论文信息
论文解读
AI 辅助整理- 方法
- 用Sentinel-1雷达、PlanetScope光学影像和卫星嵌入数据集,在G
- 发现
- 2023–2024年矿区扩张显著,增幅109%–1445%;卫星嵌入数据集比光学影像更可靠高效。
- 机会点
- 可将卫星嵌入数据集与雷达时序变化检测结合,构建可迁移的无管制采矿及毁林监测框架。
论文摘要
由于人类活动的扩张,尤其是无监管的金矿开采,亚马逊雨林的退化正在加剧。过去十年间,全球金价上涨以及影响采矿业的政策与监管相关事态的变化,进一步加剧了这些压力。鉴于该议题的敏感性以及对透明和可复现信息的需求,本研究评估了遥感数据集在检测无监管采矿及探究采矿扩张时空动态方面的适用性,包括Sentinel-1(S-1)合成孔径雷达数据、PlanetScope(PS)光学影像以及卫星嵌入数据集V1(SED)。数据集主要在Google Earth Engine中处理,并针对各数据集采用相应的方法。对SED和PS影像采用了监督式定量分类方法,覆盖2017年至2024年期间。对S-1数据则实施了序贯变化检测(SCD)方法。分析聚焦于厄瓜多尔东部三个已有无监管活动报道的采矿热点区域。结果表明,所有研究区域内采矿范围及相关森林砍伐显著增加,尤其在2023年和2024年期间,各研究区相对于研究起始年2017年的采掘扩张增幅在109%至1445%之间。分类结果比较表明,持续云覆盖和时间不一致性限制了光学PS数据的有效性,而SED数据集为年度评估提供了可靠且高效的替代方案,且预处理需求极低。SCD分析揭示了详细的扩张动态,表明采矿通常沿主要河流起始,并逐步向支流及周边森林区域扩展。这种多方法途径进一步实现了结果的交叉验证,其结果与记录类似空间格局和趋势的独立报告相一致。无监管采矿造成的严重环境后果及对当地社区的威胁,凸显了建立系统化、可移植的遥感监测框架的重要性,以支持环境保护,并为环境治理和决策提供及时、可获取的报告。
全文
Earth ObservationEarth Obs., 1, 177–194, 2026
https://doi.org/10.5194/eo-1-177-2026
© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.
Research article
Remote sensing capabilities of detecting spatio-temporal
dynamics in unregulated gold mining hotspots in Ecuador
Inga Lammers1, Christian Geiß 2,3, Jose Jara-Alvear4, and Valerie Graw1
1Institute of Geography, Ruhr University Bochum (RUB), Bochum, Germany
2German Remote Sensing Data Center (DFD), German Aerospace Center (DLR), Wessling, Germany
3Department of Geography, University of Bonn, Bonn, Germany
4Energy Sciences Research Group (CIENER), Universidad del Azuay (UDA), Cuenca, Ecuador
Correspondence: Inga Lammers (inga.lammers@rub.de)
Received: 1 April 2026 – Discussion started: 10 April 2026
Revised: 22 October 2026 – Accepted: 28 October 2026 – Published: 16 November 2026
Abstract. Degradation of the Amazon rainforest is increas-
ing by expanding human activities, especially unregulated
gold mining. These pressures have intensified over the past
decade due to rising global gold prices and changes in polit-
ical and regulatory-related developments affecting the min-
ing sector. Given the sensitivity of the topic and the need for
transparent and reproducible information, this study assesses
the suitability of remote sensing datasets, including Sentinel-
1 (S-1) Synthetic Aperture Radar data, PlanetScope (PS) op-
tical imagery, as well as the Satellite Embedding Dataset V1
(SED), for detecting unregulated mining and investigating
the spatio-temporal dynamics of mining expansion. Datasets
are processed mainly in Google Earth Engine with dataset-
specific methodologies applied. Supervised quantitative clas-
sification approaches were used for the SED and PS im-
agery, covering the period from 2017 to 2024. For S-1 data,
a Sequential Change Detection (SCD) approach was imple-
mented.
The analysis focuses on three mining hotspots in eastern
Ecuador, where unregulated activities have been reported.
The results show a significant increase in the extent of min-
ing and associated deforestation across all study areas, par-
ticularly during 2023 and 2024, with increase of extractivism
ranging from 109 % to 1445 % across different AOIs com-
pared to the initial year of the study, 2017. Comparison
of classification results indicates that persistent cloud cover
and temporal inconsistencies limit the effectiveness of op-
tical PS data, whereas the SED dataset provides a reliable
and efficient alternative for annual assessments with mini-
mal preprocessing requirements. The SCD analysis revealed
detailed expansion dynamics, demonstrating that mining typ-
ically initiates along major rivers and progressively expands
toward tributaries and surrounding forest areas. This multi-
method approach further enables cross-validation of results,
which are consistent with independent reports documenting
similar spatial patterns and trends. The severe environmen-
tal consequences of unregulated mining and threats to lo-
cal communities emphasize the importance of systematic and
transferable remote sensing-based monitoring frameworks to
support environmental protection and enable timely, acces-
sible reporting for environmental governance and decision-
making.
1 Introduction
Although the Amazon rainforest covers only 0.5 % of the
Earth’s surface, it hosts more than 10 % of all plant and verte-
brate species, making it the most species-rich subcontinental-
scale ecosystem globally (Albert et al., 2023; Guayasamin
et al., 2021; Roy et al., 2018). It is also a critical compo-
nent of the Earth system, contributing around 16 % of global
terrestrial photosynthetic productivity and regulating major
carbon and water cycles. Despite this importance, the Ama-
zon is increasingly degraded by expanding human activi-
ties, including land-use change, hydrological alteration, and
resource extractivism. These pressures contribute to an ap-
proaching ecological tipping point, with potentially severe
consequences for biodiversity and climate regulation (Albert
et al., 2023; Flores et al., 2024; Lenton et al., 2025).
Published by Copernicus Publications on behalf of the European Geosciences Union.
178 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
In Ecuador, rich alluvial gold deposits have driven the ex-
pansion of mining activities. Policy changes, including the
opening of ∼ 13 % of national territory to mining explo-
ration in 2016, combined with rising gold prices and socio-
economic pressures, have accelerated this development (Roy
et al., 2018; Capparelli et al., 2021; Gautam et al., 2022). A
substantial share of this expansion is linked to artisanal and
small-scale gold mining (ASGM), which primarily but not
exclusively encompasses unregulated activities. This type of
mining is typically characterized by low capital investment,
labor-intensive methods, and limited regulatory oversight
(Bank et al., 2014; Protschky and Morgan, 2021). ASGM
is associated with severe environmental and health impacts.
Mining-driven deforestation, river sedimentation, and pollu-
tion from toxic elements such as mercury have increased sig-
nificantly, with deforestation linked to mining rising sharply
in recent years (Borja et al., 2023; Mena-Quintana et al.,
2025). Given the dynamic and often short-term nature of
mining activities, systematic monitoring is essential. It also
provides technical support for decision making by govern-
mental departments (Götze et al., 2016). Remote sensing
(RS) has been applied to monitor mining sites in various
ways. Often land use/cover changes have been identified us-
ing RS data. Further time series data has been used to mon-
itor changes related to ASGM around the world. Optical
datasets are commonly used to monitor the extent and associ-
ated changes of mining sites, often with the help of spectral
indices (e.g. Isidro et al., 2017; Lobo et al., 2018; Adamek
et al., 2021). Synthetic Aperture Radar (SAR) technology is
useful due to their independence of weather conditions and
for monitoring in areas with more frequent cloud cover (e.g.
Forkuor et al., 2020). For underground mining, SAR data can
be used to observe land subsidence associated with mines
by applying the Differential Interferometry SAR (DInSAR)
technique to detect and monitor surface deformations (e.g.
Sellers et al., 2022). Hyperspectral RS otherwise can be used
to monitor vegetation health around mining sites (e.g. Zhang
et al., 2012).
Due to the sensitivity of unregulated mining activities,
this study relies exclusively on multi-sensor RS data, in ad-
dition the analysis is based on freely or widely accessible
datasets, including Sentinel-1 (S-1), PlanetScope (PS), and
Google’s geospatial data repositories, and is implemented
within Google Earth Engine (GEE), enabling a transparent,
reproducible and transferable workflow. Given the rapid ex-
pansion of unregulated mining and the limited availability of
in-situ observations, RS-based monitoring is essential for as-
sessing associated environmental impacts such as deforesta-
tion and water pollution. Despite the increasing availabil-
ity of RS data, there remains a lack of systematic evalua-
tion of how different datasets perform in detecting and mon-
itoring the spatio-temporal evolution of unregulated mining
activities. Existing studies often rely on a single sensor or
approach, limiting the ability to assess trade-offs between
spatial resolution, temporal frequency and data accessibility.
In particular, the question of which datasets are most suit-
able for specific monitoring objectives, and how their us-
ability varies with user expertise, remains insufficiently ad-
dressed. Key parts of the study include (1) the identification
and assessment of suitable RS-based datasets to detect un-
regulated gold mining activity and (2) to detect and quan-
tify spatio-temporal patterns of unregulated mining in satel-
lite time series data. The study focuses on regions in the
Ecuadorian Amazon where mining activities are either show-
ing signs of non-compliance practices or occur outside for-
mal concessions, contributing to improved understanding of
spatio-temporal unregulated mining dynamics in areas where
systematic information remains limited. By linking dataset
characteristics to specific monitoring objectives and levels of
technical complexity, this work seeks to provide a structured
basis for selecting appropriate RS data in the context of mon-
itoring unregulated mining.
