In the recently released KFS report of 2021 – the National Forest Resources Assessment Report 2021 – Kenya has a tree cover of 7,180,600.66ha, which is 12.13% of the total area. Based on the national ambition to reach 10% tree cover by 2022, this is a monumental achievement. Forest cover, on the other hand, has a coverage of 5, 226, 197.79ha, a paltry 8.83% of the total area. Tree cover and forest cover are two terms that are sometimes used synonymously. However, they are different. Forest cover, as per the KFS report, refers to the amount of land area covered by a forest. Its counterpart terminology – tree cover, is a lit bit vaguer. Even FAO does not seem to have a commonly agreed definition. But for the purposes of this article, it refers to any area covered by trees, within or outside forests.
There has been a widespread assumption that each country should have a minimum of 10% forest cover, supposedly from an unconfirmed UN statement. An online search shows there has been, and still continues to be a debate on this threshold. This assumption arose from the definition of a ‘forest’ itself, in which, inter alia the ground under it must be under 15% or more canopy cover. A scholarly debate on this matter revealed it is better for countries to define their own optimum threshold. This is because each country has differing climate regimes, land availability and vegetation biomes. Thus, it best seems to leave countries to design their own fate on how much of their territory should be allocated to forest cover.
In Kenya, it is already engraved in stone. The constitution recommends at least 10% of the total area should be under forest cover.
Scholarly debates aside, we will try to corroborate, or rather, see if we can verify the findings of the KFS 2021 report on nothing else but through desktop research. We already know forest coverage is 8% of the total area, so we will check out if our remote sensing analysis comes close to this. If Eratosthenes of Cyrene was able to measure the circumference of the earth with reasonable accuracy with nothing more than mere trigonometry, surely we –2000 years down the line can do better armed with the latest software and tech. Can’t we?
To stop beating around the bush, we will use GEE, Qgis and Excel for forest cover assessment in the country. Let’s see if it comes close to the 8.83% mentioned in the report.
Are you ready to plunge in with both feet?
The Google Earth Engine
Like a time machine that can take you to, or back to different epochs of time, so is Google Earth Engine (GEE) in the geospatial world, minus the time travel. The Google Earth Engine (GEE) is a cloud computing platform designed to store and process huge data sets (at a petabyte-scale) for analysis and ultimate decision-making. Like in a time machine where one can input the date they would like to teleport to, GEE works in a similar way by providing a plethora of geospatial data – satellite images, vector datasets and curated algorithms to help the user investigate and discover various geographical phenomena for their place of interest. It has been especially useful to remote sensing scientists, ecologists, and even social researchers.
In this exercise, the Global PALSAR-2/PALSAR Forest/Non-Forest Map was used. Why this one? Basically for two reasons. It had a better spatial resolution than most. Secondly, it had the most recent coverage dating 2018-01-01. The Hansen Global Forest Change v1.9 (2000-2021) was a very tempting dataset to use, its recent coverage dates to 2021. However, it is beaten off the track in that its tree-cover data is for 2000, with the most recent statistics only showing tree loss for the last twenty years (2000 – 2021). 🙁
Armed with what we will swim, let’s take a deep dive.
Ready.
Set.
Dive.
The GEE Workflow
The entire workflow code, at least for the GEE part, is shown below.
