Revista de Biología Tropical, ISSN: 2215-2075, Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

Spatio-temporal changes for natural resources conservation

and illegal logging monitoring using Breaks for Additive Season

and Trend (BFAST) algorithm in Costa Rica

Iván Dimitri Ávila-Pérez1*; https://orcid.org/0000-0002-4093-2799

Erik Lindquist2; https://orcid.org/0000-0001-5265-8021

Cornelia Miller-Granados3; https://orcid.org/0000-0002-9460-626X

Matieu Henry4; https://orcid.org/0000-0002-1145-2390

1. National Center for High Technology, PRIAS Laboratory, San Jose, Costa Rica; iavila@cenat.ac.cr (*Correspondence)

2. Food and Agricultural Organization of the United Nations, Rome, Italy; Erik.Lindquist@fao.org

3. National Center for High Technology, PRIAS Laboratory, San Jose, Costa Rica; cmiller@cenat.ac.cr

4. Food and Agricultural Organization of the United Nations, Rome, Italy; Matieu.Henry@fao.org

Received 18-I-2024. Corrected 11-X-2024. Accepted 25-XI-2024.

ABSTRACT

Introduction: Costa Rica is committed to addressing environmental issues by involving a range of strategies and

policies, with goals of sustainability and conservation. Nonetheless, addressing many challenges remains neces-

sary, with the prominent issue of illegal activities, such as logging and land use change.

Objective: To evaluate the direct detection capacity of tree cover losses caused by logging within the various

land uses of the landscape, and their relationship with physical variables of the environment such as slope and

proximity to the road network using remote sensing techniques.

Methods: Tree cover losses were detected using time series analysis of the Normalized Difference Vegetation

Index (NDVI) from Landsat and Sentinel images (S2) through the Breaks for Additive Season and Trend

(BFAST) algorithm in The Golfo Dulce Forest Reserve (RFGD) and the Amistosa Biological Corridor (CBA).

Selected sites where logging was detected were physically visited in the field and inspected using Unmanned

Aerial Vehicles (UAVs). The results were analyzed through confusion matrices to determine the algorithms

accuracy to detect illegal logging.

Results: The study highlighted a significant relationship between NDVI change and logging activities on the

ground. In areas with major NDVI changes (less than -500), the model accuracy was greater than 75 %. In addi-

tion, there is a significant relationship between logged areas and slope, and distance to roads.

Conclusions: The proposed methodological approach allows identifying forest cover logging activities in space

and time. It could be adopted and complement field operations to improve monitoring of illegal logging.

Key words: deforestation; land use change; SEPAL; NDVI; remote sensing; earth observation.

RESUMEN

Cambios espacio-temporales para la conservación de recursos naturales y el monitoreo de la tala ilegal

mediante el algoritmo Quiebres por temporada y tendencia aditiva (BFAST) en Costa Rica

Introducción: Costa Rica es un país comprometido en tratar temas ambientales, involucrando estrategias y polí-

ticas con objetivos de sostenibilidad y conservación. Sin embargo, aún se requiere trabajar en soluciones a varios

problemas, entre los que se destacan actividades ilegales como la tala y el cambio de uso de suelo.

https://doi.org/10.15517/rev.biol.trop..v72i1.58329

CONSERVATION

2Revista de Biología Tropical, ISSN: 2215-2075 Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

INTRODUCTION

Costa Rica has set ambitious targets for

protecting its environment and strong emphasis

on preserving its rich biodiversity. With around

702 366 hectares of secondary forest (Sistema

Nacional de Áreas de Conservación [SINAC],

2015) and more than 25 % of its territory allo-

cated to protected areas (Marx et al., 2017),

Costa Rica is internationally recognized as an

example for environmental protection. As for

example, the national system for Payment for

Environmental Services serves as a model for

other countries seeking to balance economic

development with environmental conservation

(Le Coq et al., 2015). However, to consolidate

the progress of national resources conserva-

tion, it is necessary to work on solutions to

several problems, among which the prevalence

of illegal activities such as logging and land use

change and structural problems in territories

where the rural population is concentrated and

where nature conservation is a priority (Minis-

terio de Ambiente y Energía [MINAE], 2015).

In recent years, technical advances in the

field of remote sensing and earth observation

allowed the development and use of revolu-

tionary methods for monitoring forest cover

dynamics and landscape changes (Cipta et al.,

2019; Marx et al., 2017; Yin et al., 2018). Among

the technologies used, Synthetic Aperture Radar

sensors (Hoekman et al., 2020), optical sensors

(Coulter et al., 2016), data fusion of both types

and the use of time series (Muñoz et al., 2020)

have been successfully implemented and dem-

onstrated useful results. On the other hand,

among the data used today, the Normalized

Difference Vegetation Index (NDVI) stands out

for its importance, since it allows analyzing the

vigor of the vegetation as well as monitoring

and predicting the dynamics of the vegetation

(Huang et al., 2021; Zhang et al., 2016).

