Senegal Land Cover Mapping. Ugo Leonardi FAO GLCN - Land Cover Mapping/Remote Sensing Specialist email@example.com. Presentation Topics. Basic dataset Ancillary data collected and used Photo-interpretation Effects of vegetation seasonality Field verification campaign & results
FAO GLCN - Land Cover Mapping/Remote Sensing Specialist
The images used for the interpretation work are:
Centre de Suivi Ecologique made available to the project aerial photographs of Tamba, Kolda, Salemata and Ouli regions, and the interpreted shapefiles of Tamba and Kolda regions. The legend of the former interpretation was translated in the respective LCCS classes and the polygons smaller than 5 ha have been eliminated
Before using the Tamba and Kolda shapefiles as reference base for the interpretation, they have been georeferenced again since they displayed a shift with the 2005 images.
CSE contributed to the land cover mapping, also giving interpreted shapefiles covering the Dakar area; the one of 1999 was used as base for the interpretation of the Dakar area. The codes have been translated in LCCS classes and small polygons (< 5ha) were eliminated.
USGS contributed to the Senegal land cover mapping with 758 geocoded aerial photographs, covering Senegal and Gambia, taken during the 1994 country aerial survey campaign.
The aerial photos became a reliable reference point during the photo-interpretation work, since they have been linked with the points of their position.
Aerial photographs give a better perspective compared to the field photos, showing effectively the spatial distribution and land cover features.
The original classes of the former 1984 Senegal Land Cover map (scale 1:1.000.000) have been grouped in major land cover classes.
The vector shapefile was converted in a raster file and georeferenced, giving one more interpretation tool for the photo-interpretation, since it shows the distribution of the main types of vegetation covers according to Senegal climatic zones.
In addiction to the ancillary data collected, the Google Earth freeware (http://earth.google.com/) gave an extraordinary chance to photo-interpreters to detect the land cover feature.
The implementation of Senegal Land Cover map is based on the multi-phase image interpretation approach, which was successfully used by FAO in a number of projects.
The visual interpretation was carried out using the GeoVIS software (http://www.geovis.net/), a vector-based editing system specifically designed for thematic interpretation.
It is a user-friendly system that embeds the main tools of vector drawing and editing, including topological functions, with advanced capabilities of raster management (Radex) and a direct link with LCCS (Land Cover Classification System) software.
The photo-interpretation mapping scale was 1:100,000
During the mapping activities, the GeoVIS “Multiple Windows” tool was used to visualize, at the same time, the Radex mosaic of both dates 2000 and 2005.
The digitization base was the 2005 mosaic even if it shows, in some portions, black strips due to the Scan Line Corrector failure, affecting Landsat satellite sensor from 2003 onward. Whenever the noise caused by the black strips made difficult the interpretation of the 2005 image, then the 2000 one was used as reference base.
As concern the visual interpretation, more weight was given to the image showing the driest situation, in order to avoid an overestimation of the vegetation cover, getting a more reliable interpretation. In fact, the herbaceous layer presents during the wet season most of the time covers the reflectance of trees and shrubs, making sometime a difficult task to separate the different natural vegetation classes.
In Senegal, usually the November date is the best one, since the herbaceous layer is almost dry, while trees and shrubs still have green leaves.
Concerning the agricultural areas falling inside the so called Peanut Basin, it was decided to map the agriculture present in both dates. So, the agricultural classes displayed on the final interpretation of this area will show the sum of the agricultural areas of the period 2000-2005.
In fact, the whole Peanut Basin is a big agricultural area, where fields may have a fallow period that, in this case, was considered no longer than 5 years. During the fallow period the cover consists mostly of grass and light bush vegetation. Natural vegetation mapped inside this area was detected both in 2000 and 2005 image.
In same cases the use of Google Earth produced controversial interpretations (amended during the land cover revision), due to the drastic changes in vegetation cover appearance caused by seasonality. This change is especially marked in woody vegetation which in Senegal normally is broadleaved deciduous.
