Hi there,

Depending on your term of canopy cover percentage, the analysis may varies. If you're talking about cover percentage, then, you should consider the nature (i.e. diameter, composition) of the tree that you want to identify. It will determine the specification of image that can be effectively used.

using very high spatial resolution data won't be effective because each pixel is sub-canopy extent, and thus, at per-pixel analysis, there will be many pixels that have 100% cover because the pixel consist only tree leaves.

the term percentage itself means that it is the percentage of area covered by tree canopy per unit area. Thus, you should define your unit of analysis before progressing. If not, quantifying the percentage will be difficult and the result will be difficult to compare.

If you have already finish with the aforementioned issues, there are many method that can be used to improve satellite-based canopy cover percentage identification such as vegetation index and unmixing. Furthermore, to obtain a real quantiative estimate of canopy cover, you should at least perform field survey, or if you use medium spatial resolution image, you can use higher resolution data to get canopy cover information. Afterward, you can perform empirical modeling using vegetation index, PC bands, image fractions or other approach.

And if you want to separate tree from background reflectance i.e. grass, you can try applying PCA with vegetation mask on. This will maxed out the variation of vegetation pixels and object separation will be easier.

Hope this help.

Thank you.

Check out Leave Area Index (LAI). Village level sounds like geometrically small scaled sensors.

We are using high resolution optical satellite imagery (e.g. RapidEye) to classify tree crowns in the first step. In a second step we apply a moving window to produce crown cover maps. Some of this work was presented at the ForestSAT Conference 2012: https://www.researchgate.net/publication/233427612_A_new_framework_for_standardized_forest_cover_mapping_based_on_the_FAO_forest_definition

A very critical point in crown cover assessment is the definition of the reference area on which crown cover is estimated. A topic which is discussed in detail here: https://www.researchgate.net/publication/233395082_Uncertainties_of_forest_area_estimates_caused_by_the_minimum_crown_cover_criterion__-a_scale_issue_relevant_to_forest_cover_monitoring

Conference Paper A new framework for standardized forest cover mapping based ...

Article Uncertainties of forest area estimates caused by the minimum...

Assuming you have no access to LiDAR or microwave data, see paper by Wolter et al. (2009) Estimation of forest structural parameters using 5 and 10 meter SPOT-5 satellite data, Remote Sensing of Environment (113) 2019-2036.

Thank you for sharing your answers.

You can also try it with derivig the (i) the NDVI (Normalized Differenced Vegetation Index) and (ii) based on the NDVI the FVC (Fractional Vegetation Cover) given in per cent.

I used very high resolution satellite image (GeoEye-1) for the extraction of tree crown. its very easy to delineation/extract through eCognition algorithms, for your understanding please have a look attached file.

well... using high res imagery is of-course very useful but at the same time its a bit expensive to have, then using commercial software such as eCognition is again the same thing. I have used Berkeley Image Segmentation software (http://www.berkenviro.com/berkeleyimgseg/) could be better alternate.

Thus using NDVI (and then FVC) calculated from freely available satellite data such as Landsat could be perhaps more worthy.

See our paper Schönbrodt et al. (2010) were we derived the crop cover based on NDVI and FVC. Perhaps the literature part is also relevant for you.

I agree with Zafeer. But at the same time when are taking about the accuracy and our study area is very small, Landsat or any other medium resolution is not giving good result, specially when we have to see the change in canopy for REDD MRV.

If you have very low spatial resolution imagery (like MODIS) you might want to use

"Spectral Mixture Analysis" to get within pixel green cover percentage. There are a ton of papers out there discussing either Linear Spectral Unmixing or Spectral Mixture Analysis or even MESMA. ENVI offers these tools.

If your are working on MODIS images then you can think about the subpixel classification, where you can easily see the percentage wise presence of classes (in your case Tree Crown Cover)

Tree Crown Cover very easy to delineation through Idrisi or eCognition (Definiens ) algorithms

I've used ENVI's feature extraction.

Have a look at this thesis, it's in spanish but can be usefull

http://www.tesis.uchile.cl/tesis/uchile/2006/montaner_d/sources/montaner_d.pdf

If you want to distinguish tree crown from over land cover, you need Hi Res images. NDVI and FVC would only give you rough approximation. This is not free, but in India, through NRSA it is not so expensive.

