brooksgroves.com  ·  the storysource on GitHub

SIERRA‑FLOW

Eight Sierra Nevada river gages, run through a COBOL batch job every morning and checked against 30 years of normal for the date.

LOADING

The last 60 days

Each river's daily mean flow against what that river normally does on those dates. The bright band is the middle half of 30 years (25th to 75th percentile); the faint band reaches the 10th and 90th. Flow is on a log scale, so a drought and a flood both fit.

Where the gages are

Three rivers out of Yosemite and the Stanislaus country. Four gages have no major dam above them and show what the mountains are doing; four sit below reservoirs, so their flow is mostly a release decision.

This morning's printout

The report exactly as the COBOL program wrote it: 132 columns, the width of a line printer. Every number on this page comes from it or from the file the program wrote beside it.

streamflow-report.txt 

    

  


  

    
The program, on punch cards

    
Fixed-format COBOL still keeps the shape of an 80-column card: sequence numbers in columns 1–6, the indicator in 7, code in 8–72. Here is each line of the real source, punched in IBM 029 card code. 

    

      

        
        
        
        
        
      

      

      

      

    

  


  

    
How it works

    

      

        
Every morning

        
A GitHub Action on Ubuntu installs GnuCOBOL, and a short Python script asks the USGS Water Data API for each gage's daily mean flow over the last 60 days. Then SIERRA-FLOW.cob compiles and runs: it compares each day with the normal for its date, sorts the gages with the COBOL SORT verb, writes the printout and a results file, and sets a return code: 0 when every gage reported, 4 when one is stale. The job log at the top of the page is that run's real output.

      

      

        
What "normal" means

        
NORMALS.cob builds it from 30 water years of USGS daily means, 1996 to 2025. It releases every day's flow to the sort seven times, once for each day within three days of its date, sorts about 584,000 records, then takes the 10th, 25th, 50th, 75th and 90th percentiles for each gage and day. The classes are the ones USGS WaterWatch uses. Grand Canyon of the Tuolumne's record starts in 2006, so its normals cover 19 years.

      

      

        
What changed in version 3

        
    
          
  • The old normals were one round number per gage for the whole year, so every river looked like a drought each fall. Each day now has its own.
    • 
          
  • After sorting, version 2 printed each station's median and alert thresholds against a different station.
    • 
          
  • The trend field couldn't hold a negative number, so no river could ever read falling.
    • 
          
  • The basin totals added up gages on the same river, counting the same water up to four times.
    • 
          
  • Each day's flow was the last reading of the day, not the day's mean.
    • 
          
  • The page used to recompute everything in JavaScript. It now shows what COBOL computed.
    • 
        

      

    

    
Version 1 first ran on April 1, 2026, the evening Artemis II lifted off from Launch Complex 39B at 6:35 p.m. EDT with Reid Wiseman, Victor Glover, Christina Koch and Jeremy Hansen aboard, the first crew to leave Earth orbit since Apollo 17.