2 Study area and characteristics of mining sites
Ecuador is located in northwestern South America. Span-
ning 283 560 km2, it includes four distinct geographical re-
gions: the coastal lowlands (La Costa), Andean highlands
(La Sierra), Amazon rainforest (El Oriente) and the Galá-
pagos Islands (La Región Insular). Ecuador has a population
of∼ 18.3 million, of whom∼ 750 000, including indigenous
populations, live in the Amazon (UNFPA, 2025; Villacís and
Carrillo, 2012). Mining plays an increasingly important role,
with 7813 concessions covering 22 812 km 2 (9.2 % of the
mainland) as of 2021 (Peck et al., 2024). Distinguishing be-
tween regulated and unregulated mining is often challeng-
ing, mining concession boundaries are not always transpar-
ent due to political and administrative uncertainties and con-
cessions may be granted or modified over time. Furthermore,
mining activities within officially designated concessions are
not necessarily conducted in compliance with environmen-
tal or legal regulations and may still involve prohibited prac-
tices, such as the use of mercury. Consequently, this study
focuses on ASGM activities that result in substantial envi-
ronmental degradation, irrespective of whether they occur
inside or outside officially designated mining concessions.
In Ecuador, ASGM is predominantly alluvial and concen-
trated along river systems. Operations typically involve sedi-
ment excavation and mercury-based amalgamation, resulting
in deforestation, soil erosion, river modification, and severe
contamination of water and soils (Navas et al., 2020; Mora-
Silva and Coronel-Espinoza, 2021). These disturbances are
identifiable in satellite imagery through characteristic fea-
tures such as bare soil scars, water-filled mining pits and
changes in river color.
This study focuses on three mining-affected areas in the
Amazon River Basin in Ecuador. They are located in the
provinces of Napo, Orellana, and Zamora Chinchipe and are
further named Punino (1), South Napo (2), and Shaime (3)
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 179
Figure 1. Location of AOIs, Sub-AOIs and mining concessions within Ecuador. Sub-AOIs are color-coded and numbered for easier discrim-
ination in the results. Basemap and map service sources: Earthstar Geographics; Esri; TomTom; FAO; NOAA; USGS; © OpenStreetMap
contributors and the GIS User Community; Esri, NASA, NGA, USGS.
(Fig. 1). Areas of interest (AOIs) were selected based on doc-
umented expansion of unregulated mining, primarily using
reports from the Monitoring of the Andes Amazon Program
(MAAP) and the Amazon Mining Watch (AMW). The se-
lected AOIs are further subdivided into smaller units (Sub-
AOIs) also based on prior knowledge of known mining lo-
cations, to enable detailed analysis of spatio-temporal min-
ing dynamics (compare Fig. 1). They cover a total area of
121.32 km2.
With an area of roughly 89 km 2, Punino (1) is the largest
AOI. Located in northern Ecuador, it lies approximately
50 km south of the Colombian border and extends mainly
across Orellana Province, with a smaller portion extending
into Napo Province. The Rio Punino runs through the center
of the AOI, with tributaries such as the Rio Biguno and the
Rio Acorna flowing into it. There are no major settlements
within the AOI. The nearest city, Puerto Francisco de Orel-
lana, is located approximately 20 km away. Since the early
2020s, there has been a dramatic increase in mining activ-
ity in this area and rising tensions between miners and local
residents have been reported (Plan V , 2024). A few mining
concessions have been granted in this area, most of which
are located close to Rio Punino.
With a combined size of 14.13 km 2, Shaime (2) lies in
the province of Zamora Chinchipe in southern Ecuador. The
Rio Shaime and the Rio Nangaritza flow northwards through
the region, while the Quebrada Shamataka flows in from the
west. Smaller settlements located along the rivers are primar-
ily inhabited by indigenous Shuar people. As in other areas,
gold mining provides a source of income for local workers
and external suppliers of equipment and machinery. Most of
this AOI is covered by mining concessions, with only a few
small areas in the south that are not. Reportedly, tensions
have arisen between miners and protestors, in a manner sim-
ilar to that seen in Punino (Puertas, 2018; Beltrán, 2022).
The South Napo (3) AOI covers 18.19 km2 and is charac-
terized by two major rivers: the Rio Jatunyacu, which flows
west to east across the northern part of the area and the Rio
Anzu, which runs northward from the south. The complete
AOI is covered by multiple mining concessions. Although
the region is sparsely populated overall, it has the highest set-
tlement density and anthropogenic modification among the
three AOIs. In the southern part of the AOI lies the settle-
ment of Carlos Julio Arosemena Tola.
3 Data and methods
In this study, different methodologies and datasets are em-
ployed to detect and classify mining sites and to document
their dynamics, depending on the specific characteristics of
each dataset. Two main aspects are investigated: quantitative
area classification and interannual spatial dynamics. To ad-
dress these aspects, the study integrates three RS datasets:
the Google Satellite Embedding V1 dataset (further: SED),
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
180 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
Figure 2. Methodological workflow. Input datasets include PlanetScope imagery, Google Satellite Embedding V1, Sentinel-1 SAR data and
the JRC Global Surface Water dataset. Classification maps and change detection outputs represent the authors’ own processing and analysis.
Data sources: PlanetScope (Planet Labs); Sentinel-1 (Copernicus); Satellite Embedding V1 (Google & Google DeepMind); JRC Global
Surface Water (Pekel et al., 2016). Classification and visualizations: authors’ own processing.
PS and S-1 data. All three datasets differ in their temporal
and spatial resolution. Figure 2 illustrates how each dataset
is used and outlines the corresponding methodological work-
flow. Although each dataset provides a distinct approach to
monitoring mining sites, the combination of these datasets
offers a comprehensive, multi-temporal perspective, captur-
ing both long-term trends and rapid dynamics. This allows
for a detailed characterization of unregulated mining activi-
ties.
3.1 Quantitative area classification
3.1.1 Annual quantitative classification using the Google
Satellite Embedding dataset
The SED provides an analysis-ready dataset, with a 10 m spa-
tial resolution. It represents annual surface dynamics as 64-
dimensional embedding vectors, which are derived from the
AlphaEarth Foundations Model and consist of multi-sensor
data (optical, SAR, LiDAR, and auxiliary sources). Its de-
sign removes the need for preprocessing and mitigates data
gaps in tropical regions, making it well suited for large-scale
analyses within GEE (Brown et al., 2025).
To generate an annual overview of mining-affected areas,
a supervised classification approach was applied, as unsu-
pervised clustering proved unreliable in AOIs with emerg-
ing mining activity during testing. The embeddings support
low-shot learning, meaning that even a small number of la-
beled samples (i.e. tens to hundreds of samples) can yield
high-quality classifications. For each Sub-AOI, 10 training
samples for both mining and non-mining areas were manu-
ally collected, which was sufficient for stable performance.
Classification was performed using a k-nearest neighbors
(kNN) algorithm within GEE’s Scalable Machine Learning
for Earth Engine (SMILE) framework. Permanent water bod-
ies were masked using the Global Surface Water dataset by
Pekel et al. (2016), which maps long-term surface water oc-
currence from multi-decadal Landsat observations. Despite
partial underrepresentation in densely vegetated areas, it was
retained for consistency. Training was guided by PS imagery
from August 2024 to ensure accurate identification of active
mining sites. The trained classifier was then applied to the
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 181
full SED time series (2017–2024) to map the spatio-temporal
evolution of mining activity across all AOIs.
3.1.2 Intermittent quantitative classification using
PlanetScope data
PS optical imagery was used to assess mining expansion over
multi-year timescales using high-resolution observations ac-
quired throughout each year. Sentinel-2 was initially consid-
ered, but persistent cloud cover and a lower revisit rate lim-
ited its usability. In contrast, PS provides near-daily global
coverage at 3 m spatial resolution, which results in more us-
able scenes for the analysis (Roy et al., 2021). Data were ac-
cessed via Planet’s Education and Research Program, down-
loaded from Planet Explorer, and processed in GEE. Only
images with full AOI coverage and < 1 % cloud cover were
retained and manually inspected, resulting in 431 usable
scenes over all AOIs, with improved availability after 2020
(see Fig. 3).
PS imagery was classified in GEE using a Random For-
est (RF) classifier within the SMILE framework. RF is an
ensemble learning method that constructs multiple decision
trees from bootstrapped samples and aggregates their pre-
dictions, resulting in robust and noise-resistant classification
(Breiman, 2001). RF has been widely used for deforestation
mapping applications (e.g. Grinand et al., 2013; Phan et al.,
2020).
For better results, in addition to RGB and NIR bands, two
spectral indices were included. The Normalized Difference
Vegetation Index (NDVI) (see Eq. 1) exploits the contrast be-
tween red and near-infrared reflectance of vegetation and was
used to distinguish vegetated areas from bare or disturbed
mining surfaces (Rouse et al., 1974).
NDVI= (NIR− Red)(NIR+ Red) (1)
The Normalized Difference Water Index (NDWI) (see Eq. 2)
enhances open water features using the green and NIR bands
and was used to identify water bodies and mining-related
features such as sediment ponds and disturbed river margins
(McFeeters, 1996).
