High-Frequency Estimation and Prediction of Carbon Emissions in Chinese Municipalities: A Case Study of 14 Municipalities in Guangxi Province

Abstract

In October 2024, the National Development and Reform Commission (NDRC) and other departments released the “Work Plan for Improving the Carbon Emission Statistics and Accounting System”, which explicitly proposed the promotion of municipal-level energy balance tables and the development of carbon emission prediction and early warning models. Currently, China has not yet released municipal-level energy balance tables, making it impossible to directly estimate municipal carbon emissions using the IPCC inventory-based method. This paper draws on the electricity–energy–carbon model at the industry level and conducts high-frequency carbon emission estimation for 14 municipalities in Guangxi as a case study. Based on this, the Prophet model is introduced, incorporating planned electricity consumption data to construct a carbon emission prediction and early warning model, enabling long-term carbon emission forecasting at the municipal level. The results indicate the following: First, among the 14 municipalities in Guangxi, Baise has the highest share of carbon emissions (27%), followed by Liuzhou (13%). In terms of carbon emission intensity, six municipalities exceed the regional average, including Baise, Chongzuo, and Fangchenggang. Second, the total carbon emissions in Guangxi (from energy consumption) are expected to peak by 2030, and all 14 municipalities are expected to achieve peak carbon emissions from energy consumption before 2030.

1. Introduction

In September 2020, China officially announced at the United Nations General Assembly that “China will enhance its national contribution, adopt more robust policies and measures, aim to peak carbon dioxide emissions before 2030, and strive to achieve carbon neutrality by 2060”. Since the announcement of the dual carbon strategy, the Chinese government has increasingly refined its related policies. In August 2024, the State Council General Office of China issued the “Work Plan for Accelerating the Establishment of the Carbon Emission Dual Control System”, which explicitly proposed incorporating carbon emission targets and related requirements into national planning and establishing a sound local carbon assessment mechanism. This marks the official shift in local government target assessments from controlling energy intensity to implementing dual control over total carbon emissions and carbon intensity.

In October 2024, the National Development and Reform Commission (NDRC) and other departments released the “Work Plan for Improving the Carbon Emission Statistics and Accounting System”, which explicitly proposed promoting the development of municipal-level energy balance tables and establishing carbon emission prediction and early warning models. At present, only provincial-level energy balance sheets are released in China. The lack of energy balance sheets at the prefectural level makes it difficult to calculate carbon emissions at the prefectural level, and the calculation of high-frequency carbon emissions at the municipal level is even more challenging.

Regarding carbon emission estimation, seven developed countries have established authoritative carbon emission databases that cover countries around the world (Shi et al., 2023 [1]). While these organizations’ historical carbon emission data for China can provide valuable trends (e.g., the results from CDIAC and EDGAR), they generally tend to overestimate China’s carbon emissions to some extent [2], and they lack the estimation data for carbon emissions at the municipal level. Domestically, in 2022, the NDRC’s Environmental and Resource Management Division organized the State Grid Corporation of China and other units to conduct a carbon emission monitoring study for various provinces based on big data of electricity consumption. In 2024, the Energy Division of NBS (National Bureau of Statistics) conducted a study on the classification and estimation of carbon emissions from industrial production processes across the country. Although the above work by the NBS and NDRC did not directly conduct high-frequency calculations of carbon emissions at the municipal level, it provides a foundation for high-frequency calculations of carbon emissions at the municipal level in the future.

Theoretically, the main methods for estimating carbon emissions at home and abroad include inventory-based methods, direct measurement, material balance methods, and modeling methods [3]. Among these, the inventory-based methods are the most commonly used, typically applied at the macro level for estimating carbon emissions in regions or industries. This paper builds upon this method and draws from Wang et al. (2024) to construct a municipal electricity–energy–carbon (E-E-C) model [4], based on municipal electricity consumption data, and applies it to high-frequency carbon emission estimation for the 14 municipalities in Guangxi. Additionally, to achieve carbon emission prediction and early warning for municipalities, we introduce the Prophet model, using the annual planned electricity consumption data for Guangxi, to predict the long-term carbon emissions from 2024 to 2030 for the 14 municipalities. The difference between our study and the existing research is that we apply the latest electricity energy carbon model to realize the high-frequency measurement at the municipal level, and based on this measurement and power planning data, we select the effective prediction method to achieve the accurate prediction of carbon emissions.

The main contributions of this paper are as follows: First, it introduces the industry-level “E-E-C” model to the municipal level, enabling high-frequency carbon emission estimation for municipalities. Second, it combines annual planned electricity consumption data with the E-E-C model and the Prophet model to predict and provide early warnings for municipal carbon emissions.

There are two objectives of this article: The first is to apply the latest electricity–carbon model to calculate carbon emissions at the municipal level. The second is to build a carbon emission prediction and warning model to predict carbon emissions and address the problem of “researching and establishing carbon emission prediction and warning models” required by the NDRC in October 2024. The main contributions of this article are as follows: Firstly, the improved industry-level “E-E-C” model proposed by Wang et al. (2024) [4] was introduced to obtain high-frequency measurements of carbon emissions at the municipal level. Secondly, the annual planned electricity consumption data were combined with the electricity energy carbon model and Prophet model to predict and provide early warnings for local carbon emissions.

2. Literature Review

In terms of carbon emission estimation methods, both domestic and international studies primarily use four main approaches—inventory-based methods, direct measurement methods, material balance methods, and modeling methods [3]. Among these, the inventory-based method is the most commonly used [1]. Authoritative international organizations, such as the IEA, CDIAC, EDGAR, and EIA, despite using different specific calculation methods for carbon emission estimation, generally adopt the inventory-based method based on IPCC calculation formulas [5].

Regarding the scope of estimation, the boundaries of carbon emissions in China are generally set to only include emissions resulting from fossil fuel combustion and cement production [5]. As research on carbon emission estimation has progressed, international organizations have gradually refined the estimation methods for emissions from industrial production processes. The 2021 “Global Energy Review” proposed a carbon emission scope that included emissions from energy consumption and from industrial processes like cement, steel, and other chemical products. The IEA calculates carbon emissions and greenhouse gas emissions related to energy combustion based on this standard.

In terms of estimation frequency, the existing literature mainly focuses on low-frequency (annual or quarterly) estimations, with fewer studies on high-frequency (monthly or daily) estimations. Due to the limitations of obtaining high-frequency data, most carbon emission estimation studies rely on low-frequency data, such as the annual carbon emission data for global countries published by EDGAR (1970–2018), and the annual carbon emission data for China and its provinces from CEADs (a third-party platform for carbon emission accounting supported by organizations such as China’s National Natural Science Foundation, Ministry of Science and Technology, Chinese Academy of Sciences, and the UK Research Councils). Other scholars, such as Dong Huijuan (2011), Zhao Hongyan (2012), and Shan et al. (2016), have also published annual carbon emission estimates [6,7,8]. Some scholars have conducted quarterly carbon emission estimations, such as Schulz et al. (2010) [9]. High-frequency estimations available include CEADs’ daily carbon emission data for China. Shi et al. (2023) [10] used big data from the electricity sector to construct an electricity–carbon model and conducted high-frequency carbon emission estimation for the Qinghai province. Wang et al. (2024) used big data on electricity consumption to build an electricity–energy–carbon model and conducted high-frequency carbon emission estimations for key industries in the Henan province [4].

