Worldgen: broaden desert climate distribution and resolve unused biome weights

[Alfredlang1989]

Problem

Deserts appear much rarer in-game than intended. The current worldgen does not seem to have a rendering/pass-order failure; the main issue is the shape of desert_mask itself.

data/worldgen/desert_mask.lua currently requires a location to be both sufficiently hot and sufficiently dry, then multiplies both already-selective responses with additional suppression terms:

• hot = max(0, (temp - 0.56) / 0.38)
• dry = max(0, (0.43 - rain) / 0.36)
• desert = hot * dry * nonmountain * patch

The desert surface passes only activate above desert_mask > 0.10. In practice this means moderate hot/dry combinations are damped very quickly, producing small/scattered desert areas instead of broad climatic desert zones.

Example:

• temp = 0.70 -> hot ~= 0.37
• rain = 0.25 -> dry ~= 0.50
• before mountain/patch suppression: hot * dry ~= 0.185

A fairly plausible desert climate is therefore already close to the visual activation threshold before the remaining multipliers are applied.

Additional finding: BiomeDef::weight is currently not used by terrain blending

data/biomes.json defines biome weights such as:

• plains: 3.5
• forest: 2.6
• desert: 1.7
• badlands: 1.4

However, RuntimeBiomeTerrain currently stores only the terrain definition, mask field and biome salt. adjustedSurfaceHeight() then uses the sampled terrain mask directly as the blend weight.

Therefore BiomeDef::weight currently has no effect on biome terrain contribution despite being exposed in the registry as a worldgen selection weight.

This should either be integrated with a clearly defined semantic or removed/renamed so the data model does not imply behavior that does not exist.

Desired behavior

Desert distribution

Deserts should remain climate-driven and deterministic, but occupy substantially broader contiguous regions.

Preferred direction:

• keep the existing large-scale temperature/rainfall fields;
• broaden the hot and dry response ranges rather than simply lowering the final pass threshold;
• use a softer combination than a raw hot * dry product, for example a shaped product such as pow(hot * dry, exponent) with 0 < exponent < 1, or an equivalent smooth response;
• retain mountain suppression so normal deserts do not indiscriminately cover alpine/high-mountain terrain;
• preserve desert_high_mountains_mask as the separate hot/dry mountain case;
• avoid tiny noisy sand patches. The result should look like large climate zones with soft transition regions.

A reasonable initial tuning target is roughly:

• clear desert cores: ~5-10% of representative large-area samples;
• desert transition / semi-arid influence: ~10-20%;
• badlands remain distinctly rarer than ordinary desert.

These percentages are tuning goals, not hardcoded runtime rules.

Biome weight semantics

Choose and document one of the following:

1. Use BiomeDef::weight as an explicit multiplier/normalization input in biome terrain contribution, with tests proving that changing the value affects the resulting blend; or
2. Remove/rename it for terrain selection if weights are intended only for a future/discrete biome selector.

Do not silently keep a public data-driven parameter that appears active but is ignored.

Important non-goal

Do not fix desert rarity solely by reducing the desert_sand / desert_sandstone mask threshold below 0.10. That would mainly expose weak mask tails and risks scattered thin sand patches rather than correcting the climatic distribution.

Validation

Add deterministic worldgen tests/statistics over a sufficiently large representative 2D sample window for seed 1337 and at least one additional seed.

Tests should verify:

• desert mask produces non-trivial contiguous coverage;
• ordinary temperate/wet regions do not become desert;
• desert/high-mountain separation still behaves correctly;
• no seams/regressions across ChunkGroup/Section/Region boundaries;
• any chosen BiomeDef::weight semantics are covered by a regression test;
• existing deterministic worldgen tests remain stable except for intentionally changed desert distribution snapshots/expectations.

Relevant files

• data/worldgen/desert_mask.lua
• data/worldgen/desert_high_mountains_mask.lua
• data/biomes.json
• data/worldgen.json
• src/world/registry/Registry.h
• src/world/worldgen/WorldGen.part02.inc
• src/world/worldgen/WorldGen.part04.inc
• worldgen regression tests