NDWI= (Green− NIR)(Green+ NIR) (2)
These variables were combined into a multi-layer input stack
to provide more stable and accurate results.
Permanent water bodies were masked using the JRC
Global Surface Water dataset, despite partial underrepresen-
tation in densely vegetated river systems. Two classes (min-
ing and non-mining) were defined and training samples were
manually collected for each image. Sample sizes were ad-
justed according to scene complexity, considering the relative
proportions of forest and mining areas and the heterogeneity
of mining features. To also account for variability in image
characteristics and quality over time, three reference images
(from 2018, 2020 and 2024) were selected through trial and
error and used for training. This improved the stability of the
classification across the time series. The dataset was split into
training and validation subsets with a ratio of 80: 20 and the
RF model was applied to all images from 2017 to 2024 to
derive subregional mining extents and temporal dynamics.
3.2 Interannual spatial dynamics with Sentinel-1 data
S-1 SAR imagery was incorporated within a SCD frame-
work to capture short-term mining dynamics. Unlike opti-
cal data, SAR provides consistent observations independent
of weather and illumination conditions, making it particu-
larly suitable for persistently cloud-covered regions. Level-
1 Ground Range Detected (GRD) products from S-1A with
30 m spatial resolution and 12 d revisit time were used
(Geudtner et al., 2014). The SCD approach enables the de-
tection of rapid changes, such as sudden expansion or re-
location of mining activity, which are often not resolved in
optical time series. It has been successfully applied in sim-
ilar contexts to monitor dynamic landscape changes under
frequent cloud cover (e.g. Asokan and Anitha, 2019; Graw
et al., 2022). In this study, SCD was implemented follow-
ing Conradsen et al. (2016) and Canty et al. (2020), using
the EESAR Docker container provided by Canty (2021) in
combination with processing in GEE. The algorithm evalu-
ates multi-temporal SAR observations on a per-pixel basis,
deriving statistical test values and corresponding p-values
to identify significant changes across the time series. A sig-
nificance threshold of 0.01 was applied to generate change
maps. Among the available outputs, the cmap layer was se-
lected because it indicates the timing of the most recent de-
tected change per pixel, thereby facilitating the identifica-
tion of recent mining expansion. To reduce false detections,
permanent water bodies were masked using the JRC Global
Surface Water dataset, acknowledging limitations in densely
vegetated river systems.
3.3 Validation datasets
For the initial delineation of the AOIs, as well as the val-
idation of the findings of this study used two auxiliary
datasets. One is MAAP (https://www.maapprogram.org, last
access: 20 October 2026), which documents mining expan-
sion, deforestation, and land-use change across the Ama-
zon, where some of their reports also covered areas ana-
lyzed in this study. Additionally, the AMW platform (https:
//amazonminingwatch.org, last access: 20 October 2026) had
been referred to, an initiative led by the Amazon Conser-
vation Association, Earth Genome and the Pulitzer Center’s
Rainforest Investigations Network. AMW uses satellite im-
agery from Sentinel-2 and AI to detect and map mining ac-
tivity across the Amazon Basin, producing cumulative quar-
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
182 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
Figure 3. PS data availability. Sub-AOIs are color-coded and numbered for easier discrimination in the results.
terly detections from 2018 onwards (Amazon Mining Watch,
2026).
4 Results
During the analysis, it was possible to identify different dy-
namics for each dataset due to the varying temporal and spa-
tial resolution. The increase or decrease in mining areas is
described and analyzed below. First, the results of the dataset
with the coarsest temporal resolution are described, in this
case the results of the SED. Then the results from the PS
data are looked at and compared to the SED results. Finally,
for areas and timeframes with high activity levels, the SCD
is applied to further analyze these dynamics.
4.1 Annual extent of unregulated mining based on SED
classification results
Across all analyzed AOIs, SED-derived mining extents re-
veal clear but regionally distinct expansion dynamics be-
tween 2017 and 2024 (see Fig. 4).
In the Punino AOI (1), mining activity remained low and
spatially limited until 2022, with all subregions exhibiting
values below 0.5 km2, although minor fluctuations appear in
this period, no sustained mining expansion is evident. This
was then followed by a rapid expansion of mining sites in all
subregions from 2022 onward. The Sub-AOI of Rio Punino
(1.4) emerged as a dominant hotspot, climbing from roughly
1.5 km2 in 2022 to more than 7.5 km 2 in 2024, a rise of
∼ 400 % in just two years. Punino North (1.1) also exhibits
a significant shift, with mining areas jumping from about
1.4 km2 in 2023 to ∼ 2.5 km2 in 2024 after years of near-
constant low values. Rio Acorna (1.2) remains with almost
no mining activity until 2022 before increasing to∼ 1.3 km2
in 2024. Rio Biguno (1.3) shows minimal activity up to 2023,
then mining activities expand from∼ 0.3 to∼ 1.8 km2 within
a single year. A comparison of these results with gold prices
for the same timespan reveals that the first significant in-
crease in gold prices in 2019/2020 had a limited impact on
the observed mining extents in the Punino AOI (1). The sec-
ond significant surge in gold prices in 2022/2023 coincides
with the expansion in mining operations (see also Fig. 4).
The Shaime AOI (2) exhibits more heterogenous dynam-
ics, however all subregions experienced net growth of min-
ing sites over the study period. The Shamataka subregions
(2.1, 2.2), Sub-AOI Rio Nangaritza South (2.5) and Wim-
bick (2.3) are characterized by peak-and-decline behavior,
while Rio Nangaritz North (2.4) experiences a steadily in-
creasing extent in mining area. The Sub-AOI started with
mining extents below 0.2 km 2 in 2017, which stayed that
way until 2020 where mining sites started to continuously ex-
pand with 0.4 km2 detected in 2020, up to more than 0.9 km2
of detected mining sites in 2024. In Shamataka East (2.1),
mining detection began at a low level ( ∼ 0.2 km2) in 2017
and increased steadily over the following years. The area
more than doubled by 2019 (0.45 km2) and continued to ex-
pand to 0.8 km 2 in 2021, until it reached a maximum of
∼ 0.85 km2 in 2022. A slight decrease to∼ 0.7 km2 followed
in 2023. Similar trajectories could be observed in the remain-
ing Sub-AOIs, however with different magnitudes. Rio Nan-
garitza South (2.5) and Shamataka West (2.2) started with
detected mining areas below 0.2 km 2 in 2017, which then
went up to∼ 1.5 km2 in 2021 for Rio Nangaritza South (2.5)
and∼ 1.2 km2 in 2022 for Shamataka West (2.2), both Sub-
AOIs then experienced a decrease in mining sites with more
vegetational regrowth visible, down to ∼ 0.75 km2 for Rio
Nangaritza South (2.5) and ∼ 1.0 km2 for Shamataka West
(2.2) in 2024. The two Sub-AOIs Shamataka East (2.1) and
Wimbick (2.3) show a similar trajectory, however with a
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 183
Figure 4. Results for SED-based classification and gold prices (source: https://goldprice.org, last access: 20 October 2026) in the study
period. Percentages indicate the absolute increase in mining sites for the Sub-AOI based on the initial value in 2017.
lower magnitude. For Shamataka East (2.1) mining extents
increased from 0.2 km2 in 2017 to∼ 0.85 km2 in 2022, fol-
lowing with a slight decrease in detected mining sites. For
Wimbick (2.3), mining extents increased from∼ 0.05 km2 in
2017 to 0.35 km 2 in 2022, following also with a slight de-
crease in detected mining sites in the following years. For
the Shaime AOI (2), the initial increase in gold prices corre-
sponds to an expansion of mining areas across the Sub-AOIs.
In contrast, the subsequent rise in gold prices does not align
with a consistent rise in mining activity, as evidenced by the
decline in mining extent across several Sub-AOIs. This find-
ing indicates that the observed mining dynamics cannot be
solely explained by gold prices and are likely influenced by
additional factors.
In South Napo (3), mining increased gradually in both sub-
regions, exhibiting only short-term fluctuations. Compared to
Arosemena Tola (3.1), South Napo Center (3.2) maintained
consistently higher activity levels throughout the study pe-
riod. The Sub-AOI Arosemena Tola (3.1) started with ob-
served mining extents of∼ 0.3 km2 with a slight increase in
area to∼ 0.45 km2 in 2019. 2020 showed a slight decrease
with 0.4 km2, with the following years showing a steady in-
crease up to∼ 0.95 km2 in 2024, presenting a ∼ 217 % in-
crease in mining sites. South Napo Center (3.2) showed more
slight fluctuations of observed mining extents for the period
between 2017 with initial values of∼ 1.4 km2 and 2024 with
observed mining extents of∼ 2.8 km2, showing an increase
of 100 %. The gradual expansion in both sub-AOIs broadly
coincided with the overall increase in global gold prices dur-
ing the study period, although year-to-year variations in min-
ing extent did not consistently correspond to fluctuations in
gold prices.
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
184 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
Figure 5. Results for PS-based classification for each Sub-AOI. The results show the absolute mining area detected in each image, as well
as the absolute mining area detected in the first and last available images for each Sub-AOI, along with the respective changes.
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 185
4.2 Extent of unregulated mining based on PlanetScope
classification results and comparison with SED
Results
The PS-based classification for the Punino AOI (1) reveals
temporal trajectories of mining expansion and decline across
all subregions despite data gaps in earlier years (Fig. 5).
