Berkley Pit 

Remote Sensing for The Berkely Pit: Using Remote Sensing to view the effects of the pit

Jake Echols

Institute for Environmental & Spatial Analysis, University of North Georgia 4018 Mundy Mill Rd Oakwood GA 30566 (jlecho1432@ug.edu)

Keywords: Remote Sensing, Berkeley Pit, Environmental Monitoring, Land Cover Change, Water Quality, Sentinel-2, Landsat

ABSTRACT

A growing lake of acidic water overlooks the city of Butte, Montana. Once an open-pit copper mine, the Berkeley Pit is now a Superfund site with long-term contamination issues that have affected the surrounding landscape. This study used remote sensing to monitor environmental change over time. Landsat imagery from 1994 and 2024 was analyzed to detect changes in land cover. Sentinel-2 imagery was used to calculate NDVI and assess vegetation health, as well as NDWI to evaluate changes in surface water extent. Synthetic Aperture Radar (SAR) data from Sentinel-1 was used to detect changes in surface reflectance patterns related to urbanization and terrain smoothing. These multi-sensor analyses provided cost-effective, long-term monitoring of the region. Results revealed widespread vegetation gain, increased water presence near the pit, and urban change consistent with community remediation efforts. These findings contribute to a better understanding of the long-term environmental impact of the Berkeley Pit and the effectiveness of recent restoration strategies. 

1. INTRODUCTION

The area of interest for this project is the Berkeley Pit that is located in Butte, Montana. Butte has a population of approximately 35,000 people and stands as one of America’s most historic mining towns. In the late 19th century, the town was home to one of the richest copper mines in the world. However, in 1982, the mining was closed for good.

A century prior, this location was known as “the richest hills on earth.” It was called the Anaconda Copper Mine. In the 1950s the method of mining changed from digging tunnels to just excavating the entire area and turning it into a large open pit. It was dug so deep that you could fit the entire One World Trade Center standing up in it. However, due to rain the pit filled up and the water became diluted with toxic acid. The water is extremely dangerous and kills any wildlife that gets too close.

In 2016 there was a major snowstorm in the area and a massive flock of geese were forced down into the pit where they all perished. Roughly 6,000 birds fly through the area each year due to being on an overlap of two major bird migration routes. For these birds, the Berkeley Pit would normally be an ideal resting spot. To help protect wildlife, Montana Resources has set up propane cannons that run on a timer 24 hours a day to help as a deterrent.

The rock surrounding the pit contains iron pyrite, which combined with oxygen and water creates sulfuric acid. Over the decades the water continued to grow, and Montana Resources had to start managing it to prevent further destruction of the town. They now have pumping stations where they pump out, treat, and discharge the water. Without maintaining the growth of the water, the protective water level would have already been reached, and the spread of its harmful effects would have reached the city and other protected areas.

The pit has also had major effects on the community of Butte. It was originally a major source of economic growth. Everyone either worked in the mining industry or worked to support it. Over the years, as the land expanded many homes and businesses were displaced to make room for the growing pit, especially during the 1950s-1960s. Later, as the pit became toxic, air and ground quality became a major concern. Many homes in the area must be cleaned regularly from centuries of toxic dust.

Remote sensing can see how some of the changes have taken place over time. There is a lot of data to be gathered by comparing older Landsat imagery to present day imagery. Through methods like NDVI and EVI, it shows areas that have been stressed from the pit and need recovery or replanting.

Eventually, this data can be used to help local and state agencies. It can help provide early indicators for environmental degradation that they wouldn’t be able to see from the ground. The local community of Butte will be able to use this information to help make informed decisions on land use and other city planning guidelines. This data will also serve as a support to public awareness and tourism. The pit gets regular tourists, and more information can enhance education and understanding of the ongoing risks. 

