Appendix II: Processing Summary for Joint ALCOM/WWF SADC Watershed Model

by Joe Dooley
RS/G1S Consultant
c/o USAID/Zimbabwe
P.O. Box 6988, Harare
1 Pascoe Ave, Belgravia
Harare, Zimbabwe

Overview

In February of 1997 the above consultant was hired under a joint project between the ALCOM/FAO SADC Regional Fisheries project and WWF Zimbabwe to develop a digital watershed model for the SADC region This model was to be derived from the 30 arc second digital elevation model (DEM) of Africa¹, and was to include watersheds for up to third order rivers which ultimately flowed into the Atlantic or Indian Oceans.

Processing Details

The IDRISI formatted DEM was translated into Arclnfo grid format for processing using the Spatial Analyst module of Arc View 3 0 (AVSA); note: to retain the integer format of these data required a intermediate translation of the data into a ERDAS GIS format. Using the hydrologic extension functions to AVSA the DEM was. first “filled” to smooth out areas of no drainage, then the flow direction of each cell was calculated, and next the accumulation of upslope cells flowing to each cell was derived Next a processing mask for the region was developed using the coastlines from the DCW to exclude oceans and islands from the delineation of watersheds; a mask which included major inland waterbodies and pans as well was also developed but later discarded based on unsatisfactory determination of watershed boundaries. Both of these masks extended from 11.25e,0s to 41.25e,35.25s. Additionally, the extent of processing was limited to a rectangle extending from 11.5e,0.75s and 41e,35s to limit any edge processing errors; this later proved unnecessary but unfortunately the northern boundary of the analysis was never extended back to teach the equator.

Using the above datasets, boundaries for watersheds at various cell accumulations were next calculated Boundaries for watersheds which had a minimum of 250,000, 100,000, 50,000; 35,000; 10,000, 5,000 and 1,000 cells flowing into them were delineated, the 0 0084 cell size of the source DEM was retained throughout processing. The watersheds determined based on the 5,000 cell accumulation were selected as the basis for further processing after consultations with ALCOM. This 5,000 cell ‘resolution’ delineated basically 4th order watersheds, while at the same time effectively capturing coastal lake and ocean watersheds. It also resulted in only ‘irregular’ discrepancies between the DCW rivers data layer, i.e. rivers flowing up hill across watersheds, rather than the 'gross' inconsistencies which resulted from processing at a lower cell accumulation. During these consultations it was also determined that the results of the 5,000 cell delineation would only be corrected based on other DCW layers where the results were totally inadequate, e.g. the Okavango Delta, Etosha Pan, and the confluences of major rivers and coastal areas; the decision to carry out only limited editing was later to prove overly optimistic.

The grided results of the 5,000 cell accumulation watershed model were next vectorized into the Arc View Shapefile file format. It was only after this vectorization process, that a realistic estimate of the manual editing which would be necessary could be determined. The grid model, which had delineated 1101 watersheds, had produced 1891 unique polygons when vectorized; many of these additional polygons were simple one to two cell polygon primitives which often happen during raster/vector conversion. However, while reviewing the amount of actual editing which would be needed, versus automated aggregation or elimination to remove these primitives, the true extent of the discrepancies between the DEM derived watersheds and the DCW drainage layers became evident. While it was anticipated that many pour points would not match the confluences of streams, rivers, and coastlines; it was not anticipated that over 40% of them would need major adjustments. Nor was it realized that whole stream and river networks would be flowing across or “up to” the wrong basin/direction as depicted on the ONC source or other smaller scale maps. Also, while the basins delineated in the model did correspond well with a drainage network derived from the DEM, this network compared poorly with the DCW drainage layer. As it was assumed that the contour and drainage layers for the ONC were likely to have been based on the same or similar sources, the consultant decided it was better to correct the basin boundaries favoring the DCW drainage layers rather than the DCW/DTED derived DEM.

Due to the massive amount of editing which would be needed, and after initial attempts at line editing/corrections in Arc View proved cumbersome due to almost constant screen updating, it was decided to translate the watershed boundaries into Arc-Dbase and from there into Intergraph/Bentley's CAD package MicroStation for further editing. Here the almost 5,000 lines delineating watershed boundaries were joined with the DCW coastlines and major inland waterbodies, and edited against a backdrop of the DCW drainage layers. Other than retaining coding for the joined DCW data and lines depicting the northern extent of the model, the editing in MicroStation yielded three line types:

1. Lines derived from the watershed analysis which required only minor editing of pour points, and/or line densification based in two points straddling a stream but not capturing the top or bends in a stream. While this editing did correct for the tops and bends of streams crossing watersheds, it could still be described as cosmetic ‘enhancement’ of the lines. {1900 lines}

2. Lines derived from the analysis but requiring extensive editing to pour points, and/or the location of the line itself because it allowed the DCW drainage network to impossibly flow across a drainage boundary or boundaries. Some of these lines were edited so extensively that they could have been classified as new or added lines. Edited lines tended to follow or conform to the nearest DCW drainage feature. {1210 lines}

