Gold mining in Amazonia involves forest removal, soil excavation, and the use of liquid mercury, which together pose a major threat to biodiversity, water quality, forest carbon stocks, and human health. Within the global biodiversity hotspot of Madre de Dios, Peru, gold mining has continued despite numerous 2012 government decrees and enforcement actions against it. Mining is now also thought to have entered federally protected areas, but the rates of miner encroachment are unknown. Here, we utilize high-resolution remote sensing to assess annual changes in gold mining extent from 1999 to 2016 throughout the Madre de Dios region, including the high-diversity Tambopata National Reserve and buffer zone. Regionally, gold mining-related losses of forest averaged 4437 ha yr-1. A temporary downward inflection in the annual growth rate of mining-related forest loss following 2012 government action was followed by a near doubling of the deforestation rate from mining in 2013-2014. The total estimated area of gold mining throughout the region increased about 40% between 2012 and 2016, including in the Tambopata National Reserve.
Our results reveal an urgent need for more socio-environmental effort and law enforcement action to combat illegal gold mining in the Peruvian Amazon.
Figure 1. Progression of forest loss from informal gold mining in the Madre de Dios region from 1999 to 2016. The Tambopata National Reserve is shown to the south, and the Reserve’s buffer zone is shown north of it. The inset map indicates the location of Madre de Dios within Peru with the coloring indicating relative changes in elevation. The locations of large mining landscapes are marked ML-1 (Huepetuhe), ML-2 (Delta-1), and ML-3 (Guacamayo), and major rivers impacted by gold mining are labelled.
Department of Global Ecology, Carnegie Institution for Science, United States
Forests and woodlands are a very important part of the global ecosystem through their provision of ecosystem goods and services. However, conversion to other land uses is one of the biggest threats to their existence. Remote sensing presents opportunities for monitoring such changes over wide and inaccessible areas including those areas that have no field data. In this study, we use the Carnegie Landsat Analysis System-lite (CLASlite) software and Landsat imagery to make the first spatially explicit national estimate deforestation in Swaziland. This was compared with deforestation data derived from the Global Forest Change (GFC) dataset for the period 2000-2014. The CLASlite analysis identified an estimated 46,620ha of forest and woodland lost between 1990 and 2015 resulting in a mean deforestation rate of 1704 ha yr-1. The GFC dataset, on the other hand, indicates a mean deforestation rate 1563 ha yr-1 when excluding forest regrowth. Validation of the results based on multi-year Google Earth and Landsat imagery indicated that both approaches are feasible for monitoring deforestation. The GFC data captured more forest loss within the dense plantation and wattle forests whilst underestimating deforestation within natural forests and woodlands.
Although there are inter-annual variations, the rate of deforestation is generally increasing and widespread in many parts of the country mainly concentrated in the eastern half of the country and a few western parts where agriculture (particularly sugarcane), human settlements and other infrastructure developments are dominant land uses. Acacia and broadleaf savanna are being depleted at higher rates with up to 8.1% of forest area lost since the year 2000. Forest policies and legislation need to be reviewed to respond to the observed trends and patterns with a focus on forest conservation, climate change mitigation and adaptation.
Figure 2. Forest types in Swaziland in 1999.
Payments for ecosystem services (PES) programs depend on consistent environmental monitoring methodologies for measuring, reporting and verification. The case of carbon PES programs is instructive: land cover estimates are crucial for environmental monitoring efforts, but they require a consistent estimation method. Such consistency is however likely to be affected by climatic variability such as that seen during severe droughts. This raises questions as to whether distinct methodologies for land cover monitoring yield the same estimates of land- cover change over time in the presence of climatic variability. This study compares deforestation estimates from four methodologies during a normal year (2008) with those during a period of extreme drought (2010) in the Madre de Dios region of Peru in the Southwestern Amazon. The four methodologies compared are the automated classification with CLASlite 2.2, Tasseled Cap classification with ERDAS Imagine 9.2, Bhattacharya classification with SPRING 5.1, and Spectral Angle Mapper classification with ENVI 4.7. The results show differences in the forest and non-forest estimates derived from using ERDAS and CLASlite compared with those from using the SPRING and ENVI.
