EARTH OBSERVATION / GIS





- The oil fields of Sudan / Southern Sudan -


Summary of Prins peer reviewed research results - covering latest advances and monitoring system


Background and Introduction

In 2006 Prins was commissioned by ECOS to investigate land use changes in Sudans Oil Block 5a with particular focus on the 1999-2003 period. Previously, Prins had investigated Oil Block 3 and 7 for ECOS with the same objective: to map the oil facilities and the traditional land use pattern covering 1999-2003. A cornerstone in these analysis have been Landsat data. Quality Landsat data has been available since the mid 80'ties and has ever since continuously monitored the earth with multi spectral image data. This means that the recorded areas are monitored into seven different light wavelengths (e.g., visible and Infared light areas) that in combinations or in models can tell us detailed information on what we are looking at. For example, crop types, water content, biomass or rangeland conditions etc. For the engaged layman - see also NASA description of Landsat data). For Sudan/South Sudan Prins has more than 25 yrs experience with using Landsat data (see more). For example, Prins has looked into a large part of the oil fields of Sudan / Southern Sudan. Prins has further uncovered the existence of the Darfur crisis that showed the Landsat potential for uncovering crime against humanity.

The 5a case has come a long way since it started in 2006. After the case reached international attention in 2010, I became interested in the robustness of the Landsat application for monitoring the footprint of the traditional farming system. At the same time Very High Resolution (VHR) or self-explaining satellite images were increasingly becoming available on Google Earth, which offered an alternative mean to verify Landsat analysis. Some of these results were presented at an invited talk to the Swedish Space days in Stockholm 2013 (presentation available down listed). In 2013-2015 the 5a area/Unity Sate was exposed to new humanitarian suffering. This resulted in a high amount of VHR images became available on Google Earth. This allowed a more systematic and comprehensive evaluation of the Landsat application, which was published in the International Journal of Remote Sensing in 2017. See more of the recent advances in Summary of Prins peer reviewed research results - including method development establishment of a  monitoring system.

In the following I will show how this work has unfolded. It will start off with a description of the traditional agro-pastoral land use which is key to understand the patterns that be derived from the satellite data.

For normal laymen's I will recommend to read the down listed description of the traditional agro-pastoral land use, summery of the Landsat analysis and perhaps stick to the figures and explaining text. The text as such is meant for the engaged or knowledgeable layman or even earth observation specialist. There is also a link to the results that are front-end applied research of human abuse - Earth Observation application - pictures and capture text in this pro version can perhaps also be informative for an engaged layman. 

Summery of the initial and holding findings of using Landsat data for mapping land use changes in oil block 5a:

The Landsat data could map the traditional way of life of people livening in the area by primarily showing the foot print from cattle's - this means that grazing activity exposes the soil which reflectance is captured by Landsat. The pattern and link to presence of people could be best monitored in a time slot after crop harvest where farming activities were concentrated at and  in the surroundings of the traditional homesteads. Hence, comparing images from one year to another showed the change pattern. Explaining or verifying the change pattern was linked to comprehensive ground reporting on displacements of people at the same time.    

Figure 1. The 5a area is located in South Sudan with the White Nile passing by. The area is flat, sesonal wet and flooding can occure. The background image is Landsat data recored the 27. November 1999 and set into show natural colours. Lush vegetation is green and the human footprint is withish from farming activity that are exposing soils (the observed farming activity is mainly the result of  high cattle concentrations - see more below). The map has  been applied major towns, water courses in blue and show eastern border of oil block 5a. The images was taken just before the establishment of the all-weather road from Leer over Thar Jath (oil field) to Bentiu. Grids from oil field exploration was also visible and digitized in yellow.  

Files, reports, links and pdf are found in the button of this page

The traditional land use

With the vicinity of the White Nile and surrounding swampy areas (the Sudd) the area are characterized by seasonal wetness and flooding. The landscape is flat and covered by grass and wooded grasslands that are utilized by Nuer and Dinka pastoralists.