//load the shapefile of Kenya
var Kenya = ee.FeatureCollection('USDOS/LSIB/2017')
.filter(ee.Filter.eq('COUNTRY_NA', 'Kenya'));
//show the map of Kenya
Map.addLayer(Kenya, {}, 'Kenya');
Map.centerObject(Kenya, 7);
//so far so good
//load the global PALSAR dataset
var dataset = ee.ImageCollection('JAXA/ALOS/PALSAR/YEARLY/FNF')
.filterDate('2017-01-01', '2017-12-31');
//check out the global PALSAR dataset
print("PALSAR dataset", dataset);
//show the global PALSAR dataset
Map.addLayer(dataset, {palette: ['green', 'yellow', 'blue'], min:1, max:3},
'PALSAR');
//green, yellow, blue colors represent the forest, non-forest and water bands
//of PALSAR respectively, as shown using min:1 and max:3 values, respectively
//so far so good
//show the Kenyan map above
Map.addLayer(Kenya, {}, 'Kenya');
//so far so good
//clip the global PALSAR dataset to Kenya only
//first show the forest class only
var forest = dataset.select('fnf')
//from this forest variable, choose value 1 which stands for trees
//lets first convert from image collection to image only. We will take the
//mean of 2017 as the forest cover
var forestonly = forest.mean();
//select values 1 which stand for forest from the forestonly image
var forestonly2 = forestonly.updateMask(forestonly.eq(1));
//add the forestonly image to see if non-forests have been masked out
Map.addLayer(forestonly2, {palette: ['green']}, 'Forest only');
//so far so good
//now time to clip to Kenya only
var kenyaForests = forestonly2.clip(Kenya);
//lets see forests in Kenya only
Map.addLayer(kenyaForests, {palette: ['green']}, 'Forests in Kenya');
//so far so good
// lets make these pixels of forests have an attribute of area, to help in
//calculations
print("Forest in Kenya", kenyaForests); //to get a feel of the data
var forestArea = kenyaForests.multiply(ee.Image.pixelArea());
print("Forests with area attribute", forestArea);
//the help says that the ee.Image.pixelArea() shall return a band
//called 'area'. However, our image, does not have this band 'area'.
//so far, so good, maybe
//calculate the area of forests in Kenya
var forestedArea = forestArea.reduceRegion({
reducer: ee.Reducer.sum(), //research more on reducers
geometry: Kenya, //the region to reduce data to or operating area
scale: 10, //the scale in meters to work in, went with Sentinel2
crs: 'epsg:32737', //used local projection WGS 84 UTM zone 37S
bestEffort: true, //used true so that server could determine its own best
maxPixels: 1e7 //the default of 10000000
});
print("forestedArea", forestedArea);
//get total forested area in sq km
var forestKm = ee.Number(forestedArea.get('fnf')).divide(1e6);
print('Area of forests in sq kilometers: ', forestKm);
//get total forested area in ha
var forestHa = ee.Number(forestedArea.get('fnf')).divide(1e4); //a ha = 10000sq. metres
print('Area of forests in hectares: ', forestHa);
//lets start phase 2
//load the counties dataset
Map.addLayer(counties, {}, 'Counties of Kenya');
//disaggregated the forestedArea statistics to each county
var countyForests = forestArea.reduceRegions({
collection: counties, //the uploaded counties shapefile
reducer: ee.Reducer.sum(), //calculate the sum area of forestedArea per county
scale: 10, //chose a resolution same as that of Sentinel 2
crs: 'epsg:32737' //used a local datum of WGS 84 Zone 37S
});
print('Area of forests in counties: ', countyForests);
//export the table of area of forests per county to google drive
Export.table.toDrive({
collection: countyForests, //the table with forest area statistics per county
description: 'county_forestarea_metres', //the title of the table
folder: 'test', //the folder the table will be saved into
fileFormat: 'CSV'
});
//view the forest coverage in counties
Map.addLayer(counties, {}, 'Counties');And additionally, there was a county dataset which we will explain later.
var counties = Table users/sammigachuhi-test/counties_kenyaTime for a gasp of air. Let’s take this code out bit by bit.
Loading the Kenya shapefile to GEE
We will load the Kenya administrative shapefile into GEE. The purpose of this to delineate our focus area – the Area of Interest (AOI). Even though this should not necessarily be the first step, it helps (and pays) to start simple.
//load the shapefile of Kenya
var Kenya = ee.FeatureCollection('USDOS/LSIB/2017')
.filter(ee.Filter.eq('COUNTRY_NA', 'Kenya'));
//show the map of Kenya
Map.addLayer(Kenya, {}, 'Kenya');
Map.centerObject(Kenya, 7);So far so good.
You should see the map of Kenya on your GEE canvas.
