Thursday of Week 6. Tuesday (Part A, watershed-delineation-a.md) covered what a watershed is and why water is managed by them, then the first three of the eight steps, ending on the flow direction grid. Today runs in three movements: how watersheds used to be delineated by hand, the same job done by USGS StreamStats, and then the rest of the eight steps (flow accumulation, the stream threshold, stream links, pour points and watersheds) finished in the Lab 5 ArcGIS Pro model.

stamp:begin

stamp:end

One goal per question on the closing quiz (Follow the Water), in the same order. The map at right is where Lab 5 ends up: Rock Canyon's basin cut into one subwatershed per stream link. Set the frame: first by hand, then StreamStats, then every remaining step as one raster operation reading the flow direction grid Tuesday built. Students who understand the chain can debug Lab 5; students who only memorize tool names cannot.

Part 1 is hands-on and comes first: how watersheds were delineated before GIS — by hand, from contours — then the same basin from USGS StreamStats in thirty seconds, then the two compared, on the lab computers. Part 2 then opens the box StreamStats is: the rest of the eight steps, finished in the Lab 5 ArcGIS Pro model.

One click. First the quadrangle alone: a 4 km by 4 km topographic map with Hog Pen Creek running through it. Ask the class where the divide is. Everyone will point at the contour crenulations, which is exactly the right instinct. Then click to show the hand-drawn answer: the red line is the watershed divide, drawn by hand along the ridges. The rules: the divide crosses contours at right angles, it runs through high points, it never crosses the stream except at the outlet, and it closes on itself. The 20 ft and 100 ft contours, the stream center line and the outlet are labeled. This is how every watershed was delineated before GIS.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Give this about fifteen minutes. Turn on a hillshade or terrain basemap so ridges are visible. The comparison at the end is the point: five students will produce five different boundaries, which sets up the StreamStats comparison that follows.

Capture: ArcGIS Pro 3.7.1 at 175 %, October 10, 2026, C:\Ames\Week06\Week06_Figures.aprx: the Emigration Canyon basin above Hogle Zoo on the World Topographic basemap at 1:123,239, a polygon sketch in progress (eleven vertices along the north and east ridges, the construction toolbar, the Create Features pane). The Catalog path (right-click geodatabase > New > Feature Class) was checked in the same session.

StreamStats runs this week's same eight steps on a pre-processed national DEM, then adds published regression equations for peak flows. Students are about to get in thirty seconds what took them fifteen minutes by hand.

The state has to be selected first because the regression equations and the pre-processed terrain data are organized by state study area. The red circle marks the state selector.

Two circled steps: activate the delineation tool, then place the pour point. Emphasize that clicking off the blue line gives a tiny nonsense basin. Part 2 meets the same problem with pour points in ArcGIS Pro, and fixes it by snapping.

Collect answers before moving on. Expect the automated basin to be close on the ridges and different near the outlet, where the pour point placement dominates. Ask what would change if they had clicked 100 m upstream.

The download is a zip containing the basin polygon and, depending on the options chosen, the flow-path lines. Students need this file for the comparison on the next slide.

The deliverable is one image showing both polygons over the same basemap. Symbolize one as a hollow outline so both are visible. If time allows, have them compute the area of each and report the percent difference. The example on the right: a quick hand-digitized basin (pink, 35.5 km2) against the StreamStats basin (black outline, 48.0 km2) for Emigration Creek at Hogle Zoo; the hand version missed about a quarter of the basin, mostly the south ridge toward Parleys Canyon.

Capture: ArcGIS Pro 3.7.1 at 175 %, October 10, 2026, C:\Ames\Week06\Week06_Figures.aprx. The StreamStats basin is the one its web app delineated for the outlet 40.7495 N, 111.8160 W (its delineation response, saved as C:\Ames\Week06\streamstats\emigration_basin.json, converted with JSON To Features); the hand polygon was digitized by eye in the same session.

What StreamStats just did, opened up: steps 4 to 8 on the flow direction grid from Tuesday, then the Lab 5 ArcGIS Pro model that wires all eight together.

Tuesday's last slide, as the bridge. Two minutes: have someone read one D8 code back to a compass direction before moving on. Then step 4.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Step 4. Flow accumulation walks the tree from the previous slide and counts how many cells drain into each cell.

