Reply on RC3

Question

1. What are the geographical classifications of the 31 provinces in China?

2. How does RF model handle nonlinear relationships?

3. How to estimate sulfate concentration?

4. How is wet deposition flux estimated using GAM?

Accepted Answer

The 31 provinces in China are geographically classified into six parts: North Central (NC), North East (NE), North West (NW), South East (SE), South West (SW), and the Tibetan Plateau (TP). These classifications represent diverse social-economical and geo-climatic conditions. NC and SE belong to relatively developed regions in eastern China, while NW, SW, and TP are less developed regions. The Qinling Mountain-Huaihe River Line serves as a boundary between the humid south (SE and SW) with more precipitation and the drier north (NC).

Accepted Answer

The Random Forest (RF) model is a state-of-the-art statistical method that deals with complicated nonlinear relationships between the response variable and interpretation variables. It uses ensemble learning, where the RF regression predictions are determined as the average of multiple regression trees based on the bootstrap sampling method. The model performance strongly depends on two crucial parameters, ntree (number of regression trees) and mtry (number of interpretation variables sampled for splitting at each node). Significant correlations of regression trees can be avoided by not including all interpretation variables in the node splitting process. Backward variable selection is also performed to achieve better performance. The RF model was applied to estimate the spatiotemporal pattern of dry deposition for individual N and S species, using 1000 trees and 3 interpretation variables for splitting at each node. The model was trained using the 'caret' package in R software (version 4.1.2; Kuhn, 2021).

Accepted Answer

To estimate sulfate concentration, a simple linear relationship between SO2 and sulfate concentration is used with CTM. The formula GG SSSS 4 2- = GG SSSS 2 x ff(GG CCCCCC-SSSS 4 2-, GG CCCCCC-SSSS 2) is applied, where 2 is the monthly ground-level concentration at CNEMC, and -4 2- and -2 are the sulfate and SO2 concentrations simulated by CTM, respectively. The ratio of simulated sulfate to SO2, denoted as f, is determined based on the significant positive correlation between the two (Luo et al., 2016). This method compensates for the lack of large-scale ground observation data for sulfate concentration.

Accepted Answer

Wet deposition flux is estimated using a nonlinear Generalized Additive Model (GAM) developed in previous work. The model applies the 'link' function to establish the relationship between response variables and interpretation variables. For SO4 2-, NO3-, and NH4+, observed monthly wet or bulk deposition at NNDMN serves as the response variables. Interpretation variables include precipitation, satellite-derived VCDs, PM2.5 concentrations, total column liquid water, temperature, boundary layer height, forest-cover, and urban-cover. The data sources and model performance evaluation are detailed in Zhao et al. (2022). The deposition obtained through GAM is uniformly defined as wet deposition in this study.

Accepted Answer

The RF model performances for dry deposition estimation evaluated with 10-fold cross validation show multi-year average R 2 of N and S species over China above 0.7, with RMSE less than 1 kg N/S ha -1 yr -1. However, the model tends to underestimate high deposition and overestimate low deposition. The modeling prediction performance of OXN (NO3 -, HNO3, and NO2) is better than RDN (NH4 + and NH3) and sulfur (SO2 and SO4 2-). POMINO, which reduced the bias of the default product, is satisfyingly applicable in OXN deposition prediction for China. The multi-year averages of dry deposition for different species estimated in this study are within the range between Q1 and Q3 of selected studies, except for NH3. Overall, the total deposition of N and S from RF modeling is satisfyingly closed to the median level of existing studies.

Accepted Answer

The temporal variability in OXN deposition in China shows a slight increase from 14.7 kg N ha-1 yr-1 in 2005 to 15.7 kg N ha-1 yr-1 in 2012, followed by a decline to 14.5 kg N ha-1 yr-1 in 2020. This variability is influenced by factors such as changes in Chinese agricultural practices, improved waste management in livestock farming, and the replacement of highly volatile ammonium bicarbonate with urea in fertilizer types. Additionally, the reduction in acidic gases like SO2 after 2013 has led to increased gas-phase NH3 concentrations, resulting in more dry NH3 deposition. The growth in precipitation may offset some of the benefits of emission control on wet deposition, but overall, the temporal variability in OXN deposition reflects the impact of various factors on nitrogen deposition in China.

Accepted Answer

The lagging response of deposition to emission abatement in China can be attributed to several factors. Firstly, the timing of emission controls plays a significant role. Europe and the USA started implementing emission controls earlier than the selected periods, resulting in a smaller gap between the changes in emissions and deposition afterwards. In contrast, China started emission controls later, leading to a more prominent lagging response. Additionally, the effectiveness of emission reduction measures and the duration of their implementation can impact the lagging response. Continuous efforts on emission abatement are necessary to achieve substantial reduction in deposition and mitigate ecological risks. Furthermore, the specific characteristics of the region, such as geographical factors and industrial activities, can also influence the lagging response. Overall, a combination of factors including timing, effectiveness, and regional characteristics contribute to the lagging response of deposition to emission abatement in China.

Reply on RC3

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What are the geographical classifications of the 31 provinces in China?

2. How does RF model handle nonlinear relationships?

3. How to estimate sulfate concentration?

4. How is wet deposition flux estimated using GAM?

5. What is the performance of RF model for dry deposition estimation?

6. What is the temporal variability in OXN deposition in China?

7. What factors contribute to the lagging response of deposition to emission abatement in China?

References

Random Forests

Drivers of improved PM2.5 air quality in China from 2013 to 2017.

A global assessment of precipitation chemistry and deposition of sulfur, nitrogen, sea salt, base cations, organic acids, acidity and pH, and phosphorus

Spatial and temporal patterns of nitrogen deposition in China: Synthesis of observational data

Changes of nitrogen deposition in China from 1980 to 2018.

Related Papers (2)

The Needham puzzle, the Weber question, and China's miracle: Long-term performance since the Sung dynasty

An Objective Comment on China-Europe Relations