Patterns closely align with SED results, though PS gener-
ally captures higher mining extents, such as averaged maxi-
mum values of∼ 9 km2 in Rio Punino (1.4),∼ 1 km2 in Rio
Acorna (1.2) or∼ 2.3 km2 in Rio Biguno (1.3). The analy-
sis showed that mining remains minimal until mid-2023 in
the Punino AOI (1), followed by rapid and synchronous ex-
pansion across Punino North (1.1), Rio Acorna (1.2.), and
Rio Biguno (1.3), mirroring the SED-derived shift. In con-
trast, Rio Punino (1.4) shows earlier onset and substantially
stronger growth, with both datasets capturing the pattern but
PS indicating higher peak extents, as mentioned before.
In the Shaime AOI (2), PS data indicates widespread in-
tensification between 2017 and 2024 (Fig. 5). Trends closely
match SED results, confirming a coherent temporal signal.
However, PS data shows that expansion occurs in multiple
short-lived phases, highlighting intra-annual variability not
resolved in the annual SED data. This pattern is consistent
across subregions, with earlier and stronger intensification in
Rio Nangaritza South (2.5) and more gradual but continuous
growth elsewhere.
For the South Napo AOI (3), both datasets show sustained
expansion from initially low activity before 2019 to more
extensive and persistent mining thereafter (Fig. 5). While
South Napo Center (3.2) exhibits earlier and stronger growth,
Arosemena Tola (3.1) follows a similar but more gradual tra-
jectory.
Overall, PS confirms SED-derived trends, however, direct
comparison is constrained by differing temporal and spatial
resolutions. In some cases, the effect is more pronounced,
as is evident in 2024 in the Punino AOI (1), where PS con-
sistently indicates larger extents than SED. This suggests a
systematic divergence that may reflect PS’s higher sensitivity
to small-scale features and SED’s potential underestimation
due to temporal aggregation and smoothing.
4.3 Interannual dynamics – sequential change detection
results for specific regions/timeframes
Analysis of the SED and PS results indicates that several
AOIs experienced phases of rapid mining expansion, rang-
ing from gradual development to abrupt increases. However,
optical data alone does not fully resolve the temporal evo-
lution of these phases due to frequent cloud coverage. This
is particularly relevant for fast increases in mining extent, as
observed in the Punino (1) AOI. Therefore, SCD was applied
to its four subregions, focusing on the year of strongest ex-
pansion (with both 2023 and 2024 analyzed for Rio Punino
(1.4)) to detect changes due to deforestation or the movement
of large land masses (see Fig. 6).
For Punino North (1.1), SCD results show a clear spatial
progression of mining expansion along the central axis, ex-
tending toward the northeastern and southwestern margins.
The coexistence of early-, mid-, and late-year detections in-
dicates continuous activity throughout 2024, while localized
early detections in the southeast suggest temporally limited
clearing. Newly established sites in the north further indi-
cate expansion into previously undisturbed areas. As in the
other subregions, areas already classified as active mining
sites show little to no change signal, reflecting the method’s
sensitivity to newly occurring land-cover transitions.
In Rio Acorna (1.2), SCD detections align closely with
newly identified mining areas in PS imagery and follow lin-
ear patterns along established corridors. The SCD sequence
indicates that changes in the westernmost patch occurred
mainly during early 2023, whereas adjacent areas show ac-
tivity emerging later in the year. Additional clusters exhibit
detections throughout 2023, indicating sustained expansion
rather than isolated events.
For Rio Biguno (1.3), change detections similarly corre-
spond to newly disturbed areas and follow existing mining
corridors. Continuous signals from early to late 2024 indicate
sustained expansion, while localized clusters reflect shorter-
term clearing phases. As observed elsewhere, pre-existing
mining areas show minimal change signal in the SCD results.
In Rio Punino (1.4), SCD captures extensive expansion in
both 2023 and 2024. In 2023, changes form spatially contin-
uous corridors along river systems, indicating rapid, directed
expansion. In contrast, in 2024, there are more dispersed and
fragmented patterns, suggesting a shift toward expansion into
surrounding areas and river tributaries.
Overall, SCD results confirm the spatial patterns observed
in PS classifications while providing detailed insight into the
timing and progression of mining expansion at sub-annual
scales.
4.4 Validation
Due to the nature of the study, no ground truth data were
available and the collection of in-situ validation data was not
feasible. Instead, the results were validated through a com-
parison of the independent analysis approaches, allowing the
outputs of each method to serve as mutual validation. As
shown in Fig. 7, the SED-based and PS-based classifications
exhibit a high degree of agreement, providing strong cross-
validation of the identified mining areas. For this compari-
son, the annual SED dataset for 2024 was evaluated against
the most recent available PS scene for each Sub-AOI in 2024.
While the overall spatial extent of mining is highly consistent
between both datasets, the PS-based classification captures
finer spatial details due to its higher spatial resolution. These
differences in spatial resolution also likely contribute to the
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
186 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
Figure 6. Comparison of SCD results with RF classification results based on PS data for each Sub-AOI of the Punino AOI (1). Basemap
and map service sources: Earthstar Geographics; Esri; TomTom; FAO; NOAA; USGS; © OpenStreetMap contributors and the GIS User
Community; Esri, NASA, NGA, USGS.
discrepancies in estimated mining extent observed in Figs. 4
and 5.
Derived mining extents and dynamics were also compared
with multiple MAAP reports, showing strong agreement with
the independently documented spatial patterns and devel-
opment of mining activities. For example, in the Punino
AOI (1), MAAP reports extensive mining largely outside
legal concessions, including within protected areas such as
Sumaco-Napo Galeras National Park and Kichwa Indigenous
territories (MAAP, 2024). In the Shaime AOI, it is reported
that mining occurs partly outside concessions or using unau-
thorized methods, particularly in ecologically important cor-
ridors like the Nangaritza River High Basin Protective Forest
(Villa et al., 2022). For South Napo similar trends are re-
ported, though with a smaller proportion of activity outside
the official cadaster (MAAP, 2025).
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 187
Figure 7. Comparison of SED-based and PS-based results of mining extents in 2024 for the Punino AOI (1). Basemap and map service
sources: Earthstar Geographics; Esri; TomTom; FAO; NOAA; USGS; © OpenStreetMap contributors and the GIS User Community; Esri,
NASA, NGA, USGS.
Figure 8. Comparison of SED-based classification results and areas detected by AMW. Basemap and map service sources: Earthstar Geo-
graphics; Esri; TomTom; FAO; NOAA; USGS; © OpenStreetMap contributors and the GIS User Community; Esri, NASA, NGA, USGS.
In addition to MAAP, we compared our results with the
AMW platform. Across all three AOIs, strong spatial corre-
spondence between the mining extents identified in this study
and those reported by AMW are observed, supporting the ro-
bustness of the classification framework. The convergence of
independently developed methodologies, each using distinct
data processing and classification strategies, increases confi-
dence in the reliability of observed trends. However, notable
differences exist. AMW outputs are comparatively spatially
coarse, representing mining-affected areas with larger, gen-
eralized polygons. In contrast, the SED and PS-based classi-
fications in this study delineate mining sites with higher spa-
tial precision, closely following the outlines of disturbed ar-
eas. This higher precision is further enhanced by incorporat-
ing a permanent water mask, which reduces false detections
along rivers and water bodies. Some discrepancies between
datasets can therefore be attributed to differences in spa-
tial resolution, classification thresholds and post-processing
choices.
A detailed comparison for 2024 (Fig. 8) illustrates these
differences at the Sub-AOI level. In the Punino AOI (1), par-
ticularly in Rio Acorna (1.2), AMW detected only a lim-
ited mining area, while SED classified mining activity across
much of the subregion. Similar patterns were observed in two
western Shaime Sub-AOIs (2.2, 2.3), where AMW outputs
underrepresented smaller or newly established sites. Despite
these localized discrepancies, most areas show substantial
spatial overlap, confirming that both approaches capture the
core mining expansion patterns.
From a monitoring perspective, these findings highlight
the complementary strengths of multiple methodological ap-
proaches for mapping ASGM areas. The AMW platform pro-
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
188 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
vides a coarse overview of the location of mining sites that is
particularly valuable for large-scale monitoring and interre-
gional comparisons. In contrast, the SED and PS classifica-
tion approaches enable a more detailed and spatially precise
delineation of mining features. Building on this, the SCD
approach further refines the analysis by identifying consis-
tent spatio-temporal changes, thereby supporting the detec-
tion of progressive expansion dynamics rather than static ex-
tents alone.
5 Discussion
The three AOIs exhibit distinct dynamics of unregulated gold
mining, reflecting different controls of economic drivers, ac-
cessibility and local constraints. In Punino (1), for example,
the rapid expansion of mining since 2023 indicates a strong
response to rising gold prices (see Fig. 4). This could indicate
that previously unexploited areas can quickly become lucra-
tive under favorable market conditions. In addition, the pro-
gression from the main river into tributaries illustrates how
rapidly miners exploit resources. Due to the dense rainforest,
no roads are visible in the satellite imagery. However, the
rapid clearing suggests the presence of either a large work-
force or substantial machinery for deforestation and mining.
In contrast, the heterogeneous trends in Shaime (2) im-
ply a stronger influence of local factors. Declining or stag-
nating mining extents after peak activity may indicate re-
source depletion, reduced profitability, or social and regula-
tory pressures, while continued growth in Nangaritza North
(2.4) shows that these constraints are not uniform.