It means that woody vegetation, during the dry season, is leafless so if the acquisition date of the image analyzed corresponds to the dry season, the woody vegetation almost disappears.
Therefore, the use of the Google Earth high resolution images imply a good knowledge of the area seasonality, i.e. when both dry and wet season occur, in order to give a correct interpretation of the vegetation cover
One more example of vegetation seasonality effects
On the other hand, the analysis of images with different acquisition dates, in same cases, gave the chance to determine the extension of flooded areas and to estimate the water persistence.
At the end of May 2007, after the completition of the preliminary interpretation, a field work campaign was carried out. The steps to organize the field work campaign can be summarized as follow:
Field verification campaign
The field verification work was performed by two groups in the same period. Each group had the task to reach the points uploaded on the GPS and fill the Field Verification Form for each point.
The two routes programmed, passed from the following places:
Route 1: Dakar – Thies – Kebemer – Louga – St. Louis – Richard Toll – Dagana – Salde – Linguere – Dara – Louga – Keur Momar Sarr - Dakar.
Route 2: Dakar – M’Bour – Fatick – Kaolack – Koungheul – Tambacounda – Goudiri – Kidira – Saraya – Kedougou – Niokolo Koba – Tambacounda – Dakar.
The result of the field verification campaign are a total of 171 point checked along two different Route. For each point a Field Verification form was comipled.
Moreover, 706 extra points have been taken all along Route 2.
The data collected was arranged in an Arc View shapefile where both fixed points and extra points are coded, described and hot-linked with photographs in an interactive database.
FAO, through Africover Project and Global Land Cover Network, has developed a comprehensive, standardized a priori land cover classifications system (LCCS).
This methodology was applied to shape the land cover classes of Senegal and which will be explained in detail later, with examples taken from Landsat ETM, Google Earth High Resolution imagery, aerial photograph and field photos.
Vedas (Vegetation Dynamic Assessment) software was applied in East Africa during 2007 GLCN activities and was demostrated that this procedure is able to extrapolate eco-climatic information on GLCN layer.
In October 2008, the Vedas procedure was tested on the Senegal land cover dataset, with Modis 005 remote sensed vegetation data (250 mt resolution – 16 days period) and with Spot vegetation data (1km resolution – 10 day period).
Summarizing, the average NDVI values (calculated in the 2001-2007 period for Modis, and 1999-2006 period for Spot) was extracted for each polygon of the Senegal land cover dataset, providing consistent spatial and temporal comparisons of the vegetation conditions, and monitoring vegetation activity in support of phenologic, change detection and biophysical interpretations.
NDVI profiles of different land cover classes can differ in mean values but tend to have a similar shape linked to the seasonality of local vegetation.
F.A.O. data distribution policy, provide for the creation of an aggregated dataset starting from the original one. For this reason two versions of Senegal land cover dataset exist:
The original full resolution dataset, consisting of 23,922 polygons
The spatially aggregated dataset, consisting of 21,238 polygons
The original full resolution dataset was aggregated on the basis of a spatial criteria rather than a thematic one, producing the reduction of about the 11% of the total amount of polygons.
The classes listed in the below tables have been aggregated only when occurring as single units and considering the following spatial criteria: after sorting the polygons according to their size, for each class was calculated the amount of polygons
corresponding to the 20% of the total number. The resulting number of polygons was eliminated starting from the smallest size forward.
Given that mixed units represent already a spatial generalization, they were not considered in the aggregation process, except for the classes created specifically to be used in mixed units, which are:
The above classes are always found associated with other classes in mixed units, but they have been aggregated using the same procedure explained for the single units.
The aggregation process was performed with ARCGIS software using the “Eliminate” extension, after selecting all the polygons to be aggregated. The selected polygons have been merged with the neighbouring unselected one with the largest area, by dropping the shared border.
The below single units have not been aggregated: