**2. Urban aerosols and pollution monitoring**

### **2.1 The atmospheric boundary layer**

The Atmospheric Boundary Layer (ABL) is the lowest section of the troposphere and is directly affected by the surface, responding to surface forcing within a onehour or less time scale. The ABL has turbulent properties and high variability in its daily cycle, and it is a fundamental parameter to several studies, e.g., air quality, numerical weather forecasting, climate modeling, and wind energy applications [16]. These characteristics, associated with the variations in the ABL stability, enable us to subdivide it into three main layers: The Convective Boundary Layer (CBL), the Stable Boundary Layer (SBL), and the Residual Layer (RL).

The ABL height (ABLH) is obtained from the vertical profile of some tracers like a potential temperature [17], vertical wind speed [18], relative humidity [19], and aerosols [11]. The radiosondes are the more traditional method to estimate the variation of some tracers indicated above and, consequently, estimate the ABLH. Nevertheless, in most regions, the radiosondes are launched only twice a day, which does not provide a detailed observation of the ABLH behavior. In this scenario, due to the lidar systems' high temporal and spatial resolution, the utilization of this kind of equipment to estimate the ABLH and other ABL properties had increased significantly in the last decade, mainly in South America [20–30].

Elastic lidar and ceilometers can estimate the ABLH from the characteristic reduction in the aerosol concentrations in the transition region between the Free Troposphere (FT) and ABL. **Figure 3** presents an example of the ABLH and its subdivisions, both estimated from elastic lidar data. Moreira et al. estimated the Urban ABLH to the city of São Paulo (Brazil), using elastic lidar data, from a method based on the curtain-plot of the Range Corrected Signal and Wavelet Covariance Transform (WCT), respectively [31, 32]. Both algorithms were validated by radiosonde data, resulting in high correlations during convective and clear sky conditions. Also, based on WCT, Niesperuza et al. estimated the ABLH to Medellín (Colombia), demonstrating the influence of the selections of the parameters in the Haar Wavelet performance [33]. Salvador et al. performed a comparison among the ABLH estimated from elastic lidar, SODAR, and Weather Research and

**Figure 3.**

*ABL and its subdivisions. WCT and Gradient Methods, applied in elastic lidar data, estimated the CBLH and RLH. The aerosol profiles were measured at SPU Lidar Station on 03 August 2020.*

Forecasting Model (WRF), finding a high positive correlation during the convective period [26]. Besides, elastic lidar data can also observe aerosol plumes' movement as the mixing level in the CBL region, from the skewness and kurtosis profiles. This method was applied in the city of São Paulo [34]. The methodology can be applied in studies about air quality providing a better observation about pollutant concentrations.

In comparison with elastic lidar, Doppler lidar provides more possibilities to identify the ABLH due to the capacity to obtain the wind speed profile, which can be applied as a tracer from its variance. Using Doppler Lidar data, Moreira et al. estimated the ABLH from the variance in the wind speed profile in São Paulo [24]. This method was compared with radiosonde data demonstrating a high correlation in CBL and SBL situations. The wind speed profile was used to detect low-level jets (LLJ). Then from the maximum of LLJ, the SBL height was estimated [25]. Marques et al. used the maximum variance in the Noise Ratio to estimate the ABLH. Such a result was compared with radiosonde data, reaching high correlations in stable and convective situations [28].

### **2.2 Retrievals from the LiDAR-CIBioFi station at Cali-Colombia**

In Colombia (Cali), to detect the ABL altitude, lidar signals obtained from the LiDAR-CIBioFi station at Universidad del Valle are employed. The study site is georeferenced in **Figure 4**. The methodology employs an interplay between the Gradient [32, 33] and WCT [34] Methods, as described in detail in Ref. [27].

Statistical validation of the implemented instrumentation is performed to support the data quality by contrasting atmospheric profiles retrieved by radiosondes launched at the local international airport, a few kilometers away from our station. The maximum vertical gradient level of potential temperature is used to detect the ABL top (ABLT) by employing radiosonde profiles. A linear relationship between the daily ABLT evolution retrieved by the lidar station and the radiosonde profiles goes as follows: ABLLiDAR = 0.967 × ABLRadio - 0.022. It is statistically significant at the 95% confidence level and R2 (consider the separation between the radiosonde launching site and the lidar station, see **Figure 4b**).

