1. Clear sky Net Surface Radiative Fluxes over Rugged Terrain from Satellite Measurements Tianxing Wang (txwang.rs@gmail.com) Guangjian Yan (gjyan@bnu.edu.cn) Xihan Mu (muxihan@163.com) Ling Chen (chenling8247@126.com) Beijing Normal University

2. Outline of the presentation  Background  Methods  Results and discussion  Summary

3. Background  Net surface radiative fluxes, including both net shortwave (0.3~3  m) and longwave (3~50  m) radiative fluxes, are the driving force for the surface energy balance at the interface between the surface and the atmosphere  Net radiative fluxes are key input parameters for most land models, such as GCM, hydrology and energy models, etc. http://serc.carleton.edu/eslabs/index.html  Thus, currently estimation of net surface fluxes is one of the hottest research issues in the field of global climate change.

4. Limitations Most work focuses on the derivation of downwelling SW and upwelling LW radiation, the effective methods for directly estimating land surface net radiation are highly needed, so that the error propagation can be avoided Almost all current researches ignore the topographic effect over the rugged terrain areas which account for about 2/3 of the global land To date, the available radiative fluxes products (e.g., ISCCP, GEWEX,CERES) from remotely sensed data are spatially too coarse to meet the requirements of land applications

5. Methodologies Fluxes over horizontal surfaces using Artificial Neuron Network Short wave topographic radiation model Short wave topographic radiation model Fluxes over rugged terrain MODIS L1B/MOD03/MOD07/ MOD35/MOD11/MOD04 Fluxes over horizontal surfaces (SW & LW) Topographic modeling Fluxes over rugged terrain (SW & LW) Longwave topographic radiation model Longwave topographic radiation model Correcting terrain shading

6. Reasons for using ANN  The ANN model can accept more input variables and output more desired quantities  It has been attempted by many researches in the field of radiative flux budget proving its feasibility in such topic  It’s convenient to couple the multi-output of ANN with the topographic model for retrieving fluxes over rugged area

7. Topology and training  over 50,000 samples simulated using MODTRAN4  single-hidden layer feed-forward network  BP training algorithm Shortwave ANN model Longwave ANN model Inputs of ANNOutputs of ANN Altitude LW downward radiative flux Surface emitted radiative flux Net LW radiative flux Viewing zenith angle 3 water vapor indices MODIS radiances of band 20,22,23,27~29 and 31~33 Temperature profiles Moisture profiles

8. Validation of the ANN models  Two years of 2008~2009 in situ data are collected as reference from seven U.S. SurfRad sites under clear sky Validation results of the ANN models SurfRad sites From :http://www.srrb.noaa.gov/surfrad/ The maximum root mean square errors (RMSE) of ANN models are less than 45W/m 2 (watts per square meter) and 25W/m 2 for net SW and LW radiative fluxes, with the average biases are less than 15 W/m 2 and 5W/m 2, respectively. These accuracies are better than the existing algorithms showing the effectiveness of the ANN models.

9. Topographic radiation models  Terrain shading (a), resulting in :  Flux contribution from the around sloped terrain (b)  Sky-view ratio (c) Key factors affecting radiative fluxes over rugged terrain  zero solar direct radiation  lower LST due to shadows Dubayah, R. and S. Loechel (1997) Shadow Terrain means the overlaying sphere may be obstructed by terrain, in this situation the sky-view-ratio is less than 1

10. Topographic radiation models (SW) Solar direct flux: Sky diffused flux: Reflected from around terrain: Net flux over rugged terrain:

11. Topographic radiation models (LW) Surface emitted flux: Sky emitted flux: Emitted from around terrain: Net flux over rugged terrain: Similarly, by considering the three factors, a LW topographic radiation model is also suggested 。 This model is more complex compared to the SW model, since it relate to LST in shadow area and the broadband emissivity etc.

12. Correction for terrain shading Seven shading situations It should be noted that, all inputs in the SW and LW topographic radiation models are the unobstructed fluxes, thus, the terrain shading of the outputs of ANN models need to be removed before incorporated in these models. This figure shows the seven shading situations ， including A B C D

13. Correction for terrain shading Correction for terrain shading in Shortwave band: Correction for terrain shading in Longwave band: These are the correcting formulas for those seven situations, the details of these variables can be found in the paper.

14. Results and discussion MODIS data collected on November 4, 2009 over Tibet Plateau, a typical region for terrain undulation and climate change research, are selected as our case study Net SW surface radiative fluxes for considering (left) and neglecting (right) the topographic effect the terrain texture of the topographically corrected map is rather obvious the variation of the fluxes due to the terrain undulation is much wider than that of fluxes neglecting the topographic effect fluxes estimated by assuming a horizontal surface are difficult to reflect the true status of surface radiation budget in terms of both spatial distribution and specific flux values

15. Results and discussion Net LW surface radiative fluxes for considering (left) and neglecting (right) the topographic effect although the terrain texture of the topographically corrected map is not as obvious as that of SW, they can also be visually felled the variation of the fluxes due to the terrain undulation is much wider than that of fluxes neglecting the topographic effect The white spots in the left image correspond to the shadowed area, where the LST is low, thus the net LW fluxes are relatively high

16. Total net surface radiative fluxes for considering (left) and neglecting (right) the topographic effect Results and discussion Because the net SW fluxes poses a larger magnitude of total fluxes, thus the spatial distribution of the total net fluxes are highly in line with the distribution of net SW fluxes it will attribute great errors to the estimated fluxes if the terrain undulation effect is not taken into account over rugged terrain area. The errors can reach up to 100 W/m 2, and even larger for SW fluxes.

17. Summary  Two ANN models have been developed in this paper, with which the net surface SW and LW radiative fluxes over horizontal surface can be directly retrieved with better accuracy  Coupling the outputs of ANN models, the corresponding SW and LW topographic radiation models are also proposed  The results show that the ANN models suggested here are rather effective, and the topographic effect on the net surface fluxes is so significant that the assumption of horizontal surface is not applicable over rugged terrain