We know through reading the records of the US Department of State and the DynCorp documents that each fumigation flight contains a high level of detail including specific dates, times of day, altitude, speed, volume of herbicide used. By comparing these two archives we know that data that was given to the Colombian Truth Commission by the Ministry of Defense was extremely partial. With this small portion of data, we travelled to Guaviare to verify what happened in these points. According to residents, the points we were able to visit did at one point contain coca. Many points had been coca fields that had been fumigated multiple times over the decades. After verifying, we then sought to understand what occurred in these points on the regional scale through a remote sensing study.
Response from DIRAN to the Colombian Truth Commission, and the full PDF containing 3012 “centroids”
(DIRAN, 2021)
Map 1: The centroids of fumigated zones released to the Colombian Truth Commission by the Ministry of Defense organized according to the first date when coca was detected in the point.
(Plano Negativo, 2025)
The pixels inside each centroid tell a different story of the land’s transformation. The video clip above depicts the story of one of these points. In the centroid 1520, coca was first detected in 2003. From this year onward, surrounding plots of coca appear and disappear. In 2018, deforestation is registered in the centroid and there is a large conversion from forest to cattle near the point, which grows until it covers the whole area in 2024.
Because we don’t have the date or frequency of the fumigation events in each point, the direct consequences after a fumigation event cannot be determined with this data alone on a local level. To achieve a greater understanding of how these dynamics have interacted with each other over the years on a regional scale we embarked on an archetype study. This is another way of saying patterns. We divided the department into 900 by 900 meter cells and looked for patterns in the transformation of land.
We used the following metrics to analyze each cell:
(1) Forest baseline
(2) Forest to cattle
(3) Forest to coca
(4) Cattle to coca
(5) Coca to cattle
(6) Coca to forest
(7) Cattle to forest
The metrics of each cell are used as inputs for a clustering algorithm which groups the cells according to similar spatial temporal patterns. The zones without a colour could not be classified into a group.
Film still- a 900x900 cell which could not be categorized into an archetype, surrounded by cells classified into one of four archetypes.
(Plano Negativo, 2025)
This process classified Guaviare into four archetypes that enable us to see how human activity influences and modifies the landscape over time. We found that 83% of the centroids fell in three out of the four archetypes identified. These three clusters have in common a high amount of initial forest cover and a similar percentage of forest to coca transition. The difference between them is the amount of deforestation caused because of cattle.
The research shows that within the high and medium cattle deforestation (archetypes 1 and 3) there was also a high turnover from coca to cattle. Looking at the pixels which cross each fumigation centroid in 2024, we can see that 247 are still coca, 1746 are now cattle, and 1019 are now forest (Map 8). The high amount of fumigation centroids corresponding to groups with high deforestation indicates a possible connection between fumigation and this form of destruction.
Map 6: Fumigation centroids over 900x900m cells according to their archetype.
(Plano Negativo, 2025)
In order to visualize the land transformations over time, we stacked the detection layers around the fumigation centroids on a vertical axis, producing a 3D landscape in which time flows downwards. Remote sensing analysis is normally represented in two-dimensions, with layers of time interacting in the 2D plane. Temporality when flattened in a still 2D image becomes hard to understand. The challenge of analyzing the forest using aerial or satellite images is that you are always looking back in time, at a landscape in continuous transformation. To capture and understand the constantly shifting forest fabric we need many images over time, preferably at small intervals; what is referred to as high temporal resolution. By transforming the pixels of each image into 3D objects, land transformation over time becomes more legible.
A 3D representation of pixels from a Landsat satellite images with a resolution of 30 meters per pixel, transforming from coca to cattle around fumigation centroid 1118.
(Plano Negativo, 2025)
3D representation of landscape transformation over time. Pixels as detected in Landsat satellite images are then transformed into 3D structures showing coca (green), cattle (peach) and deforestation (grey).
(Plano Negativo, 2025)
Remote Sensing in 3D
In this 3D pixel landscape, we can see how deforestation and cattle consumes the landscape, dominating the image.