In terms of the results of carbon emission estimations, the existing studies still contain some uncertainties. These uncertainties mainly manifest in the following four aspects: First, there is inconsistency in the emission inventory standards. For example, the standards of institutions such as IPCC, the U.S. Energy Agency, the Japan Energy Research Institute, the Chinese National Climate Change Program of the National Science and Technology Commission, and China’s National Development and Reform Commission (NDRC) all differ to some extent [11]. Second, there are discrepancies between the estimation results of authoritative institutions. For example, the results of the seven international authoritative institutions compared with China’s submissions in the “National Communications” overestimated emissions in 19 out of 22 years, with a maximum overestimation of 7%. In the comparison between the “Third National Communication” and the results from the Chinese Academy of Sciences, the difference was 19.3% and 12.3%, respectively [2]. Third, there are certain errors in the macro statistical data used for carbon emission estimation in China. For instance, there are significant differences in the historical energy consumption statistics at the national and provincial levels, which, despite adjustments in 2015, persist [2]. Fourth, there are also large discrepancies in the estimation results for different years from the same institution. For instance, the carbon emission results for China published by CEADs in 2021 and 2022 for 2017 differ by nearly 20%, with the 2021 data underestimating emissions by about 9% and the 2022 data overestimating them by about 10%.

In practical applications, to implement the national “dual carbon” strategy and reflect the social responsibility of state-owned enterprises, power companies in provinces such as Guangdong, Fujian, Zhejiang, Qinghai, and Jiangsu have actively conducted high-frequency carbon emission estimation studies based on big data on electricity consumption. On 24 May 2022, China News reported under the title “China’s First ’Electricity High-Frequency Data Carbon Emission’ Smart Monitoring and Analysis Platform Officially Launched” that the State Grid Corporation’s Qinghai branch had conducted high-frequency carbon emission estimation based on big data on electricity consumption. In June 2022, the NDRC commissioned a joint research group composed of the State Grid Corporation, Beijing University of Posts and Telecommunications, and seven other units to study “Carbon Emission Monitoring Research Using Big data on electricity consumption”. In June 2023, the NDRC organized a review meeting for the “National Carbon Emission Monitoring and Analysis Service Platform”, which can calculate, monitor, and analyze carbon emissions across regions and industries by month. However, there is still no research on high-frequency carbon emission estimation and prediction/early warning at the municipal level.

At the methodological level, emerging studies have significantly advanced high-frequency carbon emission estimation through machine learning integration. Some researchers employed a hybrid MODWT-SVR-DE model with noise reduction techniques to achieve daily-scale CO₂ prediction in Thailand, demonstrating the viability of high-resolution forecasting for policy-sensitive interventions [12]. This aligns with the theoretical framework proposed in the theory-guided deep neural network for emission prediction [13], which enhances the temporal resolution through physics-informed machine learning architectures. At municipal levels, the recent case studies have adopted spatially explicit approaches, utilizing interpretable machine learning coupled with multi-scale land use data to resolve city-level emission heterogeneity [14,15], while provincial analyses implemented STIRPAT-optimized models for subnational carbon peaking forecasts [16,17]. These methodologies address the critical gap in localized emission accounting highlighted in global carbon inventories.

Some contemporary reviews also systematically categorize machine learning advancements in this domain [15], particularly noting the rising adoption of partitioned LSTM architectures that account for urban hierarchy effects through scenario-specific neural network ensembles. The integration of urban system dynamics with explainable AI frameworks represents a paradigm shift from conventional top–down accounting to adaptive municipal-level prediction systems [18].

In the existing research, although Shi et al. (2023) conducted high-frequency measurements of carbon emissions at the municipal level based on an electric carbon model [10], their model had shortcomings such as not considering nonlinear models and overly single model selection criteria. Besides these shortcomings, they did not conduct predictive research on carbon emissions. Wang et al. (2024) improved the shortcomings of the electricity–carbon model by Shi et al. (2023) and constructed an industry level electric energy carbon model [4,10], but they did not conduct carbon emission measurement at the municipal level or conduct carbon emission prediction research. This article will draw on the electricity energy carbon model by Wang et al. (2024) at the model level and conduct pilot research on the carbon emissions of Chinese municipal cities at the application level [4]. At the same time, carbon emission prediction research will be carried out based on the carbon emission measurement at the municipal level, effectively solving the problem of “researching and establishing carbon emission prediction and warning models” required by the NDRC in October 2024.

3. Data Description

The data for Guangxi and its municipalities include the annual regional GDP data from 2010 to 2023, as well as the annual energy consumption data for various types of energy in Guangxi from 2010 to 2023. The data sources are from the Guangxi Statistical Yearbook for each year and the official websites of the municipal statistical bureaus.

The following five points need to be clarified regarding the above foundational data: First, due to the lack of energy balance sheets at the municipal level, it is difficult to obtain accurate annual energy consumption data for municipalities in China. Second, in addition to collecting the relevant data published on the official websites of the statistical bureaus in various cities, this article also collected historical energy consumption intensity assessment indicator data from local governments. From the perspective of collecting data through public channels, it is relatively comprehensive. Specific indicator data include energy intensity data, municipal GDP data, provincial GDP data, provincial energy intensity, and total energy consumption data. Third, some years and continuous years of data for the municipalities are missing. To address this, we employed regression imputation and geometric mean imputation methods to fill in the missing data. Fourth, the energy consumption data for each municipality were derived by integrating the energy intensity data published by the municipalities, GDP data at the municipal and provincial levels, as well as energy intensity at the provincial level and total energy consumption data at the provincial level). Fifth, when summing the municipal data and comparing it with the provincial totals, certain discrepancies exist. These discrepancies can have a direct impact on the final carbon emission estimation results. Therefore, we use the provincial total values as a reference to adjust the relevant indicators for each municipality.

The big data on electricity consumption for Guangxi and its municipalities, as well as the annual planned electricity consumption data for each municipality, are provided by Guangxi Power. All electricity consumption data covers the period from January 2018 to March 2024. The annual planned electricity consumption data corresponds to the planned figures for the years 2024–2030.

4. Models and Methods

4.1. Construction and Implementation Steps of the Municipal Electricity–Energy–Carbon Model

Drawing from the work of Wang et al. (2024) on the industry-based electricity–energy–carbon model [4], the municipal electricity–energy–carbon model we constructed consists of three main model groups, namely the data preprocessing model group, the data testing model group, and the core model group.

