GEOSPATIAL ANALYSIS OF ATMOSPHERIC HAZE EFFECT BY SOURCE AND SINK LANDSCAPE

Based on geospatial analysis model, this paper analyzes the relationship between the landscape patterns of source and sink in urban areas and atmospheric haze pollution. Firstly, the classification result and aerosol optical thickness (AOD) of Wuhan are divided into a number of square grids with the side length of 6 km, and the category level landscape indices (PLAND, PD, COHESION, LPI, FRAC_MN) and AOD of each grid are calculated. Then the source and sink landscapes of atmospheric haze pollution are selected based on the analysis of the correlation between landscape indices and AOD. Next, to make the following analysis more efficient, the indices selected before should be determined through the correlation coefficient between them. Finally, due to the spatial dependency and spatial heterogeneity of the data used in this paper, spatial autoregressive model and geo-weighted regression model are used to analyze atmospheric haze effect by source and sink landscape from the global and local level. The results show that the source landscape of atmospheric haze pollution is the building, and the sink landscapes are shrub and woodland. PLAND, PD and COHESION are suitable for describing the atmospheric haze effect by source and sink landscape. Comparing these models, the fitting effect of SLM, SEM and GWR is significantly better than that of OLS model. The SLM model is superior to the SEM model in this paper. Although the fitting effect of GWR model is more unsuited than that of SLM, the influence degree of influencing factors on atmospheric haze of different geography can be expressed clearer. Through the analysis results of these models, following conclusions can be summarized: Reducing the proportion of source landscape area and increasing the degree of fragmentation could cut down aerosol optical thickness; And distributing the source and sink landscape evenly and interspersedly could effectively reduce aerosol optical thickness which represents atmospheric haze pollution; For Wuhan City, the method of adjusting the built-up area slightly and planning the non-built-up areas reasonably can be taken to reduce atmospheric haze pollution. ∗ Corresponding author The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W7, 2017 ISPRS Geospatial Week 2017, 18–22 September 2017, Wuhan, China This contribution has been peer-reviewed. https://doi.org/10.5194/isprs-archives-XLII-2-W7-1437-2017 | © Authors 2017. CC BY 4.0 License. 1437

The results show that the source landscape of atmospheric haze pollution is the building, and the sink landscapes are shrub and woodland.PLAND, PD and COHESION are suitable for describing the atmospheric haze effect by source and sink landscape.Comparing these models, the fitting effect of SLM, SEM and GWR is significantly better than that of OLS model.The SLM model is superior to the SEM model in this paper.Although the fitting effect of GWR model is more unsuited than that of SLM, the influence degree of influencing factors on atmospheric haze of different geography can be expressed clearer.Through the analysis results of these models, following conclusions can be summarized: Reducing the proportion of source landscape area and increasing the degree of fragmentation could cut down aerosol optical thickness; And distributing the source and sink landscape evenly and interspersedly could effectively reduce aerosol optical thickness which represents atmospheric haze pollution; For Wuhan City, the method of adjusting the built-up area slightly and planning the non-built-up areas reasonably can be taken to reduce atmospheric haze pollution.

Data and materials
Surface cover categories include buildings, woodlands, waters, arable land, shrubs (as shown in Figure 1).1.

The aerosol optical thickness distribution is shown in
The mean value of the pixels in each grid is regarded as the value of the grid's center.Finally, using the correlation coefficients between the indices and AOD identify source and sink landscape.The correlation coefficients are shown in Table 2.

Landscape indices of category level meaning
Percent of landscape(PLAND) Patch density(PD) Fractal dimension(FRAC_MN)

Patch cohesion(COHESION)
Largest patch index(LPI) The ratio of the total area of a patch type to the entire landscape area.
Reflecting the distribution uniformity (fragmentation) of a landscape patch.
A non-integer dimension that describes the complexity of the patch or landscape inlay geometry，The closer to 1 the value is, the more regular of the shape。 Measure the physical connectivity of the corresponding patch type， The greater the value, the better the connectivity The proportion of the largest plaque in a category throughout the landscape There is correlation between landscape indices, which may cause multicollinearity.So the values in table 2 are not very high.Besides, the correlation analysis can only represent that the closely relation between the dependent variable and the independent variable, therefore, this paper introduces the geospatial analysis models.

Establishment and comparison of models
The aerosol optical thickness is treated as the dependent variable, and the landscape indices of each category are the independent variables.Through the spatial autoregressive model and geographic weighted regression model, the relationship between spatial distribution of haze and landscape pattern is discussed from the global and local level, which reflects the influence of source and sink landscape pattern on atmospheric haze pollution.

Spatial autocorrelation analysis of variables
The

Establish models
There is correlation between landscape indices.In Lagrange multiplier statistics should be used to determine which one is more suitable.From the Lagrange multiplier test statistical results ( Comparing the RSS, LIK, AIC and SC (Table 5) of these global models of SEM, SLM and OLS, the SLM and SEM are more applicable than OLS.
Considering all the parameters, the fitting result of SLM model is better than that of SEM for analyzing the relationship between the source and sink landscape and atmospheric haze pollution.
GWR model can explain the haze effect of landscape pattern at the local level, its ability for explain building, shrub and woodland is more superb than that of OLS.

