The Map Tells the Story
Summary
True color composite from Landsat (1990) providing a baseline view of land cover in the early study period. Limited spatial resolution (30m) offers a broad overview of vegetation, river systems, and human activity.
Landsat (2001) true color imagery showing mid-period conditions. Subtle changes in land cover may indicate early signs of deforestation or settlement expansion.
This Landsat true color image from 2010 captures a continuation of land cover trends. Vegetation fragmentation and development may become more visible in certain areas.
Landsat true color imagery from 2024 showing the latest large-scale landscape conditions. Though still 30m resolution, changes in vegetation density and water bodies are apparent compared to earlier years.
High-resolution (3m) PlanetScope imagery from 2024 offering a detailed view of land use and vegetation. Minor features such as small clearings, roads, and agricultural plots are more visible than in Landsat imagery.
-
The true color images from 1990 and 2024 provides a comprehensive overview of land cover change along the Amazon River at low water. Over time, a gradual expansion of cleared areas and increased human infrastructure becomes visible. By 2024, the true color imagery reveals increasingly fragmented vegetation patterns, particularly near expanding settlements and along riverbanks, where signs of clearing and land disturbance are visually apparent.
-
Urban and vegetation changes were further assessed using NDVI and SWIR difference imagery. SWIR-based differences highlight urban expansion and soil exposure (increased reflectance, red) and potential vegetation regrowth or water expansion (decreased reflectance, blue). Notably, some areas showing decreased reflectance near riverbanks may result from a long-term decline in surface water during low-water periods, possibly combined with regrowth. These were retained for interpretation due to their complex nature.
-
A composite raster was created by stacking NDVI and SWIR change layers using the Composite Bands tool in ArcGIS Pro, enabling pixel-level comparison. The goal was to assess whether areas of vegetation change also showed shifts in surface reflectance. A scatter plot generated from this composite shows a moderate negative correlation (R² = 0.43), suggesting that areas of increased vegetation cover tend to show lower SWIR reflectance, consistent with regrowth or increased moisture retention.
-
To quantify vegetation change, the NDVI difference raster was reclassified into three classes: loss, minimal change, and gain, based on a ±0.2 NDVI threshold. The Tabulate Area tool calculated class areas, which were proportionally corrected to account for background pixels. Final estimates indicate approximately 274.16 km² of forest loss, 32,659.41 km² of unchanged vegetation, and 773.15 km² of forest gain, consistent with matrix-based outputs.
-
While NDVI is a useful vegetation index, it cannot distinguish between types of vegetation or surface changes. For example, NDVI gain could mean regrowth, crops, or invasive cover; NDVI loss could mean clearing or seasonal variation. When used alone, NDVI-based area summaries risk oversimplifying change dynamics. Incorporating SWIR provides crucial context about surface moisture and soil exposure, helping clarify whether change is anthropogenic or seasonal.
-
With this, NDVI and SWIR change rasters were reclassified and combined into a nine-class matrix capturing interactions between vegetation and moisture change. This approach reveals 513.91 km² of forest loss and 137.06 km² of vegetation gain, with varied moisture trends. Stable vegetation dominated, covering over 32,623.61 km². These patterns highlight a mostly stable landscape with zones of regrowth and degradation near human-modified or hydrologically shifting areas.
-
- Consistency Table
-
- EVI
SWIR change detection map (Band 7) showing reflectance differences between 1990 and 2024.
Increased reflectance linked to urban growth
Decreased reflectance from vegetation regrowth or water expansion
SWIR
SWIR (Shortwave Infrared Reflectance) analysis was conducted using Landsat 4–9 surface reflectance imagery from 1990 and 2024. The SWIR values were extracted from Band 6 (SWIR1) for Landsat 8 and 9 and matched accordingly for older sensors and used to evaluate changes in surface moisture and soil exposure. SWIR reflectance increases are typically associated with drier conditions, exposed soils, or built surfaces, while decreases can indicate increased moisture or vegetation cover.
This raster was reclassified into three categories: SWIR Gain (> +0.2), interpreted as drying, exposed soil, or built-up areas, No Significant Change (–0.2 to +0.2), and SWIR Loss (< –0.2), indicating increased moisture or regrowth.
