Determining Mining Rates in Karamoja, Uganda

Author

Remote sensing team

Published

August 3, 2024

Classification of LULC using Sentinel 2 data


Classification using Sentinel 2 for 2017,2018, 2019,2020,2021,2022, and 2023. This clasification is done in Google Earth Engine and Google colab to determine the mining rates in the chosen 9 mining sites in Amudat, Kaboong, Moroto and Nakipiripirit.

The process:

Data preparation:

  • We create a training dataset for supervised classification in Google Earth Pro corresponding to different land cover classes like tree cover, other vegetation (shrublands, grasslands, croplands), bare soil, built-up areas, water and herbaceaus wetland, mining and other classes altogether (michen, moss, ice, snow).
  • Preprocessing steps include masking unwanted pixels (e.g., clouds, cloud shadows), scaling reflectance values, and calculating various remote sensing indices like normalized difference vegetation index, normalized difference built-up area index, urban index, etc. using Sentinel 2 images.

Supervised Classification:

  • Different supervised classification algorithms are used, including Random Forest (RF), Classification and Regression Trees (CART), and Gradient Tree Boosting (GTB).
  • These algorithms learn patterns in the training dataset to classify pixels in the satellite imagery into different land cover classes.
  • For instance, the Random Forest algorithm in which we’ll use for now builds multiple decision trees using subsets of the training data and random subsets of input features. Each tree “votes” on the class of a pixel, and the class with the most votes is assigned to the pixel.
  • Random Forest is an ensemble learning method that constructs multiple decision trees during training.
  • Each decision tree is built using a random subset of the training data and a random subset of the input features.
  • During classification, each tree “votes” on the class of a pixel based on the features it observes.
  • The final class assigned to the pixel is determined by the majority vote among all trees.
  • Random Forest is robust against overfitting and generally provides accurate results for classification tasks, especially in remote sensing applications.

Visualization and Export:

  • The classified images are visualized on the map to help assess the accuracy of the classification visually.
  • Classified images are exported to Google Drive for further analysis or sharing with others.

Extracting the areas:

  • We calculate areas under each land cover class within each mining site boundaries in hectares.

The collected training data comprises:

  • Bare land, encompassing barren surfaces and murram roads.
  • Built-up areas, incorporating manyattas (semi-permanent structures), permanent village settlements, market buildings, industrial zones, places of worship, and structures adjacent to/or are the mining sites.
  • Mining areas, including quarries and excavation sites.
  • Vegetation, consisting of tree cover, shrublands, grasslands, and cultivated fields.
  • Water bodies, comprising permanent and seasonal features such as rivers, lakes, dams, and reservoirs.

Note: values in the data that are missing (NaN) or zero means that class was not detected and the resulting mining rate will be NaN or infinity (inf), respectively.

Variables in the output data

As we refine our classification process to better distinguish the classes, we merge the extracted regions of mining and built-up areaa. This merging might stop as we continue to enhance our classification techniques. The data below shows the areas in hectares occupied by each class in the years 2017 to 2023.

