ABSTRACT

A study was conducted to estimate the forest cover change, quantify and map tree above-ground carbon stock using Remote sensing and GIS techniques together with forest inventory. Landsat images of 1980, 1995 and 2010 acquired during dry season were used in the estimation of cover changes. Supervised image classification using Maximum Likelihood Classifier was performed in ERDAS Imagine software to analyze the images and further analysis was performed in Arc GIS 9.3 software. Stratified sampling procedure was used to select concentric inventory plots in Pugu Forest Reserve (PFR) and Kazimzumbwi Forest Reserve (KFR). Plots were laid according to NAFORMA, and the tree parameters in each sampling plot were collected. A Microsoft Excel spreadsheet was used to compute the above-ground biomass for each plot using an empirical equation relating wood basic density and tree height. The above-ground carbon was calculated using a conversion factor of 0.49. Geostatistical method in ArcGIS was used to analyze and map carbon. Results revealed that for the periods 1980-1995 and 1995-2010, Closed Forest in PFR decreased by 4.5% and 25.3% respectively, while for KFR, Closed Forest decreased by 11.9% and 31.3% respectively. The mean carbon density for PFR and KFR were respectively 5.72 tC/ha and 0.98 tC/ha while carbon stocks were 14 730.41 tC and 7 206.46 tC respectively. The revealed low carbon densities were attributable to decline in area under Closed Forest in the two Forest Reserves. The study recommends concerted efforts to enhance proper management of the forests so that the two forest reserves may contribute to REDD initiatives.

1. Introduction

Global warming which is largely contributed by increased anthropogenic Green House Gases (GHGs) represents significant development challenges for this 21st Century. It is well known that the main cause of global warming is the increase of anthropogenic Carbon dioxide (CO₂₎ emission, which accounts to 77% of all GHGs [1].

Reducing carbon emissions is of great important in this era of climate change. Several mechanisms have been developed by United Nations Framework Convention on Climate Change (UNFCCC), these includes cutting down CO₂ emissions from Annex 1 countries, and reducing deforestation and forest degradation in developing countries. Although UNFCCC requires cutting down of CO₂ emissions, tropical deforestation contributes to 20% - 25% of annual global emissions of CO₂ [2] and it is said to be the second largest global source of anthropogenic carbon dioxide emissions [3].

Reducing Emissions from Deforestation and forest Degradation (REDD) concept requires developing countries to be involved in the management of forests including reforestation and afforestation. The REDD concept is proposing to finance all activities that contributes to the improvement of forests condition, whereby developing countries that contributes to the reduction of carbon emissions may be able to sell carbon credits to the international carbon markets. In order to finance for carbon emission reduction, quantification of forest carbon is of great important.

This paper presents an attempt to integrate satellite imageries and ground based inventories in the estimation of tree above ground biomass and carbon for the Pugu and Kazimzumbwi Forest Reserves (PKFR) in the coastal areas of Tanzania. PKFR are among of the important biodiversity hotspots in Tropical Africa [4], but they are under threat of being depleted due to anthropogenic activities including deforestation and forest degradation. Due to these anthropogenic activities, it is not well known how much forest covers have changed and the quantities of the available tree above ground carbon stocks are not known.

2. Materials and Methods

2.1. Study Area

Pugu and Kazimzumbwi Forest Reserves (PKFR) are located in Kisarawe District in the Coast region of Tanzania (Figure 1). Kisarawe District is located at 7˚10'0S, 38˚49'60E and bordered by the following Districts: Ilala to the East, Kibaha to the North, Morogoro to the West and Rufiji to the South. Kisarawe District receives annual rainfall of about 1236 mm, which is 20% more than Dar es Salaam. The area experiences two rainfall regimes. It receives short rains (Vuli) from October to December and the long rains (Masika) from March to May. Temperature in the District ranges between 24˚C - 31˚C [5]. According to 2002 Census, the population of Kisarawe District was 95 615 [6]. Currently, the population has continued to increase as a result of human natural birth and immigration. The PKFR are adjacent to each other (Figure 1), one to the north and the other one to the south. The altitude ranges between 100 and 350 m above mean sea level for PFR and 120 and 270 m above mean sea level for KFR [5].