When analyzing the NDVI through time

series, it is possible to identify three categories

of changes: seasonal, gradual and abrupt (Ver-

besselt et al., 2010). The seasonal changes are

due to the natural variability of the reflectance

according to the phenological cycles of the

plants. Gradual changes occur due to slow envi-

ronmental events, such as climate change, pest

and diseases or anthropogenic actions such as

“socola” (Wu et al., 2018). Finally, abrupt chang-

es are due to sudden events that affect coverage,

such as landslides or forest exploitation.

The detecting and monitoring land use/

land cover (LULC) changes is crucial (Ngadi,

et al., 2023). These changes can be detected by

the algorithm of Breaks for Additive Season

Objetivo: Evaluar la capacidad de detección directa de las pérdidas de cobertura arbórea dentro de los diversos

usos de suelo de un paisaje debido a la tala, y su relación con variables físicas del entorno como pendiente y

proximidad a la red vial mediante técnicas de teledetección.

Métodos: Las pérdidas de cobertura arbórea se detectaron mediante series temporales del Índice de Vegetación

de Diferencia Normalizada (NDVI) en imágenes Landsat y Sentinel a través del algoritmo Breaks for Additive

Season and Trend (BFAST) en la Reserva Forestal Golfo Dulce (RFGD) y el Corredor Biológico Amistosa (CBA).

Los sitios seleccionados donde se detectó tala fueron visitados físicamente en el campo e inspeccionados utili-

zando vehículos aéreos no tripulados (UAVs). Los resultados se analizaron mediante matrices de confusión para

determinar la precisión del algoritmo en detectar la tala ilegal.

Resultados: El estudio resaltó que existe una relación entre el cambio de NDVI detectado y la existencia de tala

en campo. En las áreas con cambios en el NDVI inferiores a -500 se obtuvo una precisión global superior al 75 %

de modelo. Además, existe una relación entre las áreas taladas y la pendiente del terreno, así como su distancia

a las vías de acceso.

Conclusiones: La metodología propuesta permite identificar actividad de tala de la cobertura forestal en espacio

y tiempo. Adoptar esta metodología e integrarla en las labores de monitoreo dentro de las instituciones guberna-

mentales permitirá un seguimiento más eficiente de la tala ilegal.

Palabras clave: deforestación; cambio de uso de la tierra; SEPAL; NDVI; sensores remotos.

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

and Trend (BFAST) which allows the analysis

of time series by decomposing them into trend

and seasonality components to detect and char-

acterize changes in a definite period of time

(Anoma-Kouassi et al., 2022). BFAST itera-

tively estimates the time and number of abrupt

changes within time series and characterizes

the change by its magnitude and direction (Ver-

besselt et al, 2010). Using this algorithm, time

series can be analyzed from various sensors

(for example, Landsat and Sentinel) and can be

applied to disciplines such as hydrology, clima-

tology, and econometrics (Geng et al., 2019).

BFAST method detects significant seasonal

changes in dense satellite image time series

(SITS) by combining a high-speed algorithm

with a time series similarity metric. Tested on

various datasets, BFASTm-L2 effectively identi-

fies changes, especially those caused by large-

scale land acquisitions, outperforming existing

algorithms in accuracy and speed crucial (Ngadi

et al., 2023). BFAST can provide analysis of for-

est disturbances and resilience (Zhu et al., 2024)

indicating that 20 % of forests in East, South,

and Southeast Asia experienced disturbances

from 2000 to 2022, with Southeast Asia being

most affected. Notably, 95 % of forests showed

robust resilience, often recovering within a few

decades, with stronger resilience observed in

forests experiencing greater disturbances.

This study evaluates the capacity of remote

sensing applications to detect tree cover losses,

logging and illegal logging in different land uses

and landscapes by integrating time series analy-

sis of the Normalized Difference Vegetation

Index from Landsat and Sentinel (S2) using the

Breaks for Additive Season and Trend (BFAST)

algorithm and field data. In addition, the study

explores the potential of using remote sens-

ing information and physical variables such as

slope and accessibility, to predict areas that are

more vulnerable to illegal logging and then may

benefit from stronger field and remote sensing

monitoring. This aims at supporting national

entities in their effort to prevent illegal logging

and improve nature conservation.

MATERIALS AND METHODS

Study site: The study area corresponds to

the Golfo Dulce Forest Reserve (RFGD), locat-

ed in the Osa conservation area (ACOSA) and

the Amistosa Biological Corridor (CBA), that

connects the Osa conservation area (ACOSA)

and the Amistad-Pacific conservation area

(ACLA-P) (Fig. 1). The RFGD (8º23’37”-

8º49’45” N & 83º15’53”-83º43’42” W) has an

official extension of 59 915 ha. The climate of

this area is characterized by an average annual

temperature between 26 and 28 °C (Tapia,

2011a) and an average annual rainfall that

varies between 3 000 and 6 000 mm (Tapia,

2011b). In the less humid parts, up to a dry

month can be recorded.