Viewing the Global PALSAR data
Now that we have been able to define our AOI, we can take a step further and view the global PALSAR dataset. This will be like moving from sipping tea to taking the first bite of a cookie. Unlike loading the Kenya shapefile which was relatively straightforward, this one has some nuances.
//load the global PALSAR dataset
var dataset = ee.ImageCollection('JAXA/ALOS/PALSAR/YEARLY/FNF')
.filterDate('2017-01-01', '2017-12-31');
//check out the global PALSAR dataset
print("PALSAR dataset", dataset);
//show the global PALSAR dataset
Map.addLayer(dataset, {palette: ['green', 'yellow', 'blue'], min:1, max:3},
'PALSAR');
//green, yellow, blue colors represent the forest, non-forest and water bands
//of PALSAR respectively, as shown using min:1 and max:3 values, respectively
//so far so good
So far so good.
Notice that some new methods such as .filterDate, print() and Map.addLayer() were included in this code block. GEE has various methods for conducting geospatial analyses. The reader is advised to look them up.
Next, in order to have a feel of the forest cover in the country, we shall overlie the Kenya shapefile on top of the Global PALSAR dataset. This is why it was not necessary to load the Kenya shapefile in the first step. We could have loaded it here. However, we wanted to start simple, that’s why.
//show the Kenyan map above
Map.addLayer(Kenya, {}, 'Kenya');
//so far so good
So far so good.
Mask out non-forest
Time to remove the unnecessary food items from our plates – the shell from the egg, and the peel from the fruit. In our case, let’s remain with forests only.
Already, if you are keen, you may have noted that some of the forest areas, which are green, and the waterbodies (in blue) are highly exaggerated. Since when did we have a water body the size of L. Albert, and a forest the size of Aberdares in North Eastern? Take note of this.
Let’s first remove the non-forested areas.
//first show the forest class only
var forest = dataset.select('fnf')
//from this forest variable, choose value 1 which stands for trees
//lets first convert from image collection to image only. We will take the
//mean of 2017 as the forest cover
var forestonly = forest.mean();
//select values 1 which stand for forest from the forestonly image
var forestonly2 = forestonly.updateMask(forestonly.eq(1));
//add the forestonly image to see if non-forests have been masked out
Map.addLayer(forestonly2, {palette: ['green']}, 'Forest only');
//so far so good
Clip forest coverage to Kenya only
Like carving out the silhouette from a canvas, we want to extract the forest cover in Kenya only.
//now time to clip to Kenya only
var kenyaForests = forestonly2.clip(Kenya);
//lets see forests in Kenya only
Map.addLayer(kenyaForests, {palette: ['green']}, 'Forests in Kenya');
//so far so good
Calculate the area of forest cover in Kenya
Now to the real deal. Time to calculate the area of forest cover in Kenya. For one, we already know the KFS 2021 report put this at 5,226,191.79ha or 5,2261.9179 km2. What will we get?
// lets make these pixels of forests have an attribute of area, to help in
//calculations
print("Forest in Kenya", kenyaForests); //to get a feel of the data
var forestArea = kenyaForests.multiply(ee.Image.pixelArea());
print("Forests with area attribute", forestArea);
//the help says that the ee.Image.pixelArea() shall return a band
//called 'area'. However, our image, does not have this band 'area'.
//so far, so good, maybeThe method ee.Image.pixelArea() mentions that it will return the image output with a band called area but this was not the case.
Print the area statistics for forest cover in Kenya
They say, put your money where your mouth is. We shall say, put a number on what the area is.
Here is the script for printing the area statistics of Kenya’s forest cover in metres. Remember the method ee.Image.pixelArea() assigned the value of every pixel with its corresponding area value in square metres.
//calculate the area of forests in Kenya
var forestedArea = forestArea.reduceRegion({
reducer: ee.Reducer.sum(), //research more on reducers
geometry: Kenya, //the region to reduce data to or operating area
scale: 10, //the scale in meters to work in, went with Sentinel2
crs: 'epsg:32737', //used local projection WGS 84 UTM zone 37S
bestEffort: true, //used true so that server could determine its own best
maxPixels: 1e7 //the default of 10000000
});
print("forestedArea", forestedArea);This was our result:
fnf: 126804959158.53334fnf was the band for forest cover only in the Global PALSAR dataset.