A large part of south Sudan are covered by flood plains and marshes that are suited for the agro-pastoral lifestyle, which have resulted in there is more cattle than humans. Both the Nuer and Dinkas lifeform is focused on the cattles.


Their permanent settlements are typically placed on higher grounds that is typically old natural river levees, which is just elevated enough to avoid normal seasonal flooding or wetness. During the wet period (May-November) both people and cattles are concentrated on these slightly elevated homesteads where crops are grown and harvesting is typically over by the start of the dry season in November. During the dry season the cattles are driven to the large wet or river areas to find fresh grasses. This is a kind of transhumance agro-pastoral farming system that is widespread across the Sudano-Sahel region of Africa.




Figure 2. High concentrations of Nuer cattles (abigar) which is adapted to study area. Nuer is also called 'the abigar people'. People are living in Tukuls (huts) that can be seen in the background.




Figure 3. Cattles grazing activity leaves a footprint via vegetation degradation and soil exposure that can effectively be observed in multi-spectral satellite images (photo credit: @ Peter Lawson).



Figure 4. During dry season cattles are driven far to major wetland areas close to the main rivers where fresh grasses are found (photo credit: @ Nakachew Minuye and Nuer study).  


In the study area the traditional settlements are predominantly located on sandy levees along streams (fig 5). People are living in Tukuls - circular single room mud huts with thatched roofs. Cattles can also be kept in huts that usually are larger that those populated by humans. The settlements are typically spread along a string along streams. Cropping takes place just around the housing (see below).

Figure 5. Typical settlement area viewed from an air plane. The settlement huts are typically located along slightly elevated natural levees of streams (Upper right - green overgrow in dry season). The recent cultivated farm field areas are the most bright grayish - dry grass or fallow areas are yellowish. Photo credit/Copyright © : Sharon Hutchinson.


The effect of pastorals or pastoral farming is that grazing activity produces a seasonal resource degradation of the vegetation. However, during the rainy season the vegetation normally recovers unless an area has been severely overgrazed for years or otherwise degraded (e.g. permanent housing) so the vegetation will take more seasons to recover. This type of ecosystem degradation, are typically occurring around water points, for example areas surrounding boreholes in Sahel or in this case the center of a permanent settlement that can have under hard land use for several years that grasses can take more than a  season to reestablish. However, this ecosystem degradation phenomenon was only observed at limited extent in this study (see e.g. fig of Boaw).


Monitoring of the Agro-pastoralnen Land Use from Satellites

Landsat satellite data was chosen as the key data source to detect the human footprint in the area. More specifically, it was found that the footprint from the farming activity could be observed at and around their traditional homesteads after harvest. At this time in the early dry season a bright signature was left from cropping activity and a high concentration of cattles have been around for months.

This means, that if no farming signature (bright signature) or substantial reduced farming activity could be observed from the traditional permanent settlement areas the following year - then people and their cattle's have not been around.

The environmental dynamic was great in the area, with a cloudy wet season that produced a strong flush in vegetation that hides the human activity. Hence, just before the dry season (November) was found to be the most suited time-slot to trace the farming activity. At this time bush fires (see more below) where not predominant. During the focus time 1999-2003 the area were monitored by Landsat 5 and 7 that each had a revisit cycle of 16 days. However, with the exclusion of images with unacceptable amount of noise, reduced the number of suitable annual images to very few - at best. In other words for the November time-slot 1998-2002 only one or two annual images was found useful to be compared across years.  

Figure 6, show how the area (approx. 125 x 128 km) looks like in natural colors  observed from Landsat images (band 3,2,1) recorded 1994 to 2004. lt can be observed that most images are collected at about November. At this time the lush vegetation stands out in green and the human footprint in bright whitish colors (high reflectance from bare soils -higher resolution image here).  A close-up (h) show the classic farming pattern located along elevated sandy river levees where cropping takes place (see also fig 5). The surrounding areas are more moist and have dark soils (vertisols).

Figure 6. A selection of Landsat images in natural colors (band 3,2,1) covering the study area (125 x 128 km) from 1994 to 2004. The close-up (h) show the typical land use pattern of homestead settlement areas located on slightly elevated sand banks - see also fig. 5 for an arieal view at a finer scale (higher resolution image here). The large scale picture show that from about 2000 the concentration of farming activity ichanges and moves South West and South.