As for the CBA (5º33’31”-9º01’32” N &

83º13’46”-82º49’17” W), with an area of 92 913

ha (SINAC, 2018), the climate in the corridor

is characterized by an average annual tempera-

ture between 20 and 28 °C (Tapia, 2011a) and

an average annual rainfall that varies between

1 500 and 6 000 mm (Tapia, 2011b). In the dri-

est parts, from one month up to three months

of dry season has been reported.

On the other hand, forestry activities are

intense in both study sites since the diagnoses

of the regulatory plans define this region as

an area of great forest wealth, with more than

40 timber species. According to field data

campaigns, this caused 16 % of the forest to be

logged between 1980 and 1995, further dividing

an additional 3 %; being this area in great risk

of being deforested (Roman & Angulo, 2013;

Rosero-Bixby et al., 2002).

Implementation of the BFAST algorithm

in SEPAL for change detection: Through the

Sepal platform (Tondapu et al., 2018), a tempo-

ral analysis of the study area was carried out, in

the period between June 1st, 2019 and Decem-

ber 31st, 2020 using the BFAST algorithm

implemented through the BFAST GPU plugin.

In the present study, a combined analysis was

carried out between the Sentinel-2 and Landsat

4Revista de Biología Tropical, ISSN: 2215-2075 Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

8 images and the algorithm was programmed

to detect monthly changes in the Normalized

Vegetation Index (NDVI) on the surface of the

area of study.

Random selection of sampling polygons

and field control: Based on the results obtained

by the algorithm, a review of the areas identi-

fied in SEPAL was carried out through the

NICFI platform, which allows access to PLAN-

ET image mosaics. Based on this review, three

categories of change were detected: Perma-

nence of tree cover, loss of non-tree vegetation

and loss of tree cover. For each of the classes, a

random selection of the polygons with a nega-

tive change in NDVI was carried to determine

the sample that was field visited. Then, the

formula recommended by Chuvieco (2010) to

establish a validation sample for a quantitative

variable was used, to calculate the number of

polygons to visit in each class.

Once the negative polygons were deter-

mined, a random selection of the positive poly-

gons was carried out using the same number

obtained for the negative areas. In total, 228

polygons were selected within RFGD with a

total area of 133.98 ha (Fig. 1), which represents

2.81 % of the total area of the polygons identi-

fied by the algorithm in RFGD. In the case of

CBA, 276 polygons with a total area of 216.78

ha were selected (Fig. 1), which represents a

sampling of 1.32 % of the total area of the poly-

gons identified by the algorithm.

Field measurements: Once the sampling

areas were defined, each of the polygons to be

checked in the field were visited by means of

photogrammetric flights with UAVS at 120 m to

determine the existence of logging signs within

each area. According to the existence of infra-

structures for forest exploitation and reduc-

tion of the canopy in the field, each polygon

Fig. 1. Location of the study sites. A. Golfo Dulce Forest Reserve (RFGD) located in the Osa conservation area (ACOSA). B.

the Amistosa Biological Corridor (CBA).

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

was assigned a damage category according

to Table 1.

Algorithm validation and data analy-

sis: The results of the BFAST algorithm were

validated using error matrices according to

the results of the field control polygons. Any

negative change in the NDVI detected by the

algorithm was considered as a warning of loss

of tree cover, while positive changes were con-

sidered as permanence or increase in canopy in

each area. From the error matrices, the accu-

racy of the producer and the user was calcu-

lated for both classes, in addition to the total

accuracy of each study area.

To determine the variables that influ-

ence the identification of areas where log-

ging was carried out, various factors between

field aspects and computational data were ana-

lyzed. The relationship between the changes in

the NDVI detected by the BFAST algorithm

and the results obtained in the field was ana-

lyzed. For this, the mode of the pixels of the

raster obtained after the implementation of

the algorithm was calculated using QGIS 3.18

(QGIS.org, 2021).

This information was compared with the

information obtained in the field on the pres-

ence of logging signs in the polygon using a

T-Test for independent samples in “R”. In order

to study the relationship between the presence

of logging in the field and the land use prior to

logging, a supervised classification was carried

out on the SEPAL platform of a mosaic gener-

ated for the 2019 year on the same platform and

the results were compared using contingency

tables and the calculation of Pearson’s Chi

square statistic in “R” (R Core Team, 2021).

Then, differences between land uses and dam-

age categories were analyzed through changes

in NDVI using Analysis of Variances (ANO-

VAS) tests in “R”.

In order to contrast the distance from the

polygon to the routes and access roads and the

presence of signs of logging in these areas, said

distance was and classified into three different

categories: from 0 to 500 meters, from 500 to 1

000 meters and more than 1 000 meters. With

this classification, the distribution of frequen-

cies of the polygons with and without logging

signs in said categories is compared using con-

tingency tables and the calculation of Pearson’s

Chi-square statistic in “R”.