The data preprocessing model group includes interpolation models such as linear interpolation and geometric mean interpolation, frequency conversion models such as the Litterman model and cubic interpolation model, and series decomposition models like the X-12-ARIMA and TRAMO-SEATS models. R version 4.3.1 was used for data analysis.

The data testing model group includes unit root tests such as the ADF test model and the PP test model, cointegration tests such as the EG two-step test model and JJ test model, and causality tests like the Granger Causality test model. The core model group includes the ADL(p,q) model, long-term equilibrium model, ADL-TC(p,q) model for TC series, and the nonlinear Copula model.

The specific implementation steps for the municipal electricity–energy–carbon model are as follows:

Step 1: Estimate the annual total energy consumption for each municipality (E1). Let E represent the total energy consumption for the entire province, let G represent the annual value of the provincial gross value added, and let G1 represent the annual GDP of each municipality. Thus, the total energy consumption is $E1=(E/\mathrm{G})\times \mathrm{G}1$.

Step 2: Frequency conversion and series decomposition. Using frequency conversion models, the annual energy consumption series is converted into a monthly series. Simultaneously, the X-12-ARIMA model is used to decompose the original energy consumption and electricity consumption series into four components—the TCI, TC, S, and I series.

Step 3: Model selection. After interpolating the annual data, performing frequency conversion, seasonal adjustment, and conducting unit root, cointegration, and causality tests, we estimate and select the monthly electricity–carbon model for each municipality. Assume the selected model is as follows:

$$B(L)(\ln y_t)={\alpha }_0+A(L)(\ln x_t)+{\epsilon }_t,B(L)=1-{\beta }_{1}L-\mathrm{\dots }-{\beta }_p{L}^p,A(L)=1+{\alpha }_{1}L+\mathrm{\dots }+{\alpha }_q{L}^q.$$

where L represents the lag operator, y is the monthly energy consumption, and x is the monthly electricity consumption. Further, assuming p = 2 and q = 2, we can define the model accordingly.

Step 4: Determining model selection criteria. Given that the annual statistical data released by the statistical bureau are usually available in October of the following year, we adopt the criterion of controlling generalized prediction error, i.e., maximizing the minimum generalized error max (min(E), i), where E represents the generalized error, for optimal model selection.

Step 5: Using the carbon emission coefficient for standard coal (h), we calculate the monthly carbon emissions Ct for each municipality based on the list preparation method.

4.2. Construction and Implementation Steps of the Carbon Emission Prediction and Early Warning Model

There are typically two approaches for constructing a carbon emission prediction and early warning model:

Directly based on time series models, where high-frequency carbon emission measurement results are used to extrapolate the trend and predict carbon emissions.

Based on the electricity–energy–carbon model, this approach incorporates planned electricity consumption data, selects an appropriate prediction model for forecasting the monthly electricity consumption, and then uses the predicted electricity consumption data in the electricity–energy–carbon model to obtain the carbon emission prediction.

Considering that the planning department of Guangxi Power has professional expertise in the future electricity plans for each municipality, we chose the second approach to construct the carbon emission prediction and early warning model.

We compared the MIDAS (mixed data sampling) model, ADL model, machine learning models (such as K-nearest neighbor regression, random forest models, AdaBoost regression models, etc.), and the Prophet model. We ultimately chose the Prophet model to build a carbon emission prediction and warning model for the following three main reasons: Firstly, through testing and comparison, we found that the overall accuracy of the Prophet model’s predictions is higher than other models. Secondly, compared with other models, the Prophet model makes it relatively easy to implement batch prediction, making it convenient for the simultaneous prediction of multiple municipalities in practical applications. Thirdly, compared with other models, the Prophet model has good seasonal decomposition ability, which enables it to predict the results for the next 12 months simultaneously, making it easier to achieve long-term forecasting.

The basic idea of the Prophet model is to decompose the time series Y into four components as follows: trend (g), seasonality (s), holidays (h), and residual error. The general expression is the following:

$$y\left(t\right)=g\left(t\right)+s\left(t\right)+h\left(t\right)+\epsilon \left(t\right)$$

where g(t) represents the trend component, capturing non-periodic variations in the time series. s(t) represents the seasonality component, typically measured in weeks or years. h(t) represents the holiday component, accounting for whether a given day is a holiday. ϵ(t) represents the error term, also known as the residual.

The steps for implementing the carbon emission prediction and early warning model are as follow:

Step 1: Use historical monthly electricity consumption data and apply the Prophet model to predict the monthly electricity consumption for future years.

Step 2: Use the future planned electricity consumption data for each municipality to adjust the monthly prediction results from the Prophet model, obtaining the revised monthly electricity consumption predictions.

Step 3: Based on each municipality’s electricity–energy–carbon model, substitute the adjusted monthly predicted electricity consumption data into the model to obtain long-term monthly carbon emission predictions for future years.

Regarding the discussion of the two models mentioned above, the following explanations are given here. Firstly, the prediction of carbon emissions in various municipalities should be combined with the prediction of planned electricity consumption in each municipality, and the annual prediction of planned electricity consumption is given by Guangxi Power Grid Co., Ltd. based on the future economic development of various municipalities in Guangxi, which has taken into account changes in technology, policies and other factors. Secondly, the long-term forecast in this article is a long-term monthly forecast, so it is necessary to provide monthly forecast results for electricity consumption based on the annual forecast results of planned electricity consumption, and ultimately combine them with the E-E-C model to achieve high-frequency long-term forecasting of carbon emissions. And some important external shocks have already been included in the annual forecast data of planned electricity consumption, so they can be ignored in the monthly high-frequency forecast of this article. Thirdly, for the E-E-C model, its basic idea is to build the model to depict the long-term relationship between electricity data and energy consumption data, in order to achieve high-frequency measurement of energy consumption data based on high-frequency electricity data, and ultimately achieve high-frequency measurement of carbon emissions. Due to the modeling of long-term relationships in this model, short-term external shocks are uniformly treated as the random disturbance term of the model.

5. High-Frequency Carbon Emission Estimation and Prediction for Municipalities

5.1. High-Frequency Carbon Emission Estimation in Municipal Energy Use

Based on the design methodology and computational steps of the electricity–energy–carbon model, we modeled the electricity consumption data using the total energy consumption data for each municipality in Guangxi. The final model was selected from 14 municipalities. Using these optimized models and the standard coal-related carbon emission factor of 2.66 tCO₂ per ton of standard coal, published by the National Development and Reform Commission, we calculated the monthly carbon emission data for each municipality, as shown in

Table A1. Results of high-frequency measurement of carbon emissions from energy consumptions in 14 municipalities in Guangxi based on the electricity–energy–carbon model (unit: 10,000 tons of CO₂).