Follow one path with a finger: the ridge cells are 0, and the count grows downstream. Multiply a cell's value by the cell area to get drainage area in ground units. Note the zeros on the divides; that is the giveaway for the Esri convention.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Step 5. This is the step with a judgment call in it: the threshold is chosen, not computed. In ArcGIS Pro it is one Raster Calculator expression on flow accumulation, Con("Flow_Accumulation" > 5000, 1): 1 where the test is true and NoData everywhere else. Lab 5 adds a second test so only cells inside the delineated basin qualify. Leave out the third argument; Con(test, 1, 0) writes 0 instead of NoData and the next tool treats the whole DEM as stream.

The outlined cells are the ones whose flow accumulation is 5 or more (the cell reading exactly 5 is outlined, so the test in this figure is >=, not >; in ArcGIS Pro that is Con("Flow_Accumulation" >= 5, 1)). That set is the stream network for this threshold. Nothing about the terrain changed; only the number we compared against.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Two ways a cell joins the network: it passes the threshold, or it lies downstream of a point the user forced in (a gage, a culvert, a discharge location). The second rule is how you make the network include a channel the threshold would have missed.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Lab 5's own data, rendered in ArcGIS Pro: the Rock Canyon basin with a stream network at 500 cells (0.05 km² of contributing area on 10 m cells) and at 5,000 cells (0.5 km², the Lab 5 baseline). The counts are the lab's measured values: 290 segments and 83.2 km of stream at 500 cells, 29 segments and 22.9 km at 5,000. Nothing about the terrain changed; only the number we compared against. Ask what the "right" answer is; there isn't one, which is what Lab 5 Step 14 makes them test.

TODO(instructor): decide whether to add a scale/resolution sensitivity question here — e.g. how the delineated network and watershed change between a 30 m, a 10 m, and a 1 m lidar DEM, and whether the same cell threshold should be used.

Step 6 is two tools. Stream Link gives every segment of the stream raster its own number (the next two slides). Stream to Feature then turns that numbered raster into line features that follow the flow direction, one line per link, so the network can be attributed, measured, and used by hydrologic models. Keep the Stream Link raster: it is also the input for the subwatersheds in step 8.

The link is the unit of the vector stream network: one line feature, one record in the table. Junctions are where two links meet; every link has exactly one downstream end.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Each color is one stream link, and every cell in that link carries the same identifier. Left panel: colors. Right panel: the numbers actually stored in the raster. This identifier is the key that ties the raster network to the vector network on the next slide.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Lab 5's Stream_Links raster (one color per link) under its Stream to Feature lines, labeled with their grid_code field, rendered in ArcGIS Pro around a junction in the middle of Rock Canyon. Each line's grid_code is the value of the raster cells it was traced from: the vector line carries the link ID from the raster. That shared key is what lets you join raster-derived attributes, such as contributing area, to the line features.

Step 7. A pour point is any cell you ask the Watershed tool to find the drainage area of. For one basin, it is the outlet you care about, moved onto the channel with Snap Pour Point (Lab 5 uses a 50 m snap distance). For subwatersheds you do not extract points at all: the Stream Link raster itself is the pour-point input, and each link, numbered in step 6, acts as the outlet of its own subwatershed.

Say the definition slowly: a watershed is defined relative to a point. Change the point and you change the watershed. There is no such thing as "the" watershed of an area without an outlet.

The figure plays on entry: A's watershed fills first, then B's, which swallows A's. Ask before B appears: if I move the point to B, what happens to the boundary? Diagram, not a real basin.

Two sources of pour points: the ones the network gives you for free at the end of every link, and the ones you supply because you care about a specific location. In ArcGIS Pro, user-supplied points must be snapped onto the flow-accumulation network first, or the tool will return a watershed of a handful of cells.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

The algorithm is the reverse of flow accumulation: start at the pour point and walk upstream through the flow-direction tree, collecting every cell that eventually reaches it. The boundary falls out; it is never digitized.

Credit: figure adapted from the watershed delineation course materials of D. Maidment, University of Texas at Austin.

Step 8, the payoff. The next three slides show what it looks like on real terrain.

Consistent, not necessarily more accurate. Two analysts hand-delineating the same basin will disagree; the algorithm will give the same answer every time from the same DEM and the same pour point. Change the DEM or move the pour point and the answer changes. The map is Lab 5's result on the topographic basemap: the basin above the Rock Canyon trailhead (magenta), its streams at 5,000 cells (blue) and the snapped outlet (red), rendered in ArcGIS Pro. Ask them to check the divide against the contours on the basemap.

TODO(instructor): decide whether to add a validation step here — comparing the delineated network and basin against the NHD or against a StreamStats basin for the same outlet — and what students should report when they disagree.