The DynCorp papers is a collection of declassified documents compiled by the National Security Archives for the Colombian Truth Commission. This archive helps us articulate how aerial fumigation cannot be seen in isolation to Colombia’s history of counterinsurgency, by placing the practice at the intersection between the War on Drugs and the internal armed conflict.
Dyncorp was a military contractor that the US government used to support the aerial fumigation campaign in Colombia. Beginning in 1991, DynCorp provided aviation services, including pilots, aircraft, training and maintenance. In the late-1990s, DynCorp began to support the operations of the US-supported Colombian Army Counter-Narcotics Brigade. By the early 2000s, DynCorp was providing direct support to the most sensitive Colombian military operations, including many aimed at killing or capturing top insurgents and narcotraffickers.
By comparing these two archives we know that data that was given to the Colombian Truth Commission by the Ministry of Defense was extremely partial.
Coca cultivation is often blamed as a main driver of deforestation. In this analysis of tracking land transformation around each point, we demonstrate how this is a misconception, one that leads to the demonization of the plant. We show how cattle ranching is the main force driving deforestation in the region, and how this deforestation is a part of the continuum of destruction produced by forced eradication.
The high amount of fumigation centroids corresponding to groups with high deforestation indicates a possible connection between fumigation and this form of destruction.
UNODC Data: A War of Pixels
Authors: Hannah Meszaros Martin (Plano Negativo)
Paulo Murillo Sandoval (Universidad de Tolima)
Jamon Van Den Hoek (Oregon State University)
Figure 1: An enlargement of satellite image subsets from the decision-tree in the methodology section of the UNODC’s Annual Coca Census of 2010 (full image on the right). While the decision-tree itself is presented in too low a resolution to be able to be read, we deduce from reading the methods that this image is meant to demonstrate a contrast between coca and non-coca in the initial phase of field identification.
(UNODC, 2011)
The success or failure of the US-funded War on Drugs in Colombia has revolved around a singular metric: the hectare. Each year, SIMCI (Sistema Integrado de Monitoreo de Cultivos Ilícitos), an United Nations Office on Drugs and Crime (UNODC) program, produces a Coca Census, which reports the total number of hectares under cultivation. Their analysis has played a major role in how the success or failure of the War on Drugs is perceived by policy makers in the Colombian and US governments, media outlets, and the general public, as each year’s numbers were quoted widely in the press, referenced in governmental debates, and cited as evidence for new policies. However, this analysis has also been the subject of debates and recent controversies, centering around its core methodology and how this impacts the countries policies and strategy regarding counternarcotics
The SIMCI Method
The SIMCI program was initiated in 1999 and has used remote sensing technologies to monitor coca cultivation from this date to the present. Their method combines satellite imagery and data from aerial surveys to map and track the cultivation of coca, marijuana and poppy.
We collected and analyzed all publicly available documentation - in the form of reports and censuses - produced by SIMCI on coca cultivation from 1999 to the present to establish an evolution of the program’s methodological approach over the years (the yearly coca census reports are available from 2002, which details the crop monitoring from 2001).
The first thing that we noticed was that their methodological descriptions in each report lack detail, are convoluted in their writing and often contain ambiguous images - such as the ones we include in figures 1, 3, 4, and 5 - which do not allow the replication of their approach for identifying and quantifying coca. It is also important to state that a proper assessment of their methods would need full access to the datasets used, which they do not provide to the public.
Figure 2: Comparison of 30 to 2.5 meters pixel spatial resolution over a 5m resolution PlanetScope satellite image mosaic in Guaviare department from 2025.
(Plano Negativo, 2025)
Figure 3: Images and text taken from UNODC 2012 Colombia Crop Monitoring report comparing the spatial resolution of two satellites over the same sector.
(UNODC, 2013)
Figure 4: Image from the methodology section of the UNODC Annual Coca Cultivation Survey. The image appears for the first time in the 2007 census. The yellow lines are the automatically registered flight paths of the fumigation planes, which SIMCI superimposes over the satellite image used to identify the coca plots. The image is meant to demonstrate how adjustments were made according to the data of aerial fumigation provided by DIRAN.