Reclassification allowed the SWIR data to be more easily compared with vegetation indices and was especially useful for identifying discrepancies between canopy greenness and moisture availability across the landscape.
Using SWIR-based difference visualization, urban expansion is shown through diverging color schemes, where positive values (red) indicate an increase in SWIR reflectance concentrated near riverbanks and roadways, indicating substantial urban development, deforestation, or soil exposure, and negative values (blue) show decreased SWIR reflectance, suggesting vegetation regrowth, denser vegetation, or surface water expansion. From 1990 to 2024, there is a consistent growth in red-highlighted regions, particularly near transportation corridors and riverbanks, reflecting infrastructure development and urban encroachment. This can be further assessed with the use of a NDVI composite image.
NDVI
NDVI (Normalized Difference Vegetation Index) was calculated using historical Landsat 4–9 surface reflectance imagery, with scenes selected from both the early 1990s and 2024 to capture long-term vegetation trends. The index was computed using the formula: (NIR – Red) / (NIR + Red), where the NIR band corresponds to Band 5 and the Red band to Band 4 on Landsat 8 and 9 (adjusted accordingly for older sensors). NDVI is a widely used indicator of vegetation greenness and photosynthetic activity. For this study, NDVI values were compared between years to detect areas of decline, stability, or recovery.
To simplify interpretation across the landscape, the resulting NDVI difference raster (2024 minus 1990) was reclassified into three categories: Strong Gain (> +0.4), Moderate Gain (+0.2 to +0.4), Stable/No Change (–0.2 to +0.2), and Loss (< –0.2). This reclassification allowed for consistent comparison between NDVI, SWIR, and RENDVI outputs and provided a clearer spatial framework for identifying areas of concern or improvement.
The NDVI composite image provides a visual representation of vegetation distribution and health across the study area between 1990 and 2024. In this false-color rendering, bright green/yellow areas indicate high NDVI values, typically associated with dense, healthy vegetation such as intact tropical forest. Yellow to dark orange tones reflect progressively lower NDVI values, suggesting open water, degraded vegetation, agricultural land, or bare surfaces. Dark orange or reddish patches likely correspond to urban development or heavily disturbed land. While there is a growth in green-highlighted areas, this is a result of a decline in surface water from 1990 to 2024. These areas were left in the analysis as satellite imagery suggests that there may be a combination of anthropogenic activity and significant forest growth in these areas, reflecting a long-term decline in river water levels.
The spatial pattern reveals a distinct fragmentation of vegetated areas over time, particularly near major access routes and settlements, with NDVI signals weakening near these zones. In contrast, some surrounding regions maintain strong NDVI signatures, indicating pockets of preserved or regenerating forests or vegetation, as well as zones of encroaching vegetation where the river previously existed.
False-color NDVI composite showing vegetation changes from 1990 to 2024. Light green indicates continuing dense, healthy vegetation
Degraded or cleared areas
Urban surfaces or water
Dense, healthy vegetation
Reforestation or regeneration
+NDVI | -SWIR → Regrowth with moistening (Healthy forest recovery)
+NDVI | +SWIR → Vegetation gain with drying (e.g., cropland, invasive)
-NDVI | -SWIR → Vegetation loss but moistening (flood, wetland shift)
-NDVI | +SWIR → Vegetation loss with drying (deforestation/urban)
Scatterplot showing the relationship between NDVI and SWIR changes from 1990 to 2024 in the Manaus region. Each point represents a sample location, with NDVI change (x-axis) compared against SWIR change (y-axis).
Histogram showing the distribution of pixel counts across SWIR-based surface moisture change classes from 1990 to 2024. Most pixels fall into lower-class values, indicating limited change or moistening trends across much of the study area, with smaller peaks suggesting localized drying or surface exposure.
Histogram displaying the distribution of NDVI-based vegetation change classes from 1990 to 2024. The strong central peak reflects minimal vegetation change across the region, while the spread into higher and lower classes highlights zones of vegetation gain and loss, respectively.
