Identifying a Transition Climate Zone in an Arid River Basin using a Hydrological 1 Drought Index

Libo Zhang 1 , Yongqiang Liu 2 , Lu Hao 1 , Decheng Zhou 1 , Cen Pan 1 , Peilong Liu 1 , Zhe Xiong 3 , 4 Ge Sun 4 5 6 1 Jiangsu Key Laboratory of Agricultural Meteorology, International Center for Meteorology, 7 Ecology, and Environment, College of Applied Meteorology, Nanjing University of Information 8 Science and Technology, Nanjing, China 9 2 Center for Forest Disturbance Science, USDA Forest Service, Athens, Georgia, USA 10 3 Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China 11 4 Eastern Forest Environmental Threat Assessment Center, USDA Forest Service, Raleigh, North 12 Carolina, USA 13 14

conclude that the land-surface processes and human disturbances play an important role in altering hydrological drought conditions and their spatial and temporal variability.

Introduction
Large river basins at continental and sub-continental scales usually encompass multiple climate types related to complex topography and landscape.Climate is more humid in the upper basin near the river origins with high elevations and forest and / or permanent snow cover than the lower basin with low elevations and less vegetated lands.Climate could be extremely dry in parts of a watershed under a prevailing atmospheric high pressure system.The sub-continental Colorado River watershed, for example, is dominated by cold and humid continental climate in the upper basin of the Rocky Mountains and cold semi-arid or warm desert climate in the lower basin of the southern inter-mountains.This feature of multiple climate types is also seen in some smaller basins.The Heihe River Basin (HRB) in northwestern China, for example, has an area of 130, 000 km 2 with annual precipitation varying dramatically from about 500 mm in the upper basin of the Qilian Mountains with forest-meadow-ice covers in the south to less than 100 mm in the lower basin of the Alxa High Plain with Gobi and sandy lands in the north.Climate types change from cold and humid continental to arid desert, accordingly.
The relative high precipitation in the humid upper basin supports forests and meadows and provides source water lower reaches of the Heihe River.In contrast, water is a major limitation factor in arid lower basin.In addition, more extreme weather conditions, especially droughts, occur in arid lower basin.In the Colorado River basins, the reconstructed data show decadal periods of persistently low flows during the past centuries (Woodhouse et al., 2010).The drought severity in the new millennia has been the most extreme over a century (Cayan et al., 2010).The reconstructed precipitation series in the HRB indicates that droughts were much more frequent and lasted longer than floods in the past two centuries (Ren et al., 2010).Droughts occurred more often in the dry lower basin than the humid upper basin (Li, 2012).
The watersheds with varied topography and landscape may have a transition climate zone between the two zones.In the HRB, for example, the Koppen climate classification (Peel et al., 2007) shows polar tundra or boreal climate in the upper basin of the mountain regions in the south, arid desert climate in the lower basin in the north, and a transition zone of steppe climate  (Budyko, 1974) is one of the widely used drought indices (UNESCO, 1979;Wolfe, 1997;Onder, 2009).It is also used as an essential element in many other indices to describe actual drought conditions (Arora, 2002).The AI defines the dryness degree based on the relative magnitude of water demand and supply and calculated based on average precipitation and potential evapotranspiration, which is in turn determined by temperature.The AI therefore is a meteorological drought index.
Hydrological drought indices such as evaporative stress index (ESI) are also widely used in climate and hydrological studies.This type of indices is determined not only by meteorological conditions but also the surface hydrological properties and processes such as soil moisture, groundwater, and evapotranspiration etc.The ESI defines the dryness degree of a region based on the ratio of actual (AET) to potential evapotranspiration (PET).A relatively low ESI indicates water limitation to plants and the actual rate is way below the PET.In contrast, a relatively high ESI indicates freely available water with the AET rate approaching or close to the PET.The ESI has been long used to evaluate the irrigation need for crop growth and land classification (Yao 1974).The ESI has been used recently to evaluate water stress using remotely sensed hydrological and ecological properties (Anderson et al., 2016).This study is to understand the two different drought indices of AI and ESI in their capacity in identifying the transition climate zone in the HRB.It was made mainly by comparing the spatial patterns and regional averages.Their temporal variations were also analyzed to understand the differences in the seasonal and inter-annual variability and longterm between the meteorological and hydrological drought indices.The data from a highresolution regional climate modeling were used.