To determine the relationship between

the polygons with logging signs and the slope

percentage in these areas, firstly, the slope per-

centages were calculated from a Digital Eleva-

tion Model (DEM) obtained on the Copernicus

platform. Then the mode of the raster pixels

was associated with the slopes. Finally, it was

compared with the information obtained in

the field on the presence of logging signs in

the polygon, using a T Test for independent

samples in “R” (R Core Team, 2021).

Table 1

Diagrammatic scale of damage assessment.

Severity

categories

Damage types

1Place without loss of tree cover, there are no signs of any forest exploitation

2There is a loss of tree cover due to the felling of individuals, there are signs of forest exploitation such as

clearings and logging tracks

3There is a loss of tree cover due to the felling of individuals and the creation of forest exploitation

structures. Clearings, extraction tracks, skid trails and storage yards are observed

4There is a loss of tree cover due to the felling of individuals and the creation of forest exploitation

structures. Clearings, extraction tracks, skid trails and storage yards, loading yards and truck roads are

observed.

5The loss of tree cover causes a total change in land use in the area and the disappearance of the forest.

6Revista de Biología Tropical, ISSN: 2215-2075 Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

RESULTS

Algorithm validation: The results show

that for the Amistosa Biological Corridor

(CBA), a general accuracy of 54.51 % was

obtained by the algorithm, while for the Golfo

Dulce Forest Reserve (RFGD) the accuracy was

64.63 % when all negative changes detected in

the NDVI were considered as lost in tree cover.

For the CBA, 35 polygons that were origi-

nally considered as polygons with loss of forest

cover, in the field had indeed suffered type 2, 3,

4 or 5 damage (Fig. 2A) and 103 did not suffer

changes and had remained as forest cover. In

the case of RFGD, of the 114 polygons classified

by the algorithm as loss of forest cover, 46 have

actually suffered changes, which were verified

in the field, and 68 remained unchanged.

The cases presented where the algorithm

detected a negative change in the NDVI, but

no cut signs were found in the field, can be

explained due to various changes in the reflec-

tivity of the canopy for reasons such as stress,

seasonality, factors associated with the index,

among others. The accuracy of the user of the

tree cover loss class obtained for both CBA

(25.36 %) and RFGD (40.35 %) shows that high

possibilities are obtained of considering poly-

gons that have not suffered damage as logging

areas when only the direction of change of the

NDVI is taken into account. On the other hand,

the accuracy of the user (92.45 %) of the class

“no change” indicates that there are low prob-

abilities of classifying a polygon where a cut was

carried out as “no change”.

Regarding the accuracy of the producer,

there is a 51.24 % probability in CBA and a

43.59 % probability in RFGD of incorrect-

ly classifying polygons that did not present

changes within the logging class, when only

taking into account the direction of change in

the NDVI. However, despite the limitations

presented, the algorithm is able to identify the

logging of trees in various land uses such as

wooded pastures (Fig. 2B).

Relation of the changes in the NDVI and

the presence of logging in the field: There

were differences between the polygons with

and without signs of logging according to the

direction and magnitude of the change in the

NDVI for both CBA (p-value = 0.0002) and

RFGD (p-value < 0.0001). The polygons where

logging was carried out present much higher

negative magnitudes of change compared to the

polygons without change (Fig. 3).

Similarly, there are significant differences

between the categories of damage observed in

the field and the change in the NDVI calculated

by the algorithm in both study areas (p-value

= <0.0001). The greatest magnitude of change

detected was present in the polygons under

category 3 (Fig. 4).

Fig. 4 shows how the detected NDVI

change increases from category 1 to category

3 and then decreases from category 4 to 5. This

Fig. 2. Polygon with signs of logging detected in the field.

A. Logging signs in forest polygon. B. Logging signs in

wooded pastures.

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

may be due to the fact that categories 4 and 5

have almost or completely lost their tree cover,

which means that they have a reflectance simi-

lar to non-forested areas and that the change in

NDVI is more stable and tends towards 0.

Most of the categories that imply the pres-

ence of logging (2, 3, 4 and 5) show NDVI

changes below -500, which implies that areas

with a value above -500 are more likely to

represent polygons without changes (Fig. 4).

Therefore, a validation was carried out again,

considering as classified with the possibility of

logging only those polygons where the algo-

rithm detected a negative change in the NDVI

below -500. Using this new limit, the global

accuracy increases to 75.82 % in the case of

CBA and up to 75.82 % in the case of RFGD.

Relation of physical variables of the envi-

ronment and the presence of logging: There

are no significant differences between the pres-

ence or not of logging according to the land

use of each polygon (CBA: p-value = 0.8008,

RFGD: p-value = 0.1205); however, the fre-

quency distribution shows that most of the cuts

detected in the field occur within the forest

category, followed by the pasture class.