South Napo (3) shows more moderate and temporally vari-
able dynamics, suggesting either stronger limiting factors or
a slower development trajectory. The delayed surge in 2024
suggests that mining expansion may occur in pulses, poten-
tially triggered by changing economic conditions or spillover
effects from other regions. The findings of this study demon-
strate the capability of RS-based change detection to effec-
tively identify the extent of mining expansion in remote trop-
ical regions. However, the examination of the fundamental
dynamics underlying this expansion is constrained by limita-
tions due to the absence of available validation for possible
assumptions without conducting interviews or relying on in-
sider knowledge.
5.1 Challenges, limitations and advantages of the
different datasets
The analysis faced challenges due to differences in sensor
characteristics and dataset properties, which complicates di-
rect comparisons but also highlights the individual and com-
plementary strengths of each approach.
The SED dataset, provided as an annual 10 m product,
requires no preprocessing, enabling rapid mapping and vi-
sualization of mining-affected areas and making it accessi-
ble to users with limited RS-data processing expertise. Its
coarser resolution, however, limits the detection of small-
scale features. In addition, AI-generated embeddings, such
as the SED, may vary in input data across years, although in
this study no systematic inconsistencies were observed and
classification proved reliable in all observed years. SED is
therefore best suited for providing broad overviews of min-
ing sites and their long-term evolution. In this case, the SED
proved to be a reliable tool for detecting mining sites with
minimal effort. However, it should be noted that due to the
study area’s geographical characteristics, seasonality had no
impact on the imagery, thereby ensuring that no seasonal
variations influenced the results.
PS imagery offers higher spatial resolution of 3 m and
the potential for finer temporal analysis. Image availability,
however, is affected by cloud cover, satellite coverage and
scene quality. The higher resolution allows more precise de-
lineation of mining features and is valuable for validation and
detailed spatial analyses. As discussed in Sect. 4.2, PS im-
agery provides valuable insight into intra-annual variability
under favorable weather conditions, which the annual SED
product does not capture. However, variations in image qual-
ity between individual PS scenes were found to impact the
classification results and the spectral variability necessitated
the use of multiple reference years for training, increasing
computational demands. Despite these efforts, differences in
classification performance between scenes were evident dur-
ing the visual assessment of the results. This limitation is pri-
marily related to the methodology itself because supervised
classification approaches are sensitive to variations in image
quality, illumination conditions, atmospheric effects and sen-
sor characteristics. Consequently, some degree of inconsis-
tency between individual observations cannot be avoided en-
tirely. Another limitation is that the data are not freely avail-
able and require user registration or a licensed account for
access. In addition, processing these data demands greater
prior knowledge because of the more complex preprocessing
workflow compared to more readily accessible RS datasets,
such as the SED.
SAR-based monitoring using the SCD approach provides
near-continuous observations independent of weather con-
ditions, with a 12 d revisit interval. The method effectively
detects deforestation associated with mining expansion and
shows high correspondence with optical datasets. However,
the 30 m spatial resolution limits the detection of small-scale
changes, which results in undetectable changes within previ-
ously deforested areas, restricting the approach to identify-
ing newly occurring land-cover transitions or the movement
of large land masses. Despite these limitations, the SCD ap-
proach is well suited for the continuous monitoring of the
spatio-temporal dynamics of mining expansion. Although
the dataset is freely available, its application requires sub-
stantial prior knowledge because of the relatively complex
processing workflow.
The study initially aimed to compare multiple methodolo-
gies for monitoring spatio-temporal changes in mining ex-
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 189
tent. However, the analysis demonstrated that each dataset
has unique strengths and limitations. Thus, the suitability of
a dataset depends largely on the analysis’s objectives and the
user’s prior knowledge. Furthermore, the results highlight the
added value of combining multiple datasets rather than rely-
ing on a single source of information. Based on this study’s
findings, the SED dataset is recommended as a starting point
due to its ease of use, broad spatial coverage and ability to
easily identify mining areas. Due to the limited availability of
PS imagery and the time- and computationally intensive pro-
cesses of image selection, training data collection and classi-
fication, PS data are best used selectively to validate and re-
fine observations derived from the SED and S-1 datasets. For
applications requiring higher temporal resolution, S-1 data
are recommended due to their frequent acquisition rate, con-
sistent coverage and ability to provide additional information
on surface changes independent of cloud cover. Therefore,
an integrated workflow combining the SED and S-1 datasets
with targeted PS observations is recommended for monitor-
ing mining expansion. This approach leverages the comple-
mentary strengths of each dataset while minimizing their in-
dividual limitations. Additionally, due to the sensitivity of
the methodologies presented, the analysis is most successful
when combined with coarse, known areas of mining sites,
such as provided by the AMW.
As the results of this study are based on AOIs selected
using prior knowledge of known mining locations derived
from multiple independent sources, there is a high degree of
confidence that the detected land cover changes primarily re-
flect mining expansion and associated deforestation. Further-
more, PS imagery was used to visually verify the presence
of characteristic mining features, such as exposed soil sur-
faces and mining ponds. However, it is not possible to con-
clusively determine that all observed changes are exclusively
attributable to mining activities, as some land-cover trans-
formations may result from other anthropogenic or natural
processes. Definitive attribution would require in situ obser-
vations and ground-truth data, which were unavailable within
the scope of this study due to logistical and safety constraints.
Moreover, acquiring such data would go beyond the primary
objective of this research, which is to demonstrate a repro-
ducible RS workflow for monitoring the spatio-temporal evo-
lution of mining sites.
While the methods capture mining extent and associated
deforestation dynamics, they do not quantify broader eco-
logical or social impacts. Unregulated gold mining often in-
volves mercury amalgamation, releasing toxic elements into
soils, rivers and the atmosphere, which can bioaccumulate
and pose risks to ecosystems and human health (Esdaile
and Chalker, 2018; Gibb and O’Leary, 2014). Although the
present study can indicate areas with a heightened poten-
tial for mercury exposure based on mining-related land-use
change, a comprehensive assessment of ecological and soci-
etal impacts requires additional investigations. In particular,
in situ measurements, long-term monitoring and interdisci-
plinary studies are necessary to quantify contaminant path-
ways, exposure levels and associated risks to affected popu-
lations.
5.2 Methodological comparison and broader
applicability
ASGM is not only a major challenge in the Amazon but also
a widespread global phenomenon, as it has become an in-
creasingly important economic activity in many developing
countries, providing livelihoods for millions of people while
simultaneously causing substantial environmental degrada-
tion (Ngom et al., 2023). Consequently, numerous studies
have investigated the detection and monitoring of ASGM us-
ing RS techniques across different regions worldwide. Most
existing studies rely on a single RS data source tailored to
a specific application. For example, LaJeunesse Connette et
al. (2016) employed Landsat 8 and Google Earth imagery
to identify and map ASGM activities in Myanmar, while
Gallwey et al. (2020) applied deep learning techniques to
Sentinel-2 imagery for the detection of gold panning sites.
Similarly, Kimijima et al. (2022) utilized S-1 time-series data
to monitor the dynamics of mobile mining camps in Indone-
sia. These studies collectively demonstrate the considerable
potential of RS for monitoring mining activities. However,
they also highlight that the suitability of a particular dataset
strongly depends on the study objectives, environmental con-
ditions and regional characteristics. Each platform provides
distinct advantages while simultaneously presenting inherent
limitations regarding spatial resolution, temporal coverage,
cloud susceptibility, or sensitivity to surface characteristics.
In contrast to previous studies that primarily evaluate a
single sensor or methodological approach, the present study
focuses on the comparative assessment of multiple comple-
mentary datasets and analysis techniques. Rather than iden-
tifying a single optimal approach, the objective is to evaluate
the respective strengths and limitations of multiple datasets
and methodologies and to demonstrate how their integration
can improve the monitoring of mining dynamics. The results
show that combining complementary datasets increases the
robustness of mining detection and allow for the observa-
tion of spatio-temporal dynamics on multiple levels. An im-
portant aspect of this study is the availability of the AMW
dataset. However, such reference datasets are rarely available
in many other mining regions and field-based validation is
often not feasible because of limited accessibility, security
concerns and logistical constraints. Consequently, the devel-
opment of robust validation strategies that do not rely on ex-
tensive in-situ observations remains a significant challenge
for transferring RS-based mining monitoring frameworks to
other regions. The findings also reflect the increasing im-
portance of RS for monitoring unregulated mining activi-
ties, which have expanded considerably worldwide over the
past decade (Fritz et al., 2018). In addition, previous studies
(e.g. Coral et al., 2024) have shown that mining conflicts can
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
190 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
stimulate new governance structures, strengthen civil soci-
ety organizations and support legal action for environmental
protection. Continuous observation of mining activities us-
ing available RS data can be useful by providing indepen-
dent and regularly updated information that is accessible to
researchers, NGOs and decision makers.
Beyond ASGM monitoring, the methodologies presented
in this study have broader applicability to a wide range of
land-cover and land-use change analyses. The integration of
AI-based, optical and SAR RS data could support the detec-
tion and monitoring of illicit crop cultivation, agricultural ex-
pansion, deforestation, urban sprawl and other anthropogenic
landscape transformations. Across these applications, sim-
ilar trade-offs between temporal resolution, spatial detail,
data availability, cloud sensitivity and computational require-
ments are expected. Consequently, the framework presented
in this study demonstrates the broader potential of integrating
complementary RS datasets to improve environmental mon-
itoring and change detection across diverse geographic and
thematic contexts.