Study region
The study region was the HRB and the adjacent areas (Fig. 1).The Heihe River origins from the Qilian Mountains in the northern edge of the Tibet Plateau and flows northward to the China-Russian border.The HRB spans between 98°~101°30′E and 38°~42°N.The upper HRB is within the mountains elevated 2300~3200m mainly covered with forests and mountain meadows.The middle HRB is along the Hexi Corridor elevated 1600~2300m mainly covered with piedmont steppe grass, crops, and residence and commercial uses.The lower HRB is in the Alxa High-Plain elevated below 1600m mainly covered with Gobi and desert sands.Annual precipitation is over 400mm in the upper basin, with the maximum of 800mm at extremely high elevations, about 100~250mm in the middle basin, and below 50mm in many lower basin areas.The annual precipitation in the upper basin has high seasonal variability, and nearly 70% of the total annual rainfall occurs from May to September (Gao et al., 2016).
The upper basin generates nearly 70% of the total river runoff, which supplies agricultural irrigation and benefits the social economy development in the middle and lower basin reaches (Yang et al., 2015;Chen et al., 2005).Annual mean temperature is about ̶ 4 o C in the upper basin, 7 o C in the middle basin, and nearly 9 o C in the lower basin.

Drought indices
The meteorological drought index is defined as AI = P / PET, where P and PET are daily precipitation and potential evapotranspiration, respectively.AI is a variant of the index originally defined by Budyko (1974), which is the ratio of annual PET to P. The average AI values were used to classify the arid, semi-arid, semi-humid (sub-humid), and humid climate with the ranges of AI ≤ 0.2, 0.2 <AI ≤ 0.5, 0.5 < AI ≤1.3, and AI > 1.3, respectively (Ponce et al., 2000).
The hydrological drought index is defined as ESI = AET / PET, where AET is daily actual evapotranspiration.The ranges of average ESI values of ESI ≤ 0.1, 0.1 < ESI≤ 0.3, 0.3 < ESI ≤ 0.6, and ESI > 0.6 were used to classify the arid, semi-arid, semi-humid, and humid climate, respectively (Yang, 2007).This approach agrees with Anderson (2011), which showed that the ESI values varying gradually from 0 to 1 correspond to several USDM drought levels from exceptional to no drought for each month from April to September across the continental U.S.
Two methods were used to estimate PET (mm/d).One was the energy balance based FAO-56 Penman-Monteith Equation (Allen et al. 1998): where   and G are net radiation and soil flux on the ground (MJm -2 d -1 ); T is air temperature ( o C); e s and e are saturation and actual water vapor pressure (kPa); u 2 is wind speed at 2m above the ground (ms -1 ); Δ is the rate of change of e s with respect to T (kPa/ o C); γ is the psychrometric constant (kPa/ o C).The other method is the temperature based on Hamon formula (Hamon, 1963): where k is proportionality coefficient = 1; N is daytime length.e s is in 100 Pa here.
Monthly PET, precipitation and actual evapotranspiration, obtained based on daily values, were used to calculate the drought indices.It was assumed that daily PET=0 if daily T<0 o C. Their monthly PET was not used if PET=0 for more than 10 days in a month.In this case, no The data used in calculation and evaluation of the drought indices are listed in Table 1.

Regional climate modeling
The climatic and hydrological data used to calculate the drought indices were created from a regional climate modeling using the Regional Integrated Environmental Model System  The spatial pattern of the simulated annual T averaged over the simulation period is featured by the large changes between basin reaches, increasing from about -15 o C in the tall mountains of the upper basin to over 10 o C in the deserts of the lower basin (Fig. S1).The simulated average annual P shows an opposite gradient, decreasing from about 2.5 mm/d in the mountains to less than 0.25 mm/d in the deserts (Fig. S2).The simulated average annual AET has a similar pattern to precipitation (Fig. S3).The spatial variability is much larger within the upper basin than the lower basin.