Now, to put the final nail in the coffin.
Get area statistics of forest cover in km2 and ha
To calculate the area statistics in km2 and ha, the below scripts were run.
//get total forested area in sq km
var forestKm = ee.Number(forestedArea.get('fnf')).divide(1e6);
print('Area of forests in sq kilometers: ', forestKm);
//get total forested area in ha
var forestHa = ee.Number(forestedArea.get('fnf')).divide(1e4); //a ha = 10000sq. metres
print('Area of forests in hectares: ', forestHa);The results were as follows:
Area of forests in sq kilometers:
126804.95915853334
Area of forests in hectares:
12680495.915853335Using just the ha statistics, it’s like Kenya has 7 million hectares of forests more than those reported in the KFS 2021 report – 5,226,191.79ha. Using land area, this is like 22% of the country. Lol.
Why the discrepancy? It had something to do with the wrong classification of the Global PALSAR dataset or its 25-m resolution. A more likely explanation is that the Global PALSAR dataset also classified scattered trees and clumped this up with larger forest blocks indiscriminately.
End of Phase 1.
Beginning of Phase 2.
Disaggregating the forest cover area to counties
Below is the code that was used to disaggregate forest area statistics for each county. A new attribute column sum was added to the original forestArea data.
//disaggregated the forestedArea statistics to each county
var countyForests = forestArea.reduceRegions({
collection: counties,
reducer: ee.Reducer.sum(), //calculate the sum area of forestedArea per county
scale: 10, //chose a resolution same as that of Sentinel 2
crs: 'epsg:32737' //used a local datum of WGS 84 Zone 37S
});
print('Area of forests in counties: ', countyForests);You will now notice that the output countyForests data has a column sum which is the result of the reducer ee.Reducer.sum(). Once again, research on the term reducer. You will be grateful in that it will reduce some fuzziness on this term. In the case of feature property 0 which stands for Baringo county, the sum attribute, or the forest area cover for this county is 7300173368.107181 square metres.
ADM1_EN: Baringo
...
sum: 7300173368.107181
...You may have asked why the forestedArea data was not used, since it was the last result just before applying the reducRegions() function. Using this resulted in an error shown below.
Line 98: forestedArea.reduceRegions is not a function
In order to get around this problem, the forestArea data, which came before ee.Image.pixelArea() was applied, was used. It still contains the same data as for forestedArea, except that the pixels have not calculated their area values yet.
Export results to google drive
The GEE has functionalities that enable the user to send data to various cloud storage platforms.
Here is the script to export the area statistics table to your google drive folder. It will be worth the effort to read about the Export.table.toDrive() method of GEE.
Export.table.toDrive({
collection: countyForests, //the table with forest area statistics per county
description: 'county_forestarea_metres', //the title of the table
folder: 'test', //the folder the table will be saved into
fileFormat: 'CSV'
});Calculations of county forest cover in km2 and ha were done in google sheets and Qgis.
Calculate the percentage of forest cover in each county
Some counties may be large but with small forest coverage. Conversely, some small counties have high forest coverage. To normalize our results, percentages of forest coverage were done for each county.
The methodology was as follows:
- The spatial area of each county in km2 was calculated in Qgis. This attribute was then joined to that of the google sheets (after km2 and ha have been derived).
- The joined table was saved as an excel workbook.
- The percentage calculation of each forest coverage per county was done through the following equation in excel.
(A/B) * 100 where:
A is the forest coverage are for each county in km2 (Area_km)
B is the spatial area for each county in km2 (County_Area_km)The KFS 2021 report shows the forest cover percentage for all the counties. The top five counties are Nyeri, Lamu, Kilifi, Nyandarua and Bomet.
In our results, however, it is a paradox. The top five counties are Baringo, West Pokot, Elgeyo Marakwet, Nyeri and Murang’a. Where could we have gone wrong?