These findings have implications for forest monitoring efforts for PES programs such as Reduced Emissions from Deforestation and Forest Degradation (REDD+) in the context of climate change.
Figure 1. Study site location in Madre de Dios within the Landsat 5 path 003/row 068.
In mid-January 2005 a convective squall line traversed 4.5 × 106 km2 of Amazonia from southwest to northeast. As seen in Landsat images, this atypical convective storm left blowdown imprints with diffuse geometry, unlike the fan-shaped wind disturbance of much more frequent east-to-west propagating squall lines. Previous work reported 0.2% of the forest area damaged by this one relatively rare event within one Landsat image and assumed similar disturbance across the entire traverse. We mapped convective wind damage impact to the region in 2005 by identifying large-scale (>4 ha) blowdown imprints in 30 Landsat images. The diffuse-type imprints associated with this single squall line contributed up to 60-72% of total 2005 wind-disturbed area detected across the region, but damage was highly concentrated in central Amazonia. Consequently, the distribution of large wind damage patches in 2005 across Amazonia was very different from long-term average. Regional distribution of wind-driven tree mortality for smaller patch sizes remains unknown.
Figure 1. Method workflow showing (a) 30 Landsat TM satellite image areas outlined in black, spread over the 4.5 × 106 km2 of the mid-January 2005 squall line’s traverse, which is the grey area covering most of Amazonia; (b) delimitation of blowdown imprint vicinities (many small red outlines) within the 15,000 km2 sample of one Landsat image (single black outline); (c) 10 × 10 km search grid and fixed scale of 1:80.000 used in step b; (d) subpixel fractional cover by photosynthetic vegetation—PV, nonphotosynthetic vegetation—NPV, and bare soil, obtained by CLASlite SMA (spectral mixture analysis) from the six Landsat TM bands of each pixel; (e) close-in view of RGB color composite of Landsat bands, showing an area with topographic relief shade; (f) overestimate of wind damage (concentration of red pixels) within this same area of relief shade if using an NPV fraction threshold; (g) no bias in estimate of wind damage in this same area of relief shade when using a PV fraction threshold; and (h) PV threshold and classified wind damaged pixels inside blowdown imprints.
Assessing and monitoring forest degradation under national Monitoring, Verification and Reporting (MRV) systems in developing countries have been difficult to implement due to the lack of adequate technical and operational capacities. This study aims at providing methodological options for monitoring forest degradation in developing countries by using freely available remote sensing, forest inventory and ancillary data. We propose using Canopy Cover to separate, through a time series analysis approach using Landsat Imagery, forest areas with changes over time from sectors that report a “stable condition”. Above ground Biomass and Net Primary Productivity derived from remote sensing data were used to define thresholds for areas considered degraded. The approach was tested in a semi-deciduous tropical forest in the Southeast of Mexico. The results showed that higher rates of forest degradation, 1596 to 2865 ha year-1, occur in areas with high population densities. The results also showed that 43% of the forests of the study area remain with no evident signs of degradation, as determined by the indicators used. The approach and procedures followed allowed for the identification and mapping of the temporal and spatial distribution of forest degradation, based on the indicators selected, and they are expected to serve as the basis for operations of the Reduction of Emissions from Deforestation and Forest Degradation (REDD+) initiative in Mexico and other developing countries, provided appropriate adaptations of the methodology are made to the conditions of the area in turn.
Figure 1. Study area “Menda 1” in the geographical context of the Yucatan Peninsula, Mexico.
Avian species persistence in a forest patch is strongly related to the degree of isolation and size of a forest patch and the vegetation structure within a patch and its matrix are important predictors of bird habitat suitability. A combination of space-borne optical (Landsat), ALOS-PALSAR (radar), and airborne Light Detection and Ranging (LiDAR) data was used for assessing variation in forest structure across forest patches that had undergone different levels of forest degradation in a logged forest-agricultural landscape in Southern Laos. The efficacy of different remote sensing (RS) data sources in distinguishing forest patches that had different seizes, configurations, and vegetation structure was examined. These data were found to be sensitive to the varying levels of degradation of the different patch categories. Additionally, the role of local scale forest structure variables (characterized using the different RS data and patch area) and landscape variables (characterized by distance from different forest patches) in influencing habitat preferences of International Union for Conservation of Nature (IUCN) Red listed birds found in the study area was examined. A machine learning algorithm, MaxEnt, was used in conjunction with these data and field collected geographical locations of the avian species to identify the factors influencing habitat preference of the different bird species and their suitable habitats. Results show that distance from different forest patches played a more important role in influencing habitat suitability for the different avian species than local scale factors related to vegetation structure and health. In addition to distance from forest patches, LiDAR-derived forest structure and Landsat-derived spectral variables were important determinants of avian habitat preference.
