**4. Thematic products**

*Current Practice in Fluvial Geomorphology - Dynamics and Diversity*

forest, macrophytes, and other habitats (**Table 1**) [21].

the Amazon region (10b is the most sensitive; **Table 2**).

its size and strategic location, is Lake Coari.

**3. Study area**

pipeline (**Figure 1**).

NOAA's environmental sensitivity classification system is based on the knowledge of geomorphological characteristics of intertidal regions or river and lake limits. In the case of the Amazon basin, because of the need to map the areas under the influence of petroleum production and transportation, it was necessary to hierarchize its ecosystems, after listing its fluvial features. This is necessary because the yearly water level variation results in the flooding of a large portion of the alluvial plain [19, 20]. This area comprises a complex ecosystem composed of lakes, flooded

Subsequently, taking into account local characteristics of specific features, Araújo et al. [21] defined the different degrees of fluvial sensitivity to oil spills of

The study area embraces the Urucu Petroleum Province; the vicinity of the city of Coari, located near the petroleum terminal (TESOL); the Solimões River stretched up to the Petrobras refinery in Manaus (Reman), in which transportation of crude oil and LPG takes place; and part of the area occupied by the terrestrial gas

Because it is a region devoid of infrastructure that is subject to a strong seasonal variation of the water level in the river plain, travel between Manaus and the adjacent municipalities is possible only by air or waterway. The main waterways present in the study area include the Tefé, Urucu, Coari, Manacapuru, Purus, and Solimões rivers, in the stretch from Coari to the confluence of the Rio Negro in Manaus (**Figure 1**). Some lakes are present in this region; the most expressive, in account of

*Study area location on the JERS-1 SAR LHH image msaic (high water). Red rectangle corresponds to Figure 8;* 

**66**

**Figure 1.**

*red circle is Coari city.*

## **4.1 Fluvial oil spill sensitivity index maps**

Environmental sensitivity index maps should reflect the impact of landscape change as a result of flooding. The JERS-1 SAR image mosaics in the low water and high water seasons [22] are input to the unsupervised semivariogram textural classifier (USTC). This algorithm performs image classification, thus recognizing upland forest, flooded forest, flooded vegetation (low biomass above water), and water bodies. A by-product of the classification procedure is the map depicting classes of change, which expresses the difference between dry and flooded seasons.

## *4.1.1 The USTC classifier*

When texture is more important than spectral information, remote sensing image classification must rely on spatial structure. This is the case for the JERS-1 SAR system, which operated using single frequency (L-band) and single polarization (HH). Therefore, the classifier should take into account a pixel in the context of its neighborhood. One possible approach is to analyze texture by means of the semivariogram function [23].

JERS-1 SAR data have already been successfully submitted to semivariogram classification as an aid to vegetation mapping [17, 24]. The unsupervised semivariogram textural classifier (USTC) considers the spatial structure of remote sensing data to carry out image classification. Both textural and radiometric information are combined in this algorithm.

Consider an image dataset X = {x(t) ∈ R,t = 1,…,N}, where each element is a pixel that represents the radiometric information conveyed by the Frost filter digital number (DNdsp) value. Textural information is captured by the semivariogram function γ(t,h) (Eq. (1)):

$$\gamma(t,h) = \frac{1}{2n} \sum\_{l=1}^{n} \left( \boldsymbol{\kappa}\_h(t,i) - \boldsymbol{\kappa}(t) \right)^2 \tag{1}$$

where x(t) is the pixel value; *h* is a parameter that controls the extent of the neighborhood, such that the pixels x*h*(*t,i*) lie inside the circle of radius *h* centered at the pixel x(t); and *n* is the number of pixels of the neighborhood.

The value x*h*(*t,i*) in (1) is the value of the neighborhood pixel for i = 1…*n*, in a neighborhood of *h* defined by the circular semivariogram function γ(t,h) as the texture descriptor. Each pixel in the image is thus transformed into the *h* + 2 dimensional vector (Eq. (2)):

$$\mathbf{z}(t) = \left[ \mathbf{x}(t), \mathbf{y}(t, 1), \mathbf{y}(t, 2)..., \mathbf{y}(t, h), \sigma\_h^2(t) \right] \tag{2}$$

where σ*<sup>h</sup>* 2 (*t*) is the variance of the area *xh*(*t*,*i*),*i* = 1…*n*.

The classification procedure is accomplished based on all components of this h + 2 dimensional vector, calculated for each pixel location. The training set for the unsupervised classification is the set *<sup>Z</sup>* <sup>=</sup> {*z*(*t*) <sup>∈</sup> *<sup>R</sup><sup>h</sup>*+2 , *t* = 1,…,N}.

The clustering algorithm known as ISODATA [25] is utilized to perform the unsupervised classification of this set of vectors. At last, clustering results are interactively merged together to define aggregates of one or more classes capable of bearing thematic significance.