(UNODC, 2008)
Figure 5: Images from the UNODC 2004 Annual Coca Cultivation Survey showing the collection of data during verification flights.
(UNODC, 2005)
The core workflow of coca detection and assessment of SIMCI involves five steps that have changed little through time. They consist of: 1) image collection, 2) land cover classification, 2) visual interpretation, 3) verification, and finally a phase they call corrections or adjustments (5).
First, images from publicly available (non-commercial) satellite images are collected across the country to provide as complete coverage as possible for the census year. Their method relies on moderate resolution earth observation data from USGS/NASA Landsat satellites (30m resolution; USA) with support from satellites and sensors such as SPOT (France), ALOS (Japan), Terra ASTER (Japan/USA), and IRS (India) to achieve greater spatial and temporal coverage. These satellites provided spatial resolutions ranging from 10 to 30 meters – the width and height of an image pixel as if it were projected on the ground – and collected multispectral data where each image was composed of multiple bands corresponding to near-infrared, red, blue, or green wavelengths.
Steps 2 and 3 are the image analysis, which use two main methods:
A. Supervised classification: this is an algorithmic classification, automatically identifying areas of coca growth based on training data
B. Visual interpretation within these already classified areas.
In the fourth phase (verification), manually detected coca plots were validated with field data collection and verification flights; in later years, follow-on visual interpretation of high resolution satellite imagery that better captured small-scale coca plots was used. This phase produces a percentage estimate of the accuracy of remote coca plot detection. Verification throughout Plan Colombia regularly relied on aerial images taken from fumigation planes tasked with eradication. Until 2015, when fumigation was suspended, verification data was provided by the Colombian National Police (DIRAN) to SIMCI each year. Data gathered by DIRAN during manual eradication (ground missions) was also incorporated into the verification process. Since eradication data was relied on for verification of the SIMCI/UNODC method, any attempt to replicate their approach would need access to this data, and as we described in Part I, this information regarding the specific flight paths and dates is not publicly available.
The fifth component is what SIMCI calls corrections, which are a series of methodological adjustments to account for challenges associated with the presence of clouds in satellite images that obscure visibility and the temporal gaps between images and eradication events (when the collected images did not capture the effects of eradication). From 2010 SIMCI also applied corrections to account for small crops (ranging from below .1 to .25 ha) that could not be detected using their standard methodology. Temporal gaps always present an issue throughout the census reports. Since satellite imagery was often acquired many months before the census end date of 31 December, coca estimates needed to be forecast in line with expected coca growth through the end of the year. To do so, analysts estimated the trend in coca expansion across imagery collected over several months and then extrapolated this monthly growth rate across the months between the imagery date and the census end date.
The cumulative effect of these corrections was a consistent increase of reported coca area ranging between 8% and 17% between 2002 and 2014, while the 2015 census showed the smallest correction at 2% increase. The role of these corrections in inflating coca area is important to highlight, as they were applied after verification was conducted yet represent a sizable increase in the estimated area under cultivation each year.
Given the substantial limitations to their methodology (image availability and resolution, clouds, small plots, etc.), SIMCI’s claims of high coca detection accuracy are problematic. Their reported accuracy hovers around 89% throughout the reports, which is an exceptionally high accuracy especially considering that far less complicated analyses of annual deforestation in Colombia, for example, typically yield similar or even lower accuracies in detecting forest loss.
What is clear through reading the descriptions of SIMCI’s methodology over the years, is that the analysts repeatedly go to great lengths to avoid reporting that they lacked the data necessary to conduct their analysis. This is significant as it leads to a complex, often convoluted, and sometimes problematic system of corrections to account for the gaps in their datasets. Gaps in data may be spatial (lacking images), temporal (lacking images at specific times in the growth and harvest cycle of the coca plant), cloud-related (lacking clear views of coca) or eradication-based (lacking images that capture and verify the effects of eradication or regrowth). Missing data, including the factors listed above, are common in satellite image analysis, and might be communicated as contributing to the overall uncertainty of an analysis rather than being mitigated at all costs, sometimes on shaky conceptual grounds. Yet the SIMCI methodology requires that data gaps are effectively filled in and covered over, without consideration for how those corrections influence the margin of error of subsequent estimates.