The models derived using MaxEnt were used to create an overall habitat suitability map (HSM) which mapped the most suitable habitat patches for sustaining all the avian species. This work also provides insight that retention of forest patches, including degraded and isolated forest patches in addition to large contiguous forest patches, can facilitate bird species retention within tropical agricultural landscapes. It also demonstrates the effective use of RS data in distinguishing between forests that have undergone varying levels of degradation and identifying the habitat preferences of different bird species. Practical conservation management planning endeavors can use such data for both landscape scale monitoring and habitat mapping.
Figure 1. Dongsithouane production forest area located in Savannakhet Province of Laos (With Different Patch categories).
Understanding the trans-boundary deforestation history and patterns in protected areas along the Belize-Guatemala border is of regional and global importance. To assess deforestation history and patterns in our study area along a section of the Belize-Guatemala border, we incorporated multi-temporal deforestation rate analysis and spatial metrics with survey results. This multi-faceted approach provides spatial analysis with relevant insights from local stakeholders to better understand historic deforestation dynamics, spatial characteristics and human perspectives regarding the underlying causes thereof. During the study period 1991–2014, forest cover declined in Belize’s protected areas: Vaca Forest Reserve 97.88%–87.62%, Chiquibul National Park 99.36%–92.12%, Caracol Archeological Reserve 99.47%–78.10% and Colombia River Forest Reserve 89.22%–78.38% respectively. A comparison of deforestation rates and spatial metrics indices indicated that between time periods 1991–1995 and 2012–2014 deforestation and fragmentation increased in protected areas. The major underlying causes, drivers, impacts, and barriers to bi-national collaboration and solutions of deforestation along the Belize-Guatemala border were identified by community leaders and stakeholders. The Mann-Whitney U test identified significant differences between leaders and stakeholders regarding the ranking of challenges faced by management organizations in the Maya Mountain Massif, except for the lack of assessment and quantification of deforestation (LD, SH: 18.67, 23.25, U = 148, p > 0.05).
The survey results indicated that failure to integrate buffer communities, coordinate among managing organizations and establish strong bi-national collaboration has resulted in continued ecological and environmental degradation. The information provided by this research should aid managing organizations in their continued aim to implement effective deforestation mitigation strategies.
Figure 1. Study Site: Belize-Guatemala border.
In Belize, the lack of forest degradation and socioeconomic data results in the inability of forest management organizations to make timely assessments and decisions for sustainable forest resource management. This study uses CLASlite algorithms and social surveys to identify drivers, measure, analyze and map deforestation, and forest degradation that occurred in Toledo’s ecosystems and Protected Areas as a result of the increased anthropogenic activity reported in 2010–2012. The social surveys indicated that land and institutional policy, distance to markets and lack of alternative livelihoods are the main drivers of deforestation and forest degradation. Of importance are the strong significant differences that exist between communities that were less than 2 km from a protected area (CL2K) and communities that were more than 2 km from a protected area (CM2K) regarding property rights (Cramer’s V = 0.562, p < 0.001), selective logging (Cramer’s V = 0.499, p < 0.001) and soil quality (Cramer’s V = 0.434, p < 0.001). The results of the deforestation and forest degradation analysis indicate that in 2009–2011 and 2011–2012 the annual rates of deforestation were 0.75% (2480 ha) and 1.17% (3834 ha) respectively and the annual rates of forest degradation in 2009–2011 and 2011–2012 were 0.09% (307 ha) and 0.33% (1110 ha) respectively. In 2009–2011 only 9.34% of forest loss occurred inside protected areas in comparison to 2011–2012 where 23.97% of forest loss occurred inside protected areas. In 2011–2012 out of the 1110 ha of degradation 30.38% occurred in Lowland broad-leaved wet forest and 19.39% occurred in Sub-montane broad-leaved wet forest. The maps and statistics generated in this study pinpoint in which ecosystem types and protected areas major forest change and forest disturbance occurred.