The study shows that there is a significant

relationship between logging and the distance

to the roads in CBA (p-value= <0.0159) and

RFGD (p-value= <0.0001) (Table 2).

The areas with the presence of logging

decrease as the distance to the access roads

increases, which coincides with the dynamics of

logging on the site, since vehicle access is neces-

sary to carry out the extraction of the wood.

When analyzing the behavior within the

class from 0 to 500 m, the same trend is

observed in both zones, since the majority of

polygons with presence of logging are located

Fig. 3. Changes in the average NDVI according to the presence of logging signs. A. Amistosa Biological Corridor. B. Golfo

Dulce Forest Reserve.

Fig. 4. Changes in the average NDVI according to damage category. A. Amistosa Biological Corridor. B. Golfo Dulce Forest

Reserve.

8Revista de Biología Tropical, ISSN: 2215-2075 Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

between 0 and 100 m, decreasing the frequency

as the distance to the tracks increases (Fig. 5).

Regarding the relationship between the

percentage of slope and tree cuts, the p-value

obtained (< 0.0001) allows us to determine that

land with a lower slope is more likely to be log

(Fig. 6).

DISCUSSION

By associating the information generat-

ed through the methodology exposed in this

paper, with the logging licenses granted in the

area, it would be possible to create a digital

monitoring system that allows to determine

loggings that were carried out illegally. This sys-

tem could even include the information from

the date of logged detected by the methodology,

to the period of each logging license, so that it

can be determined if, once the time of a license

expires, illegal logging continues and could

result in a possible change of land use.

Efforts to create similar systems are

recorded worldwide. For example, Miomir et

al. (2018), worked with the Minister of Science

and Technological Development of the Repub-

lic of Serbia on forest inventories and illegal

logging prevention in this country by analyzing

NDVI changes in Landsat images from 2006 to

2014. Also, Isaienkov et al. (2020), used Deep

Learning in Sentinel 2 images to detect regular

changes in Ukrainian forest ecosystems which

Fig. 5. Relationship of the presence of signs of logging and distance from the area to the access roads within the category of

0 to 500 m. A. Amistosa Biological Corridor. B. Golfo Dulce Forest Reserve.

Fig. 6. Percentage of average slope according to the presence

of signs of logging.

Table 2

Relationship between the presence of logging and the distance to roads.

Distance from roads CBA RFGD

Logging No changes Total Logging No changes Total

0 - 500 m 32 104 136 39 63 102

500 – 1 000 m 6 35 41 7 22 29

More than 1 000 m 5 62 67 5 71 76

Total 43 201 244 51 156 207

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

develop in periodical and effective monitoring

of changes in forest cover.

On the other hand, it is important to

understand the limits of remote sensing tech-

nologies. While the models have the potential

to be adopted and offer operational functions,

it is important to highlight some characteristics

and interpretation of NDVI in different condi-

tions. For example, that the algorithm detected

negative changes in NDVI without trees being

cut may be due to the seasonality of the area,

that is, the changes that occur in the trees such

as the loss of leaves, flowering and fruiting,

depending on the season of the year in which

it is found among other climatic and phys-

iographic characteristics (Olivares & López-

Beltrán, 2019). In addition, under drought and

pest vegetation stress (Rimkus et al., 2017),

the reflectance is changing (Sun et al., 2018).

Therefore, this difference in spectral response

can lead to negative changes in NDVI over time

(Díaz, 2015). In addition, the spatial heteroge-

neity of the NDVI values in a particular region

could also be influenced by the topography and

environmental variables (Gómez, 2019).

Related to the factors associated with the

index that can cause the identification of log-

ging that have not undergone any change, is

that the NDVI has the drawback of being sen-

sitive to different climatic and environmental

factors (Dong et al., 2021) like the reflectivity

of the soil on which the vegetation is located,

which limits its discrimination potential. For

example, in an area with low density of veg-

etation, the reflectivity of a pixel would be

determined mainly by the ground, with a small

variation due to the presence of vegetation. In

fact, this problem occurs more frequently when

the vegetation cover is less than 50 % (Muñoz,

2013). This would explain some of the cases

where negative NDVI values were obtained

in polygons where the trees had considerable

distances, for example, in wooded pastures, or

there were natural clearings within the forest,

but no cutting of trees was detected in the field.

However, despite some limitations, the

algorithm was able to identify tree logging even

within wooded pastures. This shows why the

NDVI continues to be recommended and is

widely used to describe the spatio-temporal

variation of vegetation cover in different ter-

restrial ecosystems (Meera et al., 2015), due

to the long time series and global coverage,

which makes this index one of the most suit-

able for characterizing the vegetation cover of

a particular area (Ayhan et al., 2020) and also

allows identifying changes and trends in a given

period (Gómez, 2019; Usman et al., 2015).