5.3 Contribution of extractivism-driven deforestation to
the Amazon Rainforest tipping point
The primary regional-scale drivers of the Amazonian habi-
tat degradation include land-use change, hydrological alter-
ation and climate-induced aridification, all of which interact
to destabilize the forest–rainfall feedback system that sus-
tains the Amazon. As approximately half of regional pre-
cipitation is recycled through evapotranspiration, continued
forest loss increases the risk of a critical transition from a
forested to a non-forested state, with a wide range of con-
sequences (Berenguer et al., 2021; Wunderling et al., 2021;
Cooper et al., 2020).
While large-scale agriculture remains the dominant driver
of deforestation (Ferrante and Fearnside, 2022), the results
of this study demonstrate that resource extractivism can act
as a rapid and spatially targeted catalyst of forest loss. In par-
ticular, the observed river-centered expansion of mining in
the Punino AOI (1) highlights how extractive activities pen-
etrate previously inaccessible regions, following waterways
and subsequently expanding into tributaries. These dynamics
indicate that mining can initiate new deforestation frontiers,
potentially facilitating further land-use change. The intense
landscape alterations that are caused, as well as the introduc-
tion of toxic substances into the soil and water systems create
a condition that makes forest restoration measures even more
complicated and harmful for humans carrying out restoration
measures.
These findings underline the importance of RS-based ap-
proaches, as they enable the timely detection of rapidly
emerging extractive hotspots in remote regions. Conse-
quently, extractivism-driven deforestation should be consid-
ered not only as a secondary driver, but as a significant and
dynamic contributor to processes that may push the Amazon
rainforest closer to a tipping point.
6 Conclusions
This study demonstrates the unique values of single, as well
as multi-sensor RS approaches for mapping and monitoring
unregulated gold mining in the Amazon. Across the three
AOIs (Punino (1), Shaime (2) and South Napo (3)) distinct
patterns of mining expansion were observed with increases in
mining sites ranging from 109 % in South Napo Center (3.2)
and 1,445 % in Rio Punino (1.4), during the study period.
Punino (1) exhibited rapid, recent intensification along the
main river and tributaries beginning mostly in 2023, closely
aligning with rising global gold prices. Shaime’s (2) subre-
gions displayed heterogeneous dynamics, with some areas
peaking in activity around 2021–2022 and others, such as
Nangaritza North (2.4), showing steady growth. South Napo
(3) experienced more gradual expansion, with notable in-
creases after 2021 in both subregions.
The integration of SED, PS and S-1 SAR datasets al-
lowed both spatial and temporal refinement of mining dy-
namics. SED provided an accessible, coarse overview of an-
nual trends, while PS enabled higher-resolution mapping and
was found to be more effective for validation purposes. S-1-
based SCD analysis complemented optical imagery by de-
tecting short-term changes every 12 d, revealing patterns of
progressive deforestation and progressive site expansion that
optical datasets alone may overlook. While each dataset al-
lows for a unique approach for capturing both the extent and
temporal dynamics of mining sites, depending on the spe-
cific use case and prior knowledge of the user. Combining
multiple datasets is an even more effective way to see the en-
tire scope of these dynamics and have additional validation.
The combination of these methods consistently identified de-
forestation associated with mining, even in remote and inac-
cessible areas, highlighting the rapid pace and spatial com-
plexity of extractive activity. Comparisons with independent
datasets, including MAAP and the AMW platform, confirm
the robustness of the proposed classification framework. Al-
though minor discrepancies arise due to differences in spa-
tial resolution and methodological design, the strong spatial
agreement underscores the reliability of the detected trends.
The presented workflows are particularly well suited for inte-
gration with platforms such as AMW, enabling more detailed
analyses of identified sites and supporting further investiga-
tion of mining dynamics.
A key outcome of this study is that the presented method-
ologies vary in complexity while relying largely on freely
available or easily accessible data and resources. This
demonstrates that RS enables the safe and independent mon-
itoring of mining extent and dynamics without requiring di-
rect access to mining sites. The accessibility of these work-
flows allows researchers, stakeholders and decision-makers
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 191
to independently assess mining activities, validate reported
mining extents and monitor unregulated operations. Conse-
quently, RS can enhance transparency and provide valuable
evidence for regulatory oversight, environmental manage-
ment and policy development.
Finally, while these RS approaches effectively map the ex-
tent and dynamics of mining, they do not capture the broader
ecological and social consequences, including mercury con-
tamination, biodiversity loss and impacts on Indigenous terri-
tories. Integrating RS with in-situ monitoring and interdisci-
plinary assessments is essential to fully understand the envi-
ronmental and societal implications of unregulated gold min-
ing in the Amazon and worldwide.
Code and data availability. All datasets used in this study are
publicly available or accessible through research programs. The
S-1 data were obtained from the European Space Agency (ESA)
and accessed via GEE. The SED was accessed through Google’s
geospatial data catalogue within GEE. PS imagery was provided
through Planet’s Education and Research Program and downloaded
via the Planet Explorer. The JRC Global Surface Water dataset
used for masking permanent water bodies is publicly available
within GEE. Outputs from the AMW platform, are accessible online
(https://amazonminingwatch.org/, last access: 21 October 2026).
All data processing and analysis were conducted within GEE us-
ing custom scripts. The code used to generate the results of this
study is available from the corresponding author upon request.
Author contributions. IL conducted the conceptualization,
methodology development, formal analysis, visualization and
writing of the original draft. VG and CG provided supervision
and contributed to reviewing and editing the manuscript, offering
critical feedback and guidance throughout the study. JJA con-
tributed local expertise and contextual knowledge on Ecuador and
supported the interpretation of the results.
Competing interests. The contact author has declared that none
of the authors has any competing interests.
Disclaimer. Publisher’s note: Copernicus Publications remains
neutral with regard to jurisdictional claims made in the text, pub-
lished maps, institutional affiliations, or any other geographical rep-
resentation in this paper. The authors bear the ultimate responsibil-
ity for providing appropriate place names. Views expressed in the
text are those of the authors and do not necessarily reflect the views
of the publisher.
Acknowledgements. We acknowledge the contributions of Jan-
Erik Feldt for the delineation of the AOI shapefiles. Linguistic im-
provements were supported by AI-based tools, including DeepL.
Review statement. This paper was edited by Ce Zhang and re-
viewed by Ricardo Paris and one anonymous referee.
References
Adamek, K., Lupa, M., and Zawadzki, M.: Remote sensing
techniques for tracking changes caused by illegal gold min-
ing in Madre de Dios, Peru, Misc. Geogr., 25, 205–212,
https://doi.org/10.2478/mgrsd-2020-0028, 2021.
Amazon Mining Watch: Amazon Mining Watch, https:
//amazonminingwatch.org (last access: 15 March 2026),
2026.
Albert, J. S., Carnaval, A. C., Flantua, S. G. A., Lohmann, L.
G., Ribas, C. C., Riff, D., Carrillo, J. D., Fan, Y ., Figueiredo,
J. P., Guayasamin, J. M., Hoorn, C., de Melo, G. H., Nasci-
mento, N., Quesada, C. A., Ulloa Ulloa, C., Val, P., Arieira,
J., Encalada, A. C., and Nobre, C. A.: Human impacts out-
pace natural processes in the Amazon, Science, 379, eabo5003,
https://doi.org/10.1126/science.abo5003, 2023.
Asokan, A. and Anitha, J.: Change detection techniques for remote
sensing applications: a survey, Earth Sci. Inform., 12, 143–160,
https://doi.org/10.1007/s12145-019-00380-5, 2019.
Bank, M. S., Vignati, D. A. L., and Vigon, B.: United Na-
tions Environment Programme’s Global Mercury Partner-
ship: science for successful implementation of the Mina-
mata Convention, Environ. Toxicol. Chem., 33, 1199–1201,
https://doi.org/10.1002/etc.2592, 2014.
Beltrán, J.: Illegal mining threatens one of the last forest links
between the Andes and Ecuador’s Amazon, Mongabay,
https://news.mongabay.com/2022/06/illegal-mining-threatens-
one-of-the-last-forest-links-between-the-andes-and-ecuadors-
amazon/ (last access: 2 March 2026), June 2022.
Berenguer, E., Armenteras, D., Lees, A. C., Smith, C. C., Fearn-
side, P. M., Nascimento, N., Alencar, A., Almeida, C., Aragão,
L. E. O., Barlow, J., Bilbao, B., Brando, P. M., Bynoe, P., Finer,
M., Flores, B. M., Jenkins, C. N., Silva Junior, C. H. L., Souza,
C., and García-Villacorta, R.: Chapter 19: Drivers and ecolog-
ical impacts of deforestation and forest degradation, in: Ama-
zon Assessment Report 2021, edited by: Nobre, C. A., Encal-
ada, A., Anderson, E., Roca Alcazar, F. H., Bustamante, M.,
Mena, C., Peña-Claros, M., Poveda, G., Rodriguez, J. P., Saleska,
S., Trumbore, S., Val, A. L., Villa Nova, L., Abramovay, R.,
Alencar, A., Rodríguez Alzza, C., Armenteras, D., Artaxo, P.,
Athayde, S., Barretto Filho, H. T., Barlow, J., Berenguer, E., Bor-
tolotto, F., Costa, F. A., Costa, M. H., Cuvi, N., Fearnside, P. M.,
Ferreira, J., Flores, B. M., Frieri, S., Gatti, L. V ., Guayasamin,
J. M., Hecht, S., Hirota, M., Hoorn, C., Josse, C., Lapola, D.
M., Larrea, C., Larrea-Alcazar, D. M., Lehm Ardaya, Z., Malhi,
Y ., Marengo, J. A., Melack, J., Moraes, R. M., Moutinho, P.,
Murmis, M. R., Neves, E. G., Paez, B., Painter, L., Ramos,
A., Rosero-Peña, M. C., Schmink, M., Sist, P., ter Steege, H.,
Val, P., van der V oort, H., Varese, M., and Zapata-Ríos, G.,
United Nations Sustainable Development Solutions Network,
https://doi.org/10.55161/AIZJ1133, 2021.