Column Name Explanation
site_name mining site’s name
Latitude mining site’s Latitude
Longitude mining site’s Longitude
geometry parameter polygon shape around Long-Lat with buffer of 5 kilometers
total_lulc_area total area of the buffer
{year}_vegetation total area under tree cover, vegetation (grasslands, shrublands & croplands)
{year}_bare total area under bare land
{year}_builtup_areas total area under built-up areas
{year}_mining total area under mining
{year}_water total area under water and herbaceaus wetland
{year}_mining_bare total area under mining and bare land
Observations
  1. Dry Season:
    1. Kaboong Site One: Minimal mining activity is observed, indicating limited operations.
    2. Moroto Sites: A noticeable increase in mining activities is observed around 2020, especially in Moroto Site Two. A slight decline in mining activity post-2020 suggests some stabilization or reduced operations.
    3. Nakapiripirit Sites: Relatively stable mining operations with minor fluctuations across the years.
  2. Rainy Season:
    1. Kaboong Site One: Consistent minimal mining activity, similar to the dry season.
    2. Moroto Sites: Mining activities peaked around 2020, with fluctuations indicating varying levels of activity.
    3. Nakapiripirit Sites: Low levels of mining, showing controlled operations with minor variations.
  3. Yearly Average:
    1. Kaboong Site One: Maintains a consistently low mining level throughout the year.
    2. Moroto Sites:Significant increase around 2020, aligning with trends in both dry and rainy season. Post-2020 stabilization with minor fluctuations.
    3. Nakapiripirit Sites: Stable mining operations with slight variations.
Key Insights
  • Expansion in Moroto: Both Moroto sites experienced significant growth in mining activities, particularly around 2020. Monitoring and sustainable practices are recommended.
  • Kaboong sites maintain minimal mining activity.
  • Nakapiripirit sites demonstrate stable mining practices with consistent low levels of activity.
  • Amudat Sites: Low mining activity overall, with slight variations
Compare the different seasons
  1. Kaboong Sites:
  • Kaboong site one - mining activities remain low across all seasons with minimal variation year to year.
  • Both dry and rainy seaons treands align closely with yearly averages.
  1. Moroto Sites:
  • Moroto site one - Significant increases in mining activities during 2020 for both dry and rainy seasons, with a noticeable peak in the rainy season.
  • Moroto site two - Mining activities are more pronounced during rainy seasons, especially in 2020 and 2021. Slight decrease post-2020, with yearly trends following the rainy season closely.
  • Moroto site three - Peaks observed in 2020 for rainy and dry seasons, followed by stabilization.
  1. Nakapiripirit Sites:
  • Nakapiripirit site one - Minimal changes across all seasons.
  • Nakapiripirit site two - very low activity and consistent across all years and seasons, indicating controlled operations.
  1. Amudat Sites:
  • Amudat site one - Minor increases in activity post-2020 in both dry and rainy seasons.
  • Amudat site two - Minimal activity across all years and seasons, indicating limited mining operations.
Site 2017_rainy_bare 2017_rainy_builtup 2017_rainy_mining 2017_rainy_water 2017_rainy_vegetation 2018_rainy_bare 2018_rainy_builtup 2018_rainy_mining 2018_rainy_water 2018_rainy_vegetation 2019_rainy_bare 2019_rainy_builtup 2019_rainy_mining 2019_rainy_water 2019_rainy_vegetation 2020_rainy_bare 2020_rainy_builtup 2020_rainy_mining 2020_rainy_water 2020_rainy_vegetation 2021_rainy_bare 2021_rainy_builtup 2021_rainy_mining 2021_rainy_water 2021_rainy_vegetation 2022_rainy_bare 2022_rainy_builtup 2022_rainy_mining 2022_rainy_water 2022_rainy_vegetation 2023_rainy_bare 2023_rainy_builtup 2023_rainy_mining 2023_rainy_water 2023_rainy_vegetation
0 Amudat_site_one 3143.65 0.01 176.01 NaN 4433.83 6306.39 107.74 211.60 NaN 1127.77 5451.80 145.31 95.55 NaN 2060.84 1897.66 41.51 45.14 NaN 845.35 3310.17 9.12 96.97 NaN 4337.24 4405.86 70.61 299.49 NaN 2977.54 5008.47 323.93 62.13 NaN 2358.97
1 Amudat_site_two 2517.33 33.41 116.02 NaN 5087.88 3574.96 137.85 130.92 0.18 3910.73 3259.50 142.74 26.85 0.19 4325.36 3360.78 102.13 59.22 4.71 4227.80 3653.35 218.54 46.97 7.46 3828.32 2981.81 190.85 140.64 5.66 4435.68 3557.13 52.15 53.07 0.09 4092.20
2 Kaboong_site_one 7.50 NaN NaN NaN 7747.04 6.90 0.26 NaN NaN 7747.38 63.30 0.94 NaN NaN 7690.30 489.36 14.69 9.17 0.08 7241.24 1717.37 0.93 0.02 NaN 6036.22 1623.91 0.35 0.03 NaN 6130.25 7.30 0.13 0.11 NaN 7747.00
3 Kaboong_site_two 5.35 0.57 0.01 NaN 7748.62 16.96 1.43 0.04 NaN 7736.12 138.50 4.18 2.52 NaN 7609.35 171.13 28.55 44.02 NaN 7510.85 398.14 0.17 0.02 NaN 7356.22 478.58 0.31 1.96 NaN 7273.70 5.85 0.39 0.32 NaN 7747.99
4 Moroto_site_one 1586.92 35.32 21.71 NaN 6110.67 3189.53 37.37 14.84 NaN 4512.88 3111.20 250.24 8.19 0.08 4384.91 2992.04 82.09 12.94 0.10 4667.45 1828.67 48.72 8.95 0.16 5868.12 2885.28 29.06 21.49 0.41 4818.38 2511.52 78.69 17.05 0.16 5147.20
5 Moroto_site_three 4075.39 124.89 25.76 NaN 3528.53 3461.46 138.89 38.52 0.03 4115.67 3128.56 129.41 26.82 NaN 4469.78 3469.64 92.40 81.36 0.90 4110.27 2014.55 26.96 19.92 0.05 5693.09 2210.09 15.90 42.71 0.04 5485.83 1610.20 44.16 13.35 0.04 6086.82
6 Moroto_site_two 4116.72 174.23 15.31 0.18 3447.06 3487.35 44.51 18.38 NaN 4203.26 3242.43 106.10 29.90 2.15 4372.92 2828.22 53.86 49.82 2.93 4818.67 1729.42 14.05 52.94 2.37 5954.72 2576.93 15.58 76.80 0.99 5083.20 2343.65 24.26 54.12 0.80 5330.67
7 Nakapiripirit_site_one 35.13 1.80 0.47 NaN 7716.12 314.40 7.46 5.16 NaN 7426.50 997.29 9.39 0.11 NaN 6746.73 983.49 54.18 74.65 6.20 6635.00 398.06 0.17 0.24 NaN 7355.05 137.41 0.02 0.12 NaN 7615.97 13.68 0.09 0.17 NaN 7739.58
8 Nakapiripirit_site_two 137.21 9.51 41.95 0.59 7565.23 483.90 33.71 46.21 0.80 7189.87 1190.17 25.41 35.12 0.55 6503.24 3988.54 109.35 93.98 2.51 3560.11 765.92 7.46 40.58 0.71 6939.82 1120.14 39.95 74.64 0.63 6519.13 638.00 70.73 52.44 0.47 6992.85

Progress of mining activities and their rate of change over time