Analysis results at the global and local levels
From the global level analysis, the regression coefficients of SLM model of each category shows that the PLAND and COHESION of building play a role in increasing AOD, and the PD play a negative role.That is, in a certain area the lager the percent and the patch cohesion of building, the higher the haze concentration is.For landscapes of shrub and woodland, the COHESION is proportional to AOD, and the PLAND and PD are the opposite.That is, in a certain area, increasing the percent of sink landscape area, and distributing them more dispersed, the haze will evacuate faster.
From the local level analysis, the spatial distribution discrepancy of variables is described.Figure 4(a) shows that the regression coefficients of PD are negative in the main urban area and are positive in the suburbs area.Although the patch density of building is low, its proportion is very large, so the corresponding values of AOD are high.In suburbs, the building is mixed with other categories, so its patch density increases, at the same time the building area is just a small amount of total area, so the corresponding values of AOD are low.
Fig. 5 (a) and (b) show that the patch density in the main urban area is high, but the percent of the shrub area is very small, so PLAND plays the dominate role in effecting AOD.In the suburbs, the percent of shrub is not so much, so the PD holds the main factor.
For woodland, Figure 6 shows that the effect of woodland on haze is almost the same as that of shrub, but in the main city its distribution is limited.(1) The source landscape of atmospheric haze pollution is the building, and the sink landscape are shrub and woodland.
(2) The SLM model is superior to the SEM model in this paper.Although the fitting effect of GWR model is more imperfect than that of SLM, the influence degree of influencing factors on atmospheric haze of different geography can be expressed clearer.
(3) Reducing the proportion of source landscape area and increasing the degree of fragmentation could cut down aerosol optical thickness; And distributing the source and sink landscape evenly and interspersedly could effectively reduce aerosol optical thickness which represents atmospheric haze pollution.
(4) For Wuhan City, the method of adjusting the built-up area slightly and planning the non-built-up areas reasonably can be taken to reduce atmospheric haze pollution.
the rapid development of urbaneconomy, urban haze occurs frequently.Haze is an atmospheric phenomenon traditionally where dust, smoke and other dry particles obscure the clarity of the sky.For the haze pollution, Landscape ecology pointed out that, the source landscape increases the concentration of haze, and the sink landscape inhibits the haze concentration from ascension and cuts down the air pollution.In the process of urbanization, the unbalanced distribution of source and sink landscape can affect the diffusion rate and concentration distribution of pollutants in the city.And it leads to the change of urban atmospheric composition, which can cause the deterioration of urban atmospheric environment.Reasonable planning of the source and sink landscape pattern in city reduce the haze pollution and restrain the atmospheric environment from negative changes caused by the city construction.At present, most of the atmospheric haze pollution studies focused on the following three aspects: the influence of meteorological factors on the formation and development of haze; the concentration and chemical composition of haze; and the transmission trajectory of persistent haze pollution.While the study on the relationship between source and sink landscape pattern and atmospheric haze in urban is insufficient.Aerosol optical thickness (AOD) can well be used to estimate the haze concentration in the atmosphere.This paper takes the aerosol optical thickness and landscape pattern of Wuhan as the research objects to study the atmospheric haze effect by the source and sink landscape.Due to the spatial dependency and spatial heterogeneity of the data used in this paper, it is necessary to take into account these spatial characteristics when use the analysis model to make the simulation more accurate.Spatial autoregressive model (SAR) is global model which introduce the items that describe the spatial autoregressive based on the regression model, which greatly improves the prediction accuracy.Geographic Weighted Regression Model (GWR) is a local linear regression model, which embeds the spatial position into the regression parameter.That is, it uses the weight matrix to represent the adjacency relation of the spatial units.The regression model can be adjusted with the change of the spatial position of the sampling points.This model can well reflect the non-stationary space of the dependent variable caused by the independent variable.Spatial autoregressive model and geo-weighted regression model are suitable for analyzing the relationship between the least relevant landscape indices and AOD.Through the selection and application of these models, the relationship between spatial distribution of haze and landscape pattern can be discussed from the global and local level, which reflects the influence of source and sink landscape pattern on atmospheric haze pollution.The results of the analysis provide theoretical reference for reasonable planning of urban source and sink landscape pattern and improves urban atmospheric environment.

Fig. 3
Fig.3 shows that normalized heterogeneity indices reached the maximum at 6km except SHEI reached the

Fig. 3
Fig.3 normalized heterogeneity index at different scales order to avoid multicollinearity and improve the fitting accuracy of the model, the least relevant landscape indices are selected based on the correlation coefficients for the following analysis.The correlation between landscape indices of different categories proves that PLAND, PD, and COHESION are well suitable to analyze with AOD for each category.Spatial autoregressive model includes spatial lag model (SLM) and spatial error model (SEM).
Figure 4(b) shows that the PLAND coefficients are almost all positive.And in figure 4(c) the COHESION coefficients are all positive.It is due to the large patch of building in the main urban.
Fig.4 geographical weighted regression analysis results of Building category

Table 4
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W7, 2017 ISPRS Geospatial Week 2017, 18-22 September 2017, Wuhan, China determined, so these two models should be used for regression analysis with AOD.
) of the traditional regression model (ordinary least squares OLS), the superiority of the two models can't be The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W7, 2017 ISPRS Geospatial Week 2017, 18-22 September 2017, Wuhan, China

Table 5
Comparison of Model Parameters