Borja, M. O., Aguilar, C., Verdesoto, G., Villa, J., Mamani, N.,
Finer, M., and Josse, C.: MAAP #198: Expansión de la min-
ería en la Amazonía de Ecuador, Monitoring of the Andes
Amazon Program (MAAP), https://www.maapprogram.org/es/
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
192 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
expansion-mineria-ecuador/ (last access: 20 February 2026),
2023.
Breiman, L.: Random forests, Mach. Learn., 45, 5–32,
https://doi.org/10.1023/A:1010933404324, 2001.
Brown, C. F., Kazmierski, M. R., Pasquarella, V . J., Ruck-
lidge, W. J., Samsikova, M., Zhang, C., Shelhamer, E., La-
hera, E., Wiles, O., Ilyushchenko, S., Gorelick, N., Zhang,
L. L., Alj, S., Schechter, E., Askay, S., Guinan, O., Moore,
R., Boukouvalas, A., and Kohli, P.: AlphaEarth Founda-
tions: an embedding field model for accurate and efficient
global mapping from sparse label data, arXiv [preprint],
https://doi.org/10.48550/arXiv.2507.22291, 29 July 2025.
Canty, M.: EESAR Docker container, GitHub [code], https://github.
com/mortcanty/EESARDocker, 2021.
Canty, M. J., Nielsen, A. A., Conradsen, K., and Skriver,
H.: Statistical analysis of changes in Sentinel-1 time se-
ries on the Google Earth Engine, Remote Sens., 12, 46,
https://doi.org/10.3390/rs12010046, 2020.
Capparelli, M. V ., Cabrera, M., Rico, A., Lucas-Solis, O., Alvear-
S, D., Vasco, S., Galarza, E., Shiguango, L., Pinos-Velez, V .,
Pérez-González, A., Espinosa, R., and Moulatlet, G. M.: An
integrative approach to assess the environmental impacts of
gold mining contamination in the Amazon, Toxics, 9, 149,
https://doi.org/10.3390/toxics9070149, 2021.
Conradsen, K., Nielsen, A. A., and Skriver, H.: Determin-
ing the points of change in time series of polarimet-
ric SAR data, IEEE T. Geosci. Remote, 54, 3007–3024,
https://doi.org/10.1109/TGRS.2015.2510160, 2016.
Cooper, G. S., Willcock, S., and Dearing, J. A.: Regime shifts occur
disproportionately faster in larger ecosystems, Nat. Commun.,
11, 1175, https://doi.org/10.1038/s41467-020-15029-x, 2020.
Coral, C., Plieninger, T., Sieber, S., and Graw, V .: Extractivism
triggering new forms of governance for the rights of na-
ture: the case of Northwest Ecuador, Geoforum, 156, 104111,
https://doi.org/10.1016/j.geoforum.2024.104111, 2024.
Esdaile, L. J. and Chalker, J. M.: The mercury problem in arti-
sanal and small-scale gold mining, Chem. Eur. J., 24, 6905–6916,
https://doi.org/10.1002/chem.201704840, 2018.
Ferrante, L. and Fearnside, P. M.: Countries should boycott
Brazil over export-driven deforestation, Nature, 601, 318,
https://doi.org/10.1038/d41586-022-00094-7, 2022.
Flores, B. M., Montoya, E., Sakschewski, B., Nascimento, N., Staal,
A., Betts, R. A., Levis, C., Lapola, D. M., Esquível-Muelbert, A.,
Jakovac, C., Nobre, C. A., Oliveira, R. S., Borma, L. S., Nian, D.,
Boers, N., Hecht, S. B., ter Steege, H., Arieira, J., Lucas, I. L.,
Berenguer, E., Marengo, J. A., Gatti, L. V ., Mattos, C. R. C., and
Hirota, M.: Critical transitions in the Amazon forest system, Na-
ture, 626, 555–564, https://doi.org/10.1038/s41586-023-06970-
0, 2024.
Forkuor, G., Ullmann, T., and Griesbeck, M.: Mapping and
monitoring small-scale mining activities in Ghana using
Sentinel-1 time series (2015–2019), Remote Sens., 12, 911,
https://doi.org/10.3390/rs12060911, 2020.
Fritz, M. M., McQuilken, J., Collins, N., and Weldegior-
gis, F.: Global trends in artisanal and small-scale min-
ing (ASM): a review of key numbers and issues, In-
ternational Institute for Sustainable Development (IISD),
https://doi.org/10.13140/RG.2.2.16756.45444, 2018.
Gautam, R., Kim, Y ., Topal, E., and Hitch, M.: Cor-
relation between COVID-19 cases and gold price
fluctuation, Int. J. Min. Reclam. Env., 36, 574–586,
https://doi.org/10.1080/17480930.2022.2077542, 2022.
Gallwey, J., Robiati, C., Coggan, J., V ogt, D., and Eyre, M.: A
Sentinel-2 based multispectral convolutional neural network for
detecting artisanal small-scale mining in Ghana: applying deep
learning to shallow mining, Remote Sens. Environ., 248, 111970,
https://doi.org/10.1016/j.rse.2020.111970, 2020.
Geudtner, D., Torres, R., Snoeij, P., Davidson, M., and Rommen, B.:
Sentinel-1 system capabilities and applications, in: Proceedings
of the 2014 IEEE Geoscience and Remote Sensing Symposium,
1457–1460, https://doi.org/10.1109/IGARSS.2014.6946711,
2014.
Gibb, H. and O’Leary, K. G.: Mercury exposure and health impacts
among individuals in the artisanal and small-scale gold min-
ing community: a comprehensive review, Environ. Health Persp.,
122, 667–672, https://doi.org/10.1289/ehp.1307864, 2014.
Götze, C., Beyer, F., and Gläßer, C.: Pioneer vegetation as an in-
dicator of the geochemical parameters in abandoned mine sites
using hyperspectral airborne data, Environ. Earth Sci., 75, 613,
https://doi.org/10.1007/s12665-016-5367-1, 2016.
Graw, V ., Dedring, T., Hiby, R., Jara-Alvear, J., Guzman, P., and
Juergens, C.: Regressive erosion at River Coca in Northeast
Ecuador: landslide monitoring with Sentinel-1 to support disaster
risk management, PFG – J. Photogramm. Remote Sens. Geoinf.
Sci., 90, 457–471, https://doi.org/10.1007/s41064-022-00221-z,
2022.
Grinand, C., Rakotomalala, F., Gond, V ., Vaudry, R., Bernoux,
M., and Vieilledent, G.: Estimating deforestation in tropi-
cal humid and dry forests in Madagascar from 2000 to
2010 using multi-date Landsat satellite images and the ran-
dom forests classifier, Remote Sens. Environ., 139, 68–80,
https://doi.org/10.1016/j.rse.2013.07.008, 2013.
Guayasamin, J. M., Ribas, C. C., Carnaval, A. C., Carrillo, J. D.,
Hoorn, C., Lohmann, L. G., Riff, D., Ulloa Ulloa, C., and Al-
bert, J. S.: Chapter 2: Evolution of Amazonian biodiversity, in:
Amazon Assessment Report 2021, edited by: Nobre, C. A., En-
calada, A., Anderson, E., Roca Alcazar, F. H., Bustamante, M.,
Mena, C., Peña-Claros, M., Poveda, G., Rodriguez, J. P., Saleska,
S., Trumbore, S., Val, A. L., Villa Nova, L., Abramovay, R.,
Alencar, A., Rodríguez Alzza, C., Armenteras, D., Artaxo, P.,
Athayde, S., Barretto Filho, H. T., Barlow, J., Berenguer, E., Bor-
tolotto, F., Costa, F. A., Costa, M. H., Cuvi, N., Fearnside, P. M.,
Ferreira, J., Flores, B. M., Frieri, S., Gatti, L. V ., Guayasamin,
J. M., Hecht, S., Hirota, M., Hoorn, C., Josse, C., Lapola, D.
M., Larrea, C., Larrea-Alcazar, D. M., Lehm Ardaya, Z., Malhi,
Y ., Marengo, J. A., Melack, J., Moraes, R. M., Moutinho, P.,
Murmis, M. R., Neves, E. G., Paez, B., Painter, L., Ramos,
A., Rosero-Peña, M. C., Schmink, M., Sist, P., ter Steege, H.,
Val, P., van der V oort, H., Varese, M., and Zapata-Ríos, G.,
United Nations Sustainable Development Solutions Network,
https://doi.org/10.55161/CZWN4679, 2021.