Time	Nanning (Y109)	Liuzhou (Y110)	Guilin (Y111)	Wuzhou (Y112)	Beihai (Y113)	Fangchenggang (Y114)	Qinzhou (Y115)	Guigang (Y116)	Yulin (Y117)	Baise (Y118)	Hezhou (Y119)	Hechi (Y120)	Laibin (Y121)	Chongzuo (Y122)
2019M01	170.21	411.48	94.52	37.37	135.42	335.84	200.19	197.30	91.29	458.31	73.73	57.88	114.57	132.70
2019M02	116.05	336.65	72.14	26.14	103.90	239.14	148.67	149.47	67.48	411.04	39.90	48.81	91.74	104.19
2019M03	145.86	364.82	215.18	36.67	124.02	258.14	158.42	176.26	79.16	387.09	65.79	62.03	97.89	149.50
2019M04	162.76	357.08	72.75	40.93	144.31	238.56	176.39	194.16	89.52	439.70	82.47	60.65	98.95	86.10
2019M05	168.11	374.40	73.78	43.78	149.55	230.61	185.98	212.29	92.29	534.76	76.45	59.37	114.68	112.50
2019M06	197.20	397.48	80.59	49.55	168.78	236.63	198.24	237.78	104.38	414.91	89.64	64.55	120.18	125.11
2019M07	206.76	435.48	86.16	50.92	176.22	216.35	205.96	248.93	111.46	506.46	84.85	71.09	118.89	124.97
2019M08	219.43	472.25	105.88	53.29	171.50	216.05	220.84	261.80	113.29	544.61	90.26	78.69	131.02	126.58
2019M09	195.83	431.08	91.79	47.79	162.02	187.01	214.19	256.23	106.62	573.58	79.60	65.62	124.05	114.12
2019M10	171.44	405.10	82.38	43.02	159.87	182.15	219.70	231.47	99.05	581.08	82.43	73.11	123.92	114.81
2019M11	161.47	396.77	80.03	43.23	142.52	173.42	217.00	208.08	92.48	625.10	77.61	71.63	128.15	123.32
2019M12	175.22	441.03	100.10	42.44	148.25	196.08	239.25	219.04	99.68	653.46	97.15	70.39	202.59	138.91
2020M01	133.92	354.63	91.94	31.23	111.14	176.05	225.50	182.88	69.16	612.10	51.57	72.86	117.79	122.11
2020M02	95.93	267.18	58.74	15.44	88.58	134.10	173.29	113.55	77.87	600.11	37.37	52.86	98.09	92.04
2020M03	125.51	331.86	71.77	28.83	107.72	160.81	204.77	156.53	73.85	643.09	54.51	57.01	103.85	89.31
2020M04	131.83	339.91	71.95	38.88	109.18	172.13	222.10	182.07	80.13	654.60	84.60	55.39	113.26	103.51
2020M05	180.63	394.15	90.21	40.73	142.80	231.89	298.61	217.55	88.63	709.69	83.93	61.49	115.81	112.33
2020M06	189.62	395.26	94.09	48.13	154.28	245.13	285.96	212.31	120.05	664.80	84.18	55.62	115.11	129.75
2020M07	214.45	440.01	108.56	50.29	157.68	276.43	320.39	240.76	111.20	737.84	95.51	71.77	125.18	139.08
2020M08	192.24	430.29	111.64	54.71	137.41	273.89	298.06	230.65	108.61	758.14	93.89	68.80	132.10	133.87
2020M09	177.05	393.82	106.20	53.75	166.56	258.08	261.82	212.06	99.13	746.47	89.96	64.75	119.88	124.26
2020M10	139.90	349.79	110.59	57.84	132.33	233.49	231.63	187.79	87.83	806.35	78.01	61.25	116.00	109.48
2020M11	149.27	398.40	102.39	48.04	143.42	248.90	240.79	197.66	87.80	771.16	89.20	65.29	132.61	145.06
2020M12	182.70	469.71	78.16	58.47	158.47	302.44	280.88	233.16	98.09	912.23	102.04	80.27	160.04	119.64
2021M01	166.38	418.99	118.37	54.67	142.16	260.82	253.97	209.28	99.71	799.76	98.10	79.10	146.46	156.94
2021M02	97.90	287.69	72.89	34.14	110.80	189.90	175.20	122.18	77.19	725.33	53.84	57.67	115.50	121.00
2021M03	160.59	353.52	90.93	47.21	154.38	263.89	233.53	221.28	88.82	795.91	88.75	70.14	154.66	116.06
2021M04	140.48	341.97	81.52	52.47	129.57	247.97	194.94	190.78	86.64	808.46	95.25	66.56	125.30	121.68
2021M05	184.75	387.95	88.33	59.07	208.62	289.30	220.47	238.65	103.09	834.02	99.80	69.90	150.43	127.04
2021M06	196.91	414.22	100.24	66.75	136.34	283.53	233.84	247.71	118.76	877.77	107.47	76.86	152.97	141.57
2021M07	222.32	459.16	119.26	60.62	190.41	289.21	238.70	279.00	129.90	903.15	107.03	78.89	157.72	150.15
2021M08	208.78	438.01	106.97	69.65	167.54	267.44	218.91	255.82	125.94	900.06	107.21	77.62	144.82	145.04
2021M09	198.36	441.33	104.81	66.69	173.76	219.92	191.60	245.97	116.17	813.90	104.38	76.37	135.71	138.79
2021M10	161.18	385.50	86.01	68.37	155.07	234.17	174.25	215.32	106.72	784.61	102.13	72.63	125.15	121.02
2021M11	138.92	380.32	83.48	58.63	153.07	232.25	167.88	206.78	98.74	780.07	97.91	68.44	112.16	123.18
2021M12	165.19	433.65	102.62	64.18	179.03	239.42	188.20	225.07	116.16	746.99	109.15	84.43	134.32	167.82
2022M01	155.01	380.41	90.06	48.85	168.38	215.49	179.09	202.94	98.94	805.07	73.02	79.44	120.28	143.76
2022M02	144.48	370.74	86.67	52.00	158.68	190.28	180.89	177.66	101.35	664.87	66.18	78.05	110.53	117.96
2022M03	151.30	375.35	83.78	56.29	167.01	242.42	181.57	204.49	98.56	701.80	89.73	78.19	110.14	114.76
2022M04	152.89	345.21	80.76	65.61	168.68	238.52	187.00	203.30	96.63	741.42	85.80	70.47	109.99	105.65
2022M05	158.53	359.57	78.27	60.00	176.28	252.18	194.90	199.92	105.86	749.55	74.77	72.74	105.52	122.14
2022M06	198.53	394.58	121.00	90.85	198.30	269.48	230.77	210.06	156.09	810.05	82.25	74.96	108.30	211.84
2022M07	230.60	444.49	112.42	65.94	200.97	269.53	266.47	259.63	138.38	841.86	83.80	84.50	122.36	152.62
2022M08	230.94	439.36	119.80	64.91	188.94	264.22	276.36	268.11	130.50	839.66	104.71	83.51	116.62	125.30
2022M09	216.17	428.79	113.18	71.22	203.63	262.48	271.74	255.50	123.89	876.31	134.09	99.75	114.82	128.08
2022M10	176.28	369.96	99.66	61.00	195.35	228.22	234.19	223.15	113.21	768.37	87.27	75.85	105.41	127.64
2022M11	154.20	325.62	84.71	50.70	172.79	195.90	225.11	198.74	95.05	847.94	88.68	65.81	99.54	117.32
2022M12	169.98	363.43	95.61	50.36	177.12	239.17	246.78	211.52	86.53	865.87	71.02	73.69	119.61	144.05
2023M01	147.70	308.30	90.59	32.44	166.05	230.61	226.32	175.69	95.16	890.01	59.91	68.12	108.61	150.19
2023M02	150.94	325.79	90.12	41.20	163.03	210.74	211.66	193.75	88.87	864.33	85.40	66.23	108.31	157.36
2023M03	174.65	347.78	93.75	53.76	180.92	250.36	229.88	236.11	105.94	940.42	109.45	68.30	127.04	163.00
2023M04	168.11	315.06	80.66	50.01	169.77	238.50	224.47	216.26	103.03	981.94	98.34	63.22	110.90	148.02
2023M05	207.26	340.54	87.82	54.79	197.20	259.42	240.64	229.69	121.41	995.50	90.04	66.46	123.62	159.85
2023M06	237.40	382.02	99.23	56.15	207.42	273.10	246.15	266.92	137.24	1027.38	106.23	68.23	122.95	162.07
2023M07	277.87	404.68	121.98	62.06	240.24	317.49	263.97	321.89	152.25	1101.91	115.96	81.97	143.25	170.44
2023M08	240.05	398.14	111.18	60.74	219.23	289.56	235.34	294.12	134.99	1097.81	119.63	75.55	111.28	146.71
2023M09	218.72	365.49	97.24	53.18	200.41	257.37	210.80	256.26	123.81	1095.01	98.80	72.99	112.91	133.58
2023M10	196.20	334.17	89.82	43.43	203.58	275.17	195.71	256.78	115.41	1082.44	98.97	63.29	134.57	143.35
2023M11	175.08	316.72	88.32	42.79	182.87	253.56	164.60	223.07	101.94	1113.90	105.72	62.27	105.93	137.10
2023M12	195.28	366.37	103.94	51.78	199.65	288.87	190.82	258.49	113.57	1158.38	122.62	73.07	126.87	180.65
2024M01	199.10	368.23	108.19	45.59	193.68	280.08	191.67	259.03	116.30	1181.84	108.37	75.71	124.24	176.15
2024M02	159.29	319.67	91.37	32.82	178.04	196.23	167.13	209.69	101.04	1164.85	67.42	74.10	103.20	146.74
2024M03	194.38	348.82	87.48	56.69	209.03	249.99	183.41	254.97	113.21	1161.17	102.72	73.98	118.79	143.23