At last, by integrating different data sourc-

es such as road network and terrain slope, it is

possible to equip illegal monitoring activities

with models that helps cost-effective methods

and approaches and support nature conserva-

tion. In terms of the relation between forestry

use and the distance to the road network, this

study shows that most logged areas appear on a

range from 0 m to 500 m which coincides with

the average travel distance of 500 m necessary

for an average productivity in wood harvesting

labor (Cadei et al., 2020). The results obtained

also coincides with López-Alegría et al. (2018),

who analyzed and modeled spatial patterns of

deforestation in the La Unión micro-basin in

the municipality of Chiapa de Corzo, Chiapas,

Mexico; and determined that there is a rela-

tionship between the distance of the roads and

the probability of logging, since, the greater

the road network, the greater the population

density and therefore greater pressure on the

production of the forests. This study also coin-

cides with that published by Pérez (2021), who

determined that there is a relation between the

distance to the road network and deforestation

in the Pómac Forest Historic Sanctuary, Peru;

and its buffer zone.

In terms of the relation between forestry

use and the terrain slope, this study coincides

with that reported by Putz et al. (2019), who

studied the behavior of forest harvesting pro-

cesses in Indonesia, Gabon, Democratic Repub-

lic of Congo, Republic of Congo, Suriname and

Mexico and determined that loggers tend to

avoid steep slopes, largely due to the fact that

said topography increases exploitation costs

by requiring techniques and machinery that

allow overcoming gravitational forces. Also,

10 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

Quispe (2021), studied deforestation patterns

in the district of Nueva Requena (Peru) and

coincide with this study since he determined

that on slopes less than 20 % there was greater

deforestation, with the trend on slopes being

even more evident in areas with less than 11 %.

In conclusion, this study demonstrates how

the implementation of the BFAST algorithm in

the SEPAL platform allows for repeated infor-

mation on the location and period of forest

cover logging in the study areas. The integra-

tion of the methodology used in this paper

in the monitoring labors within institutions

such as the National System of Conservation

Areas of Costa Rica would allow governments

to carry out more efficient monitoring of log-

ging within conservation areas and therefore

decrease the country’s deforestation rate.

Ethics statement: the authors declare that

they all agree with this publication and that

they have made contributions that justify their

authorship; that there is no conflict of interest

of any kind; and that they have complied with

all relevant ethical and legal requirements and

procedures. All funding sources are fully and

clearly detailed in the acknowledgments sec-

tion. The respective signed legal document is in

the archives of the magazine.

ACKNOWLEDGMENTS

We thank the Regional of the Osa Con-

servation Area (ACOSA) from the National

Conservation Areas System (SINAC) for the

valuable accompaniment and assistance in the

field. The SEPAL team for the assistance and

development of the base code. The FAO costar-

ican country office for doing the link between

PRIAS and the Forestry Office and FAO for the

financing of this project.

REFERENCES

Anoma-Kouassi, C., Khan, D., Achille, L., Omifolaji, J.

K., Espoire, M. M., Zhang, K., & Yang, X. (2022)

Assessing change of Lamto Reserve Area Based on

the MODIS time series data and bioclimatic factors

using BFAST algorithms. American Journal of Plant

Sciences, 13(4), 517–540. https://doi.org/10.4236/

ajps.2022.134034

Ayhan, B., Kwan, C., Budavari, B., Kwan, L., Lu, Y., Perez,

D., Li, J., Skarlatos, D., & Vlachos, M. (2020). Vegeta-

tion detection using deep learning and conventional

methods. Remote Sensing, 12(15), 2502. https://doi.

org/10.3390/rs12152502

Cadei, A., Mologni, O., Röser, D., Cavalli, R., & Grigolato,

S. (2020). Forwarder productivity in salvage logging

operations in difficult terrain. Forests, 11(3), 341.

https://doi.org/10.3390/f11030341

Cipta, S., Suprihatin, S., Tarigan, S., & Effendi, H. (2019).

Land use classification based on object and pixel using

Landsat 8 OLI in Kendari City, Southeast Sulawesi

Province, Indonesia. IOP Conference Series: Earth and

Environmental Science, 284, 012019. https://doi.org

/10.1088/1755-1315/284/1/012019

Coulter, L., Stow, D., Tsai, Y., Ibanez, N., Shih, H., Kerr,

A., Benza, M., Weeks, J. R., & Mensah, F. (2016).

Classification and assessment of land cover and land

use change in Southern Ghana using dense stacks of

landsat 7 ETM+ imagery. Remote Sensing of Envi-

ronment, 184, 396–409. https://doi.org/10.1016/j.

rse.2016.07.016

Chuvieco, E. (2010). Teledetección ambiental: La observa-

ción de la Tierra desde el espacio. Editorial Ariel.

Díaz, J. (2015). Estudio de índices de vegetación a partir de

imágenes aéreas tomadas desde UAS/RPAS y aplicacio-

nes de estos a la agricultura de precisión [Master’s tesis]

Universidad Complutense de Madrid. España.

