1 of 63

No peace for the forest:

Land changes in the Andes-Amazon region following the Colombian internal conflict

Paulo J. Murillo-Sandoval

2 of 63

Conflict is common.

Source: PRIO Institute, Conflict Catalog,

HYDE population dataset and UN

Circle size represents number of deaths

Rate of deaths per 100.000 people

Conflict Catalog (1400-2000)

PRIO Institute

(1946-2013)

3 of 63

Conflict impacts urban areas

Germany, Nuremberg, 1945

Source: National Archives and Records Administration

4 of 63

Conflict impacts forested and agricultural areas

Source: National Geographic

Illegal wood / coca farming

Agricultural land abandonment

Source: https://globalforestatlas.yale.edu/amazon

5 of 63

Colombia - Amazon

Myanmar - Asia

D.R Congo - Africa

Source: UCDP/Google Earth Engine

Conflict occurs in the tropics

Forest cover

Conflict events

6 of 63

Colombia is an excellent place for studying armed conflict influences on land system dynamics.

~60% of Colombia remains as forests.

Rich in biodiversity

Total deaths:

  • 260,000

Internally displaced persons (IDPs):

  • 7 Million

Civilians:

  • 214,000

Combatants:

  • 46,500

By guerrillas:

  • 35,000

By Paramilitaries:

  • 95,000

FARC

Germany

7 of 63

Peace agreement:

Colombian Government and FARC

November 2016

Former president: Santos

Nobel Peace Prize 2016

FARC leader:

Timochenko

Political participation in Colombian

Congress

8 of 63

Source: ACLED https://blogs.lse.ac.uk/latamcaribbean/2020/09/29/como-entender-la-ola-de-asesinatos-de-lideres-sociales-en-colombia-durante-la-pandemia/

“Peace” is fragile

Number of social leaders killed 2019-2020

Source: Murillo-Sandoval, August 2019

La Julia (Meta-Colombia).

Juan de Jesus Arroyo: Leader ~102 ex-FARC combatants

9 of 63

Overarching question:

How does armed conflict and the current post-conflict situation impact land system in the Colombian Andes-Amazon region?

10 of 63

The relationship between conflict processes and land system is difficult to study:

  1. Lack of adequate spatiotemporal datasets.

3. The impact of illicit land activities (i.e., coca farming) on landscape during conflict and post-conflict periods are unknown.

2. Mostly speculative, causal effects or mechanisms are not well understood.

11 of 63

Andes-Amazon transition belt

Strongholds of FARC

Specific massacres by Paramilitaries

IDPs

Historical land grabbing & land abandonment resulting in complex mosaic of forest, forest recovery and agricultural expansion

12 of 63

Land system

Land cover and land use types

Conflict → A fluid process, not a discrete series of isolated events

Analytical model

13 of 63

1.

Did the 2016 declaration of peace increase forest disturbance?

14 of 63

1. Methodology

bfastMonitor: Intrannual forest disturbances (2010-2018)

Spatial representation

Source: Murillo-Sandoval et al., 2020

Disturbancess

Forest stable

15 of 63

1. Methodology

Ethnographic data: Collected by Kristina Van Dexter (2016-2018) and Paulo Murillo (2019). n = 80.

Source: Murillo-Sandoval (La Julia, August, 2019)

Spatial representation

Workshop with FARC ex-combatants, campesinos, WWF and national park managers.

Disturbancess

Forest stable

16 of 63

Ch1. Methodology

Juan de Jesus Monroy (aka “Albeiro”) explains the agricultural projects, location and future goals.

Source: Murillo-Sandoval (La Julia, August 2019 )

17 of 63

Chapter 1. Results

1. Results: Choropleth map

18 of 63

1. Forest disturbances

Murillo et al., Env Research Letters, 2020.

19 of 63

1. Results:

Protected Areas

20 of 63

1. What is happening?

Wartime

2010-2012

Post-peace Agreement

2017-present

Negotiations

2013-2016

Everyone respects the environmental laws created by FARC

<2ha per year

Keep 30% of farm as forest.

Uncertainty laws

Campesinos afraid from new government rules.

People opt for coca

> 7ha

More than 60% of forest farm could be cleared

Land tenure is not the goal/

Capitalize land and get rent or interest on the money.

New outsider investors, large landholders, campesinos and FARC dissidents

21 of 63

1. Implications

Small sample (2010-2018) and no info about other land cover / land use types….

Forest was not central element in the peace accord.

Forest is a victim of “peace.”

Massive deforestation in sensitive biodiversity hotspots.

22 of 63

2.

How did conditions during the conflict and post-conflict period affect agricultural uses and forest land cover dynamics (1988-2019)?

23 of 63

2. Methodology

Source: https://emapr.github.io/LT-GEE/landtrendr.html

Land cover mapping using LandTrendr + RandomForest

  • Intensity analysis
  • Sankey plots
  • Landscape metrics

24 of 63

LandTrendr video over Tinigua!

25 of 63

LandTrendr video over Tinigua!

26 of 63

2. Connecting maps with conflict events

Armed confrontation between two groups led to at least one person killed

27 of 63

2. Connecting maps with conflict events

Armed confrontation between two groups led to at least one person killed

28 of 63

2. Connecting maps with conflict events

Armed confrontation between two groups led to at least one person killed

We use a diff-in-diff framework to estimate the causal effect of a conflict event on land cover change and landscape metrics for agriculture.

29 of 63

2. Lots of change from Forest to Agriculture

Map difference: 1988-2019

Fast

Slow

30 of 63

Loss of stable forest

Murillo et al., Global Env Change, 2021.

31 of 63

Significant conversion to agriculture

Murillo et al., Global Env Change, 2021.