(see Appendix A). The annual bar chart comprising each municipality is illustrated in Figure 1.

Based on the above results, we provide the following two policy recommendations: Firstly, from the perspective of carbon emission control, among the 14 municipalities in Guangxi, Baise has a much higher annual total than other municipalities and shows an overall growth trend in each year, which needs to be taken as a key control object. Secondly, from the perspective of carbon emission intensity control, six cities in Guangxi, including Baise and Chongzuo, etc., need to strengthen the management and monitoring of carbon emission intensity to ensure that the entire Guangxi region can meet the national carbon emission intensity assessment requirements.

Specifically, for the high emission city of Baise, we provide the following discussion: The main reason for its high carbon emissions is that its total energy consumption ranks first in Guangxi, and its high energy consuming industries such as electrolytic aluminum and cement are dense, resulting in its high total energy consumption. The electrolytic aluminum production in Baise accounts for over 70% of the entire province of Guangxi. Due to the high demand for electricity in the electrolytic aluminum industry, Baise can reduce the dependence of electrolytic aluminum enterprises on thermal power by vigorously developing green electricity in the future, thereby reducing the total carbon emissions without reducing the total energy consumption. On the other hand, to improve the accuracy of the carbon emission calculation results, the statistical departments at the municipal level should attach importance to the compilation of energy balance sheets and the use of related data.

5.2. Validation Analysis of the Measured Results of the Prefectures and Cities

Since the individual prefectures and municipalities have not released any energy balance sheet data, the annual summed results of the prefectures and municipalities are used here to validate the results and ensure that the annual total results measured by the 14 prefectures and municipalities in Guangxi have a certain degree of accuracy. The idea of the validation analysis was borrowed from Shi et al. (2023) [10], with reference to data from China’s National Communication and CEADs. The specific steps are as follows:

First, based on China’s First Biennial Update Report (BUR) submitted in 2017, the Second BUR and Third National Communication (NC3) submitted in 2019, respectively, and the Third BUR and Fourth National Communication (NC4) released in December 2023, the data on China’s total carbon emissions from 2010 to 2023 are collated and computed, and the specific results and descriptions are shown in

Table 1. Calibration table of the results of measuring total carbon emissions from energy consumptions in 14 cities in Guangxi (unit: billion tons).

Year	Authorities Result for China	CEADs Result for China	Error Rate	CEADs Result for Guangxi	Our Result	Lower Bound of Our Result	Upper Bound of Our Result
2010	86.99 ***	79.05	−9.13%	1.53
2012	98.81 ***	90.81	−8.10%	2.08
2014	102.55 ***	94.51	−7.84%	2.12
2015	104.16 *	92.54	−11.16%	2.03
2016	105.58 *	92.56	−12.33%	2.17
2017	106.86 *** (107.03 **)	94.08	−11.96%	2.28
2018	108.92 *** (108.50 *)	96.21	−11.67%	2.32
2019	109.98 * (110.57 #)	97.95	−11.42%	2.46
2020	111.49 * (112.25 #)	98.80	−11.99%	2.68
2021	113.96 #	103.56	−9.12%	2.88
2022	115.69 #	104.02 #	−10.08%	3.30 ****	3.68 ##	3.13	4.23
2023	117.44 #	104.49 #	−11.03%	3.43 ****	4.29 ##	3.65	4.93

① *** indicates that data are from the National Communication; ** indicates that data are from the Emissions Database for Global Atmospheric Research (EDGAR); * indicates that it is obtained by extrapolating the geometric mean growth rate from 2014 to 2017, and ^# indicates that it is obtained by extrapolating the geometric mean growth rate from 2014 to 2018. **** indicates that the total carbon emissions in Guangxi were calculated based on the results of the first column, after revising the results of CEADs. ^## denotes Guangxi energy consumption plus emissions from the cement production process, in order to maintain consistency in the calibre of the validation. ② Total carbon emissions for 2010, 2012, and 2014 in the Second and Third Biennial Updates of the Information Circular are also slightly different, in which the three-year data published in the former are 87.07, 98.93, and 102.75, respectively, and the three-year data published in the latter are 86.99, 98.81, and 102.55, respectively. ③ CEADs is a third-party data platform on carbon accounting supported by China’s National Natural Funding Committee, the Ministry of Science and Technology, the Chinese Academy of Sciences, and the Research Councils of the United Kingdom, bringing together scholars from research institutions in China, the United Kingdom, Europe, and the United States.

