This past June, and yes in this year, the National Forest Resources Assessment Report 2021 was released. This news was announced by the Chief Conservator of Forests, Julius Kamau, on June 3rd 2022 from the Kenya Forest Services headquarters in Karura. Unlike previous sentiments where we have been treated to grim statics of tree cover loss over the years, this time round there was some good news. Kenya had just surpassed the 10% tree cover aimed for 2022 and had just reached 12.3%. Wow!
However, looking at the Kenyan territory from above using Earth Explorer or Google Earth may seem to be an antithesis of some sort. Large portions of the country are arid. But that’s the report from the forest caretakers of this country. And guess who was responsible for the bulk of the tree cover assessment? Gis specialists, remote sensing specialists, cartographers, foresters and land surveyors. A community of forces of some sort.
Before we head to how the forest cover assessment report was done, it will be improper if we did not give a sneak peek of how GIS first got involved in forest mapping. Firstly, GIS was unheard of until 1963 when a Canadian Geographer invented the term. But of course, there must have been forest mapping being done before then. So how did GIS in forest mapping all come into force?
History of GIS in Forest Mapping
GIS, as mentioned above, was first coined in 1963. As with most disciplines, some early groundwork had already been put in place but it was in 1963 that the word GIS became official. This can be attributed to the work of Roger Tomlinson, a mapping specialist of Spartan Air Survey based in Ottawa, Canada. One of his first projects that led to the development of GIS as a discipline (and technique) was to identify the location of a tree plantation in… wait for this, Kenya! This was based on a contract given by his employer–Spartan Air services. His GIS concept reduced the workload from an initial (drum rolls please) 3 years and CAN $ 8 million to several weeks and CAN $ 2 million. Sounds inspiring, right?
History of forest mapping using GIS in Kenya
Before GIS was employed in forest cover mapping, or assessments in Kenya, whatever the case, the government often relied on inaccurate expensive field surveys. For a two-year period beginning in 2011, the government of Kenya, and with the collaboration of the Japanese Space Agency, conducted the first-ever national forest mapping exercise using satellite imagery. It was a ground-breaking exercise. The project involved combining Landsat images with the Japanese Advanced Land Observing Satellite (ALOS) from 1990 to 2000 and from 2000 to 2010. They killed two birds with one stone. Not only were they able to deduce the area covered by forests in Kenya, but they could also analyse the trends over time and extract many other derivatives.
The above chronology may be a bit inaccurate. According to the newly released KFS report, estimates of Kenya’s forest cover date back to the 1970s. The first estimation was spearheaded by the Rangeland Ecological Monitoring Unit. Landsat images of 60m resolution were used for the exercise. Today, the department goes by the name of the Directorate of Remote Sensing and Resource Survey (DRSRS).
The National Forest Assessment Report 2021
Over the years, the accuracy of some open-source satellite imagery has improved. For most Landsat satellites, the resolution remains 30m. The European Space Agency (ESA) went a notch higher with its Sentinel-2 satellites to achieve 10m accuracy in the RGB bands. Unlike previous forest assessments that used medium resolution imagery, the forest assessment of 2021 used high-resolution imagery-30cm to be exact. The main disadvantage with the medium resolution is that scattered trees, trees and woodlots in lands under 0.5ha are not captured. The same also goes for deciduous trees, especially in seasons of leaf shedding.
To carry out this herculean task, the KFS and other stakeholders incorporated highly trained personnel, cutting-edge technology and ground-truthing. Of course, this was a multi-agency approach, since if you want to go far, you go in a team.
The following were the steps undertaken to map out the forest resources in Kenya, as well their types. Please note that this is a miniaturized version of the more detailed process found in the 2021 report.
Step 1: Image Classification
This is the process of categorizing and labelling groups of pixels or vectors within an image based on specific rules.
Satellite images with less than 15% cloud cover were identified and downloaded. For those with higher cloud cover percentages, images from earlier years were used to mask out the clouds. The images were then rectified from their world datum references to the local Arc 1960 datum. Finally, the image pixel size was resampled from 0.3 metres to 1.5 metres. All these were done using QGIS and Orfeo Toolbox plugin.
K-means classification using the Orfeo toolbox was used to cluster the different classes within each satellite image.
Through the help of knowledgeable experts, polygons were drawn on those classes that best looked like tree cover. Comparison between the polygons drawn and high-resolution imagery was constantly reiterated to ensure correct classification.
Step 2: Ground-truthing
Ground truthing is a process of validating the classified image by observing what is on the ground in reality. The following components were critically examined while handling the field data: thematic content, sampling design, sample data collection, data processing and analysis.
Step 3: Accuracy assessment
Image classification accuracy assessment refers to the process of validating results obtained from the image classification process against ground truth data. At this stage of the tree cover assessment process, a total of 1544 sample location data were randomly selected and paired against high-resolution imagery for verification. The indicators used in measuring the accuracy of the classification results were: producer’s accuracy (PA), omission error (OE), user’s accuracy (UA), commission error (CE), overall accuracy (OA) and the Kappa coefficient. The report, however, did not specify the thresholds used for these indicators.
Step 4: Tree cover classification
Outputs that had passed step 3 were subjected to another round of classification (reclassification). This process was to correct for classification errors arising from ground-truth data. The reclassified tree cover data was then mosaicked for each of the 47 counties and some statistics were calculated.
Step 5: Forest cover classification
Forest land was derived from tree cover collected in the previous steps. Tree cover blocks of less than 0.5ha were eliminated to leave those above this threshold. This was to keep in line with the part definition of forests –that the block of trees must cover an area larger than 0.5ha, at least 2m height in situ and have a canopy cover of not less than 15%.
Step 6: Forest canopy density
To determine forest canopy density, the following steps were undertaken:
- The forest cover layer was vectorized.
- Grids measuring 10m by 10m were then generated on the forest cover layer.
- The two layers were then intersected, and the percentage of forest cover in each grid was computed.
- The average of each forest block was computed and the appropriate canopy density was derived using the following criteria
- Above 65% canopy density – dense forest
- 40 – 65% canopy density – moderate forest
- 15-40% canopy density – open forest
Step 7: Forest type
The output of the forest cover classification exercise above was then categorized into the following four forest types: plantations, bamboo, mangrove and natural forest. This was aided by the use of auxiliary GIS data such as existing forest boundaries and the like.
Step 8: Map production
Doesn’t the title speak for itself? In this stage, a user-friendly and understandable map, covering all the necessary information was drawn.
There ends the complex task of mapping the entire fragmented and disparate forest resources of our country. Despite initially sounding like a monumental task, after reading the steps above, it sounds so easy. Furthermore, the use of local labour rather than contracting foreign companies saved the exchequer millions of shillings.
The use of GIS in forest mapping is not that new. GIS has also been used in many areas in Kenya such as mapping water hyacinth, mapping mangroves etc. However, it has made the collection of data that would have otherwise been done in a complex process relatively easy. Furthermore, the data collected in GIS is not ‘static’. It can always be retrieved and modelled to predict future patterns using the latest data and with a high degree of accuracy to that effect. This is being reinvigorated by GIS tools becoming more sophisticated as the days go by, aided by the likes of Machine Learning and other programming concepts. The implication is that data collected is more reliable, reusable, and saves on costs. Who knows what we might use next in a future tree cover assessment.