• No se han encontrado resultados

Texto Unificado de Legislación Ambiental Secundarias (TULAS)

2.5. ANÁLISIS AMBIENTAL

2.5.1. MARCO LEGAL

2.5.1.4. Texto Unificado de Legislación Ambiental Secundarias (TULAS)

The highest flooded area was computed by superimposing the water class from the series of classified images. Multiple SAR images thus enabled the identification of maximum flood extent. Temporal assessment of water extent indicates that the areal coverage of water along the riparian region is highest in the month May which is consistent with the precipitation pattern of the area (Figure 2.3). The farm located up to 3 km from the Kilombero River to the west of Ifakara and Minepa wards (circle 1), south of the Kilombero River in Minepa

69

ward (circle 2) and up to 6km to the East of Kiberege ward (circle 3) has the highest likelihood of submergence in the rainy season (Figure 4.5). Farms located in this area, on one hand, are at risk of a reduced cultivation period and a reduced yield. On the other hand, temporary floods transport necessary nutrients onto the fields and improve the soil conditions. The resulting map time series provides useful information to plan general management strategies by the authorities and to individually support farmers in this area where rainfed subsistence agriculture is the main economic activity. In addition, setting up insurance policies for farmers currently adopting a balance between risk and chance of crop growth is required. The analysis at hand is based on two-year data, which makes it prone to general variability between the years, e.g. according to teleconnection processes such as El Nino Southern Oscillation. Therefore, the study needs to be extended to gain general decadal flood probabilities as soon as the SAR record is long enough.

Figure 4.5: Flood-prone extents derived from the multiple SAR images. The number of images used in generating the cumulative land cover map were 26. The circles show example regions with floods extending more than 3km from the Kilombero River.

70

The continuous monitoring of land cover change with the cloud independent SAR aids in assessing the vegetation patterns, which in turn are linked to the yield expected. The study has shown the potential of Sentinel- 1 in assessing the patterns of land cover in a region where data on dynamics is scarce. Land cover maps and patterns are useful to individual farmers as they provide a means of inventorying areal coverage and patterns of land use on their farms. This provides a basis upon which evaluations on suitable mechanisms to improve agricultural yield are adopted. At the management level, the land cover maps are preliminary databases providing information on vegetated land. With further analysis, cropped areas and rates of growth in areal coverage and their link to yields produced give insight into establishing measures of sustained use of the wetland.

The flooding extents aid in the identification of flood susceptible fields. For the farmers, the information aids in preventing absolute loss by delaying the planting time till after the heavy precipitation. The agricultural support organizations in the area would consequently develop flood-combating mechanisms helpful to farmers to prevent total loss of food crops. Organizations could also offer technical advice and training to farmers in flood-prone areas on alternative crops and planting times.

4.5 Conclusions

Utilization of texture features in delineating land covers is tested in the Kilombero Valley, Tanzania. Texture features were extracted from consistently available single polarized Sentinel-1 VV time series to monitor their performance in land cover mapping. Moreover, PCA of the texture features was generated to assess the effect of the use of data whose dimensionality has been reduced while retaining its variation. SVM, NNET and RF classification algorithms were applied on the single polarized, texture and PCA images. Due to the scarcity of optical images from which reference data is sought, ratioing the VV images corresponding to optical image dates was adopted to derive non-change areas. Reference data within the no change areas for each subsequent image in the time series was applied in classification. Single polarized images yielded low classification accuracies as compared to the texture and PCA images. PCA had higher overall accuracy values (84%) as compared to the texture features (77%). This indicates that the reduction in dimensionality of the data has a positive effect on the accuracies. The three classification algorithms had similar and higher performances for the PCA suggesting that reducing the dimensionality of the texture features into components containing most variance is beneficial for improved classification. Additionally, irrespective of the classifier applied for the PCA features, the ensuing performance is similar. In terms of correct identification of classes, built up and water had high sensitivity and positive predictive values. This is explained by their distinctively high and low backscattering coefficients values respectively in SAR imagery. The relatively lower bare and vegetation sensitivity and positive predictive values indicate other factors such as soil moisture, surface roughness, and the vegetation attenuation had an influence of the SAR backscatter recorded. Therefore, the training sites selected from image ratioing could have contributed to interclass confusion during the land cover classification. Entropy maps indicate the level of uncertainty between the classification algorithms applied. The areas of most disagreements were in the riparian

71

zone, near the Kilombero River. Submerged vegetation is a common occurrence in this zone thus contribution to variations in backscatter values subsequently causing confusion in the classifications. Areal extents of flooding indicated that maximum flooding occurs up to 3 km from the Kilombero River. The implication to the farmers is the high probability of loss of crops due to flooding. Flood extent information is useful to the authorities in terms of offering support to farmers whose crops are affected by floods. Moreover, the inundation patterns are valuable as inputs in flood modeling geared towards reduction of crop loss. With respect to the wetland and sustainable use, matching the crops that can do well given the water available is recommended. Additionally, analysis of the precipitation patterns will give more accurate information particularly with respect to inundation duration, appropriate planting and harvesting periods.

72