Exploring the Linkages Between Different Types of Drought and Their Impacts on Crop Production in Kyrgyzstan

Drought is a perilous agrometeorological phenomenon that often causes crop damage in arid and semiarid regions vulnerable to climate variability. However, accurate drought monitoring remains deficient in many countries, including Kyrgyzstan, and the interconnections between several types of drought and contributions to crop yield are still unclear. Hence, we aimed to determine the propagation time in three types of drought (meteorological drought, soil drought, and vegetation drought) for understanding interconnections of them. Moreover, we focused on comprehensively evaluation the performance of multiple drought indices for each type over the complex terrain of Kyrgyzstan, especially for drought index of synergistic land surface temperature and vegetation conditions information. The results demonstrated that standard precipitation index (SPI) effectively detected meteorological drought, while the vegetation health index (VHI) coupled with temperature data was optimal for vegetation drought monitoring in Kyrgyzstan. Furthermore, our findings indicated a 1-month response time for soil drought at a 10 cm depth to SPI, and a 4-month response time at a 40 cm depth to meteorological drought (SPI). The response time of VHI to soil drought condition index (SMCI) was approximately 1 month, regardless of whether the soil drought occurred at a depth of 10 or 40 cm. In general, the response time of VHI to SPI was 3 months. Finally, by analyzing the correlation between crop yield productivity and drought indices, we discovered that the crop yield predictions by the three types of drought were differential and complex, but VHI was the most effective index. At the same time, VHIacc(May–Sep.), SMCIr(0–40 cm)_May–Sep., and SPI5_Aug. have different contributions to crop yield variations, and these are also differences in their impacts on different crops and provinces. The synergistic effect of the three types of drought may significantly improve crop yield prediction in Kyrgyzstan in future studies. These findings may significantly contribute to drought prevention and mitigation in drought-prone Central Asian countries.