Dong, Y., Yin, D., Li, X., Huang, J., Su, W., Li, X., & Wang,

H. (2021). Spatial–temporal evolution of vegetation

NDVI in association with climatic, environmental

and anthropogenic factors in the Loess Plateau, China

during 2000–2015: Quantitative analysis based on

geographical detector model. Remote Sensing, 13(21),

4380. https://doi.org/10.3390/rs13214380

Geng, L., Che, T., Wang, X., & Wang, H. (2019). Detec-

ting spatiotemporal changes in vegetation with the

BFAST model in the Qilian mountain region during

2000–2017. Remote Sensing, 11(2), 103. https://doi.

org/10.3390/rs11020103

Gómez, D. (2019). Variación espacial y temporal de la

vegetación en Baja California Sur, con énfasis en Áreas

Naturales Protegidas [Master’s tesis]. Centro de Inves-

tigaciones Biológicas del Noroeste, México.

Hoekman, D., Kooij, B., Quiñones, M., Vellekoop, S.,

Carolita, I., Budhiman, S., Arief, R., & Roswintiarti,

O. (2020). Wide-area near-real-time monitoring of

tropical forest degradation and deforestation using

Sentinel-1. Remote Sensing, 12(19), 3263. https://doi.

org/10.3390/rs12193263

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

Huang, S., Tang, L., Hupy, J., Wang, Y., & Shao, G. (2021).

A commentary review on the use of normalized diffe-

rence vegetation index (NDVI) in the era of popular

remote sensing. Journal of Forestry Research, 32, 1–6.

https://doi.org/10.1007/s11676-020-01155-1

Isaienkov, K., Yushchuk, M., Khramtsov, V., & Seliverstov,

O. (2020). Deep learning for regular change detection

in Ukrainian forest ecosystem with sentinel-2. IEEE

Journal of Selected Topics in Applied Earth Observa-

tions and Remote Sensing, 14, 364–376.

Le Coq, J.-F., Froger, G., Pesche, D., Legrand, T., & Saenz,

F. (2015). Understanding the governance of the

Payment for Environmental Services Programme

in Costa Rica: A policy process perspective. Ecosys-

tem Services, 16, 253–265. https://doi.org/10.1016/j.

ecoser.2015.10.003

López-Alegría, A., Ríos, M. J., Flamenco-Sandoval, A., &

Farfán-Gutiérrez, M. (2018). Análisis y modelación

espacial de los patrones de deforestación (2005-2025)

en la microcuenca La Unión del municipio de Chia-

pa de Corzo, Chiapas. Sociedad y Ambiente, 18(7),

117–143.

Marx, A., McFarlane, D., & Alzahrani, A. (2017). UAV

data for multi-temporal Landsat analysis of historic

reforestation: a case study in Costa Rica. International

Journal of Remote Sensing, 38(8), 1–18. https://doi.org

/10.1080/01431161.2017.1280637

Meera, G., Parthiban, S., Thummalu, N., & Christy, A.

(2015). NDVI: Vegetation change detection using

remote sensing and Gis-A case study of Vellore

district. Procedia Computer Science, 57, 1199–1210,

https://doi.org/10.1016/j.procs.2015.07.415

Ministerio de Ambiente y Energía. (2015). Estrategia Nacio-

nal REDD+ Costa Rica. MINAE, Costa Rica.

Miomir, J. M., Milanović, M. M., & Vračarević, B. R. (2018).

Comparing NDVI and corine land cover as tools for

improving national forest inventory updates and pre-

venting illegal logging in Serbia. In A. Sebata (Ed.),

Vegetation (pp. 1–22). InTech.

Muñoz, P. (2013). Apuntes de teledetección: Índices de vege-

tación. Centro de Información de Recursos Naturales,

Chile.

Muñoz, E., Zozaya, A., & Lindquist, E. (2020). Satellite

remote sensing of forest degradation using NDFI

and the BFAST algorithm. IEEE Latin America Tran-

sactions, 18(07), 1288–1295. https://doi.org/10.1109/

TLA.2020.9099771

Ngadi, Y., Lebourgeois, V., Laques, A. E., Dieye, M., Bour-

goin, J., & Bégué, A. (2023). BFASTm-L2, an unsu-

pervised LULCC detection based on seasonal change

detection-An application to large-scale land acqui-

sitions in Senegal. International Journal of Applied

Earth Observation and Geoinformation, 121, 103379.

https://doi.org/10.1016/j.jag.2023.103379

Olivares, B., & López-Beltrán, M. (2019). Índice de Vege-

tación Normalizada aplicado al territorio indígena

agrícola de Kashaama, Venezuela. UNED Research

Journal, 11(2), 112–121.

Pérez, N. (2021). Estimación de la deforestación en el

Santuario Histórico Bosque de Pómac y su zona

de amortiguamiento mediante modelos estocásticos

y teledetección [Undergraduate’s tesis]. Universidad

Nacional Mayor de San Marcos, Perú.