2.2. Methods

2.2.1. Remote Sensing Data and Its Processing

Table 1 presents the Landsat images used for the study. All images were acquired during dry season between June and July so as to minimize seasonality and cloud

Table 1. Remote sensed data used in the analysis of forest cover changes.

TM = Thematic Mapper; MSS = Multi Spectral Scanner.

effects.

The acquired image scenes of the years 1980 and 2010 had already been geo-rectified by the supplier. To ensure accurate identification of temporal changes and geometric compatibility with other sources of information, image to image geo-correction was conducted to rectify the 1995 imagery based on 2010 image. Images enhancement was performed using a 4,5,3 color composite band combination and its contrast was stretched using the Gaussian distribution function followed by high pass filter 3 × 3 to increase the visibility of the ground control points in both images. The first order polynomial transformation and nearest neighborhood interpolation [7] was applied to geometrically rectified the 1995 imagery and registered to the UTM map coordinate system, Zone 37 South, Datum Arc 1960.

Base maps were prepared based on the image acquired on 7th July 2010 and used in ground truthing exercise. The essence of conducting ground truthing was to verify different covers types as described on the base maps and for collection of ground points for the classification accuracy assessment. Supervised classification, using Maximum Likelihood Classifier [7-9] was performed applying ERDAS IMAGINE software. Training fields were identified by inspecting an enhanced colour composite imagery. Areas with similar spectral characteristics were trained and classified. The error matrices [10] were prepared and used in computation of Kappa coefficient for the classification accuracy assessment.

To analyse the changes between different time epochs, change detection analysis was performed. Many change detection methods have been developed and used for various applications. However, they can broadly be divided into: post-classification approaches and spectral change detection approaches [8]. The post classification change detection method was applied followed by spatial overlay analysis [11] in ArcGIS environment resulting into attribute tables. The tables were exported to MS-Excel to compile area change detection matrix for 1980-1995 and 1995-2010 periods. The estimation for the rate of change for the different covers was computed based on the formulae [8].

2.2.2. Carbon Quantification

Stratified sampling technique [12] based on vegetation cover classification was performed. Based on 2010 Landsat TM imagery, the vegetation covers obtained were Closed Forest, Open Forest, Bushland and Grassland. Each cover class was considered as stratum. Settlements and other Land uses class (e.g., bareland and cultivated land) were not included during inventory, because they had little vegetation.

Concentric plot [13] was adopted and used in this study. To ease the counting process, each sample plot was sub-divided into four sub-plots (concentric plots) of radius 2 m, 5 m, 10 m and 15 m. At the radius of 2 m, all trees with dbh <2 cm and >1 cm were recorded; at the radius of 5 m, all trees with dbh >= 2 cm and <10 cm were recorded; at the radius of 10m, all trees with dbh >= 10 and <20 cm were recorded; and at the radius 15 m, all trees with dbh >= 20 cm were recorded. Tree diameters were measured using veneer caliper and tree heights were measured using Suunto hypsometer. A botanist and local people were engaged for the identification of botanical/ scientific names and local names of trees respectively. The number of sample plots and the distance between plots were determined by the formula [14];

(3)

where N = number of sample plots, = Total area of the forest, = Plot size and = Sampling intensity, while the distance between plots was determined by the formula:

(4)

where D = inter plots distance (m), Af = Area of the forest (ha) and N = number of plots.

The adopted sampling intensity was atleast 0.1%. Therefore for KFR, the total area of Bushland, Closed Forest, Grassland and Open Forest, based on 2010 landsat image classification was 4820.8 ha, making a total of 68 plots while in PFR, the total area for Bushland, Closed Forest, Grassland and Open Forest based on 2010 landsat image was 2230.1 ha, making a total of 33 plots. Transects were created, where in each transect, concentric plots of radius 15 m (0.07 ha) were systematically located at 842 m and 822 m intervals from each other in the North-South direction in KFR and PFR respectively, (Figure 2). According to [15], plots need to be allocated systematically so as to achieve a certain level of accuracy. During inventory, a GPS facilitated orienting direction to the next plots. In each plot, dbh, height and names (local and botanical names) of each tree was documented. Dbh was measured at 1.3 m above the ground level [13].

