*2.2.1. Work preparation*

reflectance models (e.g., *FLIGHT* [13], *Sprint* [14], *Raytran* [15]) in the context of the RAdiation

DART creates and manages 3D landscapes independently from the RT modeling (e.g., visible and thermal infrared spectroradiometers, LIDAR, radiative budget). This multi‐sensor functionality allows users to simulate efficiently radiative transfer products of the same landscape as being captured by various sensors. Major scene elements are as follows: urban features, trees, grass and crop canopies, and water bodies. A DART simulated tree is made of a trunk, optionally with branches created with solid facets, and crown foliage that is simulated either as a set of facets or as a set of turbid cells, with specific vertical and horizontal distribu‐ tions of leaf volume density. Trees of several species with different geometric and optical properties can be located within the simulated scene of any user‐defined size randomly or based on exact coordinates. Urban objects (houses, roads, etc.) contain solid walls and roofs built from triangular facets. Finally, water bodies (rivers, lakes, etc.) are simulated as facets of appropriate optical properties. DART can use external libraries to import and to some extent also edit landscape elements, digital elevation models (DEMs), and digital surface models (DSM) produced by other software or measured in field (e.g., translation, homothetic and rotational transformations). Most importantly, the imported and DART‐created landscape objects can be combined into virtual Earth scenes of user‐defined complexity. This allows importation of whole cities from urban databases provided by city councils and urban planners. The optical properties of landscape elements and their geometry, as well as and

properties of atmosphere, are specified and stored in adjacent SQL databases.

validated [18, 19] with simulations of the MODTRAN atmosphere RT model [20].

the optimal products required by the URBANFLUXES project.

Set of improvements had to be recently implemented in the DART model in order to provide

**2.2. Recent improvements of DART**

Atmospheric cells are used to simulate attenuation effects of satellite at‐sensor radiance and also to model influence of the atmosphere composition on radiative budget of Earth surfaces. The atmosphere can be treated just as an interface above the simulated Earth scene or as a light‐ propagating medium above and also within the simulated Earth scene, with cell sizes inversely proportional to the particle density. These cells are characterized by their gas and aerosol contents and spectral properties (i.e., phase functions, vertical profiles, extinction coefficients, spherical albedo). These quantities can be defined manually or obtained automatically from an atmospheric database. DART contains a database that stores the properties of major atmospheric gases and aerosol parameters for wavelengths between 0.3 and 50 μm. In addition, external databases can be imported, for instance, from the AErosol RObotic NETwork (AERONET; http://aeronet.gsfc.nasa.gov/) or from the European Centre for Medium‐Range Weather Forecasts (ECMWF; http://ecmwf.int/). Atmospheric RT modeling includes the Earth/ atmosphere radiative coupling (i.e., radiation that is emitted and/or scattered by the Earth and backscattered by the atmosphere towards the Earth). It can be simulated for any spectral band within the optical domain from the ultraviolet up to the thermal infrared part of the electro‐ magnetic spectrum. The Earth/atmosphere coupling was cross‐compared and successfully

transfer Model Intercomparison (RAMI) experiment [16, 17].

230 Sustainable Urbanization

The urban information used for the URBANFLUXES project was provided in the form of urban databases (for an example, see **Figure 7**). The two following adaptations had to be introduced to interface the databases with DART: (1) the format of 3D objects (i.e., triangular facets) stored in a common file format "\*. obj", and (2) the information about the vegetation, which was simulated as a turbid medium. Turbid vegetation was created from a set of characteristics, which for each tree in the urban scene provided geographical coordinates, physical dimen‐ sions, and optical properties. Those characteristics were obtained partially from external sources or partially from databases already exiting in DART (e.g., from database of optical properties for urban elements and vegetation).

#### *2.2.2. Modeling of in situ camera acquisitions*

In the frame of the URBANFLUXES project, in situ cameras are used for acquiring images of urban canopy. The major objective is to better assess the properties (i.e., spectral reflectance/ emissivity and thermodynamic temperature) of the urban elements that are observed. The images acquired by these sensors can be very useful for validating the approach that we devised in order to retrieve maps of optical properties from satellite images. In this context, we introduced the modeling of these in situ sensors into DART. The major difficulty was linked to the fact that these sensors (e.g., fish‐eye cameras) have a wide field of view (FOV). Hence, different elements of a scene are not viewed along the same direction, which strongly impacts the radiometry and geometry of the acquired images. It is interesting to note that most remote sensing models neglect this fact: They consider that sensors have an infinitely small FOV. In short, DART can know simulate in situ cameras at any location in the Earth landscape (natural or urban), with any view direction, either upward or downward. Some examples of simulated images are shown here: downward looking sensor (**Figure 3**) that is on top of an urban district (Toulouse, France) and sensors with upward and oblique view directions (**Figure 4**).

**Figure 3.** DART simulation of a fish‐eye camera acquisition above an urban district of Toulouse (France). Left: Red‐ green‐blue spectral composite in natural colors; right: a thermal infrared image.

**Figure 4.** DART simulation of in situ cameras: (a) Trees with leaves simulated as facets viewed by a camera with an upward looking direction, (b) and (c) trees with leaves simulated as turbid material in an urban environment viewed by a camera with a horizontal view direction.

Work continues in order to generalize the simulation of in situ sensors in case of any Earth surface element: fluids, atmosphere, water, etc., for any Earth scene configuration (i.e., isolated scenes and infinitely repetitive scenes with/without continuity of local digital elevation model). In addition to help in understanding and calibrating remote sensing measurements, the simulation of in situ sensors can be very helpful for a number of applications: to determine the optimal location and view direction of in situ sensors and to assess local atmosphere and pollution impact on sensor acquisitions.