I. INTRODUCTION
D ROUGHT is one of the natural disasters caused by the ongoing warming of the earth's climate.Among all natural disasters, drought affects the largest number of human being.Over the past few decades, both the frequency of drought and severity of droughts have been increasing [1].Monitoring and predicting droughts can be exceedingly challenging due to their slow onset, gradual development, and prolonged duration.
Drought can generally be categorized into four types: meteorological drought, agricultural drought, hydrological drought, and socio-economic drought.Meteorological drought is often regarded as the initial stage, triggering other forms of drought.Various meteorological drought indices exist, including the Palmer drought severity index (PDSI) [2], self-calibrating PDSI (scPDSI) [3], standard precipitation index (SPI) [1], and the standardized precipitation evapotranspiration index (SPEI) [4].The PDSI and scPDSI utilize readily available temperature and precipitation data to estimate relative dryness.Although they account for the basic impact of global warming on drought through potential evapotranspiration changes, monthly PDSI values may not adequately capture droughts on time scales less than about 12 months.The SPI, derived from probability functions of rainfall, is widely favored due to its simplicity, various timescales, spatial comparability, and strong performance in drought monitoring compared to other indices [5], [6], [7].Another extensively used index is SPEI, which incorporates a water balance model and considers both the drought sensitivity to temperature (in contrast to SPI), and multiple timescale characteristics.As such, the SPEI proves valuable for drought monitoring in both arid and humid regions [4], [8].Studies have revealed that SPEI and SPI yield similar estimates of drought characteristics over short periods, particularly in monsoon regions [9].
In recent years, more drought indices derived from remote sensing data have been developed and used to effectively detect agricultural drought conditions.One approach involves monitoring changes in surface temperature, soil moisture, rainfall, and other related environmental variables, and then establishing relationships between these changes with drought.Accordingly, precipitation condition index, temperature condition index (TCI) [10], and soil water deficit index [11] were proposed in the past two decades.Another approach focuses on detecting crops' growth and physiological information, where drought severity can be inferred from changes in vegetation information [12].Since drought is the major cause for reduction of vegetation activities in large geographic area, vegetation index and its derivatives are the common way to reflect vegetation drought from space [13].The earliest index, normalized difference vegetation index (NDVI) [14], has been widely utilized.The vegetation condition index (VCI) is very popular vegetation drought monitoring index [15], [16], [17], [18].Zhang et al. [19] proposed the visible and shortwave infrared drought index by combining the blue, red, and short-wave infrared bands.Based on near-infrared reflectance and short-wave infrared reflectance, the normalized difference water index (NDWI) [20] is sensitive to changes in soil moisture that are strongly related to vegetation drought conditions in the grass and cropland of the Oklahoma Mesonet.Consequently, NDWI-derived drought indices including the normalized multiband drought index [21] and the normalized difference drought index (NDDI) have been found effective in detecting drought [22].
Due to the complexity of drought, individual index may not adequately capture the drought onset and severity.In recent years, many researchers developed several combination drought indices.Combination drought indices in terms of vegetation stress, water deficit, and soil moisture status can describe the severity of and changes in drought better than each index in isolation.One of the early combination indices is the vegetation supply water index (VSWI), proposed by Carlson et al. [23].It utilizes the NDVI and land surface temperature (LST) to assess summer drought, because the ratio of LST to NDVI is shown to increase during drought.Sandholt et al. [24] found that the scatter of LST and NDVI data occupies a triangular space, and that an index based on this relationship (the temperature vegetation dryness index, TVDI) can be used to better monitor regional drought.A similar indicator, the vegetation TCI (VTCI), was applied to drought monitoring by Wang et al. [25].The VTCI had better performance than NDVI in classifying relative drought occurrence levels and in studying the distribution of drought occurrences.Another common combination index, the vegetation health index (VHI), was proposed by the work in [15] and [16] and was based on the combination of vegetation greenness (VCI) and temperature (TCI) indices.AVHRR-based drought indices (VCI, TCI, and VHI) were also proposed for monitoring grass conditions in Mongolia, where they effectively identified areas of intense drought and poor grass conditions.Meteorological, agricultural, and hydrological droughts are recurrent occurrences in Kyrgyzstan and other countries in Central Asia.Meteorological drought is identified when the rainfall during a specific year falls below the average annual rainfall over many years.If the meteorological drought persists for an extended period, it can lead to hydrological drought, characterized by the reduction of channel runoff and groundwater reserves.When the water supply of soil to crops is reduced to the extent that it has a negative impact on grain production, agricultural drought will occur, resulting in the reduction of agricultural production in the region.Given that agriculture in Kyrgyzstan is dominated by irrigation, the next most important risk factor is river water shortage or hydrological drought in the growing season (April-September).This phenomenon is very dangerous, especially in the spring and early summer.From 1975 to 2004, 13%-33% of the years in Chuy, Osh, Jalal-Abad, and Batken provinces experienced dry conditions during the growing season [26].Furthermore, Kyrgyzstan was in agrometeorological and meteorological drought in 2014, which had a serious impact on farmland and grassland in the country.
The frequent occurrence of drought poses significant challenges to the arid and semiarid areas of Kyrgyzstan.Previous studies [26] and feedback from local farmers have highlighted the detrimental impact of drought on agriculture in the country.Despite these observations, Kyrgyzstan currently lacks comprehensive drought monitoring and prediction models.While a few studies have attempted drought simulation experiments using long-term meteorological data and artificial neural network models, the identification of optimal indices for agricultural drought detection in Kyrgyzstan remains unexplored and unclear.This study aims to address this gap by conducting a comprehensive analysis of drought monitoring and early warning in Kyrgyzstan.
Many remote sensing-derived drought indices mentioned can be used for agricultural drought detection.However, agricultural drought is a general concept that is difficult to be defined by one or two drought indices alone.It encompasses various types, including soil drought, vegetation drought, hydrological drought, etc., at different stages, resulting in different degrees of impact on crop outputs.Therefore, we propose to separate agricultural drought into several stages.In this study, we focus on separating soil drought and vegetation drought from agricultural drought over Kyrgyzstan.Along with meteorological drought, we will first determine the propagation time in these three types of drought (meteorological drought, soil drought, and vegetation drought) to understand their interconnections.Though there were some study results of the propagation thresholds and time [27], [28], the majority of these studies have primarily focused on understanding the long-term triggers and relationships between meteorological and hydrological droughts.However, there is a notable scarcity of research concerning the propagation of soil drought to vegetation drought.In addition, there are partial research results about relations of meteorological drought and agricultural drought [29], [30], but in essence, these studies have mostly addressed soil drought's response to meteorological drought.Consequently, research pertaining to vegetation drought, derived from remote sensing data, remains limited in the context of drought propagation studies.
The objective of this article is to determine the three types of drought's interconnections, especially the response time from one to another.For this purpose, we have chosen various drought indices for each type and will compare and evaluate their effectiveness.Kyrgyzstan vulnerable to droughts is selected as the study areas.Through adaptability analysis of drought indices and the propagation time of each drought type determination, the optimal indices were confirmed and their spatial-temporal differences were analyzed.In addition, we will propose the spatial-temporal standardized crop yield index and evaluate three types of drought's performance on crop yield effects and predictions.

A. Study Area
The study area covers the whole Kyrgyzstan in Central Asia, between 39°N and 44°N and between 69°E and 81°E.The region is farther from the ocean than any other country in the world, although it does not contain the absolute farthest point from any ocean (Fig. 1).It covers a total area of 199 951 km 2 , and 94% of it is covered by mountains.This fact has greatly changed the country's geography and terrain appearance.Among the Eastern European and Central Asian countries, Kyrgyzstan is the third most vulnerable country to climate change, such as changes in weather patterns that may lead to long-term precipitation deficit and drought.In the past 20 years, their average temperature has increased from 5.8°C to 6°C, which is expected to rise by 0.2°C by 2060 and 1°C-2°C by 2100.The average annual precipitation is about 533 mm, and most of it is concentrated during the cold winter season between October and April [31].The highly seasonal pattern of precipitation and high mountains make the countries vulnerable to droughts.Climate change will have a negative impact on sensitive sectors such as agriculture.Due to global climate change, the frequency and scale of natural disasters (drought, fire, etc.) are increasing.