SWRI-NDVI
To evaluate the relationship between vegetation changes and surface reflectance, a composite raster was created by stacking the NDVI and SWIR change layers using the Composite Bands tool in ArcGIS Pro. Each input raster represented the difference in NDVI and SWIR values between 1990 and 2024. In the composite, Band 1 represented NDVI change, and Band 2 represented SWIR change. This stacking enabled spatially aligned analysis of the two indices, allowing direct comparison at the pixel level. The goal of this comparison was to examine whether areas of vegetation recovery or loss also exhibited corresponding changes in surface reflectance, particularly soil exposure and moisture conditions, as detected by SWIR. The purpose of this approach was to assess whether changes in vegetation, as detected by NDVI, corresponded with changes in surface moisture and exposure, as indicated by SWIR reflectance. NDVI is a well-established indicator of vegetation health and density, while SWIR is sensitive to surface moisture, soil exposure, and built-up areas. Together, these indices can help differentiate between ecological recovery, deforestation, and other forms of land transformation.
A scatter plot was generated using this composite raster, with NDVI change plotted on the x-axis and SWIR change on the y-axis. The distribution of points revealed a moderate negative correlation (R² = 0.43), indicating that, in general, areas experiencing increased vegetation cover (higher NDVI) also exhibited decreased SWIR reflectance. This pattern is consistent with forest regrowth or densification, where increased canopy cover retains more moisture and reduces surface exposure. Conversely, areas showing NDVI decline, and increased SWIR reflectance likely correspond to deforestation, degradation, or urban expansion, where vegetation is removed and dry, exposed surfaces dominate the landscape.
The histograms illustrate the distribution of pixel frequencies across NDVI- and SWIR-based change classes, offering insight into the overall landscape dynamics from 1990 to 2024. In the NDVI histogram, most pixels are concentrated around mid-range class values, indicating minimal change in vegetation greenness for the majority of the study area. This reflects a broadly stable vegetative cover, consistent with the finding that over 96% of the region experienced no significant NDVI shift. However, the presence of pixels in the higher and lower NDVI class ranges confirms zones of notable vegetation gain (e.g., natural regrowth, reforestation) and loss (e.g., clearing or degradation).
Similarly, the SWIR histogram shows a dominant peak at lower class values, indicating that most areas did not experience strong shifts in surface moisture or exposure. The tail extending toward higher SWIR change values suggests a subset of the landscape became drier or more exposed, likely due to urban expansion, road development, or soil exposure following deforestation. Conversely, the left side of the distribution shows moistening trends, potentially from forest canopy recovery or hydrological changes like flooding or increased vegetation density.
Together, these histograms provide a statistical overview of how vegetation and surface moisture conditions changed across the landscape. When paired with the NDVI–SWIR scatterplot and classification matrix, they reinforce the conclusion that while the Amazon region near Manaus remained largely vegetated, localized zones of transformation, both regrowth and disturbance, are evident and ecologically significant.
This integrated analysis supports the broader land cover classification findings and offers additional validation by linking vegetation trends with surface condition changes. It also underscores the value of combining multiple spectral indicators to capture both biological and physical aspects of landscape transformation over time.
Tabulate Area
To quantify land cover changes over time, the Tabulate Area tool in ArcGIS Pro was used on the reclassified NDVI difference raster. The raster was grouped into three categories representing vegetation loss (ΔNDVI < -0.2), minimal change (-0.2 ≤ ΔNDVI ≤ 0.2), and vegetation gain (ΔNDVI > 0.2). This classification was based on a threshold of ±0.2 in NDVI difference values to isolate only meaningful changes in vegetation cover. The Tabulate Area tool in ArcGIS Pro was then used to calculate the total area occupied by each class within the raster.
The output from Tabulate Area produced results in square meters, which were converted to square kilometers by dividing by 1,000,000. Since the raster had a cell size of 30 meters by 30 meters (900 m² per pixel), and the raster dimensions were 7,611 columns by 6,951 rows, the full raster represents approximately 47,637 km². However, the mapped study area covers approximately 33,707 km². This helps validate the output values.
The table summarizes the results of vegetation change across the study area, based on reclassified NDVI difference values. The majority of the study area, over 96%, showed minimal change in vegetation cover between 1990 and 2024. In contrast, only 0.81% of the area experienced a notable decline in vegetation, while 2.29% showed signs of vegetation recovery or increase. These values indicate relatively stable overall vegetation cover, with localized zones of gain and loss that may reflect land use changes, regrowth, or disturbances.