Moving into the dry season, transhumance activity with cattle herding to wet areas and bush fires can blurr the appearance of people. Hence, at this time, the dynamics in the environment and opportunistic cattle herding over relatively large distances for fresh grasses (can move 30-100 km per day / source: Nakachew Minuye) can make consistent assessment of disturbance in the farming system challenging at a fine scale (see fig 7). However, intense activity from people and cattles can overruled e.g. the bush fire noise and show the immediate concentration of people or cattles during the dry period (see fig 8).

Figure 7. Landsat image covering the Kuac settlement area (see location in fig 1 or below). The image is taken just into the dry season where bush fires scares (dark brownish areas) are present. The land use footprint from light grazing can  becoming masked at this time of year. However, immediately high concentration of particular cattles  tend to overrule the fire scare effect and again leave a bright foot print. In the image a more or less dried out stream is meandering northward of Kuac, tracks can be seen that can be linked to cattle herding that have crossed a bush fire scare and left a footprint. Hence, at this time of year, the very bright areas can be linked to concentration of people and cattles.

Hence, in spite of the area tend to swept into bush fires scares during the dry season, the large scale concentrations of people and their cattle's could still leave a clear foot print of their immediate appearance at the scale of the study area which revealed a massive from 2000 to 2002 (fig 8).

Figure 8. Four Landsat images recorded over the area into the dry season where bush fire scares are a fequent feature (higher resolution image here). The time serie show a bright footprint that have a relative consistent spatial patteren in 1995 (a) and 2000 (b). In march 2002 (c) this pattern changes and the central area is significant reduced and areas in north east and south east are growing. In the end of December 2002 (d) the concentration of the human footprint has moved south. 

Within range-land management it has for a long time been well known that satellite data offers a more effective mean than aerial photos to monitor the larger scale degradation patterns from grazing (e.g. Tueller 1989). Particular, the multi-spectral band recording from satellites are sensitive to bare soil, thus offers opportunities to analyze the effect of grazing that goes beyond what can be interpreted from only the visible light area commonly offered by Very High Resolutions (VHR) earth observation data or aerial photos. 

Figure 9. On the left, Very High Resolution (VHR) satellite data in natural colors taken in the dry season with a spatial resolution of 0.5 m - farm fields and huts are clearly visible. On the right a false-color composite of Landsat data using two infra-red and red light channels that produces an easily interpretable image with vegetation in green, fresh farm fields are whitish and wet areas dark bluish. The red quadrant is the cover of the VHR image on the left. In the Landsat data (band 7,5,3) bare harvested farm fields show whitish, which also characterize heavy land degradation from cattles.

The application of digital processing of the multi spectral satellite images and particular Landsat data has been available for about 50 years. This includes well established methods, for example deriving biophysical indices that are found very closely correlated with e.g., plant productivity, brightness, wetness etc. An example of composition of such indices can be viewed below (fig 10), which is a Tassel Cap transformation of the area that can be viewed in fig 9. This can help to interpret the biophysical character of objects of interest and can be an additional source to support image interpretation or as input digital image classification.

Figure 10. A Tassel Cap transformation of the Landsat data covering the area shown above (fig 9). Remark the transect set across the bare farm field (see also fig 9). Tassel Caps produces image layers that show biophysical processes or response that are shown on the right. Here they indicate properties of the farm field compared to its surroundings; high brightness, bare and dry.