You can read :

Linhai Jing, Baoxin Hua, Thomas Noland, Jili Li, An individual tree crown delineation method based on multi-scale segmentation of imagery, ISPRS Journal of Photogrammetry and Remote Sensing 70 (2012) 88–98

and

Le Wang, Peng Gong, and Gregory S. Biging, Individual Tree-Crown Delineation and Treetop Detection in

High-Spatial-Resolution Aerial Imagery, PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 2004, 352-358

This really depends on what resources you have available to you. As many people have mentioned, high-spatial (and spectral) resolution imagery is preferable as if makes it easier to distinguish between tree and grassy cover. However, one option that may be feasible, depend on your site, is using imagery acquired during the dry season as the grass may be browner then and it is easier to differentiate between dry ground and tree cover. The disadvantage of this is ifyou have lots of deciduous trees you will lose tree canopy cover as well. '

However, using an NDVI or a tassled-cap VI (depending on your imagery) would probably be quite effective for your study.

Regarding the choice of acquisition dates, Miranda is perfectly right. It is a very important parameter. For deciduous trees, adding LIDAR data to high spatial resolution image could help you, but the cost of acquisition is high. If the vegetation cover is low (deciduous in winter for example), you can try the SAVI index (SAVI, TSAVI, MSAVI ....). You can read (Huete,1988) but maybe it is in French, you will easily find other references.

There's some great work done in Australia on tree cover mapping. The trick is to get imagery at a time when the understory has senesced so that the only green targets in the imagery are permanent woody cover

http://www.scribd.com/doc/37459001/15arspc-Submission-64

You can get from MODIS data warehouse

Hi there,

Depending on your term of canopy cover percentage, the analysis may varies. If you're talking about cover percentage, then, you should consider the nature (i.e. diameter, composition) of the tree that you want to identify. It will determine the specification of image that can be effectively used.

using very high spatial resolution data won't be effective because each pixel is sub-canopy extent, and thus, at per-pixel analysis, there will be many pixels that have 100% cover because the pixel consist only tree leaves.

the term percentage itself means that it is the percentage of area covered by tree canopy per unit area. Thus, you should define your unit of analysis before progressing. If not, quantifying the percentage will be difficult and the result will be difficult to compare.

If you have already finish with the aforementioned issues, there are many method that can be used to improve satellite-based canopy cover percentage identification such as vegetation index and unmixing. Furthermore, to obtain a real quantiative estimate of canopy cover, you should at least perform field survey, or if you use medium spatial resolution image, you can use higher resolution data to get canopy cover information. Afterward, you can perform empirical modeling using vegetation index, PC bands, image fractions or other approach.

And if you want to separate tree from background reflectance i.e. grass, you can try applying PCA with vegetation mask on. This will maxed out the variation of vegetation pixels and object separation will be easier.

Hope this help.

Thank you.

you may get some idea from the http://www.geospatialworld.net/Paper/Application/ArticleView.aspx?aid=30547

You could have a look at this PhD thesis:

Ardila Lopez, J.P., (2012) Object - based methods for mapping and monitoring of urban trees with multitemporal image analysis. Enschede, University of Twente Faculty of Geo-Information and Earth Observation (ITC), 2012. ITC Dissertation 209, ISBN: 978-90-6164-333-3

or at the following papers which are contained in this thesis:

Ardila Lopez, J.P., Bijker, W., Tolpekin, V.A. and Stein, A. (2012) Context - sensitive extraction of tree crown objects in urban areas using VHR satellite images. In: International Journal of Applied Earth Observation and Geoinformation : JAG, 15 (2012) pp. 57-69.

Ardila Lopez, J.P., Bijker, W., Tolpekin, V.A. and Stein, A. (2012) Multitemporal change detection of urban trees using localized region - based active contours in VHR images. In: Remote sensing of environment, 124 (2012) pp. 413-426.

Ardila Lopez, J.P., Bijker, W., Tolpekin, V.A. and Stein, A. (2012) Quantification of crown changes and change uncertainty of trees in an urban environment. In: ISPRS Journal of Photogrammetry and Remote Sensing, 74 (2012) pp. 41-55.

Ardila Lopez, J.P., Tolpekin, V.A., Bijker, W. and Stein, A. (2011) Markov - random - field - based super - resolution mapping for identification of urban trees in VHR images. In: ISPRS Journal of Photogrammetry and Remote Sensing, 66 (2011)6 pp. 762-775.

Try to have a look to the following discussion at LinkedIn:

http://www.linkedin.com/groupItem?view=&gid=1317037&type=member&item=216667822&qid=af702c99-e7c1-41b6-b2d4-80d082e32340&trk=groups_search_item_list-0-b-ttl