32 of 63

Loss of secondary forest

Murillo et al., Global Env Change, 2021.

33 of 63

2. Heterogeneous effects in conflict events

Municipalities are separated by low-population (population below the 50th percentile) and high-population (population >= 50th).

Significance levels of the estimates are represented by: * p<.10, ** p<.05, *** p<.01.

34 of 63

2. Implications

How, where and what type of illicit land activities are happening….

Secondary forests are deforested faster after the peace accord

Violent actors dictated land changes during conflict

but

Wealthy actors aggressively acted during “peace”.

Low populated areas experience quick ag expansion

35 of 63

3.

How are policy regimes such as the war on drugs (i.e coca substitution and aerial fumigation campaigns), and the recent peace accord related to the expansion of illicit land activities?

36 of 63

>40% of deforestation is linked to illicit land activities (Lawson 2014), but finding them is very hard:

1. People intentionally work to obscure traces of illicit land activities.

2. Available remote sensing data is not linked with illicit land activities

3. They might have the same pattern as legal activities.

4. We do not know about the magnitude of illicit land activities (i.e., coca and illegal cattle ranching) during and after conflict.

37 of 63

Socializing the pixel

Pixelizing the social

Pattern:

e.g., Fishbone

Process:

Urban settlement

Known “pattern” with suspected processes that drive it

Tellman et al., 2020

Known “process” or “driver”

with suspected land outcome.

Process:

e.g., Forest management

Pattern:

Regular solid plots

Kennedy et al., 2015

Two pixel-based approaches

38 of 63

How do these two pixel-based approaches help disentangle coca farming and cattle ranching farms?

  1. We build hypothesis based on observable patterns from satellite data (Socializing the pixel)

  1. We discriminate these land uses (using few available field data) using deep learning algorithms (Pixelizing the social)

Note: If you have enough on the ground data you do not need socializing the pixel

39 of 63

Linking observable patterns with known historical and institutional processes

40 of 63

3. Semantic segmentation

41 of 63

Ch3. Semantic segmentation

Landsat image 2018

Field map

Deep learning model

Residual UNet (He et al., 2020, Ronneberger et al., 2015; 2016)

Six Landsat bands+slope+

elevation

Three classes: Forest, coca and cattle farms

42 of 63

43 of 63

By 2018. Coca→ 50kha ; Cattle ranching → 3000kha

Within

legal frontier

Outside legal frontier

frontier

44 of 63

3. Implications

Still, we do not know about other illicit land activities (oil palm, mining, illegal timber….)

Coca farming is NOT the initial spearhead of deforestation

Illegal cattle ranching is pushing deeper into the Amazon watershed

War on drugs is ineffective, “peace” encourages more illicit land activities

45 of 63

The relationship between conflict processes and land system is difficult to study:

  1. Lack of adequate spatiotemporal datasets.

  1. Mostly speculative, causal effect/causal mechanisms are less understood.

  1. The impact of illicit land activities (i.e., coca farming) on landscape during conflict and post-conflict periods are unknown.

46 of 63

Dissertation’s contributions to the land system science

  1. Lack of adequate spatiotemporal datasets.

Methodological contribution:

bfastMonitor/LandTrendr/deep learning + ethnography/stats/social approaches

47 of 63

Dissertation’s contributions to the land system science

2. Mostly speculative, causal effect or mechanisms are less understood.

Empirical contribution:

Conflict was beneficial (slow changes) for Amazon land system but during post-conflict (fast change) massive forest loss and ag expansion

Conflict events cause deforestation and ag expansion

48 of 63

Dissertation’s contributions to the land system science

3. The impact of illicit land activities (i.e., coca farming) on landscape during conflict and post-conflict periods are unknown.

Conceptual contribution:

First worldwide example connecting illicit land activities (i.e., coca and illegal cattle ranching) with specific policies. A framework that can be extended in other countries.

49 of 63

Source: Semana, January 8, 2017

50 of 63

Backup

51 of 63

Coca

“Balloon effect”

Pasture

“Cattle ranching”

Land grabbing

“No cows yet”

52 of 63

Ch2. Empirical Strategy

We use a diff-in-diff framework to estimate the causal effect of a conflict event on land cover and landscape metrics for agriculture.

53 of 63

54 of 63

After these shocking numbers, Colombia looks an interesting study case, but globally, Is war a common process, where/when armed conflict occurs?

Bubble’ size represents the size of each event with respect to the total casualties (raw: C, rescaled: D)

Source: Cirilo and Taleb 2016

55 of 63

Colombia - Amazon

Myanmar - Asia

D.R Congo - Africa

56 of 63

Urban areas, but also agricultural and forested regions are affected by armed conflict

Source: Baumann et al., 2016

57 of 63

Fighting and Foliage:

Land system dynamics in the Colombian Andes-Amazon region

58 of 63

Ch2. Results: Land cover pathways

59 of 63

Landsat image 2018

Field data 2018

Deep learning 2018

60 of 63

61 of 63

Ch2. Results - Heterogeneous effects

Municipalities are separated by low-population (population below the 50th percentile) and high-population (population >= 50th).

Significance levels of the estimates are represented by: * p<.10, ** p<.05, *** p<.01.

62 of 63

After these shocking numbers, Colombia looks an interesting study case, but globally, Is war a common process, where/when armed conflict occurs?

Source: UCDP

63 of 63

Conflict: Internal armed confrontation between two groups, that leads at least 25 battle-related deaths in one calendar year.

Land system: Mosaic that combine land cover and land uses

Post-conflict: difficult to actually define but when a decrease of deaths, peace accords, cease of fire are declared we can identify when this period starts.