Figure 1. Study Area Map

2. LITERATURE REVIEW

After years of mining, open pit mining sites leave behind piles of waste called tailings. Tailings fill with harmful chemicals that can be detrimental to the surrounding environment. These chemicals can make it difficult for vegetation to grow. Trees that are able to grow are still not as healthy as trees in similar areas not affected by tailings. These trees can show specific signs that they have been affected by this leftover waste, such as smaller leaves, unusual coloration, and dead branches. Levesque et al researched how they could observe this without having to manually walk through the forest. Through remote sensing they were able to study which parts of the forest were damaged due to tailings and how badly. By studying patterns in satellite imagery, they hoped to discover a better way to check on forests affected by pollution.

         The researchers chose an area called KamKotia Mine in Ontario, Canada to study how the old mine site’ tailings were affecting the trees. This is a location where the soil and water have been polluted for many years. After the mine had closed rainwater would mix with the leftover waste piles which flowed down into the nearby forests. This water became acidic and harmful. In the initial research phase observations were made that the closer they got to the mine, the unhealthier the tress and vegetation appeared. For their test, they picked six spots in a straight line leading away from the mine. Each spot was farther away than the next and the last spot farthest away acted as the control.

         So, to test their hypothesis they compared each spot based on a list of factors; the number of trees, how tall the trees were, how big the tree crowns were, how much sunlight was getting through the forest, how healthy the trees looked, and a stress score for each tree.

         Their primary method of observation was a Kodak Megaplus 1.4 digital camera mounted to small engine aircraft. For their testing, instead of using common vegetation indices, they mainly focused on texture and what you could viably see. The imagery was collected in 1993 and again in 1995 to view the change over time. The images were taken from 1 meter per pixel to show larger areas to 0.25 meters per pixel to show smaller details.        In their research they used a technique called semivariance which was how they were able to measure the change in texture and shape throughout all the testing sites. They performed two different ways of measuring, the matrix method which involved using each pixel square to measure the overall texture and then also the transect method where they drew lines across the image to see changes from one spot to another.  

         In conclusion, after all their testing was completed, they found that their hypothesis had been correct that damage was much worse closer to the mine. Healthier trees appeared to have much more uniform textures, while the less healthy trees had much more variation.  They also concluded that arial images and semivariance analysis worked well in this type of research. It was useful for detecting tree health changes caused by acid min drainage.

Mines are created to extract specific minerals and natural resources from the earth. However, when these mines close down, they leave a scar on the landscape, Gordon et al did a study on the effects of these mines and how they recover back to a more natural state. Scientist used different types of sensors on satellites to collect imagery that provided the specialized data to study these areas over time, specifically closed mines in Canada. Through different remote sensing methods these researchers were able to see how quickly vegetation grew back.

         These areas where mines once operated have very unhealthy soil and water and it can take many years for nature to recover on its own. Reclamation is the process of helping speed up this process and bring a more natural landscape to the area. There are two main ways of reclamation, passive and active. Passive reclamations is allowing nature to fix itself without human intervention. Although this method is much cheaper, it take a very long time, and it doesn’t always work well. The other option is active reclamation, where humans plant new vegetation and provide needed nutrients to the land.  Over the past decades remote sensing has been able to track both these methods and study their results using hard data.  

         Multiple methods of remote sensing were used to observe and track this data.

First, the primary one used was NDVI where they were able to track the health and greenness of the vegetation. Second, they utilized the random forest method to help sort the different types of land use: forest, grassland, water, bare areas. Third, post-classification change detection allowed them to see how the land changed over time. Fourth, the regrowth index compared plant growth at old mine sites to plants in areas that were never disturbed. Fifth, linear regression was used to examine whether the vegetation is increasing or decreasing. Lastly, they collected imagery from reference sites to compare the mine sites to nearby healthy areas to see how well the plants were growing back.

         In the study Wapisiw Lookout Mine was observed. They looked at satellite imagery from 1997 to 2021. Using the remote sending methods listed above their studies showed that actively planting trees and greases around the site was a big increase in health plant cover proving the hypothesis that active reclamation efforts significantly speeds up the reclamation process. The results from Wapisiw Lookout showed that NDVI increased at a rate of 0.0121 per year compared to Pine Point Mine, a passive reclamation site, at a rate of 0.0011 per year.

