Watershed Analysis & Stream Models

The Hawaiian Island of Kauai is a volcanic landscape that receives tremendous amounts of precipitation throughout the year, like all of the Hawaiian Islands, but is generally more forested and less developed than its fellows.  Creating an accurate model of the watersheds on the island presents a unique challenge because of the excess of rain, heavy and ultimately greatly varied amounts of forest cover, and a landscape surface composed primarily of volcanic material.  All of these factors can effect the amount of surface water present, and all contribute to the size and location of the island's many rivers and streams.

One of the aims of this week's final map was to depict the visual differences between the model-produced streams and watershed layers and the vector graphics of what actually exists.  The map above shows quite clearly that there are a myriad of discrepancies between both the stream and watershed locations produced by the ArcGIS model and those actually present.  One of the user-specified parameters involved in both models is the "accumulation threshold," or the drainage area required to produce an actual stream.  If this threshold is set too high the model output will contain too few streams, if it is set too low the output will contain too many. Upon comparison between a few generated outputs created using different specified thresholds with both aerial photos and the existing vector layers, it becomes evident that creating a model that produces results accurate to reality is no easy feat.  The above was produced with a rather low threshold, which allowed for more streams in the output- presumably logical, given the climate and terrain of the island.  One outcome affected by this, though, is the size of the model-generated watershed. Clearly the model's output (in orange) falls quite short of what is recognized as the actual watershed boundary (the yellow).  This exercise was a good lesson in the potential challenges involved with creating an accurate spatial model, and how the choices of data input can drastically affect the model's results.