Simulated climate and hydrology
An interesting feature is that both T and P in the middle basin are very close to their corresponding values in the lower basin but much different from those in the upper basin; the AET difference between the middle and upper basin reaches however is much small.
As expected, the regional AET values averaged over the simulation period are higher in summer than in winter (Fig. S4).In the upper basin, for example, T increases from about - The inter-annual variability of regional T and P is similar between the middle and lower basin reaches (Fig. S5).A few dry years (e.g., 1990, 2001, and 2008) and wet years (e.g., 1981, 1989, 2002, and 2007) can be found.The amplitude of variability is larger for P than T, especially in the upper basin.The variability of AET is also similar between the lower and middle basin reaches, but it differs from that in the upper basin during some periods (e.g., around 1985).The differences in AET between the middle and upper basins are much smaller in the magnitude than those for the meteorological properties.
The above features of close values and similar inter-annual variability in the simulated T and P between the middle and lower basin reaches are also seen in the observations (Fig. S6).
The simulated T in all basin regions and P in the middle and lower basin reaches are close to the observed ones.However, the simulated P is about 0.4 mm/d higher (about 1.6 mm/d for simulation vs. 1.2 mm/d for observation).The weather site in the upper basin is located in relatively flat and low valley, while the simulation grids have many points at high elevations where P is larger than at the valley locations.
The simulated P increases around 50% over the simulation period, statistically significant at p<0.01in all basin reaches (Table S1).The simulated AET also increases, but at a smaller degree of around 20% and p<0.01only in the upper basin.The simulated T shows increasing trends, but insignificant in all reaches.The simulated P trends are close to the observed ones in the middle and lower basin reaches, but opposite to that in the upper basin.The simulated T underestimates the observed warming, which was about 2 o C at p<0.01.

Spatial patterns of drought indices
PET calculated using the Penman-Monteith method is mostly 1.7-2.25 mm/d in the upper basin (Fig. 2).It increases to above 3 mm/d in the middle and lower basins.There is little difference between the two regions.The meteorological drought index, AI, shows a similar pattern but opposite gradient (Fig. 3).It is as large as 1.4 in the upper basin, but reduced to less than 0.2 in two other basin regions, indicating increasing aridity from the upper to lower basin.The hydrological drought index, ESI, has the same gradient as AI, but with different spatial pattern (Fig. 4).It is as high as 0.9 in the upper basin and reduced to mostly below 0.1 in the lower basin.However, the values in the middle basin is as high as 0.6, much larger than that in the lower basin.PET calculated using the Hamon method has the same pattern as the one using the Penman-Monteith method, but with smaller magnitude (Fig. 5).PET is mostly about 1 mm/d in the upper basin and increases to about 1.5-1.75mm/d in the middle basin, and further to 1.75-2.25 mm/d in the lower basin.
The different spatial patterns between AI and ESI seen above are also found for the Homan method.AI is mostly above 0.6 in the upper basin (Fig. 6).It is below 0.2 in the middle and lower basins without apparent differences between the two regions.In contrast, while ESI remains large values of mostly above 0.9 in the upper basin and low values of below 0.2 in the lower basin, the values in many areas of the middle basin are 0.4-0.9,much different from those in the lower basin (Fig. 7).

Seasonal cycle
For the Penman-Monteith method, PET is the highest in summer and smallest in winter (Fig. 8).Note that winter PET in the upper basin is not shown because T is below zero in too many days.The amplitude in the middle basin is close to that in the lower basin, but much larger than that in the upper basin.Different from the upper basin where AI and ESI are also the largest in summer, AI is the largest in fall, while ESI is the largest in winter in the middle basin (as well as lower basin).The seasonal variations of PET, AI and ESI estimated using the Homan method are similar to those using the Penman method.

