When I settled on the current game design idea a central piece was an environment simulation that would effect player experiences and that player choices could in turn effect - creating a feedback cycle.
The goal wasn't (and still isn't) to create an environmental simulation that was "physically accurate." The goal was to create a simulation that was an abstraction of reality or at least had some of the key features seen day to day and year to year. I wanted the player to be able to alter the world around them in a way that reflects reality but that doesn't need to be 100% realistic.
- Day to day temperature swings
- Seasonal temperature swings
- Wind to transport moisture
- Water/Rain cycle
- CO2 effects
- Pollution accumulation and transport
- Runs off the main thread
The first step in creating the simulation was to break up the environment into a series of blocks. Each box would have it's own set of environmental variables (temperature, humidity, ground water, wind, etc). These "climate blocks" would interact with it's neighbors to create localized weather and also to allow local variances such as terrain height and cloud density to impact the weather.
As the simulation developed and more complex interactions were add a few changes/additions to the climate blocks were needed. The first critical piece was the addition of an "atmosphere block" on top of each climate block. The atmosphere block is simulated air space above that often has a different temperature than the air on the ground/ocean. This allowed for up and down drafts that could help move clouds and cause an on or off shore wind depending on the season (more about this later). Without an onshore wind clouds and more importantly rain rarely make it to the land.
A second key addition was the classification of terrain types. At the time of writing this the simulation is only using three types of terrain ocean, grassland and mountains. This largely controls the "target" temperature of climate block and in the case of the ocean maintains a constant saturation of ground water. Oceans have moderate target temps with smaller seasonal and daily fluctuations. Grasslands have a higher target temp than the mountains but both have larger daily and seasonal fluctuations than the ocean.
Daily and Seasonal Temperatures
Seasonal sun intensity is controlled by an animation curve that is sinusoidal in shape. Max values occurring during the "summer" and of course minimum values occurring during the "winter." A second curve was used for daily fluctuations. The curves are currently identical... That may change in the future.
At first these curves were used and tuned to be a true energy flux with energy being added during the day and energy lost during the night. While more physically accurate and potentially allowing for more emergent weather, it also made it far more difficult to tune and balance. After significant testing and tuning this method was abandoned for a more controlled method.
Each terrain type has a median temperature that serves as a target temperature. There are also values for seasonal and daily fluctuations of that temperature. This means that median temperatures + seasonal flux + daily flux is roughly the high temp and likewise median temperature - seasonal flux - daily flux is roughly the low temp. Variance from this is caused by the connection of neighbors and other factors.
One of the other factors that adds some variability is the average height of the terrain. In the real world the higher the terrain the cooler the air. To reflect this each climate block samples the terrain height at multiple points and calculates an average height. From this the median temperature is adjusted slightly. This provides variation in the climate that (hopefully) lends itself to more interesting and emergent weather.
Wind in the simulation does not have a physical presence and is simply a Vector3 that stores the wind velocity. The wind for an individual climate block is calculated based on the temperatures of the neighboring climate blocks, including the upper atmosphere. The inclusion of the upper atmosphere allows up and down drafts which allows for onshore or offshore breezes (as mentioned above).
To calculate the wind the simulation calculates a temperature difference between the climate block and a neighbor and then multiplies that by a vector that goes from the climate block to the neighbor. Final direction (sign) of the wind is based on blowing from hot to cold.
The simulation sums all the winds due to neighbors to arrive at a (nearly) final wind value.
A prevailing wind is then added. This allows larger global winds to be added to create larger and different weather patterns. The strength of the prevailing wind is a tune-able variable. Current testing is with an in/out pattern that helps to blow clouds away from land in the summer.
After the prevailing wind is added the vector is normalized. There was some debate about normalizing the wind... In the end the choice was made to keep the normalization as this meant that clouds always would have some wind. It also prevents unrealistically large or small winds that become problematic when trying to transport clouds/moisture.
The final step is a check on the vertical component of the wind. Since the upper atmosphere is often at a very different temperature it can therefore be the dominant contribution to the wind. This is a problem due to only simulating two layers of atmosphere rather than more layers or the ideal of a continuous atmosphere. Effectively the vertical component is clamped to a range between -0.5 and 0.5. This helps to keep clouds generally moving horizontally.
Additional controls are needed to avoid the clouds going too high or going through the ground. The altitude of the clouds is mostly visual as the horizontal location determines where clouds can deliver rain. Because of the this a simple min and max altitude are set. Clouds check against these values and degrees of freedom are restricted in the rigidbody to prevent clouds from going outside the bounds. Checks on vertical velocity are also necessary to allow the clouds to move away from the limits.
In Part 2 I'll discuss the most difficult part (so far) of the simulation that being the Water and Rain Cycle. Or more specifically the balancing of that cycle.