. For comparison, we extracted data from CEADs [a third-party data platform on carbon emission accounting that bring together scholars from research institutes in China, the UK, Europe, and the US and are supported by China’s National Natural Science Foundation of China (NSFC), the Ministry of Science and Technology (MOST), the Chinese Academy of Sciences (CAS), and the UK’s Research Councils]. The website extracts and organizes the data on the carbon emissions measured by it for China and Guangxi from 2010 to 2021 prior to analyzing its results, as shown in Table 1. Third, based on the carbon emission data released by China, we revised the data of CEADs and projected the total carbon emissions of Guangxi in 2022 and 2023 based on the revised results. Fourth, using the total carbon emissions of Guangxi in the past two years projected in the third step as the standard, we calculated the lower and upper limits (±15%) of our results, respectively, and found that the lower limit of our results was closer to the standard, and the relative error rates of the two years were at 5.21% and 6.31%, respectively. The relative errors of our direct estimation results in 2022 and 2023 are 11.51% and 25.07%, respectively, and the error value of 19.3% that once appeared in the Chinese Academy of Sciences (CAS) is between the two years’ error values of our estimation.

Although the lower bounds of our results are relatively close to our reference standards, the relative errors of our direct measurements are slightly larger overall for the following three reasons: First, the underlying data on Guangxi’s carbon emissions that we used as a comparison came from CEADs, and given the underestimation of national carbon emissions data by CEADs, it is likely that its underestimation of Guangxi’s carbon emissions is similarly low, even though we have made some adjustments for this underestimation. This underestimation has been adjusted to some extent. Second, the basic data of Guangxi used in our estimation may be overestimated at the provincial level with large errors. Although China’s provincial-level carbon emission accounting methodologies were established one after another during the 12th Five-Year Plan period to meet the conditions of each province, there is no standardized system for regular operation and improvement, nor is there a system for testing consistency with national data, with the exception of the carbon intensity target set for the 13th Five-Year Plan, which has been used for this purpose. On the other hand, there is a large discrepancy between the national and provincial energy consumption statistics, and the Energy Statistics Yearbook before 2014 shows that the discrepancy between the sum of provincial energy consumption and the national total energy consumption during 2005–2012 was 12–23%, and it has been increasing year by year; in 2015, however, after a systematic review, the discrepancy between provincial energy consumption and national total energy consumption was 12–23%. The difference still exists, although it was systematically adjusted in 2015.

5.3. Long-Term Prediction Results of Carbon Emissions in Prefectures and Cities

Based on the annual electricity consumption planning data from 2024 to 2030 provided by Guangxi Electric Power, we conducted long-term carbon emission forecasts for each municipality using the prediction and warning model constructed earlier. The forecast results are shown in

Table A2. Forecast results of carbon emission based on planned electricity consumption data of Guangxi (unit: 10,000 tons of CO₂).

