To try and solve the shortcomings of the previous model, we carried out a matrix model by converting the Veg7 map layer into a matrix in which each square cell of 100 x100 m is assigned the depth that is characteristic of the vegetation association in the cell. It is then possible, for any depth value, to draw curves of "equal flood depth" over the whole Delta and thus to derive maps of potentially flooded areas for each flood depth. The map obtained for the “0” depth value corresponds to the potentially flooded area for the reference flood of 6.60 m at the Mopti gauge. The validity of this approach can be evaluated by comparing the result with the spatial extent of open water and flooded vegetation in Landsat images. This model does not intend to represent the propagation of the flood wave across the Delta, but allows us to estimate the potentially flooded area in relation to the flood measured at the Mopti gauge, year after year. Four Landsat images cover the Delta: 197_050 spreads over most of the Delta, 197_051 over the southern edge (Pondori), 197_049 over the northern edge (Débo lake) and 198_050 over the southwestern edge (from Ké Macina to Diafarabé). However, it should be noted that for a flood height between 6 and 6.60 m at the Mopti gauge, the flood is delayed by 15 to 25 days between Ke Macina and Mopti, and by 30 to 45 days between Mopti and Akka, at the exit of the Débo lake. Similarly on the Bani River, the delay is of 19 days on average between Douna (upstream of the Delta) and Sofara; and of 10 days between Sofara and Mopti (for a 6.30 m flood at the Mopti gauge ). So the maximum extension of the flood might be reached by mid-October in the south and not be reached until the end of November, or even in December, in the north of the Delta. Moreover, while Landsat images have been available since 1984 – the lowest flood of the century – the collection of these images is incomplete. It is uncommon to have a series of good quality images over the flood season, from September to December, for each flood year. Flooding from Mopti to Akka results from the combination of the floods (height and date) at Ké Macina on the Niger and at Beneny Kegni (or Sofara) on the Bani. As we will see further, each flood is unique and the same flood height in Mopti can be reached in different ways, sometimes with a stronger flood on the Niger and a lower one on the Bani, or the opposite, or else with an exceptional lag time between the flood of the two rivers. The 5.10 m flood of 1990 is a good example as we will see when analyzing this flood. Finally, the available Landsat images were often taken one month apart from one another. They make the discrimination of water and green vegetation possible. Analysis is based on R.O.I (Region of interest) tests carried out in the different environments (clear water, turbid water, green vegetation with high reflectance, vegetation on the Sahelian margins, bare soil) that can be identified on a colored composition such as Layer 753 (IFR SW2, near IFR, green) for Landsat 8 or 743 (IFR, Red, green) for ETM The classification relies on maximum likelihood and is reported within the limits of vegetation units in VEG7. Each unit is visually checked against a classical colored composition (543). Whenever possible, this check is performed at different dates for the west, south, center and north of the Delta, either separately or complementarily, with the same unit being returned to at different dates. Besides, the strong development of woody trees on the toggere – unflooded uplands – especially in the south of the Delta, leads us to arbitrarily consider these areas as non-floodable and leave them out of VEG7 A paper submitted in February 2020 on "The drought resilience of floodplain vegetation of the Inland Niger Delta of Mali" by Hiernaux P., Turner M. D., Eggen M., Marie J. and Haywood M., followed a field trip by Pierre Hiernaux and Matthew Turner made during the 2014 flood. They revisited the vegetation sites observed between 1979 and 1986. Their analysis of vegetation change included a classification of numerical data derived from the Landsat images based on the NDVI index and reflectances in the mean Infra-red bands and was performed at the University of Wisconsin. The article demonstrates a large amount of resilience of the Delta vegetation. It can remain at a very low level of production through a succession of very dry years, then resume its full development when better floods return without much change in the composition of species and the arrangement of vegetation associations. The model maps the potentially flooded areas for a given flood height at the Mopti gauge, compared with the flooded and / or heavily vegetated areas shown on the Landsat images for a reference year. The area flooded in both model maps and Landsat images provides a confidence ratio for the model. The differences (either way) between model output and image can also be precisely localized. However, these values ​​should be taken with caution because the transposition of the raster image within the limits of the vector cover of the vegetation map (Veg7) raises a problem: the vegetation layer contains some localization errors of up to 500 m on the ground. If these errors have little impact on large units of several thousand hectares, they have a large one on small areas and / or on narrow shaped units. Because of this shortcoming, it proved necessary to visually check nearly 14,000 vegetation units when overlaying the "Landsat" information on Veg7, especially when part of the unit appeared flooded and / or heavily vegetated. The analysis of the model output for water heights of 6.60 m, 6.21 m, 5.97 m, 5.10 m and 4.40 m allows us to assess the validity of the model "by levels". Without fundamentally questioning the model approach, it suggests its output should be qualified and consolidated by a study of each large internal basin in the Delta. It is to be noted, for example, that a 6 m flood at Mopti corresponds to a slightly weaker flood in the south of the Delta and a little stronger one in the north. This trend seems to be even more marked for lower floods. Finally, in addition to the logic of flood patterns in relation to levels and to basins, local factors must be taken into account: for example, a deep-flood vegetation (VB) unit isolated between weakly-flooded vegetation units could be less flooded than predicted by the model, as observed on the Landsat image, due to threshold effects. However, despite all these reservations, the spatial relationship between flood levels at the Mopti gauge and flooded areas for each year allows for plant production estimates, regarding spontaneous vegetation and rice crops under uncontrolled irrigation. The practical realization of the model In the Veg7 map layer, the item called "PROFOND" carries the depth of each vegetation association. In the model, the flood depths are coded with two digits: for example 66 for level 6 as for vegetation association (B), and 65 as for mosaic B / VOR. The flood-depth calculation is as follows: each vegetation association is assigned its maximum depth: for example B = -2.80 m. and VOR = -1.50 m. In the mosaics, the assigned depth is the arithmetic mean of the depths of their component vegetations. For example B / VOR is therefore assigned a depth of - 2.15 m. The off-Delta environment and never-flooded uplands within the Delta are assigned the arbitrary value +1 m. However, for mosaics combining flooded vegetation with never flooded uplands, the latter are assigned the value "0". Thus the AG / TA mosaic is assigned a depth of -0.30 resulting from the depths for AG = -0.60 and for TA = 0 (instead of +1 m when TA is by its own). The water streams (Niger, Bani, Diaka and the great Débo lake….), which were initially mapped as a unique polygon named “RIVER”, were later split into a series of sections to which we allocated flood depths based on data from the Mathematical Model of the Niger River, a study carried out in the 1980s by ORSTOM – for the hydrology survey – and by IGN for the topography. These surveys enabled SOGREAH to develop the "CARIMA" model simulating the flow of the river from its source to the border between Niger and Benin, and Nigeria. Moreover, the MB mosaic defined as "the mosaic of the river banks", is a complex mosaic whose flood depths range from +1 to -2.80 m and has been modified. In reality, MB covers several types of situations: it may extend on the river bank and therefore have flood gradients from "+1" to -2.80 m, or extend mostly on channels between banks, with an average flood depth of -2.80 m. It may also – and this is the most frequent case – extend over a complex series of parallel channels and levees below the main bank. On the basis of an analysis of the rectified aerial photo mosaic built for the Niger River Mathematical Model (IGN), the 188 polygons mapped as MB vegetation were subdivided in two categories: MB1, for the channels with a depth of - 2.80 m and MB2, for the mosaic of levees and channels with depths ranging from 0 to -2.8m, and therefore assigned a mean depth of “-1.40 m”. (see and download Table 1: relationships between water depths and plant formations). The modified layer is named Veg7 and was converted into a flood depth matrix named VEG7 by overlaying a 100*100m grid, where cells are assigned the maximum flood depth of the overlaid vegetation unit. To attenuate the gap between neighboring vegetation with different flood depths, we applied to the matrix a smoothing Gauss filter of 500m x 500m, whose results approximate the topographic profile of the basins as shown in the example profile produced over 3500 m transect. The curves of equal flood depth are then calculated on the smoothed matrix VEG7K3 under ArcInfo. After cleaning the "hanging arcs", each curve is transferred to ArcGis and smoothed using the Peak algorithm with a 300 m resolution. The corrected curve is transformed into a polygon named NIV_XXX. For example, NIV_660 figures the potentially floodable areas for a 6.60 m flood at the Mopti gauge. After the removal of polygons smaller than 1 ha, the flood contour curve is established, based on the remaining polygons. The contour curve corresponding to flood NIV_660 is named L_660. The issue of the Gaussian filter When two adjoining areas have very different flood depths, smoothing out the difference by using a 500x500 Gaussian filter results in a shifting of boundaries, so that the areas calculated as flooded by the model are marginally modified. Fig 3 : The raw Model Fig 4 : The corrected model

