3.1. Drought Detection with SPI Obtained Based on Data from Rain Gauge Stations
Figure 3 shows the mean values of the monthly standardized precipitation index (SPI-1) of the north–central region of the State of Mato Grosso for the period from 1985 to 2017. June, July, and August were disregarded from the analysis because they were in the normally dry period and presented a rain value equal to zero, as shown in
Figure 2.
The analysis of the average monthly SPI values (
Figure 3) showed that more than 91.9% of the monthly accumulated precipitation occurring in the study region can be categorized by SPI as normal precipitation (NP), while 4.7% was classified as extremely dry (ED) and moderately dry (MD) and 3.4% as moderately wet (MW) and extremely wet (EW).
Figure 3 shows that the extreme drought (ED) occurred in three months of three distinct years of the historical series, that is, January 1993, December 2015, and February 2016. These events occurred exactly during the soybean development period in the study region. The 2015/2016 soybean-growing season was one of the most affected due to drought. An 11% reduction in production was observed in this growing season compared to the previous growing season (2014/2015), even with an increase in planted area by 1% (
Figure 4). There was a production increase in 2015/2016, even with drought greater than 2012/2013, due to an increase in planted area. The location of the study region is in the north and central west of Brazil, where the exploitation of land by primary production is still expanding, with the removal of vegetation cover for the implementation of agriculture.
Agricultural production in the region is conducted under rainfed conditions and the sensitivity of SPI for detecting droughts and its impact on the grain-growing season in the region is evident. The study in [
20] used SPI to monitor droughts in the north and northwest regions of the State of Rio de Janeiro, Brazil, and concluded that the monthly SPI is efficient for detecting extreme droughts. The results of this research corroborate those obtained for the north–central region of the State of Mato Grosso regarding SPI, but the data of
Figure 3 and
Figure 4 also show that this index is related to a decrease in soybean production, such as that which occurred in the 2015/2016 growing season (
Figure 4).
The study in [
45] used the monthly SPI to monitor droughts in the State of Espírito Santo, Brazil, and found that its values could detect droughts, leading to a reduction in coffee production. This result and that obtained for the region under study demonstrate that the monthly SPI can detect droughts and their consequences for different crops.
The trend analysis of the monthly SPI values (
Table 4) by the Mann–Kendall hypothesis test indicated that the series has stationarity, that is, the SPI values are invariant regarding the chronology of their occurrences, except for random fluctuations. Thus, there is no trend to increase or decrease dry and humid events in the north–central region of the State of Mato Grosso for the analyzed period.
3.2. Drought Detection with SPI Obtained Based on Orbital Remote Sensing Data
Table 5 shows the validation of different rain estimations by remote sensing to detect droughts in the study region using SPI. The simplified SPI scale is shown in
Table 1 with the classes With drought and Without drought. The results in bold indicate an adequate performance of the products for detecting droughts according to criteria established in the methodology. June, July, and August were also disregarded from the analysis, as these months are usually dry in the region under study.
The results in
Table 5 show that a single product to estimate the rain by remote sensing, used to calculate SPI, could not detect droughts in the region for all months of the year, requiring the use of other products to characterize better the drought regime in the region. Drought estimation with rain estimated by PERSIANN-CCS presented good performance for detecting droughts in April. CHIRPS-2.0 was the precipitation product that resulted in the best performance in January, February, March, May, September, and November. The product GPM-3IMERGMv6 presented a moderate performance for detecting droughts in December.
The study in [
22] concluded that the PERSIANN-CDR and CHIRPS data were suitable for monitoring droughts with SPI in eastern mainland China. However, the results for western China indicated their inadequacy for monitoring droughts. In the case of the north–central region of the State of Mato Grosso, these products were effective in detecting droughts in January, February, March, May, September, November, and December, but only CHIRPS was valid for January and May (
Table 5). The study in [
47] used CHIRPS data to monitor drought and humidity throughout the semi-arid region of Midwest Argentina. According to these authors, CHIRPS is an adequate tool to monitor precipitation anomalies for periods longer than one month. The results obtained in the present study diverge from that research since the CHIRPS data were used for monthly drought monitoring.
The H0 hypothesis of the McNemar test was rejected for all products in October indicating a significant difference between the drought detection with data from rain gauge stations and products derived from remote sensing. Therefore, drought monitoring using satellite information is not recommended during this month.