B. Datasets
In this study, we have utilized various geospatial and observation datasets, which include satellite data from the TERRA/MODIS, data assimilation system, and meteorological station data (Table I).The spatial-temporal distribution of these datasets is visually presented in Fig. 1.
1) Satellite Data: The daily surface reflectance and thermal data (MOD09GA and MOD11A1) were derived from the moderate resolution imaging spectroradiometer (MODIS) on TERRA satellite launched in December 1999, which can cover the whole earth's surface in one to two days.The data is available from the Google Earth Engine platform (GEE).On the GEE platform, according to the data quality marks of MOD09GA and MOD11A1 data, the daily reflectance data and surface temperature data are directly synthesized into months.In this article, the remote sensing-derived drought indices for May to September were calculated by using the long time series data from 2000 to 2019, compared with the meteorological drought and soil drought indices, and contacted with crop yields.
2) Meteorological Station Data: The long-term dataset of meteorological stations, including precipitation, air temperature, and multilayer soil moisture, is derived from the archival data of the hydro-meteorological Bureau of the Ministry of emergency situations of the Kyrgyz Republic (Kyrgyz hydrometeorological Bureau).Fig. 1 shows the distribution of agrometeorological stations in the country, which can measure precipitation and air temperature meteorological variables and soil moisture of four layers.Attributes of all agrometeorological stations can be found in Table II.The precipitation and air temperature data were collected at 30 agrometeorological stations covering Kyrgyzstan from 1981 to 2019.The multilayer soil moisture including four depths of 10, 20, and 50 cm were collected at 5 agrometeorological stations over whole country from 2000 to 2019.Due to a significant lack of data on 100 cm soil depth, it has been excluded from this article.These parameters were employed for the main indicators of meteorological and soil drought monitoring and evaluation.
3) Other Data: The MCD12Q1 V6 product provides global land cover types at yearly intervals derived from GEE.It is derived using supervised classifications of MODIS Terra and Aqua reflectance data.The supervised classifications then undergo additional postprocessing that incorporate prior knowledge and ancillary information to further refine specific classes.The Land Cover Type 1 from the Annual International Geosphere-Biosphere Programme (IGBP) classification was extracted, then the data range was from 2001 to 2019, and was clipped into Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

TABLE II ATTRIBUTES OF AGROMETEOROLOGICAL STATIONS FOR KYRGYZSTAN
Kyrgyzstan.It was used for yearly masking the cropland areas, and it introduced data to mask out other land cover types.
Multilayer soil moisture raster came from the Global Land Data Assimilation System (GLDAS), with the spatial resolution of 0.25°.The depths of the four soil layers are as follows: 0-10, 10-40, 40-100, and 100-200 cm.Here, we just select first two layers to calculate and analyze for crops.The quality of the GLDAS dataset was assessed against available observations from multiple sources [32], [33].
Yield data (unit:ton/ha) for each crop grown in Kyrgyzstan was obtained from the National Statistical Committee of the Kyrgyzstan.The crop yield from 1991 to 2018 came from territory's annual data Center (oblast-region) covering an area of 1 hectare, and the crops provided included main grain, wheat, barley, and corn.

A. Meteorological Drought Indices
The SPI is a widely used meteorological drought index that quantifies precipitation deficits over various time scales.It is a dimensionless index and can be calculated for different accumulation periods, such as 1, 3, 6, 12 months, etc., depending on the study's needs [1], [7].The SPI is based on a long-term meteorological dataset and is one of the common indicators for global drought situation monitoring.The SPEI is an extension of the widely used SPI.It integrates the impact of temperature on the water cycle through evapotranspiration [34], which is considered as an improved drought index of SPI, especially suitable for analyzing the impact of global warming on drought conditions [8].The SPEI is based on a climate water balance, which is determined by the difference between precipitation and potential evapotranspiration in a certain period.Potential evapotranspiration was calculated based on the Hargreaves-Samani equation [35], [36], which only required the monthly maximum temperature and monthly minimum temperature.Precipitation data were employed to accumulate into water deficits or surplus at different time scales [36], [37].
Considering the advantage of SPI and SPEI, together with usage in the previous study for Kyrgyzstan, which were introduced in this article for expressing meteorological drought.We calculated the SPI and SPEI by precipitation and temperature data in long time series (seen in Table I), as well as standardized it based on simple water balance.The definition and specific classification of drought based on SPI or SPEI are shown in Table III.

