Linking drought indices to impacts to support drought risk 1 assessment in Liaoning province , China 2

Drought is a ubiquitous and reoccurring hazard that has wide ranging impacts on society, agriculture and the 10 environment. Drought indices are vital for characterizing the nature and severity of drought hazards, and there have been 11 extensive efforts to identify the most suitable drought indices for drought monitoring and risk assessments. However, to date, 12 little effort has been made to explore which index(s) best represents drought impacts for various sectors in China. This is a 13 critical knowledge gap, as impacts provide important ‘ground truth’ information. They can be used to demonstrate whether 14 drought indices (used for monitoring or risk assessment) are relevant for identifying impacts, thus highlighting if an area is 15 vulnerable to drought of a given severity. The aim of this study is to explore the link between drought indices and drought 16 impacts, using Liaoning province (northeast China) as a case study due to its history of drought occurrence. To achieve this 17 we use independent, but complementary, methods (correlation and random forest analysis). Using multiple drought indices – 18 Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI), Soil Moisture (SoilM) 19 and the Normalized Difference Vegetation Index (NDVI) – and drought impact data (on crop yield, livestock, rural people and 20 the economy) correlation and random forest analysis were used to identify which indices link best to the recorded drought 21 impacts for cities in Liaoning. The results show that the relationship varies between different categories of drought impacts 22 and between cities. SPEI with a 6-month accumulation (SPEI6) had a strong correlation with all categories of drought impacts, 23 while SPI12 had a weak correlation with drought impacts. Of the impact datasets, ‘drought suffering area’ and ‘drought impact 24 area’ had a slightly strong relationship with all drought indices in Liaoning province, while ‘population and number of livestock 25 with difficulty in accessing drinking water’ had weak correlations with the indices. Based on the linkage, drought vulnerability 26 was analyzed using various vulnerability factors. Crop cultivated area was positively correlated to the drought vulnerability 27 for five out of the eight categories of drought impacts, while the total population had a strong negative relationship with drought 28 vulnerability for half the drought impact categories. This study can support drought planning efforts in the region, and 29 provides a methodology for application for other regions of China (and other countries) in the future, as well as providing 30 context for the indices used in drought monitoring applications, so enabling improved preparedness for drought impacts. 31 https://doi.org/10.5194/nhess-2019-310 Preprint. Discussion started: 30 September 2019 c © Author(s) 2019. CC BY 4.0 License.

occurs, as has frequently been the case in Liaoning province, it causes a significant reduction in agricultural production (Yan et al., 2012). According to the SFDH, between 2000 and 2016 the average annual yield loss due to drought was 1.89 million 6 20cm and 30cm using frequency domain reflection soil moisture sensors, which are based on the principle of electromagnetic 137 pulse. Soil moisture data were not available between November and February at most stations due to freezing conditions.
(http://www.gscloud.cn/); the daily maximum data were used to derive the monthly average NDVI. In contrast to many other countries, China has a systematic, centralized drought impact information collection system. Drought 143 statistics include drought impacts, drought mitigation actions and benefits of action to agriculture, hydrology and civil affairs.

144
During a drought event, impact statistics are collected from every day to every three weeks, according to the drought warning 145 level (Wang, 2014). When a drought warning is not triggered, drought impact data are collected after an event has ended which 146 could be several months afterwards; and no data are collected when there is no drought event. Statistics for eight drought 147 impact types were collected from the SFDH between 1990 and 2016, and aggregated to annual totals, the impact types used 148 are listed in Table 1.

164
The SPI, in its default formulation, assumes that precipitation obeys the Gamma (Γ) skewed distribution, which is used to 165 transform the precipitation time series into a normal distribution.

178
Using the daily soil moisture of 10 cm, 20 cm and 30 cm depths, the daily average soil moisture for each station was calculated   10  , 20  and 30  are the measured value at different depths 183 (10cm, 20cm and 30cm).  is the average soil moisture.
i h is the thickness of the i-th layer of soil, and H is the total 184 thickness of the measured soil.

185
Some of the daily soil moisture data were missing, however, this was limited to 17% of total soil moisture data. In some cases 186 there were missing data for one depth of soil moisture measurement. In these cases, the average soil moisture of the other two 187 layers was calculated, and where there was only one layer of soil moisture available it was used to represent the average soil 188 moisture. The annual average soil moisture was calculated based on the available daily soil moisture (March to October) and 189 was analyzed with the annual drought impact data during 1990 to 2006. As each city has more than one station, the annual soil 190 moisture of each station was calculated and then averaged into one value for each city.