The link identifier from the stream-links raster carries straight through into the watershed raster. That is why "same cell value" is written on the figure: the number on each subwatershed is the ID of the link it drains to. The map is Lab 5's 29 subwatersheds at 5,000 cells, each labeled with its gridcode, with the stream links in blue, rendered in ArcGIS Pro.

Every link in the network gets its own subwatershed, and together they tile the whole basin with no gaps and no overlaps. This is exactly the input a rainfall-runoff model wants: a set of subbasins, each with an area and a routing connection to the next one downstream.

The figure is Lab 5's own result on imagery: the basin above Rock Canyon (yellow) cut into one subwatershed per stream link, with the streams in orange and the outlet in red, rendered in ArcGIS Pro.

All eight, start to finish. Ask the class to name the input and output of each step before moving on.

Same chain as the eight steps, now readable because they know every box. Trace it row by row. Surface: Fill, then Flow Direction (D8), then Flow Accumulation. Basin: Snap Pour Point moves the outlet up to 50 m onto the highest-accumulation cell, Watershed collects everything that drains to it, and Raster to Polygon turns Basin_Raster into one polygon. Streams: the threshold enters at Raster Calculator, Con(("%Basin_Raster%" >= 0) & ("%Flow_Accumulation%" > %Threshold%), 1), which keeps cells inside the basin whose flow accumulation exceeds the threshold; then Stream Link numbers the segments and Stream to Feature draws them as lines. Subwatersheds: a second Watershed uses the Stream Link raster as its pour points, so every link gets its own subwatershed, and a second Raster to Polygon with Create multipart features checked makes one polygon per link. Two things to point at: the threshold is applied to flow accumulation, never to the Watershed tool, and Flow_Direction feeds four different tools. The P marks are model parameters; students set them in Lab 5 Step 12. This is the Lab 5 model exported from ModelBuilder (Lab 5 Figure C); the gray oval is Flow Direction's optional drop raster, left empty.

The layout at right is the Lab 5 example map, the baseline run of the model. The single most common lab failure is an unsnapped pour point producing a watershed of a few cells. Say so now.

The bridge to Week 7. The Great Salt Lake subregion (HUC4 1602) on the map at right is the basin Rock Canyon belongs to. A terminal lake turns the water-balance slide into one number students can see from the freeway: the water surface elevation. Lab 6 is built on that idea.

Four minutes, in pairs, then a show of hands on the consistent-versus-accurate item: most of the room will say the automated basin is more accurate, and the deck's claim is narrower than that. The threshold item is the other one worth a word: raising it thins the network, and nothing about the terrain changed. If the room has no signal, put the URL on the board; the items read aloud just as well.

Revision notes (2026-10-05, later): the two parts were swapped so class runs manual delineation, then StreamStats, then the ArcGIS Pro steps. Part 1 is now "Manual Watershed Delineation" (the two Hog Pen Creek slides merged into one, the hand-drawn divide revealed on a click, then the Emigration Creek activity and the StreamStats slides); Part 2 is steps 4 to 8, the Example Model and Back to Lab 5. The StreamStats goal moved to the top of Today's Goals, and its question moved to the top of the Follow the Water quiz, so goals and questions stay in the same order.

Restructure notes (2026-10-05): this deck now opens at step 4 (flow accumulation), carrying steps 4 to 8, the Example Model and the Back to Lab 5 slide over from Part A, and keeps the hand-versus-StreamStats delineation. Its old Part 1 (what a watershed is, Powell, anatomy, nesting, water balance, functions) moved to the start of Part A. See the restructure notes in watershed-delineation-a.md.

Split notes (2026-10-01): this deck is Parts 2 and 3 of the original Week 6 deck (source "CE 414 Week 6 - Watershed Delineation.pptx"; its conversion notes stay in watershed-delineation.md). New here: the title and goals, the "Watersheds Nest" map (USGS WBD units around Rock Canyon, replacing a 490 px management-units diagram), the water-balance figure (on Lab 5's basin outline, replacing two small hydrologic-cycle images), the terrain-processing panels rebuilt from Lab 5's data (replacing a 584 px nine-panel figure), and the bridge to Week 7. Figures from tools/week06_figures.py and tools/week06_water_balance_svg.py. Still third-party and small: the Orange County watershed diagram (450 px) and the processes-and-functions diagram (600 px); kept because they are sourced illustrations, flagged for replacement.