Defining activity data is considered to be the most difficult part when a greenhouse gas (GHG) inventory is implemented, particularly for the areas where land use class detected to be changed (for instance, deforestation / reforestation areas). Usually, countries use available ground-based information (e.g., national statistics for agriculture, forestry, wetland and urban areas;
vegetation and topographic maps, climate data) with remote sensing data (e.g., aerial photographs, satellite imagery etc.), with the application of GIS-based methods (GOFC-GOLD 2016). A separate monitoring system is an alternative way to extract independent information. Some of good examples are found in Brazil, India and Congo – the Brazilian system generates annual deforestation estimates (Morton et al, 2005) in Amazon, the Indian National bi-annual forest cover assessment (FSI, 2013), and a sampling approach applied in the Congo basin (GOFC-GOLD 2016).
The Brazilian National Space Agency (INPE) produces estimates through annual national monitoring program PRODES (Wheleer et al, 2014). The PRODES is working starting from 1988, with the minimum mapping unit of 6.25 ha. Furthermore, the project is carried out once a year in order to use dry conditions and cloud free time for deriving estimates, it provide with results of foreseen in December. PRODES uses imagery from TM Landsat imagery, DMC satellites , and CCD sensors, with a spatial resolution of 20 to 30 meters (Wheleer et al, 2014). In India, The assessment of XII cycle applied satellite imagery from the Indian satellite IRS P6 (Sensor LISS - III with 23.5 meters resolution). Only imagery with less than 10% clouds were selected for the 313 LISS - III scenes covering India.(GOFC-GOLD 2016).
As one example of a regional research is project implemented in Congo. A systematic sampling approach with Landsat imagery was performed to the entire Congo River basin to estimate deforestation. The survey formed sample plots of 20×20 square kilometres systematically distributed every 0.5° in the whole forest area of Central Africa. Then, 547 sample plots were
established over the Congo Basin. Data were utilised from Landsat TM and ETM+ imagery of 1990, 2000, and 2005. The satellite imageries were processed using unsupervised classification procedures. The results were formed into a change matrix for each sample plot with four land use change classes – deforestation, reforestation, forest degradation, and forest recovery. Plots, where change was detected, were classified into 10 land classes, such as dense, degraded forest, long fallow and secondary forest, forest/agriculture mosaic, agriculture and short fallow, bare soil and urban area, non-forest vegetation, forest - savannah mosaic, water bodies, and no data. Degraded forest was detected spectrally using imagery.(GOFC-GOLD 2016).
In the local scale mapping, Tokola and Hou (2012) has mentioned that Scandinavian methods are applicable in tropical forests despite different climate zone and growing conditions.
Airborne laser scanning data are widely utilised in Scandinavia. If optical satellites are compared with LiDAR data, there are some disadvantages, for instance, limit of two dimensions of satellites (Tokola and Hou, 2012).
A forest inventory will produce more accurate results if conducted as a combination of remote sensing techniques and field data. Remote sensing can solve problems in case of biomass inventory on local level, for instance, when some areas are not accessible (Tokola and Hou, 2012). Among the variety of methods, Tokola (2015) mentions a sampling approach using visual interpretation plots. Imagery and tools are freely available for this type of forest cover assessment (openforis.org). The method concerns recording canopy cover (from 0 to 100%) of sample plots through visual interpretation; and to expand it to the whole target territory.
However, clouds, season, or unavailable (low quality) satellite imagery affect the accuracy and bring some form of bias. One of the solutions to obtain better results and detection of forest cover changes with freely available optical data is to analyse images of different dates.
Although factors such as seasonality are important to consider for tropical forests (Tokola, 2015).
Among all available data and methods, those which are most accurate are frequently relatively expensive, such as conducting forest inventories using LiDAR (laser detection and ranging).
Freely available satellite data of 30 metres resolution will produce a higher bias without any field inventory data. Imagery with the resolution of 10 metres allows for the detection of change in forest cover without field data. However, if the use of satellite imagery is combined with LiDAR data for canopy cover mapping in tropical forests, it will produce reliable results in comparison with field data (Tokola, 2015). This is of great importance for inaccessible areas.
