Dataset: We aggregated data from the United States Geological Survey (USGS) and the Flood Control District of Maricopa County (FCDMC), Arizona, USA. The database contains information specific to individual storms monitored at 15 catchments during the period from October 1991 to February 2005. Rain was recorded on 148 days from varying numbers of catchments per day, yielding 394 runoff events (where event refers to runoff from a single catchment during a single storm). Storm attributes, runoff hydrology, and runoff chemistry were separately recorded for each event. The five storm attributes were total precipitation, number of rain-free (dry) days preceding the storm, storm intensity, storm duration, and the season in which the storm occurred. Each catchment is described by six landscape characteristics: contributing area, the proportion of area with impervious land cover, and the proportions of area under residential, commercial, industrial, and undeveloped land use.

The catchments were selected to span a range of land-use types. Classification was based on aerial photographs and ground truthing by walking each catchment. Portions of a catchment that did not contribute to runoff were not included in contributing area. The channels flowing out of the bases of these urban catchments are made of impervious paving material. These channels are thus concrete canals and roadway curbsides. This channel structure means that channel features are easily measured (e.g., channel cross sections are simple polygons) and do not erode, shift, or otherwise change substantially over the decadal time scale of this study. Additionally, because they comprise paved material above and at the point of catchment exit, the entire outflow channel of any catchment is a flow-control structure in its own right. The installation of control structures such as V-notch weirs was unnecessary.

At each catchment, a tipping-bucket rain gage was used to measure precipitation. While runoff efflux was occurring, an instantaneous measurement of discharge (L/s) was made every 60 s. The mean of two consecutive measurements was multiplied by 60 s for an estimate of discharge for each minute. Total runoff was obtained by summing across all minute-specific estimates of discharge.

Our acquisition of instantaneous discharge data needs to be considered in two steps. First, a stage-discharge rating curve was developed for the outflow channel of each catchment. Such a curve shows the relationship between stage (i.e., water depth above the channel bottom) and discharge. To develop the rating curves, stage and discharge were estimated independently at several levels of discharge. Stage was measured using either a float and tape system or a thermally insulated pressure transducer that was calibrated to stage. Discharge was either directly measured using Price Type AA and Pygmy current meters (USGS Hydrologic Instrumentation Facility, Stennis Space Center, Mississippi, USA) or was calculated from the Manning equation based on the slope-conveyance method of setting rating curve shape. The Manning equation describes discharge as a function of channel cross-sectional area and the factors that influence flow velocity (slope, roughness, and area-to-perimeter ratio of the wetted cross section). The parameters for the Manning equation were uniquely determined for each catchment, and were either established via survey (e.g., of slope and cross section) or taken from commonly accepted values (roughness coefficients for concrete). These parameters were assumed not to change over the time span of the study, given that catchment outflow channels were made of consolidated paving materials (e.g., concrete). Rating curves fit well to the stage and discharge data used in building them (r = 0.99).

The second step is the acquisition of discharge data in the actual runoff events for which we report runoff chemistry in this study. After the rating curve had been developed for any given catchment, discharge in subsequent runoff events (those reported here) could be estimated by measuring stage. In events for which we report N load and concentration in the present study, stage was measured with a pressure transducer that signaled a datalogger. The datalogger, in turn, was programmed with the rating curve for its respective catchment, so instantaneous measures of pressure (stage) were converted into instantaneous estimates of discharge.

Load is a computed variable, where load (mass N/catchment area) = concentration of N in runoff (mass N/L discharge) multiplied by the area-corrected amount of discharge during the storm (L discharge/contributing catchment area). During an event, up to 24 discrete water samples were collected by automated pumping through an intake tube fixed to the channel bed in the center of flow at the efflux point where discharge was measured. Other than the intake tube, water samples were only in contact with glass, polytetrafluoroethylene (PTFE), and stainless steel. Samples from this tube were representative of the whole water column, as confirmed with manually collected, depth-integrated samples. A new, discrete sample was pumped into its own 1-L, PTFE-lined polyethylene bottle every time a specified volume of runoff had passed the collection point. Discrete samples were then combined into a composite by mixing equal volumes of each. The concentration of each N species in the composite sample is defined as the event-mean concentration. Here, load is the product of this event-mean N concentration and discharge during the period when the 24 discrete samples were collected, divided by catchment contributing area.

Composite samples were analyzed at the USGS National Water Quality Laboratory (Denver, Colorado, USA) following methods in Fishman and Friedman (1989; http://pubs.er.usgs.gov/usgspubs/twri/twri05A1). Chain of custody was documented during sample transport. Field and laboratory blanks were analyzed for sample contamination, spiked samples were analyzed for interferences, and duplicate samples were analyzed for precision. Results from these quality controls confirmed that sample collection, handling, and analysis produced reliable water chemistry data. All N species were analyzed colorimetrically with an automated-segmented flow system. Concentrations of NO2 + NO3 (hereafter, NO3 ) were determined with cadmium reduction-diazotization (USGS method I-4545-85). Concentrations of NH4 were determined with a salicylate-hypochlorite technique (USGS method I-4522-85). Concentrations of reduced N (NH4 +organic N) were determined by digesting organic compounds with sulfuric acid in the presence of mercuric sulfate and potassium sulfate and then analyzing for NH4 with the salicylate-hypochlorite technique (USGS method I-4552-85). For this paper, we calculated organic N and total N concentrations in the manner described in the metadata for those specific variables.