Deep Learning of Satellite Images – Methods and Findings
Using four departments in the Andes-Amazon region: Putumayo, Caquetá, Guaviare, and Meta, we compared the coca area from 1999 to 2019 using the reported figures from SIMCI and a recently published deep learning model by co-author Dr. Paulo Murillo Sandoval that employs Landsat satellite imagery in the Amazon watershed (Murillo et al., 2023). The model was primarily created to disentangle coca farming from pasturelands and better document patterns of deforestation and their underlying causes. The deep learning model can identify coca farming locations due to differences in spatial, spectral, and temporal characteristics and is very effective in classifying common and recurring landscape patterns. The model was trained using official coca records from SIMCI in national parks and verified using high-resolution imagery hosted by Google Earth. Our model achieved a spatio-temporal accuracy greater than 68% for coca plots.
Figure 6: Collection of graphs showing the difference between SIMCI annual coca crop figures, our deep learning (DL) results, and the annual aerial fumigation per department as reported by DIRAN.
(The Authors with Plano Negativo, 2025)
Our comparison of these results to the SIMCI figures and outputs suggests that the area of coca reported by SIMCI in these four departments is 93-901% greater than the area detected by the deep learning model in the years of peak discrepancy (1999-2005 depending on the department). This overestimate of cultivated coca is so great that it cannot be explained through the available data or interpreted as a consequence of our margin of error using our methodology. We hypothesize that SIMCI data from the early 2000s has high errors, with abrupt increases and decreases in coca between 2000 and 2002, associated with methodological approaches that overestimated coca detection and may amplify incorrect estimates from previous years.
SIMCI data from 2005 is somewhat more consistent with our estimates, with an exception of the coca area in Putumayo, which in the SIMCI figures shows a sharp rise between 2014-2017, and in Caquetá where they estimate a less exaggerated increase between 2013-2018. Both of these increases are not consistent with our results.
Figure 7: Country-wide hectares of coca as reported by SIMCI and US Department of State, compared to a corrected version of SIMCI using or deep learning method in 4 departments.
(Plano Negativo, 2025)
If we zoom out to the national scale, we can see the impacts of the overestimates of these four departments in the country-wide totals. By subtracting the overestimates of the four departments of this study from the country-wide coca hectares reported annually by SIMCI, we can see how these discrepancies in just four of the 32 departments produce a stark difference in the narrative of dramatic increases and decreases of coca at the state level. For the reported figure of 160,119 hectares in 1999 by SIMCI for example, we found a 95,420 hectare discrepancy, which renders more than half of the reported hectares as nonexistent (assuming that the numbers in the rest of reported departments were not overestimated as well).
The political implications of inaccurate coca data
Consider the political implications that are produced by a vast overestimate of coca in the beginnings of Plan Colombia - when aerial fumigation and militarization in coca-growing regions was intensified - at the turn of the century. At this politically significant crux, the SIMCI data showed that the massive escalation of the fumigation campaign was working, as the total area fell each year.
Trends in coca cultivation have shown a steady increase that can be traced back decades in Colombia. However these trends do not support the theory of sharp increases, or of booms in coca production, which in turn have justified violent counter narcotics policies, such as fumigation and militarized incursions, and most recently, the decertification of Colombia by the United States. Our study is further evidence for the need for an independent inquiry of the SIMCI/UNODC’s methods to provide transparency and accountability for their annual report-making. Despite the SIMCI/UNODC’s longstanding methodology, this method tends to fill data gaps aggressively, leading to erratic fluctuations and likely overestimations, particularly when satellite coverage or spatial resolution is limited. While technical refinements have gradually improved the detection of coca plots, the precise methodological details remain poorly documented. After the most recent controversy surrounding their methods, this also casts further doubt on the accuracy of the data even after these supposed technical improvements.
