Apuí, Brazil  Land Cover Change Analysis

Abstract

The Amazon rainforest is one of the most rapidly deforested regions in the world, driven by pressures from illegal logging, agricultural expansion, population growth, and fire. The municipality of Apuí, located in the state of Amazonas, Brazil, is among the areas experiencing the fastest rates of land-cover change, largely due to the expansion of pasture land for cattle grazing. This study uses remote sensing data from 2015 and 2025 to quantify land-cover change in Apuí and identify the primary drivers influencing these shifts. Landsat 8-9 imagery obtained from the USGS Earth Explorer platform was classified into six major land-cover categories and evaluated for accuracy. These classified images were then used to construct a contingency table to quantify transitions among classes and reveal dominant patterns of landscape transformation.

The results show a substantial increase in developed land, indicating ongoing population growth and infrastructure expansion. Pasture land also increased significantly, with most new pasture converted from forested areas. Together, these findings demonstrate that agricultural expansion remains a major cause of deforestation in this region.

Overall, the study highlights a clear intensification of human pressures on the landscape in Apuí. By documenting both the scale and underlying drivers of recent land-cover change, this analysis provides insight into the local factors contributing to Amazonian deforestation and underscores the importance of promoting sustainable land-use alternatives that reduce pressure on the remaining forest.

Introduction

Deforestation in the Amazon rainforest remains one of the most pressing environmental challenges worldwide, with the Brazilian Amazon accounting for the greatest share of forest loss across the basin. Major drivers of deforestation include agricultural expansion, cattle grazing, fires, illegal logging, and the broader impacts of climate change (Amazon Conservation, 2025).

This study focuses on Apuí, a municipality in the state of Amazonas, Brazil, which has experienced rapid land-cover change over the past decade. The expansion of pasture for cattle ranching has made Apuí one of the leading deforestation hotspots within the state. In response, local initiatives are promoting a transition from conventional pasture systems to coffee agroforestry, where coffee is intercropped with native tree species as a strategy to curb deforestation and restore ecosystem function (reNature, 2023).

To better understand what is occurring in this region, this study investigates the land-cover change across Apuí over the past decade. Landsat 8-9 satellite imagery is used to classify and quantify changes among major land cover classes, including forest, pasture, developed areas, water, burn scars, and bare earth. Supervised classification methods and accuracy assessments are used to generate reliable land-cover maps for 2015 and 2025. A contingency table is then constructed to evaluate transitions among land-cover classes, providing a statistical basis for identifying dominant patterns of landscape change. To further support interpretation, the results are visualized using a Sankey diagram, a bar plot illustrating total land-cover change by class, and a pie chart illustrating the proportional contributions of each transition. 

Methods

Earth Explorer Data Acquisition:

Two Landsat 8-9 OLI/TIRS C2 L2 images were obtained from Earth Explorer (Figure 1). The first image selected was from August, 2015 (ID: LC08_L2SP_230065_20150803_20200909_02_T1). The second was from August, 2025 (ID: LC09_L2SP_230065_20250822_20250823_02_T1). These dates were chosen due to the relatively low cloud coverage.

Preprocessing in RStudio:

Downloaded .tar archives containing Landsat 8–9 imagery were imported and extracted in RStudio. Each scene was visualized and cropped from the full Landsat footprint to a smaller spatial extent representing our study area (Figure 2). After cropping, the 2015 and 2025 images were exported as new .tif raster files for subsequent processing in ArcGIS Pro. 

Supervised Classification in ArcGIS Pro:

The cropped 2015 and 2025 .tif files were imported into ArcGIS Pro for supervised classification. Six classes were chosen to determine the land cover change in the area: forest, developed, water, pasture, burn scar, and bare earth. Training samples were collected for both years, and classifications were preformed using a Random Forest classifier with 50 trees and a maximum of 1000 samples per class (Figure 3). The training schema created for both years was saved and exported into RStudio to determine accuracy assessments of the classified images. The resulting Random Forest classified image raster datasets were also exported into RStudio to analyze the land cover change.

