TerraSAR-X Based Monitoring of Mangrove Reforestation in Senegal
von Poncet, Felicitas; Peterson, Keith
Astrium Services, GERMANY
The Livelihoods Fund has been supporting the replanting of mangrove forests around vulnerable coastlines in the Sine Saloum and Casamance estuaries within the framework of a CDM. The difficulty that the Fund faces is the continual monitoring of these plantations. At present, solely ground based sampling are used to collect the necessary information which is both expensive and limited by accessibility. A remote sensing based approach can potentially provide wall-to-wall coverage of the plantation areas and increase the level of efficiency in which the project is evaluated by optimizing the amount of field data needed for an accurate assessment of the plantations.
The main objective of the current study is to consider the potential for using TerraSAR-X for the routine monitoring of young (1-5 years old) mangrove plantations and to determine if tree density (trees/ha) can be accurately derived from the data. In addition, the acquisition mode of TSX was examined to determine the optimum data parameters. The study investigated several groups of plantations, established between 2008 and 2012, in the Casamance estuary in south western Senegal.
To evaluate the potential of TSX, images were acquired in 3 different modes ranging in resolution between 1–3m. For all modes, repeat pass acquisitions at 11 day time intervals were acquired to study temporal backscatter variation and the influence of water level changes. The images were acquired during the beginning of the dry season and parallel to the 2011 and 2012 field campaigns.
Field samples were collected within the plantations using GPS. The field team collected detailed information on planted tree density, height and diameter within a 10m radius (314m2) plot area.
The relationship between backscatter and plant characteristics was investigated with a priori and field information. Parameters such as planting year, plant density, height and DBH were used to separate data into different collections and analyse their behaviour with respect to specific criteria. Additionally, the influence of tide level was indirectly assessed by a) comparing scenes of the same area acquired at different dates and b) comparing samples located in test sites acquired with high water level to others showing lower water level. Tide level information available for 2 gauges along the Casamance River and some water level measurements within the plantations support the interpretation. Further analysis investigated if interferometric coherence is related to plant parameters and water-level changes. The above described investigations were based on mean backscatter and coherence extracted for the sample plot.
The high spatial and temporal variation of the water level is a major challenge for the data analysis. Water level measurements at the time of image acquisition are available for 6 sites with varying plant size and density which supports the interpretation. The presence of water significantly affects the SAR backscatter. The most important scattering mechanisms for monitoring young mangrove plantations are double-bounce signals between the water and the vertical tree structures. In this case, the backscatter intensity is governed by plant density, provided that a relevant proportion of the vertical plant structure emerges from the water.
To accurately establish a relationship between the field-measured tree density and the TSX backscatter, zonal statistics computed for the field samples had to be filtered and separated into groups representing different operating conditions. The samples are grouped into 3 water-level-based conditions present at the time of acquisition: 1) high tide, where only a small part of plant structure is exposed; 2) low tide, where the rough soil surface is exposed; and 3) low to medium water level, where a relevant proportion of the vertical plant structure in above the water. Condition 3 proved optimal to retrieve information on plant parameters, as the water surface provides the ideal background to measure small plant structures. The relationship improves with increasing plant height, as a larger proportion of the plant is above water, thus offering a sufficient backscattering surface. A higher coherence level was observed for larger plants, given T1 and T2 are both flooded. Since the 2012 field campaign focused on the assessment of freshly planted sites, most of the samples stem from very small saplings, resulting in submerged conditions under the given high tide level for many of the investigated acquisitions. Thus, field plots with < 40cm tree height have been excluded from the sample set, leaving samples which were split into equal sized datasets for model development and validation by random selection. The relationship of backscatter to plant density for group 3 (representing favourable conditions) yielded a R2=0.68. A comparison of measured versus predicted values resulted in a RMSE of 1455 plants/ha representing 12% of the observed range of plant density.
Conclusions & Outlook The results demonstrate that, given the appropriate operating conditions, a relationship exists between TSX backscatter and tree density in young mangrove plantations up to 4 years in age. Radar backscatter explains between 68% (2012 results) and 90% (2011 results) of the plant density variance measured in the field depending on acquisition conditions and plant age. The higher uncertainties for the groups of non-optimal conditions can be compensated by more field samples in a stratified sampling approach, combining remote sensing and field survey data. In the next steps, objective criteria derived from amplitude and phase information will be defined to assign homogeneous plantation areas to one of the 3 groups.The efficiency of the stratification will depend on the areal proportion of plantations showing a strong correlation to plant densities.
Figure 1: field photos representing the varying operating conditions: high tide, low tide and low-medium water level.
Figure: 2 a) density vs backscatter for the modelling points in comparison to control points; b) estimated vs field measurements with an RMSE of 12%.
While this study provides a sound analysis of the potential of high resolution X-band radar for the monitoring of young mangrove plantations, a substantial amount of work remains in order to refine the results for the development of an efficient, transferable methodology for the quantification of tree density variation within all the plantations associated with the project.
Conclusions & Outlook
The results demonstrate that, given the appropriate operating conditions, a relationship exists between TSX backscatter and tree density in young mangrove plantations up to 4 years in age. Radar backscatter explains between 68% (2012 results) and 90% (2011 results) of the plant density variance measured in the field depending on acquisition conditions and plant age. The higher uncertainties for the groups of non-optimal conditions can be compensated by more field samples in a stratified sampling approach, combining remote sensing and field survey data. In the next steps, objective criteria derived from amplitude and phase information will be defined to assign homogeneous plantation areas to one of the 3 groups.The efficiency of the stratification will depend on the areal proportion of plantations showing a strong correlation to plant densities.