Categorical classification of satellite images is also well known for decades and has traditionally been performed by supervised or unsupervised classifications. The supervised classification utilizes training areas from well known sites as input to a classification algorithm that eventually categorize the image into the trained classes. The unsupervised classification can be considered as an early machine learning algorithm that, for example, automatically classifies the image based upon spectral clusters of the selected bands and a user defined number of classes. As such, the supervised classification are considered as a stronger approach - however, to obtain high accuracy requires that the training data is representative for the objects your want to class, which in some cases can be difficult to threshold. Hence, there are some examples where unsupervised classifications have shown particular useful that includes the effect of grazing (Tueller 1989; Lemenkova 2021). This can be linked to the multi-spectra properties of e.g. Landsat bands are sensitive to soil reflectance. Hence, Landsat can respond to even light grazing activity which can be difficult to interpret from very high resolution platforms or categorize on the ground (Booth and Tueller 2003). The tracks in figure 7 illustrates this, although the Landsat data has 30 m spatial resolution the change in soil reflectance from tracks that are not more than few meters wide can predominate the image element. This is one of the cases where the lower spatial resolution Landsat data appear stronger or more effective than visual interpretation of VHR imagery to trace the effect of grazing at larger scales. 

Hence, Landsat data has several qualities that make it suitable information source to document land use changes over time; despite the 30 m spatial resolution, it has the multi-spectral information and the data covers large areas so that analysis can be made from e.g. one image with consistent properties. Furthermore, archives of good quality data are freely available from the mid 80'ties.


For the Prins 2009 report the aim was to verify the large scale change pattern in traditional land use by linking ground reporting of attacks or displacement of people from settlement areas. In other words, to provide evidence of change in land use patterns - if there is a significant consistency between ground reporting of displacement of people and change  in traditional land use patterns during the 1999-2003 time-slot.


Landsat L1T calibrated data was predominantly used in the Prins 2009 report. The procedure was quite simple and easy to reproduce; late growing season images were preferred to monitor change pattern in the traditional land use. The logic in this was; At this time the farming activity should be concentrated at the homestead areas and image noise are limited, thus if no activity then no people. The same geo-subset was cut out of all images and these were individually classified into 60 classes by the unsupervised ISODATA classifier in ERDAS Imagine pro software - using defaults settings. The unsupervised classifier was chosen as it worked quite well in response to the visual interpretation of the images and there were limited means to verify training areas for a supervised classification when most of work was done (2006-7). The results of the iso-classifications were visually interpreted and in nearly all cases the two brightest classes were merged into one representing the traditional land use from each recording and excluding areas in oil block 4, north of Bahr el-Ghazal river. The land use area was derived from each image and Boolean change maps were composed as visual products for showing the history and spatial pattern of change. In addition, some dry season images were used and processed in a similar manner. These were used to collaborate the extent of the traditional land use before the development of the oil industry (see e.g. fig. 6 and 11) or support ground reporting's of the massive displacement of people following the reported major clash in the Nhialdiu area. Few images contained cloud contamination, which was removed and considered to have marginal effect on the overall mapped land use and area pattern. For exploring the historical land use activity and explaining change patterns the least cloud contaminated Landsat images from early dry season were collected in the 1987-2006 period.

Mapping of the consistency in farming activity covering the area in early dry season was preformed (e.g. 1994-95 - see fig 11 and interpret fig 6 and 8) as a reference of the traditional land use pattern before reporting of fighting and development of oil industry.

Figure 11. Example of the result of  an unsupervised ISO classification that sort the image into classes after spectral properties of the bands. In this case a 60 class classification was specifyed and land use came out in one or two classes and used to show the extent of the seasonal tradition land use. In this case, it shows two senarious of late 1994 - early 1995  - one month apart that show the consistent high concentration of farming activity or homesteads before the onset of oil industry development. The farming pattern is consistent with the extent of settlement patterns that can observed in Russian maps from the 80'ties.

Figure 12. An assemblage of the unsupervised ISODATA classifications showing footprints of the location of people observed during late 1999 to early 2003. With reference to the ground reporting's the footprint are interpreted as follows; A) show the concentration of traditional land use concentration in the Nhialdue area (see location in fig 1). B) Some month later than A, into the dry season, showing a strong footprint and collaborate reporting's of more people were coming into the Nhialdue area following attacks on villages e.g. the northern end of the oil road. C) Just after the reported major attack on the Nhialdue the footprint are pushed south west that collaborate that people flee in the first place against Wicok and Chotchar (see also fig 1,6 and 8). D) Show that the farming footprint is now strongest in south-south-west that is consistent with the reporting's that the majority of people ended up here. E) into the dry season a couple of month later, bushfires have swept over the Nhialdue area and the major footprint is now in south-south-west were it were reported people were ended up. F) The same pattern as E - one month later.

