Content uploaded by Ewa Gromny
Author content
All content in this area was uploaded by Ewa Gromny on May 30, 2022
Content may be subject to copyright.
Framework:
•Area of interest is located in North-western Tanzania,
in the Kigoma region of the Kakonko district (30°53'15"E and 3°25'36"S)
•The camp has been established in 2016 and closed at the turn of 2020 and 2021
Objectives:
•The aim is to investigate the interaction between the environmental changes
and refugee camp inhabitants
•We are checking if there is a spatial pattern of changes as a function of the
distance from the camp. Analysis are performed depending on the distance
from the camp - 5 km buffers
•Development of unified classification methodology for each year of the time
series (2016 - 2020)
Challenges:
•LC legend Definition complexity: fulfilling end user requirements, taking into
consideration local environmental peculiarities and existing LC databases.
•The structure of landscape is not well defined. Land cover classes appearing in
the area of interest do not have a clear boundaries between each other,
especially vegetation classes such as grass, shrub and tree cover
•Wild fires occurring every dry season
Land cover change analysis
around the Mtendeli refugee camp in Tanzania
Gromny E.1, Haarpaintner J.2, Aleksandrowicz S.1, Jenerowicz-Sanikowska M.1, Woźniak E.1,
Pesquer L.3, Chułek M.4, Starczewski D.5, Pawlak Z.5
1CBK PAN Space Research Centre of the Polish Academy of Sciences, 2NORCE –Norwegian Research Centre, 3GRUMETS Research Group, CREAF,
4Centre of Migration Research (CMR), University of Warsaw (UW), 5UNEP/GRID-Warsaw Centre,
Research performed under ARICA project: NOR/POLNOR/ARICA/0022/2019-00, funded under the Norway Grants, POLNOR2019, co-financed from the State budget, Applied Research.
The European Union’s Horizon 2020 research and innovation programme under EOTIST project, grant agreement No 952111
Summary:
•Classification legend defined based on local experts knowledge,
existing databases and the characteristics of the area of interest
•The use of the mask of burned areas allowed to overcome
difficulties connected with obstacles related to seasonal wild fires
•Developed methodology allows to achieve overall accuracy for LC
2020 map - 83,5%
•The most problematic class is shrub, due to its similarity and
not clear boundary with grass and tree cover
•Presence of refuges decreases the quality of environment
•The amount of classes connected with human activity such as
croplands and artificial surfaces increases, while the natural
vegetation classes area is decreasing
Project website: https://arica.gridw.pl/en/
•The amount of changes decreases with the distance from the camp
•The most relevant transitions differ depending on the distance
form the camp
•Relevance of transition from croplands and shrub to artificial
surfaces increases as the distance to the camp decreases
•Transition from shrub to croplands is the most visible change
and increases as the distance to the camp decreases
Artificial
surfaces
Croplands
Vegetation
regularly
flooded
Tree cover
Grasslands
Shrub
Bare areas
Open Water
permanent
Total
User's
accuracy
Artificial surfaces 31 5 0 0 1 1 16 054 57%
Croplands 2 564 4 1 19 17 1 0 608 93%
Vegetation regularly
flooded 0 5 71 2 2 2 0 2 84 85%
Tree cover 0 1 4 441 21 51 0 0 518 85%
Grasslands 0 49 13 6346 25 0 0 439 79%
Shrub 0 18 030 47 185 0 0 280 66%
Bare areas 0 0 0 0 0 0 42 042 100%
Open water permanent
0 0 1 0 0 0 0 66 67 99%
Total
33 642 93 480 436 281 59 68 2092
Producer's accuracy
94% 88% 76% 92% 79% 66% 71% 97% OA 83,5%
f score
0,71 0,90 0,80 0,88 0,79 0,66 0,83 0,98
Land cover class description Training samples
Artificial surfaces
Areas with artificial abiotic constructions or artificial
materials, Includes shelters for refugees
Vito LC database 2020
filtered with NDVI, manual
samples
Croplands
Areas where different types of crops are being cultivated
WorldCover erode 5x5
Vegetation regularly
flooded