B. Soil Drought Indices
Soil drought was evaluated by soil moisture deficit, soil moisture deficit in different root layers may result in variable drought severity.Two datasets were introduced in this article, which are raster soil moisture (SM) from GLDAS and point SM from agrometeorological stations.Considering with main crops in study areas, we constructed the two soil drought indices (SMCIr(0-10 cm), SMCIr(10-40 cm)) based on soil moisture data of long time series in the first two soil layers (0-10 cm and 10-40 cm).Soil moisture raster data derived from GLDAS was from 2000 to 2019.In addition, other two soil drought indices (SMCIp(0-10 cm), SMCIp(10-50 cm)) based on stationed soil moisture were constructed.Soil drought indices built are following separately: Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

SMCIp(10 − 50 cm)
where SMr10 i is average SMr from surface to 10 cm depth in a certain month, the SMr10 max and SMr10 min are the maximum and minimum values of GLDAS-based SM from surface to 10 cm depth on the long-term sequence (seen in Table I), respectively.SMr40 i is average GLDAS-based SM from 10 cm to 40 cm depth in a certain month, the SMr40 max and SMr40 min are the maximum and minimum values of GLDAS-based SM from 10 cm to 40 cm depth on the long-term sequence, respectively.SMp10 i is average SMp from surface to 10 cm depth in a certain month, the SMp10 max and SMp10 min are the maximum and minimum values of station-based SM from surface to 10 cm depth on the long-term sequence, respectively.SMp50 i is average station-based SM from 10 cm to 50 cm depth in a certain month, the SMp50 max and SMp50 min are the maximum and minimum values of station-based SM from 10 cm to 50 cm depth on the long-term sequence, respectively.

C. Vegetation Drought Indices
Increasing the density of meteorological stations is a difficult and expensive task due to the complexity of Kyrgyzstan's terrain (mountainous area accounts for 94%).Therefore, the effective use of remote sensing data is the most meaningful and efficient.
Kogan et al. [16] proposed a typical drought index derived from remote sensing data, namely VHI, which is based on the combination of vegetation greenness (VCI) and temperature (TCI).VCI, TCI, and VHI were also validated by Kogan to monitor drought conditions, and Chang et al. [12] conducted comprehensive evaluation and comparison analysis of three indices in Mongolia, and confirming their superior performance compared to other drought indices derived from remote sensing data.The VWSI well describes the changes of soil moisture in agricultural land and is a fast and cost-effective method to monitor drought conditions [23], [38].The SWIR is sensitive to the liquid water content of leaf.Previous research results demonstrated that the NDWI is highly responsive to the changes in vegetation water and soil moisture [20], which are closely related to vegetation drought status of the grassland and cropland of the Oklahoma Mesonet [39].The NDDI, derived from NDVI and NDWI, has been shown to effectively detect drought conditions [22].For the analysis and evaluation of vegetation drought in Kyrgyzstan, the six typical drought indices mentioned earlier have been selected.To maintain comparability with other indices in both time and space, VWSI, NDWI, and NDDI were scaled using the VCI/TCI equation.The normalized drought indices are calculated as follows: ) VHI i = 0.5 * VCI i + 0.5 * TCI i (10) where

D. Drought Propagation Time
The study utilizes the Person correlation analysis method to investigate various aspects of drought propagation time.The correlation coefficients (R) are determined between the one kind of drought at a given timescale (generally monthly scale) and the other type of drought index.
Specifically, the correlation coefficient the correlation coefficient is computed between meteorological drought indices with different accumulation periods and soil drought indices.The time scale corresponding to the maximum correlation coefficient represents the drought transmission time from meteorological drought to soil drought.Similarly, the propagation time from soil drought indices to vegetation drought indices is determined by evaluating the correlation between these two types of indices.The time scale with the highest correlation coefficient indicates the propagation time from soil drought to vegetation drought.

E. Spatial-Temporal Standardized Crop Yield Index
The crop yield can be affected by many factors, such as drought, high temperature, diseases, national policies, agricultural practices, technological levels, and so on.In this article, we employed crop yield change to convey drought characteristics and influences, so other factors' influences on crop yield should be reduced.Supposed that agricultural practices or technological levels can linearly increase crop yield with scientific and technological development, so we used the detrending method to eliminate their influences [40].Moreover, direct comparisons of crop yield among different regions can be challenging due to variations in the fertility of croplands.To address this issue, one standardized treatment means is introduced [41].By applying this treatment, we create a new spatial-temporal standardized crop yield index (abbreviated as STSYI) to eliminate the impact of other factors except climate on yield as much as possible.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.We detrended the crop trend yields by linear regression approach, then detrended yields were the equation is where Yield a is the actual crop yield, while Yield d is the crop trend yield, A and B are coefficients in linear regression formula, and N year is one year of the year series.
where STSYI is the crop yield in certain year of one province, Yield d is the average of detrended yield, and σ is the standard deviation of crop yields from 2000 to 2019.