Time	Nanning (Y109)	Liuzhou (Y110)	Guilin (Y111)	Wuzhou (Y112)	Beihai (Y113)	Fangchenggang (Y114)	Qinzhou (Y115)	Guigang (Y116)	Yulin (Y117)	Baise (Y118)	Hezhou (Y119)	Hechi (Y120)	Laibin (Y121)	Chongzuo (Y122)
2024M01	181.29	324.40	153.69	115.79	192.79	252.15	196.82	229.79	157.93	351.86	134.49	149.06	153.34	210.87
2024M02	133.48	272.95	110.45	73.23	167.68	200.48	164.01	162.58	130.27	355.63	83.91	113.47	122.81	161.32
2024M03	161.33	311.34	153.53	91.66	185.48	237.27	179.03	204.15	127.01	383.18	106.72	116.80	120.89	161.77
2024M04	157.48	299.07	104.36	101.87	184.13	231.26	183.49	199.84	125.54	419.88	122.27	107.50	116.46	138.95
2024M05	193.90	316.63	111.82	104.51	231.22	255.55	212.86	219.62	138.22	451.73	114.80	115.57	125.17	151.76
2024M06	215.84	305.04	133.44	129.94	216.09	247.64	231.77	212.75	168.69	391.24	116.24	121.39	149.41	177.73
2024M07	248.05	311.39	155.97	131.87	211.26	244.63	263.27	233.17	173.62	384.61	130.73	149.42	159.69	173.94
2024M08	249.50	281.85	174.70	155.95	234.59	232.52	270.56	225.02	179.56	341.35	150.69	166.88	177.69	170.17
2024M09	238.51	240.08	177.57	176.27	249.04	202.67	260.51	208.15	180.28	292.78	168.80	186.23	190.81	173.67
2024M10	210.77	197.34	185.05	205.26	236.52	190.10	251.33	186.72	183.52	246.04	171.12	202.99	229.20	168.06
2024M11	203.24	173.98	191.48	213.67	226.73	177.24	250.80	174.12	186.80	207.13	202.99	222.83	247.11	194.45
2024M12	230.81	173.99	230.74	263.25	239.12	187.90	275.49	185.23	214.27	176.31	241.34	281.32	382.67	227.00
2025M01	177.89	128.95	224.88	189.30	203.63	138.26	227.65	139.04	231.96	125.69	166.23	269.59	320.14	202.76
2025M02	234.68	148.68	316.02	297.85	226.83	169.97	263.81	192.19	276.60	119.73	268.03	354.89	391.16	262.94
2025M03	216.57	126.66	382.35	317.80	223.58	162.42	233.13	181.95	272.37	97.39	288.59	343.50	371.35	237.29
2025M04	208.94	119.64	278.30	392.46	218.49	163.75	223.63	187.60	291.37	95.80	378.26	333.94	364.25	220.13
2025M05	250.25	131.72	303.18	415.07	266.78	191.28	238.78	219.09	332.04	100.62	382.13	354.84	371.47	253.03
2025M06	266.15	138.68	347.77	490.97	239.64	198.97	236.20	225.62	399.72	92.46	388.11	344.49	393.00	302.42
2025M07	288.40	161.78	370.59	439.69	223.42	213.50	242.57	262.25	387.05	104.45	408.99	369.02	351.30	292.90
2025M08	270.91	173.51	360.90	428.87	235.51	221.94	225.82	266.68	360.46	114.43	414.29	340.64	312.06	275.21
2025M09	240.61	179.76	306.64	378.16	237.06	211.94	198.63	257.48	313.81	128.51	385.62	301.98	259.01	262.55
2025M10	197.41	182.36	259.54	329.70	213.96	217.03	177.57	238.18	268.93	147.67	310.55	254.77	236.47	232.08
2025M11	177.40	198.92	214.48	250.83	195.95	219.06	167.33	225.89	225.96	174.33	283.58	214.12	193.67	241.01
2025M12	189.23	243.75	205.56	224.48	199.03	248.20	177.48	240.79	212.08	209.35	255.33	207.89	231.34	249.72
2026M01	188.16	260.48	215.72	156.58	188.41	262.14	173.19	233.67	181.68	228.54	227.60	193.01	156.17	251.14
2026M02	137.03	225.78	149.16	95.71	163.86	212.36	144.12	165.82	145.68	241.78	138.00	142.78	121.69	192.06
2026M03	163.46	267.66	196.88	113.12	180.68	255.55	157.23	208.24	136.65	278.20	167.12	140.49	115.86	189.78
2026M04	156.85	270.21	126.13	117.34	178.09	253.72	160.01	204.61	129.40	330.13	179.60	121.93	106.14	159.66
2026M05	189.58	302.52	126.51	111.31	221.69	284.93	184.32	225.77	136.15	388.84	155.81	122.48	107.95	169.53
2026M06	207.11	309.14	140.84	127.48	205.28	279.55	199.63	219.61	158.89	370.90	144.05	119.65	121.92	192.02
2026M07	233.88	334.47	153.47	119.28	199.00	278.12	226.40	241.50	156.84	401.91	146.74	137.01	123.98	181.07
2026M08	231.74	319.52	160.74	130.81	219.50	264.63	233.46	233.53	156.45	391.48	152.71	142.97	132.54	170.39
2026M09	218.98	285.16	153.73	138.51	232.01	229.40	226.80	216.11	152.65	365.07	154.74	150.35	138.56	167.35
2026M10	192.08	243.11	152.16	153.19	220.05	212.63	222.02	193.58	152.41	328.97	142.89	156.30	164.56	156.23
2026M11	184.71	219.62	151.34	154.01	211.31	194.75	226.00	179.90	153.72	291.79	156.25	166.07	178.36	175.17
2026M12	210.22	222.06	177.72	186.69	223.96	201.82	254.34	190.39	176.58	256.45	173.99	206.30	282.31	199.40
2027M01	206.28	201.05	218.42	193.83	217.95	176.40	260.16	182.26	191.17	203.43	160.59	237.48	242.82	225.58
2027M02	157.86	147.35	185.30	159.70	194.59	134.91	219.76	131.84	188.27	157.76	124.06	220.90	240.02	192.83
2027M03	195.53	151.70	295.04	247.35	217.95	156.54	237.65	170.78	213.48	136.77	190.37	265.61	277.92	212.78
2027M04	192.18	137.65	221.93	321.88	215.78	153.41	234.28	174.82	237.99	127.68	254.60	270.81	293.68	198.47
2027M05	234.64	144.13	252.31	361.80	266.74	174.43	255.94	202.55	284.47	125.16	267.36	304.70	324.48	230.62
2027M06	255.38	143.81	305.39	459.25	242.67	177.69	258.19	207.95	361.38	106.15	288.36	316.03	373.46	280.68
2027M07	283.59	158.73	345.68	442.97	228.91	187.91	269.00	241.87	370.41	109.91	327.91	363.20	363.00	278.16
2027M08	273.10	161.36	358.77	465.15	243.74	193.94	252.55	247.16	365.33	110.15	362.70	359.79	349.04	268.35
2027M09	248.38	159.19	324.73	439.36	247.27	185.25	222.59	240.72	335.98	113.53	371.05	340.98	310.94	263.27
2027M10	208.17	154.91	291.60	406.67	224.31	191.17	198.08	225.39	302.55	120.67	328.75	305.20	301.00	239.26
2027M11	190.42	163.66	253.76	324.40	205.84	195.80	184.63	216.94	265.01	133.41	328.81	269.18	257.47	254.97
2027M12	205.81	196.44	253.50	299.87	208.84	226.48	192.62	235.16	256.73	152.41	321.15	270.55	315.73	270.17
2028M01	146.14	169.23	207.26	161.02	158.20	179.67	151.60	167.55	201.06	146.84	176.84	208.27	180.74	224.56
2028M02	186.38	227.95	242.22	191.44	172.55	234.43	173.91	229.74	201.45	186.42	230.35	222.82	181.86	260.97
2028M03	168.10	220.83	247.98	158.66	168.39	233.25	155.55	212.99	168.79	196.48	202.33	180.31	147.55	211.64
2028M04	160.48	229.29	156.83	158.65	164.88	239.72	154.27	212.64	157.10	240.05	220.97	152.25	129.59	177.98
2028M05	192.76	266.95	153.58	143.28	204.32	279.22	173.68	238.38	160.80	296.55	191.97	147.10	124.95	188.00
2028M06	208.38	285.21	164.73	153.89	188.61	283.39	184.18	234.28	180.73	300.96	174.47	136.58	132.50	210.26
2028M07	232.41	323.89	171.32	133.73	182.78	290.79	205.33	259.16	170.57	350.85	172.23	147.50	125.88	194.68
2028M08	227.21	325.06	170.10	135.38	202.17	284.12	209.16	250.87	161.89	370.14	171.65	144.53	125.59	179.06
2028M09	211.86	304.22	153.68	132.16	215.01	251.55	201.87	231.32	150.04	374.70	165.17	142.65	122.93	171.36
2028M10	183.59	270.79	143.68	135.19	205.85	236.64	197.54	205.57	142.46	365.64	144.14	139.62	137.64	155.63
2028M11	174.77	253.69	135.43	126.67	200.11	218.46	202.24	188.85	137.20	348.79	148.78	140.60	142.10	169.77
2028M12	197.41	263.68	151.68	144.76	215.19	226.58	230.25	196.97	151.50	326.08	156.77	167.15	217.08	188.34
2029M01	193.73	244.91	176.66	135.23	206.40	211.34	231.97	189.47	157.60	282.78	142.41	191.82	187.36	201.89
2029M02	148.11	180.60	145.88	108.55	187.51	159.39	200.20	134.61	152.13	225.95	105.55	175.14	184.30	169.56
2029M03	183.92	184.96	228.99	166.95	213.74	181.25	221.93	171.21	171.07	198.06	157.04	209.71	215.76	184.90
2029M04	181.83	165.12	172.13	219.97	215.20	173.13	224.72	172.21	191.42	183.34	206.17	216.01	234.03	171.53
2029M05	223.96	168.40	198.24	255.25	270.22	190.95	252.39	196.34	232.39	174.81	215.26	248.83	268.91	199.66
2029M06	246.50	162.27	246.21	340.48	249.25	188.04	261.65	198.79	303.14	141.65	233.81	267.32	325.31	245.18
2029M07	277.28	171.84	289.25	350.42	237.81	191.85	279.64	228.68	322.04	138.01	271.03	321.22	334.88	246.85
2029M08	270.75	166.90	314.47	397.19	255.39	190.99	268.49	231.89	331.61	128.60	308.90	334.92	342.43	243.38
2029M09	249.72	157.09	300.13	407.97	260.48	176.27	241.00	224.95	319.93	122.29	328.53	335.28	324.57	245.19
2029M10	212.12	146.00	285.18	411.72	236.74	176.36	217.29	210.58	302.85	119.54	304.59	317.11	333.29	229.59
2029M11	196.37	147.86	262.69	357.34	216.93	176.00	204.04	203.43	278.64	121.71	320.03	294.71	300.49	252.51
2029M12	214.37	171.17	276.96	356.94	219.05	199.62	213.15	222.09	282.52	128.84	328.68	310.33	384.77	276.09
2030M01	201.85	180.75	292.79	276.51	196.02	203.66	197.50	221.37	259.97	132.81	278.38	303.64	269.24	284.67
2030M02	148.19	154.37	214.03	177.56	167.73	169.19	158.76	161.91	216.64	131.77	185.84	231.62	211.36	226.21
2030M03	176.75	184.44	291.33	212.89	181.27	210.92	166.18	209.71	206.40	147.53	241.44	228.44	196.38	229.69
2030M04	168.25	191.78	187.54	216.20	174.70	218.73	161.57	212.27	193.91	176.90	270.44	193.22	170.34	196.10
2030M05	200.46	225.50	184.34	194.03	212.52	258.15	177.57	240.64	197.82	218.30	237.60	184.26	159.70	208.61
2030M06	214.82	246.02	196.49	203.82	192.53	267.19	183.78	239.46	219.19	225.45	216.32	166.98	162.68	233.74
2030M07	237.14	287.70	200.95	170.46	183.07	280.67	200.03	267.91	201.78	271.99	211.46	174.10	146.90	215.46
2030M08	229.29	299.36	194.40	163.88	198.84	281.41	199.28	261.91	185.13	301.29	206.47	163.28	138.24	196.16
2030M09	211.39	291.78	169.83	150.35	207.96	255.83	188.66	243.32	164.66	323.92	192.74	153.28	127.04	184.86
2030M10	181.16	271.03	152.68	143.56	196.24	246.86	181.80	217.25	149.32	338.31	161.87	142.19	133.41	164.63
2030M11	170.70	264.85	137.92	125.20	188.58	233.17	184.16	199.89	137.03	346.72	159.81	135.64	129.51	175.54
2030M12	191.12	286.30	147.92	133.33	201.16	246.45	208.56	208.14	144.26	348.11	160.51	153.13	187.12	190.01