#show the results of forest cover percentage for each county
#show top six counties with largest forest cover percentage
## # A tibble: 6 x 8
## ADM0_EN ADM1_EN date sum Area_km Area_ha County_Area_km
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 Kenya Baringo 1509692400000 7.30e 9 7300. 7.30e5 10889.
## 2 Kenya West Pokot 1509692400000 6.22e 9 6215. 6.22e5 9306.
## 3 Kenya Elgeyo-Marakwet 1509692400000 1.92e 9 1920. 1.92e5 3008.
## 4 Kenya Nyeri 1509692400000 1.92e 9 1920. 1.92e5 3335.
## 5 Kenya Murang'a 1509692400000 1.36e 9 1365. 1.36e5 2526.
## 6 Kenya Kitui 1509692400000 1.41e10 14146. 1.41e6 30457.
## # ... with 1 more variable: `Percentage forest Area` <dbl>
All fingers point to the Global PALSAR dataset. By adding the map.addLayer() method to GEE, you will notice that some of the arid counties–which are the notorious top three in our resultant table data–have some of the highest forest coverage per land area.
Nothing could be further from the truth.
//view the forest coverage in counties
Map.addLayer(counties, {}, 'Counties');
Conclusion
Though with unexpected results, the above was a simple workflow for forest cover assessment in a country using desktop research. Arguably, it has also brought to the fore why higher resolution will always be a call for most studies.
This is because higher resolutions reduce chances of generalization, whereby certain land covers can be classified wrongly, such as the likelihood of woodland or scattered trees still being categorized as forest.
Nevertheless, the use of powerful AI tools such as GEE enable researchers to conduct their own background studies to either corroborate, verify or refute certain studies that may have otherwise been published or publicized. Alternatively, at points where results may fall far short of what is already acknowledged as accurate, other statistical measurements such as standard deviation to identify the margin of error can be used. The results of this work can serve as a point of departure. It was not work in vain.
Forest coverage in counties
## ADM0_EN ADM1_EN date sum Area_km Area_ha
## 1 Kenya Baringo 1.509692e+12 7300173368 7300.17 730017.34
## 2 Kenya West Pokot 1.509692e+12 6215088616 6215.09 621508.86
## 3 Kenya Elgeyo-Marakwet 1.509692e+12 1919553427 1919.55 191955.34
## 4 Kenya Nyeri 1.509692e+12 1919625779 1919.63 191962.58
## 5 Kenya Murang'a 1.509692e+12 1364808462 1364.81 136480.85
## 6 Kenya Kitui 1.509692e+12 14146377144 14146.38 1414637.71
## 7 Kenya Kirinyaga 1.509692e+12 679056736 679.06 67905.67
## 8 Kenya Kericho 1.509692e+12 1141081192 1141.08 114108.12
## 9 Kenya Lamu 1.509692e+12 2564613887 2564.61 256461.39
## 10 Kenya Kiambu 1.509692e+12 1066365916 1066.37 106636.59
## 11 Kenya Mandera 1.509692e+12 10257178183 10257.18 1025717.82
## 12 Kenya Samburu 1.509692e+12 8269434150 8269.43 826943.41
## 13 Kenya Trans Nzoia 1.509692e+12 956571720 956.57 95657.17
## 14 Kenya Bomet 1.509692e+12 854856391 854.86 85485.64
## 15 Kenya Meru 1.509692e+12 2398656929 2398.66 239865.69
## 16 Kenya Nandi 1.509692e+12 947915158 