A. Response of Soil Drought to Meteorological Drought
Meteorological drought has the potential to trigger varying degrees of soil moisture deficiency, leading to the occurrence of soil drought [29].There exists a close connection and correlation between these two types of droughts [30].Inversely soil moisture changes can be directly responded by rainfall deficit, especially in rain-fed areas.Soil drought index named as SMCIr was used for representing the soil moisture deficit in different soil depth by GLDAS raster data.In Fig. 2, we present the average correlation coefficient (R) of SMCIr at different depths (0-10 cm and 10-40 cm) and meteorological drought indices (SPIx and SPEIx) across 30 meteorological stations from May to September, including 2995 data pairs(p <0.01).The analysis reveals that the correlation between SPIx and SMCIr(0-10 cm)/SMCIr(10-40 cm) is notably higher compared to the correlation between SPEIx and SMCIr(0-10 cm)/SMCIr(10-40 cm) for each timescale (ranging from SPI1 to SPI12, or from SPEI1 to SPEI12).The maximum correlation value of SPIx and SMCI(0-10 cm)/SMCI(10-40 cm) are 0.21 and 0.35, respectively, while for SPEIx and SMCIr(0-10 cm)/SMCIr(10-40 cm), they are 0.17 and 0.22.Compared with the SMCIr(0-10 cm), we have found a lower correlation between SMCIr(10-40 cm) and SPIx before the SPI3(SPI1,SPI2,SPI3), but after the SPI3, the relationship changes and the correlation becomes higher.Similarly, a lower correlation between SMCIr(10-40 cm) and SPEIx is evident before the SPEI4(SPEI1,SPEI2,SPEI3,SPEI4), then after the SPEI4, the correlation becomes stronger.This phenomenon can be explained by the fact that moisture in deeper soil layers responds more slowly to meteorological drought, resulting in a lower correlation in short timescales.
When we further took the comparison of the lags of soil drought indices to meteorological drought indices in temporal process, the maximum correlations between SPIx(or SPEIx) and SMCIr(0-10 cm)/SMCIr(10-40 cm) results can be found in Fig. 2(a).The maximum correlations were used for detecting soil drought's response to meteorological indices changes.It was examined to determine that how many months ahead of time meteorological drought can trigger the soil drought.The highest correlation was proxy to find them, Fig. 2 shows that SMCIr(0-10 cm) has the highest correlation with SPI1 of 1month time scale, while SMCIr(10-40 cm) shows the highest correlation with SPI4 at a 4-month time scale.
Further analysis was conducted to comprehensively examine the spatial distribution of response times.Fig. 3(a) and (b) display the maximum correlations spacial distributions of SPIx at different temporal scales (x = 1-,2-,3-…12-month) and SMCIr(0-10 cm) (p < 0.01).Among 30 agrometeorological stations, the drought propagation time in 19 stations is less than 3 months, with 1-month lag in 13 stations.In total, 11 stations' soil drought lagged for more than 3 months, which can be found in nonmajor crop planting area, shown by blue round dots.The correlations between SPIx and SMCIr(10-40 cm) are presented in Fig. 3(b).As to deeper soil depth of 10-40 cm, soil moisture may slowly response to meteorological drought.The results also gave the certification of them, 10 stations' soil drought response time was 4 months to meteorological drought, and the response time larger than 6 months is 15 stations.
In addition, considering that the soil drought index (SM-CIr) developed in the previous section is based on the lowerresolution gridded soil moisture data from GLDAS, there may be produced deviations in the comparative analysis at small scales or within localized areas.Therefore, in this study, we also collected and processed several soil moisture datasets from 5 field observation stations.Similarly, SMCIp derived from field stations was introduced in Section III.   and SMCIp(0-10 cm)/SMCIp(10-50 cm) are 0.36 and 0.34, respectively, while for SPEIx and SMCIp(0-10 cm)/SMCIp(10-50 cm), they are 0.25 and 0.27.The maximum correlation coefficients are very close for SMCIr and SMCIp, both of them are relatively small.One possible reason for this could be the difference in scales between the SM and agrometeorological datasets.It is shown that SMCIp(0-10 cm) has the highest correlation with SPI1 on 1-month time scale, which is the same as SMCIr(0-10 cm).While SMCIp(10-50 cm) shows the highest correlation with SPI2 at a 2-month time scale, which of the lag time is shorter than SMCIr(10-40 cm).The differences in response time for the deep soil drought indices (SMCIr(10-40 cm) and SMCIp(10-50 cm)) could be attributed to variations in the number of used monitoring stations and differences in soil depths Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.considered in the analysis.Fig. 3(c) and (d) show the spatial distribution of response times of SMCIp to SPIx.The drought propagation time in all 5 agrometeorological stations is less than 2 months for SMCIp(0-10 cm).Similarly, to deeper soil drought, the propagation time is longer.There was variable response time in different agrometeorological stations, the possible reasons of which were the climatic conditions, soil attribution, and physical geographical characteristics [42].

B. Response of Vegetation Drought to Meteorological Drought
Vegetation drought may originate from meteorological drought or soil drought, which of proxy are TCI, VCI, VHI, WCI, sNDDI, and sVWSI.In an attempt to improve our understanding of how vegetation responds to meteorological drought, Pearson's coefficients between SPIx/SPEIx and six vegetation drought indices are shown in Fig. 4 with the radar maps.The R is produced by 2995 data pairs in all 30 stations for 20 years.Generally, the R of SPIx and vegetation drought indices are higher than SPEIx in all time scale.Vegetation drought is more likely to occur in the meteorological drought expressed by SPIx.The average R value of vegetation drought indices and SPIx over 30 stations is between 0.17 and 0.55, while the value of vegetation drought indices and SPEIx is between 0.06 and 0.28.Among six vegetation drought indices, VHI is more relevant to SPIx with the R highest than others indices obviously.
VHI is the optimal vegetation drought index among the six indices mentioned for Kyrgyzstan.Further to vegetation drought's response to meteorological drought, it was identified by the maximum R between VHI and SPIx presented by correlation