The volume of discharge required to trigger the pumping of a new sample was set to divide runoff into 24 discharge-equivalent (as opposed to time-equivalent) periods. Discharge-weighting accommodates pulsed runoff, which occurs owing to transient storage zones within the catchment, or to fluctuations in rainfall intensity. The sample-triggering volume was specific to each catchment (i.e., catchments with typically more runoff need larger sample-triggering discharge). When discharge was greater than expected, water sampling stopped after the 24th sample, and the tailing end of the runoff was not sampled. We assumed that by the 24th sample, all potentially mobile material would have been exported from the catchment; thus, missing the tailing end of runoff was inconsequential for estimating load. Conductivity data (not shown) from discrete runoff samples support this assumption.

Discharge: Volume (liters) of storm-associated water (runoff) exported from catchment. This is not the total amount of water exported. Rather, it is the amount of water exported from the catchment during the period in which water samples were collected for chemical analysis. Sometimes water samples were collected during the the entire period in which runoff was occurring. In many cases, however, a slower trickle of "tailing" runoff continued long after sampling was ceased. It is assumed that most N exported by storms occurs in the first flush, and that this tailing, unmeasured discharge is dilute. Thus, the tailing, unmeasured discharge is considered unimportant for calculating load. For more commentary on how discharge was sampled, see expanded metadata commentary, or read Lewis and Grimm (2007) Ecological Application 17(8):2347-2364.

Season: summer monsoons originate in the Gulfs of Mexico and California, and occur between July and October; winter cold fronts originate in the Gulf of Alaska, occur between November and March; and spring storms between April and June can derive from either storm.

NO3 concentration (mg N / L): Event-mean concentration of N as NO2 and NO3. Total (unfiltered sample). If this variable (called "NO2NO3" in the original USGS file) was not available, and the variables "NO2" and "NO3" were separately available, than the value shown here is the sum of NO2 and NO3. For explanation of "event-mean" concentration, see expanded metadata commentary, or read Lewis and Grimm (2007) Ecological Application 17(8):2347-2364.

NH4 concentration (mg N / L): Event-mean concentration of N as NH4. Total (unfiltered sample).

Organic N concentration (mg N / L): Event-mean concentration of N as organic N. Total (unfiltered sample). This variable is the most complete compliation of organic N information. It is computed as follows. (1) If USGS-supplied variables NH4 and NH4+ORGN are available, then org N conc = (NH4+ORGN) - NH4 (variable "NH4+ORGN" removed from this file). If, by this equation, org N conc less than 0, then org N conc = 0. (2) If USGS-supplied variable ORGN is available, and, following the above criteria, org N conc less than ORGN, then org N conc = ORGN (variable ORGN removed from this file). (3) If USGS-supplied variables NH4 and NH4+ORGN are not available, and USGS-supplied variable ORGN is available (this condition is rare), then org N conc = ORGN.

Total N concentration (mg N / L): Event-mean concentration of N (all forms). Total (unfiltered sample). Calculated by me using one of the following approaches, in order. (1) N = org N + NO2NO3 + NH4, where org N is calculated as described above. (2) If NH4 is missing, and NH4+ORGN is available, N = "NH4+ORGN" + NO2NO3. (3) If NH4 and NO2NO3 are missing, and both NO2 and NO3 are separately available (rare situation), N = "NH4+ORGN" + NO2 + NO3

NO3 load (kg N / km2): Mass of N as NO2 and NO3 exported from the catchment during the storm, corrected for the contributing area of the storm. Calcuated as NO3 load = NO3 conc / (discharge x catchment area).

NH4 load (kg N / km2): Mass of N as NH4 exported from the catchment during the storm, corrected for the contributing area of the storm. Calcuated as NH4 load = NH4 conc / (discharge x catchment area).

Organic N load (kg N / km2): Mass of N as organic N exported from the catchment during the storm, corrected for the contributing area of the storm. Calcuated as organic N load = organic N conc / (discharge x catchment area).

Total N load (kg N / km2): Mass of N (all forms) exported from the catchment during the storm, corrected for the contributing area of the storm. Calcuated as Total N load = total N conc / (discharge x catchment area).

Landfactor x storm attribute interaction terms: Each one of these is calculated as follows: Interaction = (landfactor - A) x (storm attribute - B), where A and B are "centering" values, used to center the landfactor and storm attribute variables. That is, A = a value roughly in the middle of the range of the landfactor variable, and B = a value roughly in the middle of the storm attribute variable. These are derived variables used in path analysis in Lewis and Grimm (2007) Ecological Application 17(8):2347-2364.