Areas alongside seasonal rivers, temporarily flooded,
mostly cultivated
WorldCover erode 3x3,
manual samples
Tree cover
Areas covered by tree canopy
WorldCover erode 5x5
Grasslands
Areas covered by natural grasses
WorldCover erode 5x5
Shrub
Areas covered by shrubs
WorldCover erode 5x5
Bare areas
Uncovered areas, Includes e.g. bare soil, bare rock, roads
manual samples
Open Water permanent
Water bodies; Water present in the dry season
manual samples
1,0% 1,0% 1,1% 0,7% 1,0%
26,4% 27,8% 29,6% 29,9% 29,5%
2,7% 2,6% 2,0% 2,4% 2,3%
34,1% 32,5% 31,0% 31,3% 30,7%
18,2% 19,0% 18,2% 19,4% 19,7%
17,5% 17,1% 18,0% 16,4% 16,7%
2016 2017 2018 2019 2020
10 - 15 km
0,7% 0,7% 0,9% 0,4% 0,7%
20,3% 22,3% 24,4% 25,0% 24,0%
2,2% 2,3% 1,9% 2,2% 2,4%
36,8% 35,4% 32,1% 35,6% 34,1%
18,8% 21,1% 19,9%
21,3% 20,5%
21,2% 18,2% 20,8% 15,3% 18,2%
2016 2017 2018 2019 2020
1,5% 1,7% 1,6% 1,3% 1,7%
27,3% 29,9% 35,7% 39,1% 38,3%
1,3% 1,1%
0,9% 1,4% 1,3%
26,4% 25,5% 20,2%
23,7% 19,4%
23,8% 26,8% 22,8%
22,9%
21,9%
19,7% 15,0% 18,7% 11,6% 17,3%
2016 2017 2018 2019 2020
10,6%
26,4% 21,7% 22,1% 18,0%
45,3%
39,3% 49,8%
57,7% 58,9%
1,2%
0,3%
0,8%
0,8% 1,5%
18,2%
10,7% 6,1%
9,2% 6,0%
3,1% 4,6% 2,3%
2,5%
2,9%
21,2% 16,7% 17,4%
6,0% 11,2%
0,3% 2,0% 1,9% 1,5% 1,5%
2016 2017 2018 2019 2020
25,15% 28,11% 37,03% 44,44%
0%
20%
40%
60%
10 - 15 km 5 - 10 km 0 - 5 km Camp
The Amount of changes vs the distance from the camp
Analysis and Figures on the poster contain modified Copernicus Sentinel-1/2 data (2016-2020).
Analysis of changes depending on the distance from the camp
The percentage of occurrence of a given class year by year
The most relevant transitions from 2016 to 2020
Distance from the camp
10 - 15 km 5 -10 km 0 -5 km camp
Fig. 2 Location of the Mtendeli camp in Africa, Fig. 4 Land cover map of the Mtendeli camp for the year 2020
Pixel Based
Classification (RF)
Aggregation of
classifications within
one year
Random selection of training
pixels for each image based on the
burned and cloud areas mask
Thematic
transformation
Original S2 data
Selection of 10m,
and 20m bands
LC classes
legend
Existing LC databases
Stacks: S-2 band for single
date, S-1 quarterly means,
burned and cloud areas mask
Reference data
Final yearly
classification
Yearly Automated Land Cover Classification
Original S1 data
Manual
samples
Burned area
masking
Validation of single
classifications
Automatic selection of
the best classifications
Quarterly means
Sen2Cor correction
& cloud masking
Fig. 1 Yearly land cover maps, 2016 - 2020
2016 2020 Camp
Shrub Croplands 12,69%
Croplands Grasslands 0,67%
Grasslands Croplands 2,51%
Tree cover Shrub 6,42%
Shrub Grasslands 0,64%
Grasslands Shrub 0,10%
Shrub Tree cover 0,69%
Tree cover Grasslands 1,32%
Tree cover Croplands 4,46%
Croplands Artificial surf. 7,87%
Shrub Artificial surf. 2,59%
2016 2020 0 - 5 km
Shrub Croplands 6,78%
Croplands Grasslands 2,89%
Grasslands Croplands 6,08%
Tree cover Shrub 5,60%
Shrub Grasslands 2,25%
Grasslands Shrub 1,93%
Shrub Tree cover 2,14%
Tree cover Grasslands 2,03%
Tree cover Croplands 2,47%
Croplands Artificial surf. 0,61%
Shrub Artificial surf. 0,08%
2016 2020 5 - 10 km
Shrub Croplands 4,25%
Croplands Grasslands 2,90%
Grasslands Croplands 2,68%
Tree cover Shrub 3,23%
Shrub Grasslands 3,04%
Grasslands Shrub 2,47%
Shrub Tree cover 2,51%
Tree cover Grasslands 1,64%
Tree cover Croplands 1,08%
Croplands Artificial surf. 0,25%
Shrub Artificial surf. 0,08%
2016 2020 10 - 15 km
Shrub Croplands 3,87%
Croplands Grasslands 3,48%
Grasslands Croplands 3,16%
Tree cover Shrub 2,77%
Shrub Grasslands 1,61%
Grasslands Shrub 1,36%
Shrub Tree cover 1,32%
Tree cover Grasslands 1,28%
Tree cover Croplands 1,17%
Croplands Artificial surf. 0,36%
Shrub Artificial surf. 0,10%
Fig. 3 Mtendeli camp and the circular buffers:
camp, 0 - 5 km, 5 -10 km, 10 –15km
For more details about other refugee camps investigated in ARICA project,
see Haarpaintner et al. [ESA LPS poster 65239, 25 May 2022]
0 -5 km
5 - 10 km
10 -15 km