As indicated from the Landsat time series above (Fig. 6, 8 and 12) there were indication of massive displacement in the traditional land use areas, particular during 1999-2003. These changes could be linked to the reported massive human abuses that were avaliable from plenty of independent sources (most summed up in HRW 2003). For example, some of these included comprehensive quistionnaire of local people carried out by surveyors from Christian Aid and ECOS that transformed their findings into simple maps (Fig. 13).


Figure 13. Simple maps of displacement of people based upon questionnaire of local people following the major clash and reported clearing of the Nhialdue area in 2002. The maps also included escape routes the places where they ended up. In summery, it showed a displacement from Nhialdiu area and most people were first driven south west and then south.


Overall, at the large scale, three major displacement areas could be summarised from the ground reportings. These areas was digitized and used to collaborate the change pattern observed in traditional land use from the Landsat data.


Figure 14. Focus areas for change detection in traditional land use. The three selected focus areas are within oil block 5a and digitized after careful interpretation of the displacment extent from several ground reports. For example, the northern end of the oil road used a buffer of 2.5 km as it was reported that villages within half an hours walk from oil road were attacked (HRW 2003).  

When the mapping of the Landsat analysis where put together into a change map in traditional land use, it was found to fully collaborate the ground reportings of attacks and displacement of people. This did not only include the focus areas but the hole area as such (Fig 15). For example, the south ends of the oil road (e.g. around Bieh - attacked in 2002; HRW 2003; Sudanreevs ; Relifeweb 2002) and the larger area surrounding Nhialdue. In the Prins 2009 report the statistics from focus areas during 1999-2003 showed the most conservative measures of land use a reduction of 72%, while including the dry season images 99% reduction. From the road side these numbers were ranging from 61-97%. In the Stockholm presentation (fig 14 and 15) the focus areas were elaborated in more detailed and showed for the Nov. 1999 - Nov. 2002 a 90% reduction in Nhialdue and 85% for the northern road side. These figures can be hold against reportings that claims these areas to have been deserted - that the result is consistent with - and anomalies in statistics can be related to exact timing of event e.g. the road site were attacked earlier and some stayed around untill 2001 and that some core settlement areas do not recover immediately with vegetation if they have been bare for years. However, in conclusion - the result show that a massive displacement of land use could be observed and is consistent with the conclusions of the comprehensive ground reportings. In other words; no other plausible explanation could be reached from interpreating the Landsat archives and the Landsat data analysis.

Figure 15. Focus areas and events put together with Landsat analysis of land use during the 1999-2004. The two data sources supported each other by the change in the land use footprint. Statistics of traditional land use area collaborated that the two areas were reported to be deserted by the end of 2002. The reported major refugee area in south could also be confirmed by the steep increase in land use. Yellow explosions refer to identified settlement areas that have been reported attacked and green triangles refer to refugee campsites. It can further be seen that at the south end of the oil road - the area around settlement Bieh show depopulation, which also have been reported (ECOS 2002; HRW 2003; Sudanreevs;  Relifeweb 2002) to be attacked several times during early 2002.

The Landsat data also uncovered the gradual development of the oil infrastructure. Figure 16 show the status of the development in 2006 and features that were observed in the Landsat data untill 2006. Remark the elevated all weather roads traps water or disturb the natural hydrology. The backdrop image is Landsat band 7 that is sensitive to wetness. More detailed image that show hydrological disturbance close to Thar Jath be viewed in the here and in the Prins 2009 report.

Figure 16. Development of the oil infrastructure observed and digitized from the Landsat data - status October 2006.  

A generalized scenario showing development of oil industry, reported attacks or reported fights (most comprehensive source: HRW 2003) as well as refugee camp areas and the major change pattern in land use (2000-2002)  can be view in figure 17.



