Spatial-Temporal Characterization of Atmospheric Aerosols via Airborne Spectral Imaging and Growing Hierarchical Self-Organizing Maps

Neural network analysis based on Growing Hierarchical Self-Organizing Map (GHSOM) is used to examine Spatial-Temporal characteristics in Aerosol Optical Depth (AOD), Ångström Exponent (ÅE) and Precipitation Rate (PR) over selected East African sites from 2000 to 2014. The selected sites of study are Nairobi (1 ̊S, 36 ̊E), Mbita (0 ̊S, 34 ̊E), Mau Forest (0.0 ̊ 0.6 ̊S; 35.1 ̊E 35.7 ̊E), Malindi (2 ̊S, 40 ̊E), Mount Kilimanjaro (3 ̊S, 37 ̊E) and Kampala (0 ̊N, 32.1 ̊E). GHSOM analysis reveals a marked spatial variability in AOD and ÅE that is associated to changing PR, urban heat islands, diffusion, direct emission, hygroscopic growth and their scavenging from the atmosphere specific to each site. Furthermore, spatial variability in AOD, ÅE and PR is distinct since each variable corresponds to a unique level of classification. On the other hand, GHSOM algorithm efficiently discriminated by means of clustering between AOD, ÅE and PR during Long and Short rain spells and dry spell over each variable emphasizing their temporal evolution. The utilization of GHSOM therefore confirms the fact that regional aerosol characteristics are highly variable be it spatially or temporally and as well modulated by PR received over each variable.

Journal of Geoscience and Environment Protection heterogeneous field that makes aerosol characterization real challenge [1] [2].
Despite this, recent initiatives by organizations such as NASA among others have increasingly deployed a number of passive remote sensing platforms that provide systematic and accurate long-term measurements of aerosol characteristics over the globe. This initiative hasn't been reciprocated adequately by the science community since a large percentage of the data actually used is low, in part because of a lack of efficient and effective analysis tools. For example, less than 5% of all remotely sensed images are ever viewed by human eyes or actually used [3]. Therefore, the increasing quantity and type of data available for climate change research studies among them atmospheric aerosols require effective feature extraction methods such as self-organizing map (SOM) and the growing hierarchical self-organizing map (GHSOM). Additionally, accurate extraction of key features and characteristic patterns of variability from a large data set is vital to correctly monitor atmospheric processes and how they alter climate change [4].
Techniques for pattern detection i.e. clustering, classifying and feature extraction for multi-dimensional spectroscopic datasets are becoming increasingly important since the previous is growing in size and complexity. The SOM, an artificial neural network based on unsupervised learning, is an effective software tool of feature extraction [5] [6]. It provides a nonlinear cluster analysis, mapping high dimensional data onto a (usually) 2D output space while preserving the topological relationships of the input data. As a tool for pattern recognition and classification, the SOM analysis is in widespread use across a number of disciplines among the climate research [7] [8] [9].
Notwithstanding its wide applications, SOM analysis has inherent deficiencies.
First, it utilizes static network architecture with respect to the number and arrangement of neural nodes that are predefined prior to the start of training.
Second, hierarchical relations between the input data are difficult to be detected in the map display. These two issues have been addressed within a single framework of the GHSOM that is available [10] [11]. The GHSOM is composed of independent SOMs, which are allowed to grow in size during the training process until a quality criterion regarding data representation is met. This growth process is further continued to form a layered architecture such that hierarchical relations between input data are further detailed at lower layers of the neural network.

Description of Study Area
The East Africa region covers diverse land forms comprising of glaciated mountains, Semi-Arid, Plateau and Coastal regions. Details and the map illustrating the study region and specifics on each site of study are as shown in Figure 1.

Growing Hierarchical Self-Organizing Maps (GHSOM)
The inherent deficiencies of SOM are well addressed by GHSOM through the use an incrementally growing version of the SOM, which does not require the user to directly specify the size of the map beforehand and its enhanced ability to adapt to hierarchical structures in the data as illustrated in Figure 2 [10] [11].
Prior to the training process, a "map" in layer 0 consisting of only one unit is created. This unit's weight vector is initialized as the mean of all input vectors and its mean quantization error (MQE) of unit i is computed as: A new 2D array SOM is always created underneath layer 0 map so as to increase the map size so that all the spectroscopic data is well represented.
where 1 τ is the breadth controlling parameter. The GHSOM array of unit i with the largest i MQE is normally the error unit. On the other hand, the unit with the largest distance with respect to the model vector is selected and a new row or column is inserted between the two. If the inequality in the above equation is not satisfied, then the next decision is whether to expand some units in the next hierarchical level or not. If the data mapped unto one single unit i still has a large variation i.e.

Data
Level-3 MODIS gridded atmosphere monthly global product "MOD08_M3" at spatial resolution of 1˚ × 1˚ [13], was used in the current study for spatio-temporal    [20]. It is of importance to note that the GHSOM algorithm efficiently discriminated, by means of clustering between AOD, during wet and dry seasons over each variable. The dark clusters correspond to long rain spells that are associated with enhanced scavenging of AOD hence, their low values while greyish clusters correspond to a less aerosol scavenging from the atmosphere due to low PR.

GHSOM Analysis of Aerosol Optical Depth
Moreover, the white clusters reveal enhanced AOD values due to inefficient scavenging of atmospheric aerosols via dry deposition over each variable during dry season. The details of the GHSOM classification of AOD over the six variables are shown in Figure 3.
On the other hand, spatial variability in AOD is pronounced since each varia-  Figure 5. Likewise, Mount Kilimanjaro has been reclaimed in the recent past restraining the negative impacts of deforestation hence, explaining the dominance of greyish and dark clusters (low AOD values) during the study period [17].
The dominance of both greyish and white clusters for both AOD (Figure 3) and precipitation rate ( Figure 5) implies high AOD and precipitation rate over Kampala during the study period. The observed high AOD values are associated to the high vehicular emissions from growing private motorized transport over the city [16].  Aerosol particles from high energy use and emissions associated with the growth of private motorized transport over Kampala [16], dominate the 470 nm wavelength, this may explain the higher span range of 1.30 -1.83 ± 0.06 in the ÅE values observed over the site as compared to the rest of the study sites.

GHSOM Analysis of Precipitation Rate
As noted earlier, darker and white colors infers lower and high manifestation PR over each study site. It's clear that Malindi experiences the lowest PR as compared to the rest of the region. Additionally, Mbita-Kampala and Nairobi-Mau Forest Complex PR are correlated during the study period as shown Figure 5.

Conclusion
MODIS Terra monthly AOD and ÅE level 3 data from 2000 to 2014 are used to spatio-temporal characterization of AOD, ÅE and PR over selected study sites of the East African region using GHSOM algorithm. It is possible to use the neural network techniques in studying spatial-temporal characteristics over the region with enhanced efficiency. The GHSOM algorithm classification of both AOD and ÅE is attributed to various factors among them aerosol transport, diffusion, direct emission, hygroscopic growth and their scavenging from the atmosphere.
The East African region experiences diverse and highly variable aerosol characteristics as revealed by GHSOM.