947.92 94791.52
## 17 Kenya Embu 1.509692e+12 912337543 912.34 91233.75
## 18 Kenya Tharaka-Nithi 1.509692e+12 830983192 830.98 83098.32
## 19 Kenya Kisii 1.509692e+12 412830466 412.83 41283.05
## 20 Kenya Nyamira 1.509692e+12 276010302 276.01 27601.03
## 21 Kenya Nairobi 1.509692e+12 208895198 208.90 20889.52
## 22 Kenya Nyandarua 1.509692e+12 957839381 957.84 95783.94
## 23 Kenya Nakuru 1.509692e+12 2174109541 2174.11 217410.95
## 24 Kenya Kilifi 1.509692e+12 3365582568 3365.58 336558.26
## 25 Kenya Mombasa 1.509692e+12 61467500 61.47 6146.75
## 26 Kenya Garissa 1.509692e+12 9252932604 9252.93 925293.26
## 27 Kenya Narok 1.509692e+12 3563396689 3563.40 356339.67
## 28 Kenya Taita Taveta 1.509692e+12 3308754376 3308.75 330875.44
## 29 Kenya Makueni 1.509692e+12 1331737655 1331.74 133173.77
## 30 Kenya Uasin Gishu 1.509692e+12 541016675 541.02 54101.67
## 31 Kenya Isiolo 1.509692e+12 3962383357 3962.38 396238.34
## 32 Kenya Tana River 1.509692e+12 5843615791 5843.62 584361.58
## 33 Kenya Wajir 1.509692e+12 7137564402 7137.56 713756.44
## 34 Kenya Kwale 1.509692e+12 944596929 944.60 94459.69
## 35 Kenya Turkana 1.509692e+12 7816033471 7816.03 781603.35
## 36 Kenya Vihiga 1.509692e+12 56760134 56.76 5676.01
## 37 Kenya Marsabit 1.509692e+12 7679984672 7679.98 767998.47
## 38 Kenya Kakamega 1.509692e+12 297926598 297.93 29792.66
## 39 Kenya Bungoma 1.509692e+12 286252975 286.25 28625.30
## 40 Kenya Kajiado 1.509692e+12 1942974510 1942.97 194297.45
## 41 Kenya Laikipia 1.509692e+12 808356882 808.36 80835.69
## 42 Kenya Homa Bay 1.509692e+12 149993542 149.99 14999.35
## 43 Kenya Machakos 1.509692e+12 174205360 174.21 17420.54
## 44 Kenya Kisumu 1.509692e+12 72402463 72.40 7240.25
## 45 Kenya Busia 1.509692e+12 42206501 42.21 4220.65
## 46 Kenya Siaya 1.509692e+12 38714252 38.71 3871.43
## 47 Kenya Migori 1.509692e+12 30384458 30.38 3038.45
## County_Area_km Percentage forest Area
## 1 10889.2 67.0404621
## 2 9305.5 66.7894256
## 3 3008.4 63.8063422
## 4 3334.6 57.5670245
## 5 2525.5 54.0411800
## 6 30456.8 46.4473615
## 7 1475.3 46.0286044
## 8 2513.2 45.4034697
## 9 6047.0 42.4112783
## 10 2543.0 41.9335431
## 11 25979.3 39.4821262
## 12 21015.7 39.3488202
## 13 2484.7 38.4984103
## 14 2438.2 35.0611107
## 15 6992.5 34.3033250
## 16 2835.8 33.4268989
## 17 2823.7 32.3100896
## 18 2581.0 32.1960480
## 19 1313.7 31.4249829
## 20 897.2 30.7634864
## 21 707.5 29.5265018
## 22 3266.7 29.3213335
## 23 7487.1 29.0380788
## 24 12599.3 26.7124364
## 25 233.6 26.3142123
## 26 43596.0 21.2242637
## 27 17912.2 19.8937037
## 28 17131.1 19.3142880
## 29 8174.9 16.2905968
## 30 3396.8 15.9273434
## 31 25399.3 15.6003512
## 32 39192.2 14.9101607
## 33 56668.6 12.5952644
## 34 8266.0 11.4275345
## 35 70131.8 11.1447731
## 36 560.5 10.1266726
## 37 76027.8 10.1015418
## 38 3008.0 9.9045878
## 39 3016.2 9.4904184
## 40 21884.8 8.8781712
## 41 9536.9 8.4761296
## 42 4731.9 3.1697627
## 43 6044.3 2.8822196
## 44 2665.0 2.7166979
## 45 1811.6 2.3299845
## 46 3520.4 1.0995910
## 47 3144.3 0.9661928
Forest cover assessment using GEE