As the data stands to this day, it is questionable how we can even assess anti-narcotic policies without an accurate picture of how these policies have affected the cultivation of coca. Because of the potential that these reported figures have to drastically alter the political landscape, leading to the intensification of the Drug War, the methods used to assess so-called illicit crops should at the very least be subject to normal standards of scientific scrutiny and review. With the United States threatening the region with escalating violence in the name of the War on Drugs, the geopolitical stakes are high, the Colombian public deserves full transparency in how this war has been justified and perpetuated over the last decades.
Resources
UNODC Colombia coca survey collection (2002-2023)
Original publication from which the deep learning dataset was used, validated and peer-reviewed. Murillo-Sandoval, P. J., Kilbride, J., Tellman, E., Wrathall, D., Van Den Hoek, J. & Kennedy, R. E. (2023). The post-conflict expansion of coca farming and illicit cattle ranching in Colombia. Scientific Reports, 13(1), 1965.
Figure 8: Cutouts from El Tiempo newspaper article from March 18th 2003, citing a significant reduction of coca crops in departments being fumigated (Putumayo, Meta and Caquetá) between 2000-2002. Our method, while showing drastically lower numbers of hectares under cultivation, also shows a slight increase - rather than a reduction - in coca crops for all of three departments during these years.
(El Tiempo, 2003)
Consider the political implications that are produced by a vast overestimate of coca in the beginnings of Plan Colombia - when aerial fumigation and militarization in coca-growing regions was intensified - at the turn of the century.
While the United Nations and its programs are independent - with a mandate not to interfere with the politics of a particular country - the effects of these yearly coca numbers these yearly numbers of cultivated coca have impacts that go far beyond the borders of Colombia, influencing international anti-narcotics policies, the evolution of the War on Drugs and its intersections with the decades-long armed conflict. Recently SIMCI/UNODC’s methodology of coca detection has come under scrutiny for another seemingly “explosive” rise in cultivation, as they reported a 53 percent increase of the potential for the production of cocaine in 2023. A scandal emerged when it was discovered that the Pacific region's yield had not been measured since 2019, meaning that this increase did not occur over the course of a year but in fact over the previous four years. The UNODC has yet to give a clear reason for why they presented the data in this way and why they had not measured the yield of the region for four consecutive years. This controversy has provoked a public debate about the methods they employed to detect coca in recent years. Our findings suggest that the failures of their methodology in fact go back to the program’s beginnings, and have had dire consequences for the communities and landscapes across Putumayo, Caquetá, Guaviare, and Meta that have endured the violence of the policies put in place through Plan Colombia (testimonies gathered from Putumayo can be seen in Part III)..
We compared the coca cultivated area reported by SIMCI between 1999 and 2019 with our novel dataset, which uses an architecture based on artificial neural networks (deep learning) to classify satellite images and detect pixels associated with coca plots in publicly available Landsat images in Putumayo, Caquetá, Guaviare, and Meta. These are the departments in the Colombian Amazon where coca cultivation has been more concentrated both historically and in the present-day. Our results show the structural problems with the data production methodologies that have driven the War on Drugs since its very beginning, uncovering a stark overestimate of coca production in the early 2000s. Significantly, this overestimate is the most egregious during the beginning of Plan Colombia (2000-2005) with large proportional difference between approaches (with the highest differences in Putumayo and Caquetá ranging between 901% and 454%).
Plan Colombia was a multi-million dollar military and aid package from the US, which placed aerial fumigation with herbicides as a central tactic, as well as a mode of counterinsurgency. Overestimates of coca during Plan Colombia’s initiation gave an impression of an “explosion” of cultivation, fueling a logic of emergency that contributed to the justification for the intensive eradication in the region, increased militarization and the criminalization of farming communities that continues to the present day. Significantly, forced eradication of coca was directed towards the four departments of our study, making them some of the most fumigated regions in the country. The overestimate of coca cultivation during these years presents a distorted image of the eradication campaign, painting it as a success, as numbers of coca under cultivation seem to plunge during these years of intensive fumigation from an alleged peak.
What is clear through reading the descriptions of SIMCI’s methodology over the years, is that the analysts repeatedly go to great lengths to avoid reporting that they lacked the data necessary to conduct their analysis.