Figure 17. A scenario which support the link between mass dislocation of people and development of oil infrastructure. The map summarises change in traditional land use 2000-2002 that has applied, geo-located village attacks or fighting's from reports as well as observed developed oil infrastructure in 2006. Brown refers to areas which were observed as farming areas in 2000 but not in late 2002. Bright green show observed farming activity in late 2002. 




Post 2010 work

As new reference sources became available post 2010 more investigations were put into the satellite methodology for exploring potentials and consistency in the land use assessments or classification.

Post 2010, VHR images were becoming freely available via Google Earth that could be used to access the accuracy of the classification approach at much finer scales as well as using supervised classifications that small training areas for classification and verification (see fig 18). This was in first place the seasonal farming activity at the homesteads that were evaluated. Figure 19; show the placement of the first freely available VHR images within the study area (fig 9 and 10 contain part of the 2006 image).


Figure 18. Typical farming area observed from Very High Resolution satellite taken well into the dry season. Circular huts are visible as well as bare and heavily degraded areas (digitized and used for traning and verification in Landsat analysis  - see below). The very bright areas indicate that there is recent activity on the ground. The ripping pattern in lower part of images is previously cultivated areas. The dark grayish areas indicate that bush fires have swept the area since the rainy season.


Figure 19. From approximately 2010 an increase in detailed spatial data became available, including VHR images on Google Earth and databases of named settlements (yellow dots in figure). Recent land use activities were digitised from 2003 and 2006 VHR images and used to make accuracy assessment of the landsat classification at scales of 0.5 - 2 ha (see table 1 below). The two inserted images show co-occrence (white) and comission error (green) and omission error (red). The 2006 classification covers the areas shown in figure 9 and 10.


There is plenty of algorithms that can be used for Landsat image classifications - however, the important point is how you relate your result to a hypothesis or an assumption that should have source that can be verified against. Therefore, it is good practice to make accuracy assessment against an independent reference. At higher spatial resolution, this is commonly done by an error matrix which includes overall mapping accuracy (%) and class errors - omission (did not catch) and commission error (classed too much). Kappa statistics have also been used as one measure to describe  overall accuracy levels (e.g. Fagan and De Fries 2009) although this approach is still debated.


An accuracy assessment of the 2006 and 2003 showed that the unsupervised image classification method used in Prins 2009 preformed well at higher resolution (table 1). However, a machine learning algorithm (Classification Tree Analysis (CTA)) that is known to be particular effective to handle PCA transformed data (Tassel Cap) was found to perform very strong when it was evaluated with VHR images (table 1). The CTA also showed strong in a comprehensive comparison of classifications methods that was scientifically published (Prins 2018). This (Prins 2018) showed that the errors linked to the unsupervised ISODATA algorithm was commission errors. In other words, this means that the unsupervised classifier consistently grabbed more land use than the 10% threshold set out for the verification. However, it simply enlarged the active homesteads areas (see examples here) beyond what could consistently be interpreted from the VHR images source. This means that it is not necessary wrong - it is just hard to verify the light grazing.


Very high accuracy is generally more difficult to achieve at fine scale. However, as a hand rule, you can increase the scale or minimum mapping units and reduce the errors (see table 1).



Table 1. Accuracy assessment of classification methods that was tested on Landsat images and using Q-bird 2 VHR images on Google Earth as a mean of verification. ISODATA is unsupervised classification and CTA is a supervised tree classifier that have been used to threshold spectral bands or biophysical indices. Overall mapping accuracy is very high. In 2012 Kappa statistics was used to document mapping agreement - that accuracy levels have been taken from Fagan and De Fries (2009) who used it to describe overall accuracy for classification types. All classification methods showed acceptable accuracy levels and the CTA classifier allowed detailed mapping with very high accuracy at settlement level - according to the Minimum Mapping Unit (MMU) used.