Saputra et al researched how multi-modal deep learning can be used to automatically map mining and non-mining land cover areas across different locations around the world by using multispectral satellite imagery. When studying old mines, data collected from Landsat and Sentinel are used. Through different bands like true color and NIR, raster calculations such as NDVI are performed to help know specific information like vegetation health.

         Researchers use data augmentation to improve the quality of the images. Data augmentations allowed them to remove clouds and make the overall image quality sharper. The researchers used a supercomputer for deep learning which recognized patterns in the imagery. They tested different language learning models to see which ones performed the best at differentiating mine areas from non-mine areas. The models they used were U-Net, DeepLabV3+, FPN, SegFormer, and Prithvi. They wanted to find out if the programs could accurately classify different land use.

 This process is called semantic segmentation. An example of semantic segmentation in use is using an image of a mining pit and the program could identify different areas within the pit, like areas that are dug out, waste piles, water bodies, and vegetation. This method differs significantly from traditional techniques like point data classification or spectral classification. Instead of only using color differences, the computer model can learn patterns in the data, including shape, texture, and relationships between objects. However, for these programs to work well and accurately, they need a lot of training with detailed maps.

         The scientist in this study started by collecting data and making a classification system to sort each type of land into categories. To do this, they collected ground truth data by labeling 37 real mining areas around the world. Each mine that was chosen represented different types of mines, i.e. gold, coal, and silver, to help better train the computer.

All the data used came from different sources including previous studies that had already been labeled mining areas. In the study they made sure that all labels were carefully fixed and updated so they represented accurately. They even confirmed with mining experts with areas they were unsure.

         When the data was ready, they split it into two different groups, one for computer training, and the other to check if the computer was learning correctly.

         After the studies were completed, the researchers found that the Feature Pyramid Network with a DenseNet-121 backbone performed the best. The results were mostly successful however they did discover some challenges and limitations. The FPN moder was able to accurately map different types of mining land cover types. This was generally successful across multiple locations. Though, the model still struggled to differentiate between certain land cover types like mining waste areas and disturbed land. Arid and cold climate regions had worse results and lower accuracy.
        With more datasets for training, the model will only improve.

3. METHODS

To analyze this issue, four map products will be generated using a variety of remote sensing methods. Multiple sources of data will be leveraged for this exploratory analysis (Table 1).

3.1 Map 3: Land use and Land cover changes

The primary purpose of map 1 is to analyze the change in land use over the past few decades. It will see how the expansion of the urban areas has changed, and the environment has degraded.  The focus will be on highlighting the areas that have been most affected by human activity and the pit’s contamination.

Study Area:

The Berkeley Pit will be the center of the study area. The adjacent urban areas of Butte, and nearby natural landscapes and water bodies, will help show these changes to the study area.

Data Sources:

Landsat 5 and Landsat 8 (USGS Earth Explorer) – For historical and recent land use/land cover data (30m resolution).

Time Period:

The time period for Map 1 will focus on 08-15-1994 to 08-17-2024.

The major RS analysis and list of process steps include:

1. Data Acquisition

·         Download Landsat 8/9 imagery for the years 1994 and 2025 from USGS Earth Explorer

2. Preprocessing

·         Download ARD (analysis ready data)

·         Clip imagery to the study area for focused analysis.

3. Geoprocessing Steps

·         Land Cover Classification:

o    Use supervised classification (Random Forest or Maximum Likelihood Classification) to classify land into key categories—water, vegetation, barren land, and urban areas.

·         Change Detection Analysis:

o    Conduct change detection by comparing classified land cover maps from 2015 and 2025 to identify significant changes (raster differencing).

