**2.3 Production of seamless mosaic images**

Cloud cover is inevitable on the images acquired by the satellites. However, cloud patching process can eliminate the cloud covers that appear on a single-date observation data. Images of particular scenes that were acquired on different dates were used for cloud patching process as shown in **Figure 2**. F\_mask algorithm was used to perform this process [15, 16]. Seamless mosaics product (i.e. images without cloud covers and atmospherically corrected) were used as input for subsequent processes.

#### **Figure 1.**

*Landsat scenes that were used for the classification. Numbers within the scene boundary indicates path/row ID of Landsat satellites.*

**Figure 2.**

*Cloud detection and removal process. Individual Landsat scene that was captured on 26 January 2017 (a) was merged with that captured on 14 June 2017 (b), where both produced a cloud-free images for the year 2017 (c).*

#### **2.4 Images classification**

Appropriate enhancement techniques were applied to the images to make the mangroves appear better on the images [17]. In addition to the individual spectral bands of Landsat images, vegetation indices such as Normalized Different Vegetation Index (NDVI), Green Atmospherically Resistant Index (GARI), and Normalized Difference Infrared Index (NDII) were also derived from the images to improve quality of classification. The vegetation indices that were used in this study are summarized in **Table 1**.

Most spectral-based image classifications are performed using traditional methods such as maximum likelihood, linear discriminant analysis, and spectral angle mapper classifiers. These methods are applied to the spectral bands to produce a classified feature in images [18]. Instead of using these approaches, this study attempted a new approach to classify the images. R Package, which is free, open source software with the RandomForest algorithm [19] was used.

RandomForest implements Breiman's RandomForest algorithm, based on Breiman and Cutler's original FORTRAN code for classification and regression [20]. It can also be used for assessing proximities among data points without necessarily a training set. All sampling points that were collected on the ground were connected to the corresponding pixels on the image through this algorithm. Classification was done by searching the most important variables i.e. which spectral bands are used in decision tree approach [21–23]. RandomForest applies four major steps of looking at the importance of variables as follow:


**105**

erosion.

**Table 1.**

**2.5 Estimation of CO2 emission**

can be expressed as Eq. (1) as follow [24];

<sup>∆</sup>*<sup>C</sup>* <sup>=</sup> (*Ct*<sup>1</sup> <sup>−</sup> *Ct*2) \_\_\_\_\_\_\_\_

where ∆*C* is changes in carbon stock (Mg C yr<sup>−</sup><sup>1</sup>

(44/12 = 3.67) [24].

*GIS and Remote Sensing for Mangroves Mapping and Monitoring*

**Formula Description**

NDVI Commonly used to delineate vegetation

GARI Normally used for detection of green

NDII It uses near- and mid-infrared bands to

from other features on images and to measure vegetation vigor. It is sensitive to

pigment concentration and differentiate chlorophyll levels. It is more sensitive to chlorophyll concentrations than the

detect changes in plant biomass and water stress in wetlands like mangroves

atmospheric effects

atmospheric effects

decision tree. At every split, one of the mth variables is used to form the split and there is a resulting decrease in the Gini. The sum of all will decrease the

All images have been classified to distinguish mangroves from the other land uses. The classification results were transformed into vector shapefile for further refinement and editing. The accuracy of the classification results were assessed by using a number of ground truth points. The GIS platform was used to carry out post-classification analysis. Post-classification analysis is usually used for quantifying changes of land uses. Changes of mangroves were identified from the conversions of mangroves to other landuse classes, which are (i) urban, settlement, and industrial areas, (ii) agricultural, (iii) aquaculture activities, and (iv) coastal

Carbon dioxide (CO2) is defined as natural, colorless and odorless greenhouse gas that is emitted when fossil fuels (i.e. natural gas, oil, coal, etc.) are burnt. In this study, the CO2 emission is expressed as C loss, assuming that the gas is emitted when deforestation occur. The units of metric tons C was converted to CO2 by multiplying the ratio of the molecular weight of carbon dioxide to that of carbon

The CO2 resulted from deforestation is one of the important elements in greenhouse gases emissions. Therefore, it is also essential to quantify the contribution of mangrove deforestation towards the CO2 emission. Net emission as resulted from deforestation of mangroves can be estimated based stock-difference method, which

stock at time *t*1 and *t*2 (year), respectively. In this case, the *Ct*1 and *Ct*2 was quantified

from the changes analysis that have been carried out earlier this study.

(*t*<sup>2</sup> <sup>−</sup> *<sup>t</sup>*1) (1)

), *Ct*1 and *Ct*2 (Mg C) is carbon

forest due to a given variable, normalized by the number of trees.