Isidro, C. M., McIntyre, N., Lechner, A. M., and Callow, I.: Ap-
plicability of earth observation for identifying small-scale min-
ing footprints in a wet tropical region, Remote Sens., 9, 945,
https://doi.org/10.3390/rs9090945, 2017.
Kimijima, S., Sakakibara, M., and Nagai, M.: Characterizing time-
series roving artisanal and small-scale gold mining activities in
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics 193
Indonesia using Sentinel-1 data, Int. J. Environ. Res. Pu., 19,
6266, https://doi.org/10.3390/ijerph19106266, 2022.
LaJeunesse Connette, K. J., Connette, G., Bernd, A., Phyo, P.,
Aung, K. H., Tun, Y . L., Thein, Z. M., Horning, N., Leimgruber,
P., and Songer, M.: Assessment of mining extent and expansion
in Myanmar based on freely-available satellite imagery, Remote
Sens., 8, 912, https://doi.org/10.3390/rs8110912, 2016.
Lenton, T. M., Milkoreit, M., Willcock, S., Abrams, J. F., Arm-
strong McKay, D. I., Buxton, J. E., Donges, J. F., Loriani, S.,
Wunderling, N., Alkemade, F., Barrett, M., Constantino, S., Pow-
ell, T., Smith, S. R., Boulton, C. A., Pinho, P., Dijkstra, H. A.,
Pearce-Kelly, P., Roman-Cuesta, R. M., and Dennis, D. (Eds.):
The Global Tipping Points Report 2025, University of Exeter,
Zenodo, https://doi.org/10.5281/zenodo.18163977, 2025.
Lobo, F. D. L., Souza-Filho, P. W. M., Novo, E. M. L. D. M.,
Carlos, F. M., and Barbosa, C. C. F.: Mapping mining areas in
the Brazilian Amazon using MSI/Sentinel-2 imagery (2017), Re-
mote Sens., 10, 1178, https://doi.org/10.3390/rs10081178, 2018.
MAAP: MAAP #219: Illegal mining expansion in the Ecuadorian
Amazon (Punino area), Monitoring of the Andes Amazon Pro-
gram, https://www.maapprogram.org/maap-219-illegal-mining-
expansion-in-the-ecuadorian-amazon-punino-area/ (last access:
23 January 2026), 2024.
MAAP: MAAP #230: Mining in the Ecuadorian Amazon, Cen-
tral Sector – Napo Province, Monitoring of the Andes Amazon
Program, https://www.maapprogram.org/ecuador-mining-napo/
(last access: 23 January 2026), 2025.
McFeeters, S. K.: The use of the Normalized Differ-
ence Water Index (NDWI) in the delineation of open
water features, Int. J. Remote Sens., 17, 1425–1432,
https://doi.org/10.1080/01431169608948714, 1996.
Mena-Quintana, F. N., Álvarez, W., Franco, W., Moncayo, L.,
Tipán, M., and Ayala, J.: Land degraded by gold mining
in the Ecuadorian Amazon: a proposal for boosting ecosys-
tem restoration through induced revegetation, Forests, 16, 372,
https://doi.org/10.3390/f16020372, 2025.
Mora-Silva, D. and Coronel-Espinoza, B.: Minería aurífera arte-
sanal en la Amazonía norte del Ecuador: gestión e im-
pactos socioambientales en la parroquia El Dorado de Cas-
cales, provincia de Sucumbíos, Green World J., 4, 3,
https://doi.org/10.53313/gwj42003, 2021.
Navas, L. O., Robles, P. M., and Mishquero-Ullauri, D.: Conflictos
e impactos generados por minería: una amenaza al territorio de
la comunidad indígena Cofán de Sinangoe, Sucumbíos–Ecuador,
Green World J., 3, 1–12, https://www.greenworldjournal.com/
index.php/ambiental/es/article/view/317/298 (last access: 10
November 2026), 2020.
Ngom, N. M., Baratoux, D., Bolay, M., Dessertine, A., Sa-
ley, A., Baratoux, L., Mbaye, M., Faye, G., Alphonse, Y .,
and Kouamé, K.: Artisanal exploitation of mineral resources:
remote sensing observations of environmental consequences,
social and ethical aspects, Surv. Geophys., 44, 225–247,
https://doi.org/10.1007/s10712-022-09740-1, 2023.
Peck, M. R., Desselas, M., Bonilla-Bedoya, S., Redín, G., and
Durango-Cordero, J.: The conflict between rights of nature and
mining in Ecuador: implications of the Los Cedros Cloud Forest
case for biodiversity conservation, People Nat., 6, 1096–1115,
https://doi.org/10.1002/pan3.10615, 2024.
Pekel, J.-F., Cottam, A., Gorelick, N., and Belward, A.
S.: High-resolution mapping of global surface wa-
ter and its long-term changes, Nature, 540, 418–422,
https://doi.org/10.1038/nature20584, 2016.
Phan, T. N., Kuch, V ., and Lehnert, L. W.: Land cover classi-
fication using Google Earth Engine and random forest classi-
fier – the role of image composition, Remote Sens., 12, 2411,
https://doi.org/10.3390/rs12152411, 2020.
Plan V: Colombian and Ecuadorian criminal groups expand
their violent influence throughout the Amazon, InfoAmazo-
nia, https://infoamazonia.org/en/2024/06/12/colombian-and-
ecuadorian-criminal-groups-expand-their-violent-influence-
throughout-the-amazon (last access: 29 March 2026), 12 June
2024.
Protschky, S. and Morgan, R. A.: Historicising sulfur mining,
lime extraction and geotourism in Indonesia and Australia, Extr.
Ind. Soc., 8, 100881, https://doi.org/10.1016/j.exis.2021.02.001,
2021.
Puertas, M.: El Alto Nangaritza, otro bosque virgen ase-
diado por la minería en Ecuador, Mongabay Latam,
https://es.mongabay.com/2018/09/el-alto-nangaritza-otro-
bosque-virgen-asediado-por-la-mineria-en-ecuador/ (last access:
14 January 2026), 2018.
Rouse Jr., J. W., Haas, R. H., Schell, J. A., and Deering, D.
W.: Monitoring vegetation systems in the Great Plains with
ERTS, in: Proceedings of the Third Earth Resources Technol-
ogy Satellite-1 Symposium, V ol. 1, Section A, NASA Goddard
Space Flight Center, 309–317, https://ntrs.nasa.gov/citations/
19740022614 (last access: 28 January 2026), 1974.
Roy, B. A., Zorrilla, M., Endara, L., Thomas, D. C., Van-
degrift, R., Rubenstein, J. M., Policha, T., Rios-Touma,
B., and Read, M.: New mining concessions could
severely decrease biodiversity and ecosystem services
in Ecuador, Trop. Conserv. Sci., 11, 1940082918780427,
https://doi.org/10.1177/1940082918780427, 2018.
Roy, D. P., Huang, H., Houborg, R., and Martins, V . S.: A global
analysis of the temporal availability of PS high spatial resolution
multi-spectral imagery, Remote Sens. Environ., 264, 112586,
https://doi.org/10.1016/j.rse.2021.112586, 2021.
Sellers, C., Ammirati, L., Khalili, M. A., Buján, S., Rodas, R. A.,
and Di Martire, D.: The use of DInSAR techniques for the study
of land subsidence associated with illegal mining activities in
Zaruma, Ecuador, in: European Workshop on Structural Health
Monitoring, 553–562, Springer, https://doi.org/10.1007/978-3-
031-07322-9_56, 2022.
United Nations Population Fund (UNFPA): Ecuador population
2025, https://www.unfpa.org/data/world-population/EC (last ac-
cess: 5 February 2026), 2025.
Villa, J., Aguilar, C., Villacís, S., Finer, M., and Josse, C.: Min-
ing activity in the Protective Forest Cuenca Alta del Río
Villa, J., Aguilar, C., Villacís, S., Finer, M., and Josse, C.:
Mining activity in the Protective Forest Cuenca Alta del Río
Nangaritza (Ecuador), MAAP Report No. 167, Monitoring of
the Andes Amazon Program, https://www.maapprogram.org/
es/minera-nangaritza-ecuador/ (last access: 23 January 2026),
2022.
https://doi.org/10.5194/eo-1-177-2026 Earth Obs., 1, 177–194, 2026
194 I. Lammers et al.: Remote sensing capabilities of detecting gold mining dynamics
Villacís, B. and Carrillo, D.: País atrevido: la nueva cara
sociodemográfica del Ecuador, 1st edn., Instituto Na-
cional de Estadística y Censos (INEC), Quito, Ecuador,
https://www.ecuadorencifras.gob.ec/wp-content/descargas/
Libros/Economia/Nuevacarademograficadeecuador.pdf (last
access: 9 November 2026), 2012.
Wunderling, N., Donges, J. F., Kurths, J., and Winkelmann, R.: In-
teracting tipping elements increase risk of climate domino ef-
fects under global warming, Earth Syst. Dynam., 12, 601–619,
https://doi.org/10.5194/esd-12-601-2021, 2021.
Zhang, B., Wu, D., Zhang, L., Jiao, Q., and Li, Q.: Applica-
tion of hyperspectral remote sensing for environment mon-
itoring in mining areas, Environ. Earth Sci., 65, 649–658,
https://doi.org/10.1007/s12665-011-1112-y, 2012.
Earth Obs., 1, 177–194, 2026 https://doi.org/10.5194/eo-1-177-2026
论文原文与出版信息
Earth Observation