By utilizing the data generated by this study, Belize’s forest management organizations will be able to efficiently allocate resources to forested areas that are being threatened; thus, more effectively mitigate deforestation and forest degradation of important forest. Spatially explicit forest carbon (C) monitoring aids conservation and climate change mitigation efforts, yet few approaches have been developed specifically for the highly heterogeneous landscapes of oceanic island chains that continue to undergo rapid and extensive forest C change. We developed an approach for rapid mapping of aboveground C density (ACD; units = Mg or metric tons C ha−1) on islands at a spatial resolution of 30 m (0.09 ha) using a combination of cost-effective airborne LiDAR data and full-coverage satellite data. We used the approach to map forest ACD across the main Hawaiian Islands, comparing C stocks within and among islands, in protected and unprotected areas, and among forests dominated by native and invasive species.
Figure 1. Surveyed communities and protected areas.
Remotely sensed data have revealed ongoing reforestation across many tropical landscapes. However, most studies have quantified changes between discrete land cover categories that are difficult to relate to the continuous changes in forest structure that underlie reforestation. Here, we demonstrate how generalized linear models (GLMs) can predict tree height and tree canopy cover from Landsat satellite reflectance in a 109 882 ha tropical agricultural landscape of western Panama. We derived tree canopy cover and tree height from airborne Light Detection and Ranging (LiDAR) data, and related these variables to the fraction of photosynthetic vegetation (PV) in Landsat pixels. We found large gains in predictive accuracy from modeling tree canopy height with a gamma GLM and tree canopy cover with a binomial GLM, relative to modeling these variables using linear regression. Adding social and environmental covariates to our GLMs, including topography and parcel membership (representing different land owners), increased predictive accuracy, resulting in best-fit models with an R2 of 55.68% and RMSE of 23.69% for tree canopy cover, and an R2 of 51.24% and RMSE of 3.42 m for tree height. Finally, we applied the GLMs to predict tree height and tree canopy cover in Landsat images from c. 2000 to 2012, and used results to quantify changes in forest structure during this 12-year period. We found that >60% of pixels in our study area had increased in tree height and tree canopy cover, suggesting widespread forest regrowth. These increases were spatially widespread across the study area, yet subtle, with most pixels increasing <2 m in tree height. Our results suggest ecological and agricultural changes that could be overlooked if measuring land cover change with discrete forest and non-forest categories.
Overall, we show the advantages of linking LiDAR and Landsat data to quantify forest regrowth in an agricultural landscape.
Figure 5. Endmember fractions (photosynthetic vegetation, non-photosynthetic vegetation and soil) derived through spectral unmixing of Landsat 7 data for the Azuero Peninsula in epochs 2000–2001 and 2011–2013. Green color indicates photosynthetic vegetation, red indicates soil and blue indicates non-photosynthetic vegetation.
A quantitative assessment of forest cover change in the Moulouya River watershed (Morocco) was carried out by means of an innovative approach from atmospherically corrected reflectance Landsat images corresponding to 1984 (Landsat 5 Thematic Mapper) and 2013 (Landsat 8 Operational Land Imager). An object-based image analysis (OBIA) was undertaken to classify segmented objects as forested or non-forested within the 2013 Landsat orthomosaic. A Random Forest classifier was applied to a set of training data based on a features vector composed of different types of object features such as vegetation indices, mean spectral values and pixel-based fractional cover derived from probabilistic spectral mixture analysis). The very high spatial resolution image data of Google Earth 2013 were employed to train/validate the Random Forest classifier, ranking the NDVI vegetation index and the corresponding pixel-based percentages of photosynthetic vegetation and bare soil as the most statistically significant object features to extract forested and non-forested areas. Regarding classification accuracy, an overall accuracy of 92.34% was achieved. The previously developed classification scheme was applied to the 1984 Landsat data to extract the forest cover change between 1984 and 2013, showing a slight net increase of 5.3% (ca. 8800 ha) in forested areas for the whole region.
Figure 1. Location and delineation of the Moulouya River watershed upstream of the dam Mohammed V (UTM 30N projection and WGS84 reference system).