4.1. The Regional Discussions of Agricultural Fires
Spatially, the study area can be divided into seven geographical regions. The monthly time series of the agricultural fires in the seven geographical regions is presented in Figure 8
. The agricultural fires category has its own set of trends in each region across different seasons. The temporal and spatial distributions of agricultural fires in each region have been carried out based on the detailed analysis and discussion.
In the Northwest and Southwest regions, agricultural fires accounted for 5.3%, 8.9%, 5.7%, 11%, and 9% of all the fires in 2010, 2011, 2012, 2013, and 2014, respectively. In the Northwest region, where the major crops are wheat, cotton, and corn, agricultural fires were mainly localized in northwest Xinjiang, north-central Gansu, central Shaanxi, and northern Ningxia. Figure 8
a shows that cropland burning was unusually active in October 2013 and July 2014. The agricultural fires in 2013 accounted for 45% of all fires in the Xinjiang region for five years. Agricultural fires in 2014 accounted for 42% of all fires in Shanxi during 2010 to 2014. Figure 8
shows that cropland burning fires in the Southwest region were mainly concentrated in July and August, up to 74%. Agricultural fires in the Southwest region, where rice and corn is mainly planted, mainly occurred in Sichuan and Chongqing during the summer and autumn seasons. The Sichuan Basin and Yunnan-Guizhou Plateau are important rice producing areas in China. The proportions of agricultural fires were higher than the other years in 2011 and 2013 because of the large number of fires in Sichuan and Guizhou (Figure 9
The North China region contributed to an average of 12% of all the agricultural burning detected in China, and it was the third largest contributor to agricultural burning. Agricultural burning in North China was concentrated in Hebei and accounted for 76% of all agricultural fires in the region; this is likely because the province has a relatively large cultivation area, and the prohibition of crop straw burning is less strict than in Beijing and Tianjin. Agricultural burning in this region was mainly concentrated in the summer season, with June and July accounting for 57% of the number of agricultural fires, likely because this is when the harvest of the wheat and rapeseed occurs in the region. This phenomenon is consistent with the local sowing and harvest times.
The multiple-cropping rice, winter wheat, and winter rapeseed are mainly cultivated in South China. As shown in Figure 8
d, a peak appeared in August which was related to the harvesting of early-season rice [28
]. The low proportion of agricultural fires in Guangxi in 2012 led to a decreased proportion in South China for this year (Figure 9
). This region is located in the tropics, with abundant water and sunshine, and has two or even three harvests each year. However, cropland burning is not a notable occurrence in the region during the summer and autumn. The Northeastern region is located in a high latitude area, and has just one harvest in the whole year, and its main crops are spring maize, spring wheat, rice, and beans. Agricultural fires accounted, on average, for 11% of all the fires detected in the region of Northeastern China. The percentage of agricultural fires in Heilongjiang exceeded the average of the other years by 6% in 2011. The maximum fire activity in 2010, 2011, 2013, and 2014 in this region occurred in October, which is in agreement with agricultural practices. In this region, spring maize, spring wheat, and rice are sown in late April, and the sowing time for beans is mid-May. Early October is the main harvest season for almost all the crops [28
]. The agricultural fires in October 2012 only contributed 1.5% for the year's agricultural burning in the region, which can be primarily attributed to adverse weather conditions (i.e., the increased precipitation affected the farmland straw burning in October 2012 [29
Central China and Eastern China are the main agricultural zones. There are suitable climatic conditions for agricultural production in the region, and, as such, the region accounts for almost half of the country’s agricultural production [21
]. Extensive agricultural fires occurred in the region and accounted for 59%–80% of all the fires. Wheat and maize are the primary crops because of the widespread use of the wheat-maize rotation system in the area [28
]. In the region, the agricultural fires burning peak occurred in the summer season each year. We detected that agricultural fires in June accounted for 73%, 66%, 79%, 59%, and 60% of all fires during 2010 to 2014 in the zones, which is related to the burning of winter wheat and the clearing of cropland for summer maize cultivation. Winter wheat is sown in mid-October and harvested at the end of May. Summer maize is sowed in mid-June and harvested at the end of September. Figure 8
f,g shows that agricultural burning has an increasing trend in October. The agricultural fires in Anhui, Jiangsu, Henan, and Shandong provinces accounted for an average of approximately 59% of the total fires over the five years, and the largest annual average number of agricultural fires occurred in Anhui (Figure 9
In this paper, the accuracy of the results from fire monitoring is influenced by two factors. The first one is that there are some limitations associated with the use of the MODIS and the GlobeLand30-2010 data sets. The second is that the process of the GlobeLand30-2010 down sampling has a certain error in order to make the resolution of MOD14 and GlobeLand30-2010 data consistent.
The problem of the MODIS image resolution and the occurrence of many small fires falling below the MODIS detection limit for thermal anomalies led to some small agricultural fires not being recorded [30
]. Moreover, a small number of MODIS images were not complete (Table 1
), which resulted in an underestimation of agricultural fires. Weather conditions such as cloud cover and precipitation also affected the fire monitoring results.
The precision of the GlobeLand30-2010 data should also be considered as an important source of uncertainty. In our study, 2.7%, 3.1%, 2.4%, 2.7%, and 2.4% of the fire points were identified as falling on wetland and water bodies in 2010, 2011, 2012, 2013, and 2014, respectively, which is, of course, unrealistic. Based on using a multiple feature optimization method in pixel scale, the cultivated land type of GlobeLand30-2010 data was extracted [27
]. Although the classification method has been taken into account in regards to the texture, spectrum, and phenological characteristics of the cultivated land as much as possible, the results of the classification have the phenomenon of fault and leakage. The accuracy of the land classes of GlobeLand30-2010 data is 83.06% [26
], but when the land cover is misclassified, this may lead to false detection. The GlobeLand30-2010 data is divided into 10 categories, but land cover points of each category are not very detailed. Furthermore, the GlobeLand30 only has two phase data. Although the global land cover data of the GlobeLand30-2010 data has high accuracy and resolution in recent years, the cultivation areas in 2014 have change compared with 2010. In this article, the cropland data of GlobeLand30-2010 data was used from 2010 to 2014, which could impart an influence on the results from agricultural fires.