3D model of floodplains
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
NIGER   BANI   DIAKA   DEBO Lake   NIGER     after   DEBO   North LAKE   - 7.50 m   - 7.50 m   5.50 m   5 m   6 m   5 m
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
VEG7.gdb.rar contains the VEG7 shape file and the VEG7k3 matrix. The water depth table is in excel format.