Water Doesn't Code Itself: What I Learned Building Hydrological Models for NGOs

If you had told me a few years ago that I would be doing climate modeling, simulating water availability decades into the future, I would have laughed. I write web apps. I build SaaS. But software work in Nepal takes you to strange and meaningful places, and one of those places was hydrology.

How a developer ends up doing climate science

An NGO needed water resource modeling for a watershed near Godawari. They could not afford a big commercial consultancy and they could not afford expensive proprietary software with per-seat licenses priced for Western budgets. What they had was a problem, a deadline, and a developer willing to learn fast. That was me.

In a developing country, the line between domains blurs. The person who can learn the tool is more available than the specialist who already knows it.

What WEAP actually is

WEAP, the Water Evaluation and Planning system, lets you build a model of a watershed: where water comes from, where it goes, who uses it, and how that changes under different scenarios. You feed it rainfall, land use, demand, and it simulates how much water is available where, over time. For an NGO trying to plan for a community's water future, it is exactly the right tool.

The CMIP6 part, in plain words

To model the future you need future climate data. That comes from global climate models, the CMIP6 ensemble being the current generation. The problem is that global models work at a coarse scale. They are not built to tell you what happens in one specific small watershed in Nepal. Their raw output has systematic biases when you zoom into a local area.

So you do bias correction. You take the model's historical output, compare it against what actually happened locally, learn the offset, and apply that correction to the future projections. It is unglamorous statistical work, and it is the difference between a model that means something locally and a model that is just confidently wrong at the wrong scale.

• Global climate data is the raw ingredient. It is not the meal.
• Bias correction is the step that makes a global projection say something honest about a local valley.
• Get this wrong and every downstream conclusion inherits the error.

What the model said about 2050

The projections were sobering. Without going into numbers I am not qualified to defend in a journal, the direction was clear: water stress in the region gets worse as we move toward mid-century. Changing rainfall patterns, shifting timing, growing demand. Nepal sits below the Himalaya and people assume we are water-rich. At a national scale, maybe. At the scale of a specific community in a specific season, the future the model painted was not comfortable.

What it taught me

A few things stuck. First, that software skills are transferable to problems that matter far beyond software. The ability to wrangle data, automate a pipeline, and reason about a model is useful in hydrology, in medicine, in anything. Second, that the affordability gap in tools is real and harmful. This NGO did important work and was nearly priced out of the software it needed to do it. And third, that some of the most meaningful engineering I have done had no users, no dashboard, no launch. Just a model, an NGO, and a question about whether a community would have enough water in twenty-five years. That mattered more than any feature I have ever shipped.