3. Lines which were added by hand because the derived lines were grossly incomparable with the DCW drainage layer, or lines which were added because a needed line was missing and no drainage boundary had been delineated. Every effort was taken to ensure that lines added were based on a visual interpretation of the DEM, the ONC source maps where available, or other source maps where the ONC charts were not. However, even with these efforts, these lines can perhaps best be described as conformal questimates based on the DCW drainage layer. {260 lines}

The types of editing listed above arc described in order to emphasize that while the model may be derived from the DEM of Africa, it now would more closely depict any errors in the DCW Drainage layer rather than those in DEM. Basically other than the 841 lines derived from adding the DCW coastlines and waterbodies, and the extent of the analysis boundaries, each of the remaining 3370 original lines have been edited in one way or another; and over 1500 lines were deleted from the original dataset. Any inland DCW waterbodies which were wholly contained within a single watershed were also deleted from the dataset. After editing in MicroStation the line and label/text data were translated back into Arc-Dbase where topology was constructed for the 1113 resulting watershed boundaries. The Arc-Dbase coverage was then translated into Atlas BNA polygon format specified by ALCOM as a deliverable. The primary and secondary attributes used in the translation were the User_ID value and a field called Basin_Name. In the BNA dataset the Basin_Name contains only three different labels. These are: “Coastal Drainage”, used for all single basin stream networks which flow directly into a major lake or ocean, “Edge Drainage”, used for all basin polygons which were truncated by the northern extent of the analysis boundary and are not part of a basin network which originates or continues from/into the interior of the model; and lastly, “Unknown Basin”, used for the majority of the basins until naming and basin order conventions are developed by ALCOM/WWF.

Attribution of the Model

Once ALCOM/WWF codify basin order and naming conventions, each of the individual basins will need to be given a primary name - possibly using the Bartholomew 1:5,000,000 Map of Central and Southern Africa as a source - and attributed as to their membership in any larger basin orders. It is likely that, although many basins delineated are not ‘fourth order’, each will need to be coded to this or at least a third order level. An example of fourth order attribution using the Sanyati River Network originating in Zimbabwe could be:

However, this example highlights some of the problems which will need to be addressed in basin order convention and attribution; in this case what to do with Lake Kariba. The Sanyati actually flows into Lake Kariba, but the Zambezi actually flows into/through/out-of the lake. In effect, Lake Kariba is the Zambezi River. One could add a fifth order here, although this could lead to greater redundancy and possibly confusion; in fact Lake Kariba could be dropped as a fourth order itself.

To aid in classifying each basin the grided data from the 250,000; 100,000; 50,000; and 35,000 cell accumulations were also vectorized to shape-file and then BNA format. ALCOM/WWF will use these as possible guides to help in attributing the membership of each basin in larger basin networks. Unfortunately, coding and naming these larger basin delineations, and then simply overlaying them with the 5,000 cell basin model to attribute the model, would yield too many inaccuracies based on the extent of editing which took place in the model. It was decided that ALCOM/WWF would use and expand the Dbase file listed below, which contains the USER_ID and BASIN_NAME fields, as the basis for their attribution. This Dbase file will then be returned to the consultant who will then produce aggregated basins which reflect the edits made to individual basin delineations. Additionally, using the model the consultant will generate a coding scheme for the each order of basin, calculate the maximum, minimum, and mode elevations for each basin, as well as equal area measurements of each basin's hectarage. The coding, statistics, and measurements will also be present in any first, second, or third order basin aggregations requested by ALCOM/WWF.

Limitations of the Basin Model

Based on the 1:1,000,000 source used to derive most of the DEM, the basin model could be used for analyses at a scale of 1:1,000,000; although analyses at a scale of 1:750,000 should also yield satisfactory results. At no time should the model be used for larger scale analyses than 1:500,000. In terms of the positional accuracy of the edited basin boundaries all DCW and type 1) lines described above should be between 3 and 5 pixels, or roughly 3 to 5 kilometers. Line types 2) and 3), also described above, should be recognized as having a positional accuracy of between 5 to 10 pixels, or 5–10 kilometers.

Files delivered to ALCOM/WWF

* These basin bna files were modified by ALCOM to eliminate the artificial tie lines which connected Africa mainland with Islands in the oceans.

¹ The DEM of Africa used had been developed by U.S and other international cooperators at the USGS EROS Data Center in Sioux falls,SD.USA This DEM was based primarily on elevation data taken from the Digital Chart of the World (DCW) which had been digitized from the 1:1.000,000 Operational Navigation Charts by the Environmental Systems Research Institute (ESRI) for the U.S. Defence Mapping Agency (DMA). Where incomplete elevation data were indicated on the ONC Charts courser resolution data from the DTED DEM, also from the DMA, were used to fill elevation gaps in the USGS DEM. The interpolated resolution of the DEM of Africa corresponds to 0 0084 decimal degrees which equates to roughly one square kilometer pixels. The DEM provided by the FAO/FSTAU to ALCOM for the project was in 1DR1S1 format.