The study in [
48] found that the accuracy of rainfall data estimated by the TRMM depends on several factors such as region, the season of the year, time and space scales. These authors also presented the results of other studies that showed the variability of the TRMM accuracy for different regions of the world as well as for the intensity and other characteristics of the rain. The results presented in
Table 5 indicate that this variability in the accuracy of the orbital remote sense rainfall measurement is also valid for the CHIRPS, PERSIANN, and GPM products tested in this work and detailed in
Table 2.
The causes for this variability are related to different factors such as seasonality, type of rain, the influence of synoptic systems, and surface factors such as soil occupation/roughness. Because of this, further studies should be carried out in the region to identify the intervening factors in the rain performance estimated by remote sensing and in the drought estimate with the SPI.
The results show that the rain deficit in the study region can be monitored with remote sensing during practically the entire soybean development period, that is, from October/November to February/March. Moreover, soybean is the preferred crop for producers, showing the highest commercial value other than corn.
Figure 5 shows the drought maps for the period from November 2015 to March 2016, obtained from the remote sensing products validated in
Table 5. These maps corroborate with those obtained with rain gauge data measured on the surface (
Figure 3), but with higher detailing regarding the spatial variability of the drought event in the region.
Most of the region in November 2015 had no drought, but some municipalities such as Cláudia and Santa Rita do Trivelato had most of their area affected by a rain deficit. Almost the entire north–central region of the State of Mato Grosso was affected by drought in December and February. Moreover, January and March presented no rain below normal.
As previously discussed, the drought that occurred in certain locations in November and almost the entire region in December and February was the main factor responsible for the soybean crop shortfall in the 2015/2016 growing season.
Researchers from the Empresa Brasileira de Pesquisa Agropecuária (EMBRAPA) made a technical report in March 2016 on the situation of the 2015/2016 growing season in the study region requested by the Associação Brasileira de Produtores de Soja (Aprosoja). Details of this report are available in [
49]. This report confirms the results obtained in this research, as it associates a reduction in soybean production with the water deficits that occurred from 30 November to 24 December, 2015, and February 2016. The researchers highlighted that the periods with low precipitation coincided with the vegetative development and flowering and grain-filling stages of many soybean cultivations, resulting in a low development and abortion of flowers and empty pods, respectively.
In addition, [
49] associated the soybean production losses with an increase in the incidence of plants with black root rot caused by
Macrophomina phaseolina. According to the authors, the occurrence of a drought period followed by a humidity period favored the root invasion by the fungus, leading to wilt and deterioration of the woody root tissue as it progresses.
Figure 5 shows the alternation of dry and wet periods, mainly between December 2015 and March 2016, which may have favored the increased incidence of
Macrophomina phaseolina on soybean plants.
The analyses showed the possibility of monitoring droughts with SPI calculated through historical rain series derived from remote sensing. This approach, even not characterizing the drought severity, allows a synoptic analysis of this natural disaster and can optimize the direction of measures to mitigate impacts.
The density of rain gauge stations in the Amazon region, where the study area is located, is very low compared to other regions of Brazil. The promising results obtained in this study for monitoring droughts with indirect rain measurements originating from remote sensing validate an important resource to deal with the limitations of the hydrometeorological information collection network in the region.
According to [
50], the estimation of rain with a high spatial and temporal resolution by remote sensing (radar) can be a promising approach to mitigate the uncertainty resulting from the high spatial variability of precipitation in an area, particularly in combination with surface data (rain gauges). The results obtained for the north–central region of the State of Mato Grosso (
Table 3 and
Figure 5) prove that this statement is also valid for the drought estimation using orbital remote sensing products duly validated with surface data.
A comparative analysis of the results obtained in this study with those recently published show that drought indices and indirect satellite rain measurements work specifically for different world regions, as well as their correlation with the agricultural production. In this sense, studies that use procedures similar to those employed in this research and that validate a given index regarding its ability to characterize the drought of a region and its impact on the agricultural sector are essential and strategic.
The use of soil moisture estimated by remote sensing also can be used in further research to improve the drought monitoring in the study region. Works such as [
51,
52] evaluated different estimates of soil moisture derived from remote sensing and found that these products have adequate performance for this purpose. The soil moisture and other physical characteristics of the soil would make detection of droughts and their impacts on crops more robust. In this work, we do not use the soil moisture estimated by satellite due to the absence of a soil database for the validation of satellite products, such as the soil moisture and the water storage capacity in the soil measured in-situ. These are not only a limitation of the study region but many in developing countries. There are works aimed to improve the soil database conducted by the association of soy producers in partnership with universities in the northern middle region of Mato Grosso, which will allow future work to validated soil moisture products to monitor drought.