Statistical Analysis in RStudio:

Post-classification analyses were conducted in RStudio. These steps included accuracy assessment of the classified rasters, stacking the 2015 and 2025 classifications to quantify land-cover transitions, and generating visual outputs illustrating observed changes. Raster areas were converted to hectares (ha) to provide more interpretable results.

Results and Discussion

Accuracy Assessment:

The accuracy assessment of the classified images resulted in an overall accuracy of 98.7% for 2015 and 100% for 2025. The confusion matrix tables and bar graph displaying the accuracy analysis is shown in Figure 4. The misclassification occurred between bare earth and developed land. This was likely caused by both having a similar reflectance patterns. 

Figure 4: Confusion matrix tables and classification accuracy bar graphs for2015 (top row) and 2025 (bottom row).

Across the landscape, the most prominent trend is the expansion of pasture and bare-ground areas accompanied by a substantial decline in forest cover. Much of the land classified as bare ground likely represents recently burned pasture undergoing preparation for regrowth. 

Pasture exhibited the most visually apparent change and the second-largest proportional increase. Its total area expanded from 86,572 ha to 273,379 ha, representing a 216% increase. The majority of this growth resulted from forest conversion: 179,875 ha of forest transitioned to pasture, the single largest area gained by any land-cover class in the analysis. This pattern reflects continued agricultural expansion and the encroachment of cattle ranching into previously forested landscapes.

Developed land experienced the highest proportional increase, growing from 2,344 ha to 15,066 ha, a 543% gain. While some of this increase may reflect misclassification, particularly instances in which bare ground was incorrectly labeled as developed, a substantial portion appears to represent genuine land-use change. Forest contributed 8,251 ha to newly developed areas, and bare earth/burn scar/pasture contributed an additional 6,155 ha, indicating both population growth and the expansion of infrastructure.

Frequent transitions between bare ground, burn scar, and pasture land cover highlight the cyclical nature of pasture management in the region. Local agricultural practices often involve slash-and-burn techniques in which aging pasture is burned to promote new grass growth. Although burning is officially prohibited, it remains widespread and contributes not only to the temporary appearance of bare ground and burn scars but also to long-term forest loss throughout this portion of the Amazon.

Water exhibited relatively minor overall change between the two years. Approximately 184 ha of areas previously classified as water are now identified as forest, while water cover increased by roughly 1,395 ha. New water bodies appear scattered throughout Apuí, as illustrated in Figure 3. Many of these emerging features occur within recently deforested areas, likely reflecting hydrological alterations that promote the formation of standing water bodies. 

Conclusion

The analysis of land-cover change in Apuí from 2015 to 2025 demonstrates that the expansion of pasture and developed land has been a major driver of deforestation in this portion of the Amazon. High classification accuracy from Landsat 8–9 imagery, aside from minor confusion between bare earth and developed surfaces, provided a reliable foundation for interpreting patterns of change across the landscape. The results show a substantial increase in pasture area, driven largely by the conversion of forest, as well as transitions from bare earth and burn scars associated with cyclical ranching practices. Developed land also exhibited a pronounced rise, suggesting population growth and the continued expansion of infrastructure that supports agricultural activity.

Together, these findings highlight the intensification of land-use pressures in Apuí and underscore the need for sustainable alternatives to conventional pasture systems. By capturing the extent and trajectory of recent deforestation, this study provides evidence that can support efforts to promote land-use transitions that reduce forest loss. In particular, the adoption of coffee agroforestry, where coffee is cultivated alongside native tree species, offers a pathway to maintain agricultural productivity while restoring ecosystem function. Encouraging such practices may help slow ongoing deforestation and contribute to more sustainable landscape management in the region.

References

Amazon Conservation Association. (2025). Threats to the Amazon. Amazon Conservation Association. https://www.amazonconservation.org/the-challenge/threats/ 

reNature. (2023). Apuí, Brazil. https://www.renature.co/projects/cafe-apui-idesam-brazil/