(see Appendix B). To better understand the changes in carbon emissions across the different regions, we divided the 14 municipalities into three regions as follows: Southern Guangxi, Northern Guangxi and Eastern Guangxi. The forecasted emission trends for each region are presented in Figure 2, Figure 3 and Figure 4. And the meaning of Y109-Y122 in the three figures is the same as in Table A2.

From the results, we can make the following key observations: The Eastern Guangxi region, consisting of four municipalities, shows relatively consistent emission trends across the different cities. In Southern Guangxi, the five municipalities exhibit similar monthly emission levels with minimal variation. The Northern Guangxi region, consisting of five municipalities, demonstrates a wider range of monthly emission differences and shows more variability in emission patterns. Additionally, compared to the other two regions, the overall carbon emissions in Northern Guangxi are significantly higher, and therefore, this region warrants more focused attention from decision-making bodies.

The emission totals for Southern Guangxi and Eastern Guangxi are relatively similar, with annual emissions not differing significantly (within a range of around 10 million tons). In contrast, the carbon emissions in Northern Guangxi vary greatly between the cities. For example, cities such as Guigang, Baise, and Yulin have annual emissions reaching around 30 million tons, while cities like Hezhou and Qinzhou have much lower emissions, around 10 million tons annually.

From the perspective of carbon peak goals, the forecast results for carbon emissions across the 14 municipalities indicate that most municipalities will reach their peak carbon emissions before 2030. Specifically, the carbon emission peak for most cities is expected to occur between 2025 and 2027.

These findings provide valuable insights into how regional energy consumption and carbon emission trends are evolving over time. The forecast highlights key areas where emissions are likely to rise more sharply, especially in Northern Guangxi. Such information is crucial for policymakers to focus on emission reduction strategies, particularly in areas with high growth potential for emissions.

6. Conclusions

Based on external statistical data and power big data provided by Guangxi Electric Power, we constructed the electricity–energy–carbon model at the prefecture and city levels to measure the carbon emission in the energy consumption of the prefecture and city; at the same time, we constructed the prefecture and city carbon emission prediction and early-warning model, based on the future planned electricity consumption data of the prefecture and city, and measured the carbon emission of the 14 cities and municipalities in Guangxi in high-frequency and long-term prediction. Preliminarily, the following conclusions can be obtained:

First, the situation of emission reduction in the whole region of Guangxi is still severe. The total carbon emissions from energy consumption of 14 prefectures and cities in Guangxi in 2023 will be 380 million tons, an increase of 8.57% over 2022, and the growth rate of 2020 and 2021 will also be as high as 7.2% and 9.17%, respectively, and the pressure of emission reduction in the whole region of Guangxi is still large.

Secondly, among the 14 cities, attention needs to be paid to the seven cities of Baise, Liuzhou, Chongzuo, Fangchenggang, Laibin, Qinzhou, and Guigang. Among them, Baise has the highest share and its share is gradually increasing; its carbon emission intensity is much higher than the average level of the province. Liuzhou has the second highest share, but its share is slightly decreasing; the carbon emission intensities of the latter five cities are all higher than the average level of the province.

Thirdly, the results of the long-term forecast show that the carbon emissions of the 14 prefectures and cities basically show their respective peaks before 2030, with the peak point of most of them in 2025 or 2027.

Fourthly, the high-frequency carbon emission measurement and prediction methods used in this article can effectively solve the problem of “lagging behind” when municipal governments face the national carbon emission dual control target assessment. Monthly monitoring and forecasting of high-frequency carbon emissions at the municipal level is beneficial for the municipal government to anticipate the completion of annual carbon emission dual control targets in advance and adjust the relevant industrial policies in time.

The limitations of this article mainly manifest in the following three aspects: Firstly, for the high-frequency calculation of carbon emissions at the municipal level, in terms of ensuring the accuracy of the total amount calculation results in various municipalities throughout the province, the lack of official energy statistics data in municipalities may result in insufficient accuracy of the carbon emission calculation results at the municipal level. Secondly, in the long-term prediction of high-frequency carbon emissions for municipalities, the accuracy of the prediction results is limited by the accuracy of the planned forecast data, due to the reliance of the planning department of the power company on the annual planning forecast data. Thirdly, in terms of measuring carbon emissions of municipalities, due to the lack of industry production process data at the municipal level, this article only calculates the carbon emissions of energy consumption in municipalities, without measuring the carbon emissions of the production process.

The future work mainly include the following two aspects: First, in terms of application, when the energy balance sheets of the cities and regions are published, the accuracy of the measurement results of the cities and regions will be further verified and, at the same time, the methodology will be downwardly applied to the counties and cities. Secondly, in terms of methodology and theory, on the basis of the existing research, machine learning theory will be introduced to automate the optimization and preference of the model, so as to improve the intelligence level of the future carbon emission measurement in the cities and districts.