Republic of KenyaKenya Forest ServiceKenya NFMS-FIPNational Forest Monitoring System – Forest Information PlatformHealthy Forests • Resilient Ecosystems • People and Nature • Data for Action

Accuracy Assessment

Reference data, accuracy metrics and validation results.

Accuracy Metrics

The overall accuracy of the 2018 land cover / land use map is 76.04%, from 3,649 randomly generated reference points. Accuracy is used to check the maps and to estimate the uncertainty of the activity data.

The Why? panel beside the forest cover figure shows this assessment with the product's provenance.

Accuracy of the SLEEK land cover maps, 2000–2018
60%70%80%90%2000: overall accuracy 83.018%, kappa 0.743002002: overall accuracy 87.03%, kappa 0.815 (map used in the FRL)022003: overall accuracy 83.931%, kappa 0.738032004: overall accuracy 81.611%, kappa 0.705042005: overall accuracy 82.258%, kappa 0.749052006: overall accuracy 88.713%, kappa 0.828 (map used in the FRL)062007: overall accuracy 78.227%, kappa 0.697072008: overall accuracy 78.001%, kappa 0.688082009: overall accuracy 89.485%, kappa 0.851092010: overall accuracy 82.392%, kappa 0.748 (map used in the FRL)102011: overall accuracy 81.818%, kappa 0.727112012: overall accuracy 77.526%, kappa 0.705122013: overall accuracy 83.139%, kappa 0.764132014: overall accuracy 75.635%, kappa 0.7025 (map used in the FRL)142015: overall accuracy 78.87%, kappa 0.727152018: overall accuracy 76.021%, kappa 0.705 (map used in the FRL)18
Maps used in the FRL (2002, 2006, 2010, 2014, 2018)70% (the report’s acceptable limit)

All maps used in the FRL exceed 70% overall accuracy. Accuracy is lower in recent years (75.6% in 2014, 76.0% in 2018).

Accuracy of the 2018 map by class (3,648 reference points)
Dense forest63%74%
Moderate forest41%50%
Open forest34%43%
Wooded grassland87%82%
Open grassland83%69%
Perennial cropland78%73%
Annual cropland80%82%
Vegetated wetland58%82%
Open water73%83%
Other land76%83%
Producer’s accuracy (how much of the real class was mapped)User’s accuracy (how often the map is right)

Forest classes are the hardest to map: open forest is correctly mapped only a third of the time and moderate forest about 40–50%, while grassland and cropland exceed 75%. This is the main reason gross forest change is uncertain.

Accuracy of the change maps (Table 34)
Change mapOverall accuracyUncertainty
2002–200641.05%±4.94%
2006–201051.9%±4.03%
2010–201435.75%±2.17%
2014–201830.01%±2.15%

Source: Ministry of Environment and Forestry, National Forest Monitoring System in the Republic of Kenya, Version 1 (Draft), February 2021. Chapter 5.1.1 (7) and Appendix 2.

Source: Ministry of Environment and Forestry, The National Forest Reference Level for REDD+ Implementation, December 2019 (Kenya’s FRL submission to the UNFCCC). Chapter 7.1, Tables 32–34.