Once the data are validated, the ABLT levels are compared against the available local Particulate Matter (PM) concentration information. The correlations between

#### **Figure 4.**

*(a) The location of Valle del Cauca county (the gray area) and the city of Cali (white area) in Colombia (altitude values according to the color scale). (b) The light color shape represents Cali's rural area. The blue region is linked to the city's urban area. The red dots mark the LiDAR-CIBioFi system's location at Universidad del Valle (UniValle) and the radiosonde station (at Cali's international airport). (c) Political and administrative division of Cali. The blue dots place the Air Quality Monitoring System (SVCA) ground stations of the local Administrative Department of Environmental Management (DAGMA).*

daily ABLT evolution and PM concentration data from three representative city ground-stations (Cañaveralejo, Pance, and UniValle), shown in **Figure 4c**, are analyzed. A strong negative relationship for the Cañaveralejo station gives R2 = 0.79, while the Pance station exhibits an unencouraging positive slope with a correlation coefficient R2 = 0.13, meaning a PM concentration increase for higher ABLT values, with PM values above the World Health Organization limits. The UniValle station, located about 100 m away from the LiDAR-CIBioFi station, reveals a low negative correlation (R<sup>2</sup> = 0.20) for the ABLT evolution for all months, especially at the beginning of the wet season.

An innovative method for retrieving the ABL top from LiDAR signals was developed at the LiDAR-CIBioFi station. It consists of training a convolutional neural network (NN) in a supervised manner, driving it to learn how to retrieve this dynamical parameter on real, non-ideal conditions and, in a fully automated and unsupervised process [31]. The Wavelet Covariance Transform (WCT) is used as a labeling method for constructing the training data set and as a baseline method for comparison with the trained NN.

The dataset used for the model's training and tuning is composed of 15,000 signals extracted from daytime measurements taken during December 2018 and February 2019. The signals were labeled using WCT, with a custom search threshold for each one; this was done to ensure the labels' quality and, consequently, the neural network predictions [31].

It is expected that the corrected training of the model replicates the predictions of a signal-by-signal fine-tuned WCT but in a completely automated and non-supervised process. The convolutional neural network proposed for the ABLT detection is compared to WCT in a supervised variant (custom search threshold for

#### *Lidar Observations in South America. Part II - Troposphere DOI: http://dx.doi.org/10.5772/intechopen.95451*

all time evolution) and unsupervised WCT (full signal as input during all time evolution) [31]. WCT was chosen as the labeling and comparison method for its ease of implementation and well-known robustness and performance, as for the past two decades, it has been used to measure the ABLT in numerous case studies [31].

On 14 August (**Figure 5a**), clouds around 4 km height, with some formations around 2 km, were detected, with the latter being very close to the boundary layer's height, thus posing a challenge for accurate ABLT detection. Besides, some cases of changes in density were detected after 14 h [31].

The first single lidar measured profile (**Figure 5b**), taken at about 12:30 h, exhibits a well-mixed layer, making it easy to discriminate between ABL and free troposphere. This condition allows a straightforward evaluation of the predictions: the NN gives very similar results to the supervised WCT (about 1.8 km), while the unsupervised WCT located the ABLT in a cloud formation above 4 km height [31]. In contrast, the second profile (**Figure 5c**), taken at 15:40 h, gives very different results for the supervised (sup.) WCT, the unsupervised WCT, and the NN. The supervised WCT located the ABLT at 500 m, below the expected result.

The unsupervised WCT placed the result at around 4 km, in a cloud formation pattern, while the NN located the ABLT about 2.6 km, following its actual behavior [31]. The temporal evolution of the LiDAR measurement profiles of **Figure 5** clarifies that the unsupervised WCT detection locates the boundary layer position at cloud formation height, erroneously placing the ABLT in most cases [31]. The supervised WCT shows ABLT detection problems, severely underestimating ABLT for cases of proximity to clouds (see, e.g., around 12 h), and in cases of residual layers (e.g., after 14 h). Despite the drawbacks of supervised WCT and unsupervised WCT, **Figure 5** clearly shows that our convolutional NN estimation of ABLT is more resilient to nearby clouds than WCT [31].

#### **Figure 5.**

*Lidar retrievals for 14 August 2019 [30]. (a) Temporal evolution of the ABL, and (b), (c) Two selected single profiles. They point out different scenarios treated with our method: the measurements exhibit b Profile 1 (12:30 h), a well-mixed layer where NN and sup. WCT values are very similar, and c, Profile 2 (15:40 h), shows conditions where the NN estimation differs from supervised and unsupervised WCT values; the latter profile exhibits an extended ABL a height of about 2.6 km for the NN prediction. The intensity of the signals is given in arbitrary units (a. u.) [31].*

The SVCA is the local Air Quality Monitoring System, a governmental policy for continuously monitoring air quality and assessing the pollutants trend. This air quality network provides information in the medium and long term to support developing strategies to attend to the air pollution effects from a holistic view. This network is composed of nine ground stations distributed, as shown in **Figure 4**. There are six stations to measure PM10 concentration and four stations for PM2.5 concentration measurements. These stations count with automatic analyzers technology for monitoring aerosols and gases such as ozone, carbon dioxide, and others. Employing the raw data retrieved by these stations, the local Administrative Department of Environmental Management (DAGMA) makes monthly and annual reports, which can be found [31] at http://www.cali.gov.co/dagma.