4. Visualization & Map Composition

·         Create a well-formatted map layout in ArcGIS Pro, including:

o    Scale bar, legend, north arrow, title, and map elements to ensure clear communication.

o    Integrate change detection results with accompanying descriptive statistics and summary analysis.

·         Make sure that the final product highlights the key findings with color-coded categories and clear map labels for easy interpretation.

3.2 Map 2: Vegetation analysis

The Map 2 is going to be focusing on the health of the vegetation and environmental stress analysis around the Berkeley Pit. The project will use vegetation indices to see how the vegetation has been impacted by contamination and environmental changes over time. This will show areas of decline and potential ecological recovery.

Study Area:

The Berkeley Pit will be the center of the study area along with the surrounding vegetated areas.

Data Sources:

·         Sentinel-2 Copernicus

Time Period:

The time period for Map 2 will focus on 09-22-2015 to 09-09-2024.

The major RS analysis and list of process steps include:

1. Data Acquisition

·         Download Sentinel-2 imagery from ESA Copernicus Data for 2015 and 2024.

2. Preprocessing

·         Clip imagery to the study area (Berkeley Pit and surrounding vegetation).

3. Geoprocessing Steps

·         Vegetation Indices Calculation:

o    Use ArcGIS Pro’s NDVI Raster Function to calculate Normalized Difference Vegetation Index for each time period. NDVI = (NIR – Red) / (NIR + Red)

·         Multi-Temporal Analysis:

o    Compare NDVI values between 2015 and 2024 to detect changes in vegetation health and highlight areas of significant decline or recovery.

4. Visualization & Map Composition

·         Create a well-formatted map layout in ArcGIS Pro, including:

o    NDVI change map, color-coded to show increases and decreases in vegetation health.

o    Legend, scale bar, north arrow, title, and detailed description of results.

·         Use charts and statistics to support the visual representation and describe key trends

3.3 Map 3: Surface water extent and water reflectance

Map 3 will focus on monitoring surface water extent and detecting changes in water reflectance patterns in the Berkeley Pit and surrounding water bodies. The project will track how surface water has changed over time and point out the different indicators that the contamination of the pit has had.

Study Area:

The Berkeley Pit will be the center of the study area along with the surrounding bodies of water.

Data Sources:

·         Sentinel-2 Copernicus

Time Period:

The time period for Map 3 will focus on 09-22-2015 to 09-09-2024

The major RS analysis and list of process steps include:

1. Data Acquisition

·         Download Sentinel-2 imagery from ESA Copernicus Data Space Ecosystem for the years 2015 and 2025.

2. Preprocessing

·         Clip imagery to the study area (Berkeley Pit and nearby water bodies).

3. Geoprocessing Steps

·         Surface Water Mapping:

o    Use the Normalized Difference Water Index (NDWI) to identify water bodies for each time period. NDWI = (NIR – SWIR) / (NIR + SWIR)

o    Create binary water masks (water vs. non-water) to delineate surface water extent.

·         Change Detection Analysis:

o    Conduct a multi-temporal analysis to detect changes in surface water extent between ~2015 and ~2025 using raster differencing.

4. Visualization & Map Composition

·         Create a map layout in ArcGIS Pro, including:

o    Surface water change map with color-coded areas showing changes over time.

o    Legend, scale bar, north arrow, title, and annotations explaining key findings.

·         Summary statistics should be used to see changes in water extent values.

3.4 Map 4: Impact of the Berkeley Pit on urban development

Map 4 will look at how the Berkley pit has overall affected the urban environment of Butte. Using Synthetic Aperture Radar, the project will track urban expansion and see how the pit may have influenced how the city has grown and used the land of urban development. 

Study Area:

The Berkeley Pit will be the center of the study area along with the surrounding urban area of Butte, Montana.

Data Sources:

·         Sentinel-1 Copernicus – C-band SAR data

Time Period:

The time period for Map 4 will focus on 08-04-2016 to 09-06-2024

The major RS analysis and list of process steps include:

1. Data Acquisition

·         ESA Copernicus

·         Obtain Sentinel-1 SAR data

2. Preprocessing

·         Perform speckle detection ArcGIS Pro for consistency.