As the data stands to this day, it is questionable how we can even assess anti-narcotic policies without an accurate picture of how these policies have affected the cultivation of coca.
Comparing the fumigation data provided by the national police to the SIMCI data presents another dimension to the consequences of these overestimates. This is particularly evident in Putumayo, where SIMCI reported 66,022 hectares under cultivation in the year 2000 (in the census published in 2001), ten times greater than our estimate of 6,595 hectares. DIRAN reported that they fumigated 71,891 hectares in the department in 2002, a figure that is greater than the already overestimated figure from 2000. Since SIMCI publishes their coca census reports in the middle of the following year, this means that the police would have been using the year 2000 data at the beginning of the year in 2002. SIMCI would later report in the 2003 census that in 2002, the department only had 13,726 hectares of coca. In 2002, our approach detected only 8,538 hectares, a slight rise from 2000, but far from the 71,891 hectares that the police allegedly fumigated. With such extreme discrepancies, this leads us to question what exactly the police were fumigating that year.
Data used for comparison of department level coca measurements from SIMCI and our deep learning method between 1985 and 2019, as well as aerial fumigated hectares as reported by DIRAN.
Data used for comparison between yearly coca crops reported by US Department of State, SIMCI, and SIMCI minus the detected discrepancies in our four studied departments.
Falta de Luz
Falta de Luz, Hannah Meszaros Martin, 2020
Falta de Luz, 23 min. (Hannah Meszaros Martin, 2020)
"After about three days it starts to turn yellow, and it turns yellow until it becomes dry, at least for the plantain, the chiro, this is really delicate, the corn you don’t even need to fumigate it, when the spray comes down, just with the wind and the smell, it will die"
In August of 1983 The US Drug Enforcement Agency (DEA) produced a short video of its domestic marijuana eradication program using the herbicide paraquat. The video contained an experiment, filmed in the mountains of Georgia, in the Chattahoochee National Park, of marijuana plants sprayed with the herbicide dying slowly as they were guarded by the federal agents.
The purpose of the recorded experiment was to demonstrate, particularly to the government of Colombia, that eradication of so-called illicit crops using herbicides was the way to win the war over these outlawed plants and their narcotic derivatives.
Fragments of the raw footage used to compile the short video of the experiment sit in the United States National Archives collection of the records from the DEA (1915-1993). In this archive one can find the records of destruction of different criminalized plants, from marijuana, to poppy and coca. These videos show the many ways in which these plants are killed: they are burned en masse, cut down with machetes, and sometimes, sprayed with chemicals from planes.
While we can see fragments of the raw footage used to make the short video used to try and convince Colombia to begin the use of herbicides as a method of eradication, the video in its entirety is missing from the archive. Falta de Luz (2020) begins from this entry point, from the missing film of a forgotten experiment in the mountains of Georgia in 1983.
In June 1984, one year after the production of the DEA video, the newly appointed Colombian Minister of Justice, Enrique Parejo Gonzalez, gave the first authorization to fumigate in marijuana crops in the Sierra Nevada de Santa Marta, located on the northern Caribbean coast of the country, using glyphosate. This fumigation in the Sierra Nevada was also an act of experimentation, the first of many whose documentation is also missing in the archive of ecocidal violence of herbicides. Eventually, in the early 1990s, the practice of aerial fumigation was formalised and Colombia continued to use glyphosate in enforced eradication even after its use from planes was suspended in 2015 after a World Health Organization report deemed the substance as a probable carcinogen. After 2015, the practice of eradication continued on the ground level, with police spraying the substance using backpack sprayers, often producing violent confrontations between police and farmers.
Film Still: A man in a uniform, wearing aviator sunglasses speaks to a group we cannot see. The camera zooms out as he speaks, revealing another man in a uniform standing to his left. The man speaking holds a binder by his side. A helicopter is seen, parked in the background.