Hourly raw data from three ground base stations that belong to the SVCA network were used to analyze the daily behavior of PM10 and PM2.5. Due to its proximity to the LiDAR-CIBioFi station, Cañaveralejo, Pance, and UniValle stations have been picked. Specifically, the UniValle station is about 100 m away from the LiDAR-CIBioFi, and for a direct intercomparison with ABL altitude is a primary source of data. The data retrieved from these ground stations are significant since they are the only available aerosols data source that features Cali to support environmental public policy reinforcement; these are meant to control day-by-day vehicular fleet restrictions and industrial emissions.

One way to assess the impact of PM concentration on public health and radiative forcing is to study its response according to the ABL vertical and horizontal dynamics. We account for PM10 and PM2.5 concentrations retrieved by the automatic analyzers and the vertical atmospheric profiles from LiDAR-CIBioFi. The correlations for the PM as a function of the ABL altitude retrieved by the LiDAR system, the vertical response of PM for each station, according to the ABL daily evolution during July, August, September, and October 2018, are analyzed. The daily data are used as a first step to identify PM daily behavior within the ABL.

The hypothesis to understand the relationship between PM and ABL is that the aerosol's mass volume has an inverse behavior to the ABLT evolution. **Figure 6** shows how daily data from Cañaveralejo station (red dots) already comply with this hypothesis: PM10 concentration decreases with the ABL height; this behavior is a continuous trend for each of the four covered months, regardless of the temperature and solar radiation transition along the day. Especially in the mornings, the Cañaveralejo station reports the maximum concentration amongst the three stations, with values not above the local 24-hour limit.

Regarding Pance (green dots) and UniValle (blue crosses) stations, the regressions exhibit odd behavior, particularly during July and August 2018: **Figure 6a** and **b** show that Pance's PM10 concentration dangerously grows with increasing ABL height, a situation of concern since this means that the population remains exposed to high concentrations within a large air volume mass in the ABL. Since we do not have additional information about the emission source, we postulate that Pance high concentrations could be attributed to tropical forests' haze. Nevertheless, it is necessary to warn on this behavior and spot possible emission sources, and identify such aerosols' optical properties.

The Pance and Cañaveralejo stations were chosen due to their proximity to the LiDAR-CIBioFi station, but their locations correspond to a different landscape, and the ABLT dynamics could be very different from that of UniValle's. Since UniValle station is located about 100 m away from the LiDAR-CIBioFi station (and within the same university campus), it is a direct source for PM data comparison/analysis against ABLT dynamics; it is, however, necessary to keep in mind that UniValle station only makes measurements of fine (PM2.5) particles concentration. Even though UniValle station shows PM2.5 values well below those of PM10 Pance and

*Lidar Observations in South America. Part II - Troposphere DOI: http://dx.doi.org/10.5772/intechopen.95451*

#### **Figure 6.**

*(a) Linear regression between PM (*μ*g m−3) concentration and ABLT (or PBL top-planetary boundary layer top) altitude retrieved by LiDAR-CIBioFi in July 2018. (b), (c), and (d) denote the same as in (a) except for the month of data collected (August, September, and October 2018, respectively).*

Cañaveralejo stations during July, August, and September, during October 2018, Pance and UniValle registered very similar values for PM2.5 and PM10 against the ABL height; this means a high worrying indicator of fine (PM2.5) particles concentration at UniValle. The linear regression between UniValle and Pance stations and the ABL height shows similar behavior. Thus, PM2.5 concentration values are kept almost constant during the whole day and ABL height evolution. These fine particles' size allows them to stay longer in the atmosphere, explaining the regression line trend.

All data for each station in **Figure 6** and corresponding regressions were organized chronologically. Hence, regarding air quality, unquestionably, the most critical report is that for UniValle station due to the size of measured particles and the worrying linear regression balance with the ABLT daily evolution. The reported concentration levels are slightly above the local annual threshold and the World Health Organization (WHO) 24-hour PM limits.

To summarize, the ABLT results obtained at the LiDAR-CIBioFi station were compared against three PM SVCA ground-stations (Pance, Cañaveralejo, and UniValle) to analyze the behavior of the correlation with PM concentration near the LiDAR station. The Cañaveralejo station that carries out PM10 measurements shows an inversely proportional relationship with the ABLT, indicating that the population is not overexposed to PM10 concentration as higher ABLT values are reached. On the other hand, PM10 and PM2.5 concentrations retrieved by Pance and UniValle stations show a different relationship with ABLT. Unexpectedly, the linear correlation slopes for each one of these stations were quite close to zero. The slope's mean value for

Pance station was 1.67, reporting positive values for July and August, which means that PM10 concentrations increased with increasing ABLT, a word of warning due to a possible negative impact on public health. On the other hand, the slope's mean value for UniValle (PM2.5) station was −3.85, with negative monthly mean values.