These results indicate that vegetation gain far outpaced vegetation loss over the study period, with over While NDVI is a valuable indicator of vegetation health and change, using Tabulate Area on NDVI change data alone provides a limited view of landscape transformation. NDVI captures greenness but does not distinguish between different types of vegetation or surface conditions. For example, an increase in NDVI could indicate forest regrowth, crop expansion, or even invasive plant cover, each with very different ecological implications. Similarly, NDVI decline may result from deforestation, seasonal variation, or water expansion. Tabulate Area quantifies the extent of change but cannot differentiate between these underlying causes. Without additional context, such as surface reflectance from SWIR, NDVI-based classifications risk oversimplifying complex land cover transitions, reducing the accuracy of land use interpretations and the ability to detect anthropogenic impacts like urbanization or soil exposure.
Vegetation cover change based on NDVI thresholds (±0.2) from reclassified Landsat data and Tabulate Area outputs.
Vegetation loss (0.81%)
Vegetation gain (2.29%)
No change (96.89%)
Proportional bar illustrating NDVI-based vegetation change (1990–2024).Each section represents a percentage of the total study area.
NDVI–SWIR classification matrix showing land cover change categories based on combined vegetation (NDVI) and moisture (SWIR) trends. Class codes represent unique combinations of vegetation loss, gain, or stability with moisture increases, decreases, or no change.
NDVI–SWIR change classification (1990–2024), highlighting land cover dynamics such as forest loss with drying (red), regrowth with moisture increase (green), and stable vegetation (gray–purple).
NDVI–SWIR change classification summary showing the area of each land change class from 1990 to 2024. Each class represents a unique combination of vegetation (NDVI) and surface moisture (SWIR) trends, categorized as forest loss, stable vegetation, or forest gain. Areas are reported in both square meters and square kilometers.
9-Class Matrix
To gain a more comprehensive understanding of land cover changes between 1990 and 2024, NDVI and SWIR difference rasters were reclassified and integrated into a single nine-class matrix. This method was employed to evaluate both vegetation dynamics and associated changes in surface reflectance, allowing for a more holistic interpretation of environmental transformation. NDVI, a widely used index for detecting live green vegetation, was used to assess changes in vegetation health and density. In contrast, SWIR reflectance is sensitive to surface moisture, bare soil exposure, and impervious surfaces, making it a useful indicator of deforestation, drying trends, or urban development. By combining these indices, the matrix allows for simultaneous evaluation of biological (vegetative) and physical (surface) changes across the landscape.
Each pixel in the composite raster was classified into one of nine change classes based on the direction and magnitude of NDVI and SWIR differences. This dual classification revealed both consistent and divergent trends between vegetation and surface conditions, making it possible to identify areas of forest loss, regrowth, degradation, and potential land use conversion.
To determine the spatial extent of each NDVI–SWIR change class, pixel counts were extracted from the attribute table of the combined matrix raster. Each pixel represents a 30 × 30 meter area (900 m² or 0.0009 km²), corresponding to the resolution of the Landsat imagery used in this analysis. The total number of pixels in each class was multiplied by 0.0009 to convert pixel counts into square kilometers. For example, if a class contained 50,000 pixels, the area would be calculated as 50,000 × 0.0009 = 45 km². This method allowed for accurate quantification of the spatial coverage of land cover changes across the study area.
The resulting classified image highlights several key patterns. Areas showing vegetation loss with increased SWIR reflectance (Class 13), typically associated with deforestation, urban expansion, or soil exposure, accounted for approximately 63.51 km². Vegetation loss with decreased SWIR (Class 11), which may reflect localized flooding or persistently moist surfaces, covered 208.33 km², while only 2.31 km² (Class 12) showed vegetation loss without notable surface reflectance change.
The most extensive category observed was stable NDVI with decreased SWIR reflectance (Class 21), covering 30,636.23 km². This class likely represents areas of stable vegetation with increasing moisture availability or canopy densification. Additional stable classes included 897.90 km² with no change in SWIR (Class 22), and 1,125.28 km² where stable vegetation coincided with increasing SWIR reflectance (Class 23), possibly due to drying trends or minor land disturbances.