In the example above, taken from the analysis of the 5.97m flood at Mopti, one can clearly see that associations P and VH, with a depth of only -0.30m, are partly included in the calculated model, whereas VOR, which has a - 1.50m depth, undergoes a cut that is not justified by a variation in depth. Those differences, whether positive (P_VH) or negative (VOR), only marginally modify the estimation of flooded areas (by about 2%) but they introduce fragments of vegetation associations which do not belong to the model into it (P, VH, VSP/VH….) or conversely deprive the model of fragments of areas that belong to it (VOR, VB/V...)

When comparisons are made with the flooded areas shown by Landsat for the year corresponding to the flood, the risk is that the shape files showing areas that are common to the model and the Landsat pictures, as well as in those displaying the additional or missing areas, might include such “alien” vegetation associations. Despite the limited area concerned, the analysis would nevertheless be affected. We have therefore decided to use a spatial operator in order to erase such “mistakes” resulting from the smoothing out process, as shown by that very same example after the correction has been made (figure 4). For each example of flood height, we shall indicate both the “raw” area calculated by the model and the “corrected” area used for comparisons. I n the following examples (pages 44 to 48), we have “normalized” the names given to the “shape” files, taking the 5.97m flood and the year 2006 as references) NIV_597 : is the « raw » model as calculated for the 5.97 m height at Mopti NIV_597_VEG7 : is the calculated model, as re-positioned within the limits of Veg7 and corrected as to the effects of the smoothing out process VEG_2006 : shows the results of the analysis of the Landsat images for the year 2006 (corresponding to the 5.97 m flood) COMMUN_597_2006 : shows the flooded areas common to the calculated model (NIV_597_VEG7) and the Landsat images (Veg_2006). Flooded areas include water and vegetated areas with a high degree of reflectance INON_MOINS_2006 : is a shape file representing the areas calculated as flooded by the model, but not appearing as such on the Landsat images INON_PLUS_2006 : is a shape file representing the areas appearing as flooded on the Landsat images but which were not calculated as flooded by the model. SYNTHESE_597_2006 : represents flooded areas common to the model and the images, as well as those in adddition and those missing (Synthese_597_2006 = commun_597_2006 + inon_moins_2006 + inon_plus_2006). We shall invite you to download two shape files : NIV_597 et SYNTHESE_597_2006 The items in synthese_597_2006 will make it easy to extract the shape files: commun, inon_plus, inon_moins and Veg7_2006

Download

VEG7.gdb.rar contains the VEG7 shape file and the VEG7k3 matrix. The water depth table is in excel format.

Veg7.rar