*Note: NIR = near infrared, G = green, B = blue, R = red, and MidIR = middle wave infrared channels.*

*DOI: http://dx.doi.org/10.5772/intechopen.81955*

*Vegetation indices that were used derived from the images.*

**Vegetation indices**

*GIS and Remote Sensing for Mangroves Mapping and Monitoring DOI: http://dx.doi.org/10.5772/intechopen.81955*


#### **Table 1.**

*Geographic Information Systems and Science*

**2.4 Images classification**

**Figure 2.**

are summarized in **Table 1**.

the importance of variables as follow:

that was shrank.

Appropriate enhancement techniques were applied to the images to make the mangroves appear better on the images [17]. In addition to the individual spectral bands of Landsat images, vegetation indices such as Normalized Different Vegetation Index (NDVI), Green Atmospherically Resistant Index (GARI), and Normalized Difference Infrared Index (NDII) were also derived from the images to improve quality of classification. The vegetation indices that were used in this study

*Cloud detection and removal process. Individual Landsat scene that was captured on 26 January 2017 (a) was merged with that captured on 14 June 2017 (b), where both produced a cloud-free images for the year 2017 (c).*

Most spectral-based image classifications are performed using traditional methods such as maximum likelihood, linear discriminant analysis, and spectral angle mapper classifiers. These methods are applied to the spectral bands to produce a classified feature in images [18]. Instead of using these approaches, this study attempted a new approach to classify the images. R Package, which is free,

RandomForest implements Breiman's RandomForest algorithm, based on Breiman and Cutler's original FORTRAN code for classification and regression [20]. It can also be used for assessing proximities among data points without necessarily a training set. All sampling points that were collected on the ground were connected to the corresponding pixels on the image through this algorithm. Classification was done by searching the most important variables i.e. which spectral bands are used in decision tree approach [21–23]. RandomForest applies four major steps of looking at

1.Step 1: to determine the significance of the mth variable. In the left out cases for the kth tree, randomly permute all values of the mth variable. Put these new

2.Steps 2 and 3: for the nth case in the data, its margin at the end of a run is the proportion of votes for its true class minus the maximum of the proportion of votes for each of the other classes. The 2nd measure of importance of the mth variable is the average lowering of the margin across all cases when the mth variable is randomly permuted as in Step 1. Step 3 then count the margins

3.Step 4: the splitting criterion used in RandomForest is the Gini criterion, a mechanism that can measure the most to least importance of variables used in

open source software with the RandomForest algorithm [19] was used.

covariate values down the tree and get classifications.

**104**

*Vegetation indices that were used derived from the images.*

decision tree. At every split, one of the mth variables is used to form the split and there is a resulting decrease in the Gini. The sum of all will decrease the forest due to a given variable, normalized by the number of trees.

All images have been classified to distinguish mangroves from the other land uses. The classification results were transformed into vector shapefile for further refinement and editing. The accuracy of the classification results were assessed by using a number of ground truth points. The GIS platform was used to carry out post-classification analysis. Post-classification analysis is usually used for quantifying changes of land uses. Changes of mangroves were identified from the conversions of mangroves to other landuse classes, which are (i) urban, settlement, and industrial areas, (ii) agricultural, (iii) aquaculture activities, and (iv) coastal erosion.

#### **2.5 Estimation of CO2 emission**

Carbon dioxide (CO2) is defined as natural, colorless and odorless greenhouse gas that is emitted when fossil fuels (i.e. natural gas, oil, coal, etc.) are burnt. In this study, the CO2 emission is expressed as C loss, assuming that the gas is emitted when deforestation occur. The units of metric tons C was converted to CO2 by multiplying the ratio of the molecular weight of carbon dioxide to that of carbon (44/12 = 3.67) [24].

The CO2 resulted from deforestation is one of the important elements in greenhouse gases emissions. Therefore, it is also essential to quantify the contribution of mangrove deforestation towards the CO2 emission. Net emission as resulted from deforestation of mangroves can be estimated based stock-difference method, which can be expressed as Eq. (1) as follow [24];

$$
\Delta C = \frac{\{C\_{t1} - C\_{t2}\}}{\{t\_2 - t\_1\}} \tag{1}
$$

where ∆*C* is changes in carbon stock (Mg C yr<sup>−</sup><sup>1</sup> ), *Ct*1 and *Ct*2 (Mg C) is carbon stock at time *t*1 and *t*2 (year), respectively. In this case, the *Ct*1 and *Ct*2 was quantified from the changes analysis that have been carried out earlier this study.