Inter-annual variability
PET in the middle basin calculated using the Penman-Monteith method shows similar interannual variability over the period of 1980-2010 to that in the lower basin, but much different from that in the upper basin (Fig. 9).The standard deviation (SD) increases from the upper (0.12) to middle (0.21) and to lower basin (0.29) (Table 2).The coefficient of variation (CV)

Long-term trends
PET shows little trends over the simulation period (Table 3).On contrast, drought indices increased dramatically, by 60% or more for AI and 15-50% for ESI.The trends are significant at p<0.01 in the upper and middle basin reaches and p<0.05 in the lower basin.

Extreme events
The drought indices for 4 simulated dry years (1982, 1990, 2001, and 2008) and 4 wet years (1981, 1989, 2002, and 2007)   The HRB is a typical inland river basin with a strong contrast in topography, landscape, climate, and human activities from the headwater to end point along its drainage system.The water shortage and frequent droughts are the biggest environmental threat to the ecosystems and human activities in this region as well as entire northwestern China.The HRB is well studied for understanding the interactions among eco-hydrological and socio-economical processes.This comparison study provides evidence for the importance of water and energy interactions between land process and the atmosphere and between upstreams and downstreams in determining climate types in an arid climate.
Because the ESI values are related to AET that is controlled by land-surface properties and management practices (e.g., rainfall-fed crops vs irrigated crops; natural wetlands vs cultivated drained croplands), our results suggest the land-surface processes play an important role in affecting drought conditions and their variability.The landscape in the HRB, especially its transition zone, has changed remarkably in the past several decades due to urbanization, farming, and grazing activities (Hu et al., 2015).The irrigation may have caused the lower basin more water stressed (higher ESI than AI) since stream water from Heihe is intercepted and rivers go dry downstreams.The ESI should reflect this change since it is calculated partially based on the land-surface hydrological conditions.The regional landatmosphere coupled models would provide proofs for this hypothesis though modeling the impacts of land cover change, which is a driver of local land-atmosphere interactions.

Role in moderating climate
The magnitude of AI (ESI) inter-annual variability in the middle basin is (in not) very close to that in the lower basin, another evidence for the unique capacity of ESI in separating the climate zones between the middle and lower basin reaches.The magnitude of the relative inter-annual variability differs mainly between AI and ESI, larger with AI.In addition, both AI and ESI in the HRB decreased dramatically from 1980 to 2010, at greater rate with AI.
Thus, the drought conditions described using ESI is less variable, suggesting the role of local hydrological processes in moderating extreme climate events.

Future trends
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2017-310Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 16 January 2018 c Author(s) 2018.CC BY 4.0 License. in the middle.Identifying this transition zone and understanding its unique climate features are of both scientific and management significance.The complex topography in upper basin and harsh climate in lower basin make both regions unsuitable for human living.The transition zone however is relatively flat in comparison with the mountain region and less arid in comparison with the dryland region.It therefore provides a favorable condition for industrial and agricultural development.Also, the environmental conditions in this region are more dynamical and localized because of human induced rapid and fragmental landscape changes.The Koppen climate classification, one of the most widely used climate classification techniques at large geographic scales and constructed based on the properties of ecosystems, latitude, and average and seasonal precipitation and temperature, is often used for a large region with static environmental conditions.Drought indices are another useful tool to classify and monitor aridity and drought of a region.Drought indices are mainly determined by precipitation, and the ground surface temperature and evapotranspiration, all of which are strongly affected by local landscape and topography.The aridity index (AI) Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2017-310Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 16 January 2018 c Author(s) 2018.CC BY 4.0 License.Many studies have compared various types of drought indices in different climatic environments.Otkin et al. (2013) compared the ESI with drought classification used by the U.S. Drought Monitor (USDM) (Svoboda et al., 2002) and found that the ESI anomalies led the USDM drought depiction by several weeks and large ESI anomalies therefore were indicative of rapidly drying conditions.This finding was coincident with the droughts occurred across the United States in recent years.Choi et al. (2013) compared the ESI with the Palmer drought severity index (PDSI) in a watershed of the Savannah River branch in southeastern United States during 2000-2008.They found that the ability of the ESI to capture shorter term droughts was equal or superior to the PDSI when characterizing droughts for the watershed with a relatively flat topography dominated by a single land cover type.However, the differences between the meteorological and hydrological drought indices in capturing the spatial patterns and temporal variations under complex topography and environments, especially with a transition zone, are not well characterized and understood.