More advanced methods were elaborated, this included narrow inter-calibration of images that show changes in more detail by the use of continuous data instead of the categorical classification. Below (fig 20) show an example of only using change in blue light (Band 1). The narrow inter calibration of images from different years means that you can establish a monitoring system that make values comparable or consistent from year to year, which is another way of providing evidence of change over years. Many approaches were later tested as much more reference data became available for training and verification. Eventually, this was developed as a monitoring system that has strong implication for rangelands and sustainable resource utilization. Details can be viewed in the Prins (2018) and the summery

Figure 20. Change in blue light reflectance for November 1999 and 2002 image. Landsat band 1 (Blue light; 450 nm) holds limited information but is good in separation of bare soil and vegetation. Band 1 could be narrowly calibrated in-between Landsat recordings across years (see more in scientific summery).



Overall the results of the satellite mapping (Prins 2009; Stockholm 2013 presentation) fully collaborate and detail the ground reports of mass displacement of people in oil block 5a during 1999-2003 .

It was found that most consistent footprint of the traditional agro-pastoral land could be monitored after harvest (November). At this time people and cattles have been concentrated at the homestead and the land use leaves a brightness footprint from exposed soils. Into the dry season immediate high concentration of people and cattle's could also be observed although cattle herding and bush fires introduces challenges to map the extent of less intense land use.  The study used the best available seasonal Landsat data (e.g. limited cloud contamination). It was found that the anomaly in land use that was shown for 1999-2002 was supported by testing a variation of digital classification methods and establishment of a monitoring system (Prins 2018 summery). This means that the result is consistent, and it is not likely that the observed massive changes in land use patterns had not occurred.  

For mapping of the traditional agro-pastoral land use displacement of people over time the categorical ISODATA classification worked effectively to show large scale patterns. Monitoring systems of the tested classification solutions may have implications for calculation of population numbers and as such may be used in other context of humanitarian crisis and range land and sustainable land use management. 





PRINS KEY-NOTE SPEAKER AT THE SWEDISH SPACE DAYS 2013: Global reporting interpretation of speach



pdf  REPORT TO ECOS 2009: Satellite mapping of land cover and use in relation to Oil exploitation in concession 5A


Client of the 2009 report: The European Coalition on Oil in Sudan (ECOS)




[1] In general, an AUC of 0.5 suggests no discrimination (i.e., ability to predict farming activity), 0.7 to 0.8 is considered acceptable, 0.8 to 0.9 is considered excellent, and more than 0.9 is considered outstanding.



Aljazeera. 2015. Accessed 26 September 2016. 05/south-sudan-man-catastrophe-150525120843494.html

Bastin, G. N., G. Pickup, V. H. Chewings, and G. Pearce. 1993. “Land Degradation Assessment in Central Australia Using a Grazing Gradient Method.” The Rangeland Journal 15 (2): 190–216. doi:10.1071/RJ9930190.

Booth, D.T., and P.T. Tueller. 2003. Rangeland monitoring using remote sensing. Arid Land Research and Management 17 (4): 455–467.

Christian Aid. 2002. Hiding between the Streams: The War on Civilians in the Oil Region of Southern Sudan.

de Guzman, D., 2002. Depopulating Sudan’s Oil Regions. European Coalition on Oil in Sudan (ECOS).

Fagan, Matthew, and Ruth DeFries. "Measurement and monitoring of the world’s forests: A review and summary of remote sensing technical capability, 2009–2015." UMBC Geography and Environmental Systems Department Collection (2009).

Google Open Street Maps. 2015. “The Humanitarian Openstreetmap Team.” https://www.hotosm. Org

Holling, C. S. 1973. Resilience and stability of ecological systems. Annual Review of Ecology and Systematics 4:1-23

HOT. 2016. Project - South Sudan, Unity State, Bentiu. The Humanitarian OpenStreetMap Team. http://

HRW. 2003. “Human Rights Watch (November 2003). Sudan, Oil and Human Rights. 754 p. Accessed 26 September 2016.

HRW. 2015. “’They Burned It All’ Destruction of Villages, Killings, and Sexual Violence in Unity State, South Sudan: Report.” 42 p. Accessed 26 September 2016. files/report_pdf/southsudan0715_web_0.pdf

James, C. D., J. Landsberg, and S. R. Morton. 1999. “Provision of Watering Points in the Australian Arid Zone: A Review of Effects on Biota.” Journal of Arid Environments 41 (1): 87–121. doi:10.1006/jare.1998.0467.