Evidence of forest recovery was most notable in areas classified as vegetation gain with decreased SWIR (Class 31), which spanned 456.05 km² and is consistent with reforestation or natural regrowth in moist conditions. Smaller areas showed vegetation gain with no change in SWIR (174.29 km², Class 32) or increased SWIR (142.81 km², Class 33), the latter potentially indicating conversion to cropland, pasture, or low-moisture vegetation.
Overall, the NDVI–SWIR change matrix provided a detailed classification of land cover dynamics that could not be captured using a single index. By distinguishing between ecologically significant regrowth and surface-level vegetation change, this method enhances the understanding of forest fragmentation, land degradation, and the spatial footprint of human activity in the Amazon region over time.
RENDVI
The RENDVI (Red Edge Normalized Difference Vegetation Index) image was produced using PlanetScope surface reflectance imagery acquired in 2024. The index was calculated using the raster function expression (NIR – RedEdge) / (NIR + RedEdge), where the Near-Infrared (NIR) band corresponds to Band 4 (planet2c4) and the Red Edge band to Band 5 (planet2c5). These bands were selected due to RENDVI’s sensitivity to subtle changes in canopy vigor and chlorophyll content, making it especially useful in areas with moderate to dense vegetation. Prior to calculation, imagery was visually checked for cloud cover and quality. Although the PlanetScope bands were not scaled to top-of-atmosphere reflectance for this version, the unscaled RENDVI was sufficient to highlight patterns of vegetation stress, regrowth, or degradation across the study area. The final RENDVI raster was visualized using a green-to-red gradient to distinguish areas of higher vegetation vigor (green) from degraded or sparse cover (orange to red).
The RENDVI image provides a detailed view of vegetation condition across the study area, with pixel values scaled to highlight variation in canopy health. Areas in bright green represent dense, vigorous vegetation, while orange to red zones indicate low RENDVI values, suggestive of vegetation stress, sparse canopy, or degradation. Notably, major river channels and adjacent floodplains exhibit lower RENDVI values, potentially reflecting exposed soils or early successional vegetation following seasonal inundation. In contrast, intact forest regions further from disturbance corridors maintain consistently high RENDVI values, consistent with healthy, mature canopy cover. Visible patches of degradation can be observed near road networks and settlement edges, aligning with known anthropogenic influence zones. This image effectively complements NDVI and SWIR analyses by offering a more structure-sensitive index that enhances detection of subtle canopy changes, particularly in transitional or regrowth areas.
EVI
The EVI (Enhanced Vegetation Index) image was derived using PlanetScope surface reflectance imagery from 2024. The EVI was calculated using the raster function expression:
2.5 × ((NIR – Red) / (NIR + 6 × Red – 7.5 × Blue + 1)),
where the NIR band is Band 4 (planet2c4), the Red band is Band 3 (planet2c3), and the Blue band is Band 1 (planet2c1). This formulation enhances sensitivity to canopy structure and minimizes atmospheric effects, making it more effective than NDVI in areas of dense vegetation or where soil background influences may be present. For this analysis, raw digital number (DN) values were used without scaling to reflectance due to time constraints, which may slightly limit the accuracy of the absolute EVI values but still allows for relative comparisons across sites. The final EVI layer was rendered using a color ramp emphasizing vegetation density and health, helping to interpret canopy condition as a tie-breaker when NDVI, SWIR, and RENDVI presented mixed signals.
The resulting EVI values ranged from approximately –0.07 to 0.15, with most forested regions displaying positive values between 0.1 and 0.15, indicating moderate to high vegetation activity. Areas with lower EVI values, particularly those near rivers, roads, and cleared patches, corresponded to vegetation stress, sparse canopy cover, or exposed soil.
Visually, the EVI output revealed clear spatial distinctions in vegetation structure such as dense, mature forest areas showed consistent high EVI values (green in the output raster), Recently cleared zones or low-canopy agricultural plots exhibited yellow to red tones, indicating lower EVI values and reduced vegetative, and Riparian corridors and floodplain zones demonstrated slightly depressed EVI, possibly due to shadowing, water reflection, or early regrowth.