Figure 1 .
Figure 1.The study region of the Heihe River Basin, with landscape (upper left) and elevation (meter) and three provinces (upper right) (data source: Wang et al., 2014).The triangles signs in upper left are meteorological observation sites.The bottom panel shows the Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2017-310Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 16 January 2018 c Author(s) 2018.CC BY 4.0 License.drought indices were calculated for the month.It was also assumed that daily ground energy was in balance, so R n ̶ G=H+L×AET, where H and L are sensible heat flux and potential heat constant.

(
RIEMS 2.0)(Xiong and Yan, 2013).The simulation was conducted over the period of 1980-2010.The horizontal spatial resolution was 3km.A unique feature with this simulation was that the model's parameters, including soil hydrological properties, were recalibrated based on observations and remote sensing data over the HRB that greatly improved the model's performance.The model evaluation indicated that the model was able to reproduce the spatial pattern and seasonal cycle of precipitation and surface T. The correlation coefficients between the simulated and observed pentad P were 0.81, 0.51, and 0.7 in the upper, middle, and lower HRB regions, respectively (p<0.01).The historical T and P observations during the simulation period at Yeilangou of the upper basin (38.25 o N, 99.35 o E, 3300m above the sea level), Zhangye of the middle basin (38.11 o N, 100.15 o E, 1484m), and Dingqing of the lower basin (40.3 o N, 99.52 o E, 1177m) (Fig. 1) were used to compare with the simulations.
15 o C in winter to 10 o C in summer, P increased from about 0.25 to 4 mm/d, and AET from about 0.25 to 2.5 mm/d.Again, T and P are close between the middle and lower basin reaches all seasons, and AET is close between the middle and upper basin reaches during winter and spring.While AET is close between the middle and lower basin reaches during summer and fall, the differences between the middle and upper basin reaches are much smaller than the differences in T or P.

P
and AET are the highest in the upper basin and the lowest in the lower basin, while T and PET have an opposite seasonal cycle.This explains why AI and ESI are larger in the upper basin than the middle or lower basin.Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2017-310Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 16 January 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 3 .
Figure 3. Spatial distributions of aridity index with potential evapotranspiration estimated using the Penman-Monteith method.

Figure 4 .
Figure 4. Spatial distributions of evaporative stress index with potential evapotranspiration estimated using the Penman-Monteith method.

Figure 5 .
Figure 5. Spatial distributions of potential evapotranspiration (PET, mm/d) estimated using the Hamon method.

Figure 6 .
Figure 6.Spatial distributions of aridity index with potential evapotranspiration estimated using the Hamon method.