Lemenkova, Polina. "ISO Cluster classifier by ArcGIS for unsupervised classification of the Landsat TM image of Reykjavík." Bulletin of Natural Sciences Research 11.1 (2021): 29-37.

Maynard, C. L., R. L. Lawrence, G. A. Nielsen, and G. Decker. 2007a. “Ecological Site Descriptions and Remotely Sensed Imagery as a Tool for Rangeland Evaluation.” Canadian Journal of Remote Sensing 33 (2): 109–115. doi:10.5589/m07-014.

Maynard, C. L., R. L. Lawrence, G. A. Nielsen, and G. Decker. 2007b. “Modeling Vegetation Amount Using Bandwise Regression and Ecological Site Descriptions as an Alternative to Vegetation Indices.” GIScience & Remote Sensing 44 (1): 68–81. doi:10.2747/1548-1603.44.1.68.

Olofsson, P., G. M. Foody, M. Herold, S. V. Stehman, C. E. Woodcock, and M. A. Wulder. 2014. “Good Practices for Estimating Area and Assessing Accuracy of Land Change.” Remote Sensing of Environment 148: 42–57. doi: 10.1016/j.rse.2014.02.015.

Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. “Maximum Entropy Modeling of Species Geographic Distributions.” Ecological Modelling 190 (3): 231–259. doi:10.1016/j. ecolmodel.2005.03.026.

Pontius, R. G. Jr, and M. Millones. 2011. “Death to Kappa: Birth of Quantity Disagreement and Allocation Disagreement for Accuracy Assessment.” International Journal of Remote Sensing 32 (15): 4407–4429. doi:10.1080/01431161.2011.552923

Erik Prins (2018) Landsat approaches to map agro-pastoral farming in the wetlands of southern Sudan, International Journal of Remote Sensing, 39:3, 854-878, DOI: 10.1080/01431161.2017.1392634

Samain, O., L. Kergoat, P. Hiernaux, F. Guichard, E. Mougin, F. Timouk, and F. Lavenu. 2008. “Analysis of the in Situ and MODIS Albedo Variability at Multiple Timescales in the Sahel.” Journal of Geophysical Research: Atmospheres 113: D14. doi:10.1029/2007JD009174

Stehman, S. V. 2004. “A Critical Evaluation of the Normalized Error Matrix in Map Accuracy Assessment.” Photogrammetric Engineering & Remote Sensing 70 (6): 743–751. doi:10.14358/ PERS.70.6.743

Tueller, P. T. 1989. “Technology for Rangeland Management.” Invited Synthesis Paper 42 (6): 442.

UNHRC. 2016. “UN Human Rights Council.” Assessment mission by the Office of the United ations High Commissioner for Human Rights to improve human rights, accountability, reconciliation and capacity in South Sudan: detailed findings, 10 March 2016, A/HRC/31/CRP.6. Accessed 26 April 2016.

UNMISS 2015: UN Mission to South Sudan. 2015. “2015, Flash Human Rights Report on the Escalation of Fighting in Greater Upper Nile.” April/May, June 29. Accessed 26 April 2016.

Unosat. 2016. “UNOSAT LIVE Map for South Sudan Complex Emergency (CE20131218SSD).” Accessed 26 September 2016. id=cac78381476c43ebb98a7e642211809d

UNOSAT. 2015. “Rapid Damage Assessment of the Area of Rapid Damage Assessment of the Area of Ngop Ngop, Unity State, South Sudan.” Accessed 26 September 2016. http://unosat-maps.

Washington-Allen, R. A., N. E. West, R. Douglas Ramsey, and R. A. Efroymson. 2006. “A Protocol for Retrospective Remote Sensing–Based Ecological Monitoring of Rangelands.” Rangeland Ecology & Management 59 (1): 19–29. doi:10.2111/04-116R2.1.





About Us    |    References    |    Services    |    CV's    |    Contacts    |    Press     Copyright © PRINS, Eng. 2024