There is a large variety of methods to estimate canopy over using optical satellite imagery. The main focus of this research was on two statistical approaches and visual assessment. Visual assessment was necessary to create true data sets.
The statistical approaches utilised for this work are aimed at creating a model for each variable in order to predict a target value (in this case, canopy cover), where variables were vegetation indexes and satellite bands. One of objectives of this research was to explore methods of canopy cover estimation and their accuracy. When summarising the outcomes of accuracy assessment of the two approaches used, it is necessary to outline some important results. According to the multinomial test, the highest overall accuracy of 0,6838 was achieved with the Sparse Bayesian regression and 50% of threshold. Generally, the non-forest class was the most complicated to predict. According to confident intervals, where about 95% of values are predicted correctly, values from 0,34 – 0,77 are suitable for the used methodologies to obtain reliable results.
When conducting tests using 10 vegetation indexes and 6 satellite (Landsat 8) bands, some were found to be utilised more frequently than others. Sparse Bayesian method’s indexes were following (listed by repetitiveness) Band 5, MSAVI, NDVI, NBR, ARVI, NR, TNDVI, IPVI, SR, band 2, and band 7. Since it was possible to derive the value which represents the impact of each variable for the prediction, SR is important in this particular case. This index has been utilised twice, but it took part in prediction for 100% of plots. ZONIBR method’s indexes listed by repetitiveness are band 5, NBR, band 7, NDVI, MSAVI, NR, band 2, band 4, and band 6.
Overall, indexes were different depending on image and season. Although, some variables were selected for both approaches and utilised two or more times. There are Band 5, NBR, Band 7, NDVI, MSAVI, NR, and Band 2.
Concerning forest classes, the most complicated interval to predict was from 0 until 30% as the confidence interval starts from 34% and for some results it starts from 50 – 59%. Likewise, canopy cover with 70% and more was difficult to predict. For instance, if non-forest is below 10% canopy cover, open forest is in between 10% and 40 or 50% and medium dense forest is in between 40 - 50% and 70 - 80%, non-forest and dense forest classes will be harder to forecast than open and medium forest classes. Accuracy is lower for non-forest and for dense forest classes.
Visual assessment with Google Earth allows obtaining canopy cover percentage using freely available imagery, which opens many possibilities to perform land use land cover classification.
However, there are some limitations, for instance, low quality of images for some periods of time. This kind of analysis helps to record forest classes and to have an overview of the target area.
There are only two methods analysed in this case, Sparse Bayesian regression is applied mostly for Scandinavian boreal forest and ZONIBR method was performed for working with data from Laos. Tropical forests have unique features and it brings some uncertainty while estimating different variables (Achard et al., 2014). Nowadays, application of LiDAR in tropical forests is limited by its availability; despite limitation of optical data, they are freely downloadable and, consequently, widely utilised. Although, effort is put into launching global space LiDAR collection by the United States (GOFC-GOLD, 2014). It will open many opportunities for scientists.
As a conclusion to this work, it is important to outline that utilised methodologies are applicable with existing datasets with REDD+ projects. Moreover, they use data sources recommended by FAO and which can be served to produce reliable results taking into account uncertainty.
Although, limitations of application of the methods are still a concern, Improvements are necessary, especially, ground data capture. LiDAR or field measurements would be a good alternative, which can diocese uncertainly and reduce limitations. Sparse Bayesian method can be a suitable tool to derive necessary vegetation indexes to monitor changes in canopy cover and see how each of them affect the model. Finally, Sparse Bayesian statistical approach showed slightly better accuracy than ZONIBR method. It is still an unknown what results can bring presence of field data or using radar data. In terms of costs, ZONIBR method is applied in open source software. At the present moment both methods are recommended to use for crown cover modelling for REDD+ programmes purposes, depending on request and available data sources.