C. Response of Vegetation Drought to Soil Drought
Vegetation drought is caused by soil water deficit, and water deficit results in mismatching of water supply and the water required for vegetation growth.Then vegetation will adapt it and show responsive symptoms and characteristics.
Here, SMCIr(0-10 cm) related to VHI in different lag time among the whole Kyrgyzstan was shown in Fig. 6(a) (lag0 is below 1-month lag, lag1 is between 1-month lag and 2-month lag, lag2 is between 2-month lag and 3-month lag).The results are that the R of VHI and SMCIr(0-10 cm) in lag0 is highest compared with lag1 and lag2.The response time of VHI to SMCIr(0-10 cm) is within 1-month.month, and indicating that VHI is fast response to changes in soil moisture at a shallow depth (0-10 cm) and further deeper soil (10-50 cm).The different results observed at the Talas station could be attributed to the insufficient representativeness of the station data.It is possible that the point location's water stress cannot fully influence the vegetation growth at the pixel scale represented by VHI.This may lead to variations in the correlation and response time between the vegetation drought and soil drought.

D. Three Kinds of Drought Spatial Patterns in Typical Years
The study utilized the SPI and SPEI indices, derived from monthly station data, to assess the spatial distribution of drought intensity in Kyrgyzstan.The investigation focused on the spatialtemporal variation and trends of these indices during the growing season (May to September) from 1981 to 2019, encompassing the entirety of Kyrgyzstan.To generate a comprehensive understanding of drought patterns, the SPI and SPEI distribution maps for the growing season were calculated.This involved processing monthly precipitation and air temperature data collected from 30 meteorological stations across Kyrgyzstan.Then the point values were skillfully interpolated into raster data using the inverse distance weighted (IDW) method.This interpolation technique considers neighboring data points and assigns appropriate weights based on their proximity to the target location.
Based on records from Kyrgyzhydromet, the years 2014 and 2016 stood out as significant in terms of meteorological conditions, with 2014 being remarkably dry and 2016 experiencing excessive rainfall.The previous research [26] and the invaluable insights of local farmers experiences corroborated that severe impact of drought on Kyrgyzstan's cropland and grassland in 2014, a year characterized by heavy agrometeorological and meteorological drought.We selected two typical years, namely 2014 and 2016, as illustrative examples.For consistency with previous research and the observations of farmers, we focused on two specific meteorological drought indices, SPI5 and SPEI5, covering the entire crop growing period, as listed in Table II.Fig. 8 demonstrates that the SPI5 effectively captures drought and accurately portrays the contrast between dry and wet periods across the entire country from 2014 to 2016.
Fig. 9(a)-(c) presents the changes in SMCIr(0-10 cm) from June to August in 2014, which was a dry year characterized by an increase in drought severity and an expansion of the drought's extent.By July, almost the entire region experienced drought conditions.However, in 2016, a wet year [as seen in Fig. 9(d)-(f)], Kyrgyzstan encountered only mild drought conditions, with a limited scope of drought observed.Only the central region exhibited partial drought.It should be noted that the detailed strip area in the north might have been affected by anomaly soil moisture data from GLDAS, while several specific block areas were influenced by the coarse resolution of the data.Fig. 10(a)-(c) illustrates SMCIr(10-40 cm) changes from June to August in 2014 of dry year, with the severe drought happened in August.The drought conditions derived from SMCIr(10-40 cm) differed from those obtained from SMCIr(0-10 cm).SMCIr(10-40 cm) exhibited a slower response compared to SMCIr(0-10 cm) due to deeper soil layers and soil moisture movement process.In the contrasting conditions of 2016, a wet year, the soil moisture was abundant and there were no significant changes for drought severity in whole Kyrgyzstan from June to August.Some patch areas still displayed drought, particularly in the case of severe or extreme drought, which can be attributed to anomalous situations.Drought category of SMCIr was the same as VHI, but in fact, it had differences due to the water demand diversity of different crops in variable Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.growing period.Furthermore, the coarse resolution of GLDAS data contributed to some limitations in the analysis.
Next, the vegetation drought distribution maps proxy to VHI from June to August of two typical years are showcased in Fig. 11.Notably, these maps reveal a growing trend of intensified vegetation drought, encompassing a wider area, during the continuous occurrence of meteorological and soil drought in 2014.From July to August, a crucial period for summer grains, extensive regions of vegetation faced severe drought conditions.The VHI map effectively captures spatial variations and offers detailed information, providing valuable insights into the extent and severity of vegetation drought.In contrast, during the wet year of 2016, the VHI map shows only a few minor pixels with slight drought in June.From July to August, no vegetation drought was observed, which indicates a favorable condition for vegetation growth during this period.