(Hannah Meszaros Martin, 2020)
In the film, the archival footage from the DEA and collected testimonies from Colombia are divided into four parts. The first section is constructed with the fragments of the eradication event in Georgia. The fragments are bracketed with a black screen and the video noise that is slowly devouring the images. The grain of the film, and its material degradation, serve as another register of violence, a form of violence that conceals, obscures, and erases itself.
The second part of the film combines news footage in the aftermath of the controversy of the fumigation in Georgia. Outraged farmers tell news reporters that they had no warning, that they had only fifteen minutes warning to get away, that some trees and crops seemed damaged. That they felt betrayed.
Film still: Helicopter sprays over the forests of the Chattahoochee National Park
(Hannah Meszaros Martin, 2020)
One farmer, standing in the middle of his corn field, tells the camera-person: “If you could have seen this field a week ago today, you would never have dreamed that this is the same field today. It's just unbelievable, what it's done.”
Their voices are cut and disrupted by testimonies from Colombia gathered during fieldwork in the region of Putumayo, located along the border with Ecuador, and one of the most fumigated regions of the country. Alvaro, a farmer who has experienced many years of eradication explains to me in detail how his plants die:
“Como unos tres días de ahí comienza ahí a amarillear y ya, se sigue amarillando hasta que se seca, por lo menos el plátano, el chiro, eso es algo muy delicado. El maíz al menos no necesita casi ni fumigarlo. Cuando ya viene la fumiga por abajo, con la brisa o el olor, que será que lleguen él ya se va a muriendo.” (After about three days it starts to turn yellow, and it turns yellow until it becomes dry, at least for the plantain, the chiro, this is really delicate, the corn you don’t even need to fumigate it, when the spray comes down, just with the wind and the smell, it will die)
These dried, dead landscapes are a part of the structural memory of the war in the Colombian Amazon. Living in this environment meant you were subjected to almost three decades of continuous aerial fumigation.
Glyphosate acts through extreme desiccation, the plant dries to death. These dried, dead landscapes are a part of the structural memory of the war in the Colombian Amazon. Living in this environment meant you were subjected to almost three decades of continuous aerial fumigation. A Taita who was recorded in a yage ceremony in 2012, described the simultaneous experience of the plant and the human in relation to the herbicide, illustrating the continuum of violence through the human and non-human bodies:
“Uno siente también cómo se seca la plantica, uno también empieza a secarse.” (As you feel the plant drying you are also starting to dry up as well.)
In the recording, we ask: ¿Cómo es el aspecto de las plantas cuando están enfermas? (How do the plants look when they are sick?)
The Taita responds: “Ella muestra, y también se sienten, como secarse, las hojitas también como achantaditas.” (She shows, and also feels, as though she is drying up. The leaves also appear as if they are ashamed)
In the film, the black screen dissolves into another black screen, and the third part begins.
Film still: A Mobil Oil Barrel lies discarded by a destroyed cocaine laboratory.
(Hannah Meszaros Martin, 2020)
The camera moves out of the black, pointing towards the sky, scanning. There is a destroyed, burnt structure, trees and a thick underbrush. The narrator is the cameraman, he is a DEA agent. He says: “I can hear the helicopters from above and can't see 'em too much. Unless they fly in the right area, of course, if you fly over there you couldn't even see the lab.”
The moment before herbicides arrived in Colombia’s landscapes is merged with the experience of those who also live with the accumulated chemicals in their farms, water, and bodies. These two colliding worlds come together to tell a story about how the war on drugs is also a war on the natural world.
We slowly understand that this is a destroyed cocaine laboratory in an undisclosed location in Colombia. He narrates the scene, moving through the rubble, he tells the viewer the different chemicals he encounters, focusing his lens on each empty barrel as if they are pieces of evidence he is constructing in a story of chemicals in the production of cocaine. The camera pans to a discarded yellow barrel with writing that lies on the ground.
He says: “Local, local company here. We also have...Dow Chemical.” He zooms in on a Dow Chemical label, then one from Mobil Oil. He names them all for the viewer. We see many of these discarded barrels piled high in the forest.
Inside the destroyed lab, he films a yellow container in the dark. The camera pans into the darkness as the narrator speaks about what he is seeing. We the viewer can barely discern the images on the screen.
