All Over the Place: February 2012

We have done a lot of work collecting data in the Harapan Rainforest. In this blog we have an overview on how all that data is used to produce a vegetation map.

Our focus in Harapan Rainforest was to summarize the condition of the forest. Satellite imagery (from the French satellite SPOT) from 2009 was used to generate a very basic classified image. This imagery was three years old, but it was the most recent high resolution image available of the entire Harapan Rainforest boundary with little cloud cover we could find (it is very wet and cloud there).

In this map the darker the red the more developed the forest
(or at least that is the theory we want to test).

In general we view forest in shades of red, a false colour composite (FCC). This is where the images colours are shifted to allow use to use near infra-red. Healthy vegetation will reflect the near infra red (making it looking red in the resultant image). The naked eye can discriminate more shades of red than any other colour.

Red-Green-Blue image display colour corresponds to 3-Green, 2-Red, 1-Near-Infrared bands.

Red shades: Vegetation

Cyan shades: Low vegetation/ clearings

White/black: No data due to cloud/ cloud shadows

A quick classification of the SPOT imagery gave us an indication of differences we might be able to identify on the ground within the forest. We were not entirely sure what those differences were and that was why we need to see them on the ground.

Colours were associated with ‘best guess’ vegetation classes
(without any ground-truth data). Good forest in red tones,
younger forest blues and or disturbed areas into the greens.

In the forest we used pre-prepared image analysis maps (such as the classification map above) to guide our daily route of ‘plots’ through the forest. We were looking at the different blocks of colour as an indication of different forest cover levels. We also looked at changes between 2009 and 2011, which would indicate re-growth or loss.

We loaded these images on to android smart phones (we used the Samsung Galaxy S2 and Samsung Galaxy Note devices, in the field) to view, interact and set interest points using Locus Pro.

At each plot we collected data:

location coordinates

environment e.g. hill
slope, swamp etc.

percentage forest cover

size of trees

dominant tree species

dominant understory
vegetation

All the information we gathered in the field was entered into digital forms (ODK collect) on the smart phones.

We periodically sent data up to the server when we had WiFi connection back a base camp. Back in Kew the data was collated and published on the GIS Unit Expedition Maps webpage. Geotagged tweets as well as our ODK datasheets were mapped in near-real-time.

Roki records tree size and species for large trees in the plot

It was a very successful field campaign with over 300 plots surveyed with detailed canopy measurements and environmental variables with associated geo-locations, photographs and videos.

Back in the office at Kew we cleaned up the data, filled-in blanks and attached photos taken with our own cameras. The field data was broken into thirds. One third was used to train the satellite imagery into defined forest categories and the last two-thirds to test the classification.

The forest categories we used were: Old Secondary Forest / Disturbed, Young Secondary Forest / Disturbed, Very Young Secondary Forest (Thicket / Disturbed), Scrub, Cleared. After a few iterations a base map was produced for the area. Our results show that all forest levels showed good potential for regeneration from species identified in the field. There was a lot more intact forest than we suspected from our initial work with the satellite imagery. See some of the results from the map.