Real-time realistic illumination and shading

of stratiform cloudsAntoine Bouthors, Fabrice Neyret, Sylvain Lefebvre

Evasion-GRAVIR / IMAG-INRIAGrenoble, France

Eurographics Workshop on Natural Phenomena

MotivationsClouds are very complex:

Complex shapeLots of visual featuresLots of details

No one has reproduced everything yet

Cloud facts

Clouds types

Multiple scatteringNo absorption, albedo = 1Thick (up to several km) Dense (Mean free path = several meters)

→ Very high number of scattering events (up to several hundreds)

Mie phase functionStrongly anisotropicComputationally expensive

• Depends on droplet size, temperature, wavelength

49% 49% 1%

Droplet size distributionOne common phase function for the

whole cloud

Visual features Fogbow & glory

Visual features Anisotropic forward scatteringDiffusive multiple scattering

Visual features Pseudo-specular effect

Visual featuresClouds-ground inter-reflections

Ice blink Water sky

Previous workOffline approaches

Musgrave

Musgrave et. al.

Harris et. al.

Previous workInteractive approaches

Different approximations:Low albedo / low density → single scatteringSimpler phase function → not all featuresDiffusion approximation → isotropic

Volumetric modelsVery coarse

Billboards or slicesLots of overdraw

Our Model

OverviewWhat we assume:

Stratiform → locally equivalent to a slabHomogeneous density

What we want:RealisticReal-time (GPU-friendly)Allow animation

OverviewComplex behavior

Rays with different orders of scattering

Our idea: a study of the contribution of each order

OverviewOur analysis:

Narrow forward scattering only important on silhouettes (low orders)

Low orders of scattering give anisotropic features (fogbow, glory, pseudo-specular)

Higher orders become diffusive (i.e. isotropic)Sky and ground play a role

Sky illuminance = ~25% of that of the sunStandard floor reflectance = ~20%

OverviewOur analysis:

Narrow forward scattering only important on silhouettes (low orders)

Low orders of scattering give anisotropic features (fogbow, glory, pseudo-specular)

Higher orders become diffusive (i.e. isotropic)Sky and ground play a role

Sky illuminance = ~25% of that of the sunStandard floor reflectance = ~20%

OverviewOur analysis:

Narrow forward scattering only important on silhouettes (low orders)

Low orders of scattering give anisotropic features (fogbow, glory, pseudo-specular)

Higher orders become diffusive (i.e. isotropic)Sky and ground play a role

Sky illuminance = ~25% of that of the sunStandard floor reflectance = ~20%

OverviewOur analysis:

Narrow forward scattering only important on silhouettes (low orders)

Low orders of scattering give anisotropic features (fogbow, glory, pseudo-specular)

Higher orders become diffusive (i.e. isotropic)Sky and ground play a role

Sky illuminance = ~25% of that of the sunStandard floor reflectance = ~20%

OverviewScattering:

Strong forward scattering: special treatment1+2 orders: analyticHigher orders (3+) considered isotropic

Environment:Take sky into account (diffuse source)Clouds-ground radiosity (plane parallel)

Shape: Height field

Clouds representation

Clouds are stored as a height fieldWell fitted for stratiform cloudsLot of details in a small spaceProcedural & animatable

Phase functionOur Modified-Mie model

Strong narrow (<5°) forward scattering moved into extinction function

Error negligible with multiple scattering

Phase functionModified-Mie model validation

Monte-Carlo bench: generated reflection BRDFs

1234567+

1234567+

Single scattering

Slab -> Local analytical function

Double scattering

Local approximation by convolution

3+ scatteringConvolution = bad idea for high ordersAssume slab → 1D problem → column

3+ scatteringAssuming diffusion → simple

interaction between column cellsGiven scattering behavior

of a cell → analytical solutionfor the column

3+ scattering

How to know the multiple scattering behavior of one cell ?Characteristic of the « cloud material »Precompute Monte Carlo integration

→ reflectance & transmittance of a cell

3+ scattering

Issue:Diffuse hypothesis broken on top cell:

anisotropy still plays a roleSolution:

1D model correction term dependant on the sun incident angle

Correction parameters fitted on Monte Carlo simulations

Sky & ground contributions

Previous: valid for directional L,VSky & ground: Assumed diffuse source

1D model used→ gives diffuse reflectance & transmittance

Ground-clouds inter-reflections

Radiosity between two facing parallel planes → known form factors (analytic)

Heterogeneous source (clear sky + cloud bottom) Heterogenous reflectance (ground AND clouds)

Ecloud = (Sun+sky)T + Σi(Rcloud_i FF Eground_i)Eground = (Sun+sky)R + Σi(Rground_i FF Ecloud_i)

Ground-clouds inter-reflections

Reflectances and radiosities in texturesPlane-plane → Form Factor (ring to dS)

→ Using MIP-mapping

GPU-enhanced iterative algorithm

Render-to-textureHardware MIP-mapping

= -

ValidationValidation of our 1D model

ValidationComparison of our model with a Monte-

Carlo integration

100m-thick slab

5m-thick slab

Monte-Carlo bench: generated reflection BRDFs

123

123

ImplementationHeight field: advected textures [Ney03]

16km-wide landscape2km-height, 500m-thick cloud layer

Shaders + radiosity on GPU512x512 clouds textures512x512 ground textures512x512 shadows textures16x16 radiosity textures

18 to 40 FPS on current hardware

Results

Lighting contributionsbottom view

1 and 2 scattering

3+ scattering

Ground illumination(incl. radiosity)

Sky illumination

Summing it all bottom view

Lighting contributionstop view

1 and 2 scattering

3+ scattering

Ground illumination

Sky illumination

Summing it all top view

FeaturesGlory and fogbow

FeaturesGlory and fogbow

FeaturesWater sky

FeaturesPseudo-specular reflection

FeaturesPseudo-specular reflection

VideosVideos

ConclusionGood

RealisticReproduce all clouds visual featuresAccount for clouds-ground inter-reflectionsReal-timeAnimation-friendly

BadPoor lateral shading (clouds borders)3+ scattering can be improvedLimited to stratiform clouds

Questions ?