Putz, F. E., Baker, T., Griscom, B. W., Gopalakrishna, T.,

Roopsind, A., Umunay, P. M., Zalman, J., Ellis, A.

E., Ruslandi, & Ellis, P. W. (2019). Intact forest in

selective logging landscapes in the tropics. Frontiers

in Forests and Global Change, 2, 30. https://doi.

org/10.3389/ffgc.2019.00030

QGIS.org (2021). QGIS geographic information system.

Open-Source Geospatial Foundation Project. http://

qgis.org

Quispe, M. (2021). Simulación geoespacial de la tasa de

deforestación al 2030 en el distrito de Nueva Requena-

Ucayali [Undergraduate’s tesis]. Universidad Agraria

de la Selva, Perú.

R Core Team. (2021). R: A language and environment

for statistical computing (Software). R Foundation

for Statistical Computing. Vienna, Austria. https://

www.R-project.org/

Rimkus, E., Stonevicius, E., Kilpys, J., Maciulyte, V., &

Valiukas, D. (2017). Drought identification in the

eastern Baltic region using NDVI. Earth System

Dynamics, 8(3), 627–637. https://doi.org/10.5194/

esd-8-627-2017

Roman, M., & Angulo, J. (2013). Panorama socioeconómico

de los cantones de Osa y Golfito: tendencias y desafíos

para el desarrollo sostenible. Stanford Woods Institute

for the Environment.

Rosero-Bixby, L., Maldonado-Ulloa, T., & Bonilla-Carrión,

R. (2002). Bosque y población en la Península de

Osa, Costa Rica. Revista de Biología Tropical, 50(2),

585–598.

Sistema Nacional de Áreas de Conservación. (2015). Car-

tografía base para el Inventario Forestal Nacional de

Costa Rica 2013–2014. SINAC, Costa Rica.

Sistema Nacional de Áreas de Conservación. (2018). Corre-

dor Biológico Amistosa: Plan de Gestión 2018–2027.

SINAC, Costa Rica.

Sun, Y., Ren, H., Zhang, T., Zhang, C., & Qin, Q. (2018).

Crop leaf area index retrieval based on inverted

difference vegetation index and NDVI. IEEE Geos-

cience and Remote Sensing Letters, 15(11), 1662–1666.

https://doi.org/10.1109/lgrs.2018.2856765

Tapia, A. (2011a). Mapa digital de temperaturas promedio

para Costa Rica. Instituto Meteorológico Nacional.

12 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 72: e58329, enero-diciembre 2024 (Publicado Dic. 11, 2024)

Tapia, A. (2011b). Mapa digital de precipitaciones promedio

para Costa Rica. Instituto Meteorológico Nacional.

Tondapu, G., Markert, K., Lindquist, E. J., Wiell, D., Díaz,

A. S. P., Johnson, G., Ashmall, W., Chishtie, F., Ate,

P., Tenneson, K., Patterson, M. S., Ricci, S., Fonta-

narosa, R., & Saah, D. (2018). A SERVIR FAO open

source partnership: Co-development of open source

web technologies using Earth Observation for Land

Cover Mapping. American Geophysical Union, 2018,

IN21B-27

Usman, M., Liedl, R., Shahid, M. A., & Abbas, A. (2015).

Land use/land cover classification and its change

detection using multi-temporal MODIS NDVI data.

Journal of Geographical Sciences, 25(12), 1479–1506.

https://doi.org/10.1007/s11442-015-1247-y

Verbesselt, J., Hyndman, R., Newnham, G., & Culvenor,

D. (2010). Detecting trend and seasonal changes in

satellite image time series. Remote Sensing of Envi-

ronment, 114(1), 106–115. https://doi.org/10.1016/j.

rse.2009.08.014

Wu, Q., Liu, K., Song, C., Wang, J., Ke, L., Ma, R., Zhang, W.,

Pan, H., & Deng, X. (2018). Remote sensing detection

of vegetation and landform damages by coal mining

on the Tibetan Plateau. Sustainability, 10(11), 3851.

https://doi.org/10.3390/su10113851

Yin, H., Pflugmacher, D., Li, A., Li, Z., & Hostert, P. (2018).

Land use and land cover change in inner Mongolia-

understanding the effects of china’s re-vegetation pro-

grams. Remote Sensing of Environment, 204, 918–930.

https://doi.org/10.1016/j.rse.2017.08.030

Zhang, X., Wu, S., Yan, X., & Chen, Z. (2016). A global

classification of vegetation based on NDVI, rainfall

and temperature. International Journal of Climatology,

37(5), 2318–2324. https://doi.org/10.1002/joc.4847

Zhu, Y., Wang, H. & Zhang, A. (2024). Satellite remote

sensing reveals overwhelming recovery of forest from

disturbances in Asia. Atmospheric and Oceanic Scien-

ce Letters, 2024, 100511. https://doi.org/10.1016/j.

aosl.2024.100511