·         Clip imagery to the study area, focusing on Butte and the surrounding urban environment.

3. Geoprocessing Steps

·         Built-Up Area Detection:

o    Apply SAR data to detect structural changes.

4. Visualization & Map Composition

·         Create a map layout in ArcGIS Pro, including:

o    Built-up area change map, highlighting urban expansion trends and patterns.

o    Legend, scale bar, north arrow, title, and annotations explaining key findings.

o    Use summary statistics to provide insights into the extent of urban growth and its relationship with the Berkeley Pit.

4. RESULTS & DISCUSSION

4.1 Map 1: Land use and Land cover changes

Supervised classification of Landsat imagery was used to show widespread land cover change from 1994 to 2024 around the Berkeley pit. Red areas indicate change, while the gray shows no change. Expansion of barren land and shifts in vegetation near the pit reflect environmental degradation. Dual pit outlines show the spatial growth of the contaminated area.

4.2 Map 2: Vegetation analysis

This map shows vegetation change using NDVI from Sentinel-2 imagery. While urban areas display widespread vegetation loss, which is likely due to an exceptionally dry summer in 2024, The area directly surrounding the pit shows a stable to moderate gain. This would align with recent community-led revegetation and improved water treatment efforts. NDVI loss in developed areas likely reflects short-term drought stress rather than long-term degradation

4.3 Map 3: Surface water extent change detection

The NDWI index show that water has increased north of the pit. This is  Yankee Doodle Tailings Impoundment. It is an engineered surface water and tailings management system designed to capture and runoff and control sediment. The NDWI shows the effectiveness of the site and the increase of drainage and storage expanding the surface water. The Pit itself shows slightly reduced water reflectance. This indicates that ongoing pumping and treatment efforts are working.

4.4 Map 4: Impact of the Berkeley Pit on urban development

The surface reflectance in this SAR image shows changes from 2016 to 2024. The red areas indicate increased backscatter which usually show that there has been some type of urban development or hardened surfaces. The blue areas suggest vegetation growth or surface smoothing. The Berkeley pit doesn’t show much change which is to be expected. The scatter of red around Bute is mostly near or around I15 which is common for American cities. The changes in the mountainous areas are most likely due to radar distortion and snow and not urban development. 

5. CONCLUSIONS

The findings represent exactly what has been going on in the area over the past few decades. Everything the local community has done to improve the area, like planting more vegetation, cleaning the water, and pumping it out, can be seen in the images. Exploring these links to human activity and land management strategies might be something that future work would benefit from.

Figure 6 shows the total geographic area (in km²) that experienced a specific type of change. land cover (1994–2024), vegetation (2015–2024), water (2015–2024), and surface change (2016–2024). The largest transformation is from vegetation gain and loss where a significant area of land changed. This is likely due to the drought in 2024 and the replanting initiative.

REFERENCES

Gordon, S., Xu, X., & Wang, Y. (2023). Remote sensing-based revegetation assessment at post-closure mine sites in Canada. Sustainability, 15(11287). https://doi.org/10.3390/su151411287

Lévesque, J., and King, D.J. (1999). Airborne digital camera image semivariance for evaluation of forest structural damage at an acid mine site. Remote Sensing of Environment, 68(2), 112–124. https://doi.org/10.1016/S0034-4257(98)00104-7

Saputra, M. R. U., Bhaswara, I. D., Nasution, B. I., Li Ern, M. A., Husna, N. L. R., Witra, T., Feliren, V., Owen, J. R., Kemp, D., & Lechner, A. M., (2025). Multi-modal deep learning approaches to semantic segmentation of mining footprints with multispectral satellite imagery. Remote Sensing of Environment, 318, 114584. https://doi.org/10.1016/j.rse.2024.114584