E. Effects of Three Kinds of Drought on Crop Yield Index
Evaluating the impact and contribution of different types of drought on crop yield through nonlinear fitting of the determination coefficient (R 2 ) between variables is an effective method.
A higher R 2 value, closer to 1, indicates that the relation model can effectively explain the variation in crop yield, implying a more significant impact of the respective drought index on crop production.Conversely, a lower R 2 value, closer to 0, indicates weaker explanatory power of the model in explaining variations in crop yield, suggesting a lesser influence of the respective drought index on crop production.In order to minimize the influence of factors other than climate on crop yield, a novel spatial-temporal standardized crop yield index (STSYI) was proposed.This index with province scale was derived using a comprehensive dataset of long-term crop yields spanning from 2000 to 2018, computed through (11) and (12).Fig. 11 presents the determination coefficients between the STSYI and SPI for the seven oblasts (provinces) of Kyrgyzstan during the months of May to September.Specifically, SPI5 covered information in key crop growing period was put to use.After preliminary testing, the three optimal SPI5 indices were selected for a detailed analysis of their impact and contribution to yield variations seen in Fig. 12.These three SPI5 indices are as follows: SPI5 in July, August, and September.The box plot lengths vary significantly, indicating substantial differences in the explanatory power of the three meteorological drought indices on the grain yield index across different stations.Notably, Talas station consistently shows the lowest explanatory ability.Overall, SPI5_Aug.exhibits better performance, with an average explanatory power of 36% regarding grains yield variations.Among the factors contributing to wheat yield variations, SPI5_Aug.can explain 20% to 70% of the yield variability, with an average of 35%.When incorporating meteorological information for September (SPI5_Sep.),the determination coefficient for yield does not increase.This is because wheat has reached maturity in September, and the yield is already established before that time.For barley, the box plot lengths are relatively short, indicating minor differences in the explanatory power of the three meteorological drought indices on the barley yield index across different stations.Overall, SPI5_Aug shows better performance, with an average explanatory power of 43%, which contributes the most to the variability of barley yield compared to other crops.The impact of the three meteorological drought indices on corn yield is relatively small.Although SPI5_Jul.performs better than the other two indices, the average explanatory power for yield variations is low at 23%.Overall, meteorological drought indices have a limited effect on corn yield in Kyrgyzstan.One possible reason is the significant variation in irrigation, corn cultivation areas, and phenology among different provinces.Therefore, accurate crop distribution and phonological information are crucial for quantitative assessments.
When rainfall shorted or irrigation reduced, the meteorological drought propagated to soil drought, further influenced on the crops yield.Both soil moisture stress in the surface and deeper layers can impact crop yield.In this study, the weighted average of SMCI for the depths of 0-10 cm and 10-40 cm was combined to create SMCIr(0-40 cm), aiming to comprehensively reflect the overall impact on crop growth.In addition, we established three different time aggregations to characterize the water information during the main growth period of crops, namely from June to August, from May to August, and from Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.May to September.Fig. 13 presents the contribution rates of these three aggregation methods to the variations of four crop parameters.The results were showed that SMCIr(0-40 cm) from May to September can better predict the crops yield compared to other two aggregations, which average of R 2 was 0.48, 0.48, and 0.58 for grains, wheat and barley in all provinces.It means that SPI5 has the ability to explain or reflect crop yield changes about 50% or 60%.But corn yield variations cannot be very consistent to soil drought index changes, with average of R 2 was 0.25.One possible explanation is that SMCIr(0-40 cm) is derived from soil moisture data of coarse resolution(0.25°),resulting in many mixed pixels due to complex and high heterogeneity of farmland.As a consequence, its ability to accurately represent soil drought on cultivated land is limited.
Based on the analysis in the earlier sections of this article, it is evident that VHI is the optimal vegetation drought monitoring index for Kyrgyzstan.Here, we will further analyze how this index affects different crop parameters.Considering that vegetation drought can have a cumulative effect on crop yield, especially when aggregated at the provincial level, the proportion of cumulative drought is a better reflection of its impact on crop yield.VHI was accumulated over whole crop growing period named as VHIacc(May-Sep.),and over key crop growing period named as VHIacc(Jun.-Aug.).Fig. 14 illustrates the response capability of different crops' yield indices to VHIacc(May-Sep.) and VHIacc(Jun.-Aug.).For grains, wheat, and barley, VHIacc(May-Sep.)slightly outperforms VHIacc(Jun.-Aug.), but corn yield variation shows a weaker response to vegetation drought.In the box plot, there are extreme high values within the box of grains.This is due to the significantly higher coefficient of determination (R 2 ) resulting from the nonlinear fit between grain of STSYI and VHI in Chuy province, reaching 0.84.The variation in grain crop yield in Chuy province can be explained by 84% due to crop drought conditions (VHIacc(May-Sep.)).This indicates that a significant portion of the changes in grains yield in Chuy province can be attributed to the impact of vegetation drought, and the VHIacc(May-Sep.)has good predictive capability for grains yield variation in this region.
For meteorological drought, soil drought, and vegetation drought indices, the indices that contribute the most to yield variations are SPI5_Aug.,SMCIr(0-40 cm)_May-Sep., and VHIacc(May-Sep.),with average contribution rates of 34%,    48%, and 45% across all provinces in the country.The vegetation drought index(VHIacc(May-Sep.)) can directly express the crop drought condition and its impact on yield accumulation, but soil drought (SMCIr(0-40 cm)_May-Sep.)also has directly and highly contribution capabilities for regional scale.Different types of drought indices have their own advantages, and there are variations in their contributions to the changes in crop yield for different crops and different provinces.Matrix of the maximum determination coefficients between drought index and STSYI for four crops and seven provinces were shown in Fig. 15.Among the 28 combinations in this figure, there are 6 combinations with the strongest explanatory power for yield variation by SPIx, 14 combinations with the strongest explanatory power by SMCIx, and 8 combinations with the strongest explanatory power by VHIx.Specifically, the most influential factor for yield variation differs among different provinces.The strongest influential factor is VHIx in Batken, Chuy, and Jalal-Abad provinces, while it is SMCIx in Osh, Talas, and Yssyk-Kul provinces, and SPIx has a stronger impact on crop yield variations in Naryn province.Moreover, as to grains and wheat yield variation, the dominant factors are SMCIx and VHIx, it is VHIx and SMCIx for barley, and it is SMCIx and SPIx for corn.
In this article, we focus on analyzing relationships of meteorological drought, soil drought, vegetation drought in Kyrgyzstan.Our findings reveal that soil drought (SMCI) at a 10 cm depth responds to SPI with a 1-month time lag, while soil drought at a 40 cm depth responds to SPI with a 4-month time lag.The VHI showed a 1-month response time to soil drought, irrespective of whether the soil drought occurred at a depth of 10 or 40 cm.However, in general, the VHI exhibited a 3-month response time to SPI.
These results of the study may play a significant role in drought prevention and mitigation in drought-prone countries of Central Asia.Synergism of VHIx, SMCIx, and SPIx may effectively increase the prediction of crop yield in Kyrgyzstan for the next study.VHI mentioned was just average of TCI and VCI, which is whether adaptation needs to be studied in different climatic regions [43].Coarse-resolution soil moisture may result Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
in some errors, higher resolution data was necessary and should be produced in the future.