Tree above ground biomass (AGB) was computed as a product of total tree volume and wood basic density. The average wood density of 0.58g cm⁻³ for natural forest was used [16]. The volumes of trees were estimated using the formula [15]:

(5)

where = Volume of the ith tree (m³)

g = the tree basal area (m²)

0.5 = tree form factor.

The value recommended in natural forests of Tanzania without distinction of the vegetation type involved [17].

The value of tree biomass was converted to carbon using a biomass-carbon ratio of 0.49 [12,18]. The carbon density for the whole forest was obtained by averaging carbon density from each individual forest stratum. Carbon stock was obtained by summing the products of stratum’s carbon density and their corresponding cover area. The carbon stocks for 1980 and 1995 were obtained by assuming that individual cover class’s carbon densities didn’t change [19].

2.2.3. Carbon Mapping

According to [20], Ordinary Kriging using exponential semivariogram model was considered to have the best performance for AGB estimation and for examining its spatial heterogeneity. Therefore, in this study, mapping of tree above ground carbon was done by Ordinary Kriging using exponential semivariogram model.

3. Results and Discussion

3.1. Accuracy Assessment for PKFR

The results from classification accuracy assessment revealed that the overall accuracy of classification for PFR was 84.85% and that of KFR was 82.35%. According to [21] the overall accuracy is acceptable if it is greater than 80%.

3.2. KFR Land Cover Maps and Their Changes

The land cover maps for the period 1980, 1995 and 2010 are presented in Figure 3, while Table 2 presents the cover areas for respective periods and the area changes between 1980 and 1995, and between 1995 and 2010 periods. Generally, the maps indicate that there is significant variation between periods under consideration.

Results (Table 2) indicate that in 1980, Closed Forest occupied 4050.9 ha (75.7%), Bushland 728.8 ha (13.6%), Grassland 269.4 (5%), Open Forest 260.2 (4.9%), Settlement and other land uses 4.7 ha (0.1%). In 1995, Closed Forest occupied 3415.4 ha (63.8%) followed by Grassland 741.2 ha (13.9%), Open Forest 654.3 ha (12.2%), Bushland 449.1 ha (8.4%), and Settlement and other Land uses 89.7 ha (1.7%). Likewise in 2010, Closed Forest occupied 1740.55 (32.5%), Bushland 1131.7 ha (21.2%), Open Forest 1032.3 ha (19.3%), Grassland 916.5 ha (17.1%) and Settlement and other land uses 528.9 ha (9.9%).

During the period 1980-1995, the result (Table 2) revealed that closed forest decreased by 635.5 ha (−11.9%) and 1674.9 ha (−31.3%) for the period 1995-2010. Similarly, in the period 1980-1995, settlement and other land uses increased by 85 ha (1.6%) and 439.2 ha (8.2%) between 1995-2010.

3.3. PFR Land Cover Maps and Their Changes

Figure 4 presents the land cover maps of PFR for the periods 1980, 1995 and 2010 respectively, while Table 3 presents the cover areas for respective periods and the area changes between 1980 and 1995, and between 1995 and 2010 periods. During the period 1980, it was revealed that Closed Forest occupied 2106.6 ha (87.2%), Open Forest 110.4 ha (4.6%), Bushland 81.1 ha (3.4%), Grassland 21.5 ha (0.9%), and Settlement and other land uses 10 ha (0.4%) (Table 3). In 1995, Closed Forest occupied 1997.4 ha (82.7%), Open Forest 299.1 ha (12.4%), Bushland 15.5 ha (0.6%, Grassland 18.3 ha (0.8%), and Settlement and other land uses 85 ha (3.5%). Similarly,