Then a Colombian military officer asks: “¿Falta de Luz?” (You need light?)
The DEA Agent answers softly, and distractedly: “No.” We can no longer see the image in the grain of the film.
The DEA agent walks slowly through the dense underbrush. We hear the crunching of the vegetation beneath his feet. He stops to film a black tarpaulin that covers something nestled in the brush. As he films the black cover the noise of the damaged analogue tape begins to dissolve the image, covering the cover.
The image dissolves into an aerial view, the final part of the video. We are flying towards the destroyed lab, the DEA agent filming out the window. There are thin white trees that the helicopter encircles over and over again. They become a blur of yellow and white. Descending, we get closer and closer to the canopy.
In the final montage of archival footage, the worlds of the Georgian Mountains and the landscapes of the Colombian Amazon collide. The grain of the film expands, the noise consuming the image.
The film constructs an interplay of temporalities and toxic geographies. Forming a dialogue between the chemical substances used to eradicate the plants. The moment before herbicides arrived in Colombia’s landscapes is merged with the experience of those who also live with the accumulated chemicals in their farms, water, and bodies. These two colliding worlds come together to tell a story about how the war on drugs is also a war on the natural world.
Film still: The final shot of the film, recorded from a helicopter back in the mountains of Georgia, we see a team of eradicators as they move within the thick vegetation, the camera zooms in as we lose sight of them, just below the tree canopy.
(Hannah Meszaros Martin, 2020)
Team
GUAVIARE
Research led by
Hannah Meszaros Martin
Paulo Murillo Sandoval
Researchers
Pedro Sánchez
Camilo Garcia
Luis Diego Arias Campos
Satellite remote sensing research
Paulo Murillo Sandoval
Pedro Sanchez
2D and 3D design conception
Camilo Garcia
Research support
Oscar Pedraza
Nadia Méndez
Jamon van den Hoek
Gustavo Adolfo Niño Rojas
Sofia Prado
Maria Fernanda Vaca
The Pixel and the Plot (2025)
Directed by
Hannah Meszaros Martin
Script
Hannah Meszaros Martin
Camilo Garcia
Animation
Camilo Garcia
Sofia Prado
Editing
Andrés Jurado
Narration
Andrés Jurado
Sound design
Andrés Jurado
Script consultation
Oscar Pedraza
Paulo Murillo Sandoval
Andrés Jurado
Camera
Hannah Meszaros Martin
Alejandro Jaramillo
Field producer
Gerald Bermudez
Subtitles
Maria Fernanda Vaca
UNODC DATA
Authors
Hannah Meszaros Martin (Plano Negativo)
Paulo Murillo Sandoval (Universidad de Tolima)
Jamon Van Den Hoek (Oregon State University)
FALTA DE LUZ
Falta de Luz (2020)
Directed by
Hannah Meszaros Martin
Editing
Manuel Correa
Hannah Meszaros Martin
Sound design
Emil Olsen
Musical score
Mhamad Safa
Interview ‘Taita’ conducted in collaboration with
Asicaz Monzón-Aguirre
Archival consultant
Emily Coxe
SUPPORT
Made with the support of
Porticus Foundation
Antipode Foundation Ltd.
Mellon Foundation
The Center for Creative Ecologies at the University of California Santa Cruz
SPECIAL THANKS
Special thanks to
Fundación Paiz
La Comisión para el Esclarecimiento de la Verdad, la Convivencia y la No Repetición
Forensic Architecture
Michael Evans and The National Security Archives
La Vulcanizadora
Manuel Correa
Nick Masterton
Ariel Caine
Kishan San
Fernanda Barbosa
Folco Zaffalon
Alejandro Valencia Villa
Martín Martínez
Edinson Arroyo
Brett Story
Nicolás Pereda
The Center for Ethnographic Media Arts
El Consejo Regional Indígena del Cauca
Edinson Ivan Arroyo Mora
Jessica Marsden for support with the FOIA process
All those who would like to remain anonymous and helped us verify the eradication centroids in Guaviare