Figure 7 .
Figure 7. Spatial distributions of evaporative stress index with potential evapotranspiration estimated using the Hamon method.
04, falling into semi-humid, arid, and arid climate.The corresponding ESI values are 0.63, 0.22, and 0.07, falling into humid, semi-arid, and arid climate.The annual PET averaged over 1980-2010 calculated using the Homan method are 1.25, 2.33, and 2.65 mm/d for the upper, middle, and lower basin reaches.The corresponding AI values are about 1.3, 0.18, and 0.07, falling into humid, arid, and arid climate.The corresponding ESI values are 0.78, 0.31, and in the lower basin.This indicates that only the hydrological drought index is able to identify the transition climate zone in the middle basin.The difference between AI and ESI in classifying climate is related to the similar feature with the meteorological variables.Annual P is 555 mm in the upper basin, which is substantially different from 69-139 mm in the middle and lower basins.The mean T is -4.0 o C in the upper basin, which is well below 6.9-8.7 o C in the middle and lower basin reaches.The corresponding PET values fall into two groups, 299 mm in the upper basin and 672-767 mm in the middle and lower basin reaches.This explains why the AI falls into two groups.In contrast, AET is 226, 161, and 80 mm, substantially different not only between the middle and upper reaches but also between the middle and lower reaches.This explains why the ESI falls into three groups.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2017-310Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 16 January 2018 c Author(s) 2018.CC BY 4.0 License.The seasonal AI and ESI cycles are related to those of the meteorological and hydrological conditions.T, P and AET (Fig. S4), and PET (Fig. 8) all increase from winter to summer.In the upper basin, the increases in P and AET from spring / fall to summer are larger than the corresponding increases in PET, leading to larger AI and ESI values in summer.In the middle as well as lower basin, however, PET increases substantially from spring / fall, leading to smaller AI and ESI in summer than in spring / fall.

Figure 8 .
Figure 8. Seasonal variations of simulated potential evapotranspiration (PET, mm/d), aridity index (AI), and evaporative stress index (ESI) (from left to right).The top and bottom panels are for the Penman-Monteith and Hamon method, respectively.
and the averages over the dry or wet years (Fig.10) were analyzed.The annual AI values using the Penman-Monteith method are 0.4-0.5 for the first two dry years and 0.7-1.0 for the last two years in the upper valley (Fig.10).The average over the 4 years is about 0.65.In comparison, the average is about 0.9 over 1980-2010 and 1.4 over the 4 wet years.The values are very small in spring (except in 1982) and occasionally in fall (1990).The annual AI values in the middle and lower basin reaches are below 0.2 for individual dry years and average.The small values are found for individual seasons except falls of the last two years in the middle basin.In compassion, the annual values are 0.4 or above in 3 falls of the 4 wet years.The annual ESI values using the Penman-Monteith method are 0.5 or larger in the upper valley.The average over the 4 years are nearly 0.6.In comparison, the average is about 0.62 over 1980-2010 and 0.7 over the 4 wet years.The values are comparable from spring to fall, though relatively smaller in spring.This is different from AI.The annual ESI values are Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2017-310Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 16 January 2018 c Author(s) 2018.CC BY 4.0 License.about 0.2 in the middle and below 0.1 in the lower basin for individual dry years and average.Thus, the values are apparently different between the middle and lower basin reaches.This is another difference from AI.The lowest values mostly occur in summer in both basin reaches.In compassion, the annual values are 0.25-0.35 in the middle basin and 0.1 or larger in 3 of the 4 wet years in the lower basin.Same results, that is, substantially smaller AI than normal, especially in spring but no much ESI changes from normal and between seasons in the upper basin, and no much AI change from normal and wet events (small in all cases) in the middle and lower basin reaches but much smaller ESI than wet events and different between the two basin reaches, can be found for the Hamon method, though slightly larger AI and ESI values.The results suggest that ESI is better representative of extreme dry conditions in the middle basin, but less sensitive to drought in the upper basin.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2017-310Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 16 January 2018 c Author(s) 2018.CC BY 4.0 License.themiddle and lower HRB regions when the AI was used.The comparison results from this study therefore suggest that only ESI is able to identify a transition climate zone between the relatively humid climate in the mountains and the arid climate in the Gobi desert region.We conclude that the hydrological drought index ESI is a better index than the meteorological drought index AI for aridity classification in the HRB with a complex topography and land cover.Selection of the most appropriate drought index facilitates drought characterization, drought assessment and risk mitigation, and water resources management in the arid region.

Table 1 .
The data used in calculation and evaluation of the drought indices.H, AET, P, T, and e (RH) are sensible heat flux, actual evapotranspiration, precipitation, temperature, wind speed, and water vapor pressure (relative humidity).HRB stands for Heihe River Basin.

Table 3 .
Mann-Kendall trends from 1980 to 2010 of potential evapotranspiration (PET,