Weather Bot — Episode 1: How I Started Betting on Temperature With $198 on Polymarket

It was a Tuesday around 2 AM when I found something that didn't make sense. I was digging through Polymarket trade histories — the kind of rabbit hole you fall into when you should've gone to bed three hours ago — and one account kept showing up in weather markets: gopfan2.

I pulled up the profile on PolymarketAnalytics. The numbers were real. $1.48 million in verified profit. A 60.8% win rate across 1,700+ positions. And here's the part that got me: the PnL curve wasn't some lucky spike from one big bet. It was a steady grind upward over 18 months, trade after trade after trade.

I stared at that chart for a while. Then I looked at my Polygon wallet. $198. And a thought that wouldn't go away: if someone can make $1.48 million just by predicting tomorrow's temperature better than the crowd, could I automate that?

What This Post Covers

This is the first episode in a series documenting how I built an automated weather trading bot on Polymarket. I'll walk through what Polymarket weather markets actually are, the traders who convinced me this was worth trying, and one resolution rule that creates real mispricing most people miss. If you've been curious about prediction markets but aren't sure how the money flows, start here.

Polymarket Weather Markets, Explained

Polymarket is a prediction market running on the Polygon blockchain. You're not buying stocks or tokens — you're buying shares in real-world outcomes. Every market has two sides: YES and NO. Prices range from $0.01 to $0.99. Buy YES at $0.10, and if the answer turns out correct, each share pays $1.00. That's 10x. Wrong? Zero. No partial credit.

Weather markets get weirdly specific. Like: "Will the highest temperature at Esenboğa Airport in Ankara be 10°C on March 5?"

No committee decides the answer. A thermometer at that specific airport reads a number, Weather Underground publishes it, and the market resolves. That simplicity is what hooked me — there's no opinion, no panel vote, no appeal. Either the temperature hit that number or it didn't.

The Traders Who Started All This

Before writing a single line of code, I spent three days studying on-chain trading histories. I needed to know if anyone was actually making money from weather — or if it was just gambling with blockchain steps.

gopfan2 is the obvious headliner. $1.48M in total Polymarket profit with weather as a major focus. The strategy is almost stupidly simple: buy YES shares priced below $0.15, bet around $1 per position, and do it thousands of times. Not every trade wins — the overall win rate is 60.8%. But the math works because winning trades pay 6-10x while losing trades only cost $1.

gopfan2 wasn't alone. meropi automated $1-3 bets across dozens of cities and pulled about $30K in profit — with some $0.01 positions returning 500x. 1pixel turned $2,300 into $18,500 by focusing exclusively on NYC and London, including one trade that went from $6 to $590. neobrother made over $20K specializing in Buenos Aires, Ankara, and Miami, using a technique called "temperature laddering" — buying multiple adjacent temperature ranges simultaneously.

And then there's Hans323. One single bet. $92,000 on a London weather outcome priced at 8% probability. It hit. $1.11 million.

I'm not Hans323. I had $198 and a MacBook. But gopfan2's approach — small bets, high volume, systematic edge — that's the one I couldn't stop thinking about.

Why Weather Instead of Elections or Sports

I looked at other Polymarket categories first. Elections, sports, crypto prices. They all had the same issue: outcomes depend on human behavior, and humans are unpredictable in ways that models can't fix.

Weather is a different game. The resolution is completely objective — Weather Underground publishes airport temperature data, everyone can verify, no disputes. Professional forecast models are free — the best ones in the world (GFS from the US, ECMWF from Europe, ICON from Germany) are all available through Open-Meteo's API with no API key and no paywall. Most Polymarket traders aren't using them. They're checking their phone's weather app or just guessing.

New markets open daily for 12+ cities. Unlike election markets that drag on for months, weather resolves in 24-48 hours. You find out fast whether you were right.

And the bet sizes are tiny. gopfan2 bets $1 per position. With $198, I had enough runway to try this for weeks.

The Resolution Rule Most People Get Wrong

This took me too long to figure out.

Polymarket weather markets resolve using Weather Underground data, and temperatures are truncated to whole degrees. Not rounded — truncated. That difference matters more than you'd think.

• 10.9°C → resolves as 10°C
• 10.1°C → resolves as 10°C
• 11.0°C → resolves as 11°C

See the problem? If weather models predict 10.7°C, most people think "basically 11" and bid up the 11°C bucket. But the actual resolution is 10°C. This truncation rule creates a systematic gap between what people expect and what gets recorded.

My Starting Point

Here's what I walked in with: $198 USDC on Polygon as my entire prediction market bankroll. A MacBook Pro M4. Python — though not in the traditional sense. I'd been using Claude to write and debug code for months. I could read it, understand what it did, and modify pieces. Writing from scratch? Not my level. And zero prediction market experience. I'd heard of Polymarket but never placed a single trade.

The plan made sense in my head. Three weather models produce forecasts. Open-Meteo gives me the data for free. I compare those forecasts to Polymarket's market prices, and wherever the models see something different from the crowd, I buy. Automate the whole thing, let it run while I sleep.

If gopfan2 was doing this manually at scale, I figured a bot could do it with a fraction of the capital. The weather data was free. The APIs were documented. The math seemed straightforward.

Key Takeaways

• Polymarket weather markets resolve objectively — an airport thermometer reads a number, Weather Underground publishes it, done
• gopfan2: $1.48M profit, 60.8% win rate, $1 per bet. Other traders pulling $18K-$30K from weather alone.
• Free professional weather models (GFS, ECMWF, ICON) provide data most market participants aren't using
• Temperatures are truncated, not rounded. 10.9°C resolves as 10°C. Most people miss this.

What's Next

In Episode 2, I'll get into how I actually turned weather model data into trading signals — the three models, Open-Meteo's API, and why I ended up targeting cities like Ankara and Buenos Aires instead of the obvious picks like NYC or London. Where you trade turns out to matter almost as much as how.

← Previous: (Series Start)       Next: Episode 2: Turning GFS, ECMWF, and ICON Forecasts →

---

More updates on the way. If you're working on something similar or found a smarter way to do it, drop it in the comments — the more we share, the faster we all move.

Disclaimer: This blog documents my personal learning journey. Nothing here is financial advice.