#### *2.2.3. Atmosphere database*

In relation to the in situ cameras implementation, the DART atmosphere database and its management in the model were improved to offer higher flexibility of DART when dealing with different atmospheric conditions, especially in case of available in situ and/or satellite measurements of atmosphere. This atmosphere database was originally derived from simu‐ lations of the MODTRAN atmosphere radiative transfer model. Its use with DART was already validated with MODTRAN simulations. However, the simulation of aerosols was not as accurate as for gases. The DART atmosphere database was therefore completed using trans‐ mittance spectra for scattering and absorption mechanisms, derived from the atmospheric model MODTRAN. An interesting feature of this improvement is a new possibility to specify gas and aerosol amounts within the urban scene, independently of the atmosphere character‐ istics above the considered environment. This will be of a great help for assessment of local atmospheric properties and pollution impact using in situ sensor acquisitions. This improve‐ ment was accomplished by importing HITRAN [21] line‐by‐line cross section database (with specified temperature and pressure) for thermal infrared spectral domain, as well as the MPI‐ Mainz [22] cross section database for visible and near‐infrared spectral domains.

**Figures 5** and **6** show comparisons of DART and MODTRAN simulations in the visible and near‐infrared and the thermal infrared wavelengths, respectively. Both results demonstrate that the DART update and the introduction of new atmospheric database, combined with an improved radiative transfer modeling approach, brings DART simulations of atmosphere very close to MODTRAN 5.1 simulations, which is very encouraging especially for simulating ac‐ curately the in situ sensors.

**Figure 5.** DART (red) vs. MODTRAN 5.1 (blue) simulations in short wavelengths (UV, VIS, near infrared). Gas model: US standard, aerosol model: Rural, visibility: 23 km. (a) Sun irradiance, (b) bottom of atmosphere (BOA) radiance, (c) top of atmosphere (TOA) reflectance (*ρground* =0.5).

**Figure 4.** DART simulation of in situ cameras: (a) Trees with leaves simulated as facets viewed by a camera with an upward looking direction, (b) and (c) trees with leaves simulated as turbid material in an urban environment viewed

Work continues in order to generalize the simulation of in situ sensors in case of any Earth surface element: fluids, atmosphere, water, etc., for any Earth scene configuration (i.e., isolated scenes and infinitely repetitive scenes with/without continuity of local digital elevation model). In addition to help in understanding and calibrating remote sensing measurements, the simulation of in situ sensors can be very helpful for a number of applications: to determine the optimal location and view direction of in situ sensors and to assess local atmosphere and

In relation to the in situ cameras implementation, the DART atmosphere database and its management in the model were improved to offer higher flexibility of DART when dealing with different atmospheric conditions, especially in case of available in situ and/or satellite measurements of atmosphere. This atmosphere database was originally derived from simu‐ lations of the MODTRAN atmosphere radiative transfer model. Its use with DART was already validated with MODTRAN simulations. However, the simulation of aerosols was not as accurate as for gases. The DART atmosphere database was therefore completed using trans‐ mittance spectra for scattering and absorption mechanisms, derived from the atmospheric model MODTRAN. An interesting feature of this improvement is a new possibility to specify gas and aerosol amounts within the urban scene, independently of the atmosphere character‐ istics above the considered environment. This will be of a great help for assessment of local atmospheric properties and pollution impact using in situ sensor acquisitions. This improve‐ ment was accomplished by importing HITRAN [21] line‐by‐line cross section database (with specified temperature and pressure) for thermal infrared spectral domain, as well as the MPI‐

Mainz [22] cross section database for visible and near‐infrared spectral domains.

**Figures 5** and **6** show comparisons of DART and MODTRAN simulations in the visible and near‐infrared and the thermal infrared wavelengths, respectively. Both results demonstrate that the DART update and the introduction of new atmospheric database, combined with an improved radiative transfer modeling approach, brings DART simulations of atmosphere very close to MODTRAN 5.1 simulations, which is very encouraging especially for simulating ac‐

by a camera with a horizontal view direction.

232 Sustainable Urbanization

pollution impact on sensor acquisitions.

*2.2.3. Atmosphere database*

curately the in situ sensors.

**Figure 6.** DART (red) vs. MODTRAN (blue) in the long wavelengths (thermal infrared). Gas model: Tropical, aerosol model: rural, visibility: 23 km. (a) Path radiance calculated at top of atmosphere (TOA) of scattered + emitted fluxes from atmosphere. (b) Direct transmitted radiance from Earth to TOA. (c) Total TOA radiance (*Tground* =299.15 K). (d) TOA brightness temperature (*Tground* =299.15 K).

#### *2.2.4. Decomposition of a sensor image into images per type of scene element*

A common difficulty for analyzing in situ sensor images (i.e., radiance images) is assessment of radiance and area proportion per type of surface material (e.g., wall, roof, atmosphere) inside an image pixel. Knowledge of the different radiance components is valuable for the "iterative calibration" that calibrates DART with remote sensing images as presented in Section 3.1.2. Hence, DART has been improved to facilitate in addition to the original sensor radiance image *LDART,Δλ(xDART, yDART, Ωv)* the per‐pixel simulations of radiance *LDART,Δλ, n(xDART, yDART, Ωv)* and cross section *σn(xDART, yDART, Ωv)* images of each type of scene element n in discreet directions along the sensor viewing direction *Ωv)*.