V. CONCLUSION
To identify the optimal index for accurate drought monitoring and crop yield prediction in Kyrgyzstan, we conducted a comprehensive analysis of the interconnections between meteorological drought, soil drought, and vegetation drought.We selected multiple drought indices for each type and compared and evaluated their performance over complex terrain.The results revealed that there are differences in the propagation and response times of the three types of drought indices.The optimal drought indices for the three categories are SPI5_Aug.,SMCIr(0-40 cm)_May-Sep., and VHIacc(May-Sep.),respectively.
Key findings are as follows.
1) The SMCIr at 0-10 cm depth demonstrated a 1-month response time to SPI, while the SMCIr at 10-40 cm depth exhibited a 4-month response time to SPI. 2) The propagation time of VHI to soil drought was similar, occurring within 1 month, whether at a 10 cm depth or 40 cm depth.The VHI showed a 3-month response time to SPI.
3) The impacts of the three types of drought indices on crop growth and yield remain differential and complex.Though the VHI emerged as the most effective index for describing crop drought and predicting yields in most provinces.Specifically, the accumulation of VHI during key crop growing periods accurately reflected crop yield variations in the majority of provinces.At the same time, SMCIr(0-40 cm)_May-Sep.and SPI5_Aug.have different contributions to crop yield variations, and these are also differences in their impacts on different crops and provinces.The findings can provide valuable insights for agricultural planning and drought management in the region.Additionally, we recommend further research to explore the adaptability of the VHI in different climatic regions and the need for higher-resolution soil moisture data to establish a comprehensive drought monitoring model to predict crop yield.

Fig. 6 (
b) presented the relations of VHI and SMCIr(10-40 cm) in different lag time.Similar results can be found, which is that the response time of VHI to SMCIr(10-40 cm) is also within 1 month.Fig. 7 displays the correlations between station-based SMCIp(0-10 cm)/SMCIp(10-50 cm) and VHI in different lag time among 5 Kyrgyzstan' stations.The findings indicate that the correlation coefficient between VHI and SMCIp (0-10 cm) at lag0 is the highest except for Talas station.The response time of VHI to SMCIp(0-10 cm) and SMCIp(10-50 cm) is within 1

TABLE I LONG
TIME SERIES DATA USED IN THE STUDY

TABLE III CATEGORIZATION
OF DROUGHT ACCORDING TO THE SPI/SPEI i is a certain month; VCI i is the vegetation condition index in a certain month; NDVImax and NDVImin are the historical maximum and minimum values of NDVI, respectively.TCI i is the temperature condition index in a certain month; LSTmax and LSTmin are the historical maximum and minimum values of the LST, respectively.WCI i respects the moisture condition index in a certain month; NDWImax and NDWImin are the maximum and minimum values of NDWI on the long-term sequence, respectively.sNDDI i is the scaled NDDI in a certain month; NDDImax and NDDImin are the maximum and minimum values of NDDI on the long-term sequence, respectively.sVWSI i is the scaled VSWI in a certain month; VSWImax and VSWImin are the maximum and minimum values of VSWI on the long-term sequence, respectively.