It is important to note that the vertical banding visible across the EVI image is the result of merging multiple PlanetScope scenes captured on different dates and under varying atmospheric conditions. These individual images were stitched together to generate continuous coverage of the study area; however, differences in acquisition times, lighting, and sensor calibration between scenes introduced visible striping artifacts. Ideally, radiometric normalization or histogram matching would be applied to minimize these inconsistencies. Due to the time constraints of this project and the rapid processing timeline, such corrections were not implemented. While the banding does not prevent interpretation of broad vegetation patterns, it may influence pixel-level comparisons and should be considered when assessing fine-scale variation.
These patterns aligned well with NDVI and field observations but provided enhanced contrast in zones of intermediate vegetation, particularly around disturbed or recovering areas. EVI’s reduced sensitivity to atmospheric and canopy saturation effects made it useful for differentiating regrowth stages and detecting subtle stress not visible in NDVI alone.
Consistency Table
To evaluate the consistency between vegetation conditions detected by high-resolution PlanetScope imagery and long-term vegetation trends derived from Landsat-based NDVI and SWIR indices, a site-level comparison was conducted using eight field validation points (table 3). Each point was evaluated based on four metrics: PlanetScope RENDVI value from 2024, interpreted as a proxy for current canopy density and vigor; NDVI data from 2024, indicating greenness trend; SWIR data from 2024, indicating surface moisture trend; and PlanetScope EVI data from 2024, used as a tie-breaker to clarify vegetation condition in cases of conflicting signals, particularly regarding canopy structure and stress.
Agreement was determined by whether the RENDVI value at a site supported the direction of change observed in NDVI and SWIR values, with EVI included as a diagnostic tie-breaker. Sites showing high RENDVI alongside positive NDVI trends and decreasing SWIR, indicative of healthy, moist vegetation, were categorized as in full agreement. Similarly, sites with low RENDVI, decreasing NDVI, and increasing SWIR, indicative of degradation or drying, were also considered fully aligned. Sites where the trends partially aligned, such as high RENDVI but mixed NDVI or SWIR responses, were categorized as partial agreement. Clear mismatches were counted as disagreements.
Out of the eight evaluated locations, only two sites, Water Buffalo Farm and Enchanted Lake, exhibited full agreement among NDVI, SWIR, and RENDVI. Both showed NDVI decline accompanied by SWIR reduction, suggesting vegetation stress or canopy drying, and were supported by negative RENDVI values, indicating low canopy vigor.
The remaining six locations demonstrated either partial agreement or clear disagreement between indices. In cases like Plantation and Passionfruit Vines, NDVI and SWIR trends suggested regrowth under dry conditions (NDVI gain + SWIR gain), yet RENDVI indicated degradation, possibly reflecting sparse canopy or dry regrowth. Sites such as New Plantation, Regrowth, and Burnt Manioc had conflicting signals, with NDVI-SWIR combinations hinting at vegetation stress or recovery while RENDVI values were either slightly positive or strongly negative.
To resolve these inconsistencies, EVI was used as a tie-breaker.
The comparison between EVI and RENDVI revealed a nuanced relationship, with both indices providing complementary yet occasionally conflicting insights into vegetation condition. In general, RENDVI appeared more responsive to canopy structure and density, while EVI was more sensitive to greening patterns and subtle vegetation changes, making it particularly useful as a tiebreaker in ambiguous cases.
At locations such as Plantation, New Plantation, and Regrowth, RENDVI suggested sparse canopy or degradation, while EVI values pointed to some degree of regrowth or early recovery. This divergence indicates that EVI may be capturing initial photosynthetic activity not yet resulting in significant structural canopy changes detectable by RENDVI.
However, at Passionfruit Vines and Resort Fire, EVI values were near zero or negative, more closely aligning with RENDVI and supporting the notion of limited or stressed regrowth. This suggests that in certain degraded or sparsely recovering environments, RENDVI and EVI together may provide stronger validation of canopy condition than NDVI or SWIR alone. Overall, while RENDVI served as a robust indicator of canopy integrity, EVI offered valuable context for interpreting early regrowth or mixed-condition landscapes where traditional indices diverged.