191
The area-averaged NDVI at city unit was calculated based on the monthly NDVI. The critical stages of the spring maize growth 192 in Liaoning is in July, so the area-averaged NDVI in July was selected for the analysis with the annual drought impacts during

247
The Pearson correlation coefficient (r) for each city and drought impacts is shown in Figure 4. In most cases the drought index 248 is negatively correlated with the drought impacts, suggesting that the lower the drought index, the greater drought impact.

249
However, correlation strength, and direction, varied between the cities and impact types, ranging between -0.890 to 0.621. In

264
The performance of soil moisture varied significantly between cities and impact types ( Figure 4); it had a strong correlation 265 with the impacts in Chaoyang, and a weak correlation in Huludao. In Chaoyang, the correlation between soil moisture and 266 drought impacts was significant (α=0.01), whilst other cities were not significantly correlated.

268
Each drought impact type was selected as the response variable in the random forest. On average the random forests explained 269 41% of the variance observed within the drought impacts. The MSE% for each city and impact type is shown in Figure 5.

278
The variables identified MSE% from the random forest analysis generally match those with strong negative correlations. This 279 supports the statement that indices are negatively related to impacts. The threshold of impact occurrence based on the indices 280 were also identified in the RF analysis using the first splitting value. Figure 6 shows the distribution of first splitting values of 281 each decision tree within the RF. The average first splitting values for SPI18 and SPI24 were higher than those of SPI6, SPI12 282 and SPI15 (i.e. a more negative index value and more severe meteorological drought state) for all categories of drought impacts.

283
For SPEI, the results were similar (i.e.long-term deficits must be more severe to result in equivalent impacts compared to short-284 term deficits) but there was more variability between accumulations. When viewed in terms of impact types, DSA had a low 285 threshold, indicating that DSA impacts occur more readily than DA or RA, as may be expected. The impact occurrence of 286 index values increase for DSA, DIA, DA and RA; and YLD and DELA tended to occur for more severe water deficits, with     severity with SPEI6 equals -1.5 were applied to measure the drought vulnerability.

324
The methodology in this research has the following characteristics. Firstly, it takes many drought impacts, across a range of 325 sectors, into consideration. Secondly, the extensive drought impact data were systematically collected at county level, which 326 is a consistent and reliable data source enabling regional comparisons. The drought impact data used here included impact 327 variables that are rarely available in other studies such as population with difficulty in accessing drinking water, number of 328 livestock with difficulty in accessing drinking water, yield loss due to drought and direct economic loss in agriculture. Thirdly,

329
we not only considered the occurrence of drought events, but also the severity of drought and its spatial extent. Finally, the 330 drought indices-impacts linkage was applied to assess drought vulnerability in Liaoning province.

331
The biggest challenge of this study was the spatial and temporal matching between the drought impacts and indices. Drought   and NLH are related to many factors, such as drinking water source location and the quality of water resources, for example, 353 livestock can drink water from the river directly, but the water quality of the river cannot meet the human drinking needs. For 354 this reason, NLH showed least sensitivity to water deficits.

355
The random forest algorithms presented in this paper explained an average of 41% of the variance observed within the drought 356 impact data. This is relatively modest, because of the limitation of the impacts data. Collinearity of the drought indices (e.g.

357
SPI6 is correlated with SPEI6) is also a potential cause of the low MSE%. The correlation coefficients calculated for drought 358 indices and NLH in Yingkou, and PHD in Fushun were positive. This result is unexpected given the interpretation of these 359 indices as estimations of the drought severity, and the majority of reported correlation coefficients being negative. Therefore,

360
it seems likely this result is not representative of the true relationships between these indices and impacts, and instead an 18 representative of the true relationships. The availability of more data would enable a better approximation of the true 364 relationships between indices and impacts.
For all the drought impacts, Dalian and Fuxin showed the highest correlation coefficients among drought impacts and drought 366 indices in all cases. The most vulnerable cities were Fuxin, Tieling, Chaoyang, Jinzhou and Shenyang, which are all located 367 in the northwestern part of Liaoning province indicating there is a high drought vulnerability and drought risk in northwestern The authors declare they have no conflict of interest.
were supported by the NERC National Capability Official Development Assistance project SUNRISE ("Sustainable Use of Fukuda, S., Spreer, W., Yasunaga, E., Yuge, K., Sardsud, V., and Müller, J.: Random Forests modelling for the estimation of