Desiccation of the Transboundary Hamun Lakes : Natural or Anthropogenic ? 2

This paper investigates the hydrologic and water management reasons behind the desiccation of the Hamun Lakes in the Iran-Afghanistan border region. We analysed changes in Hirmand (or Helmand) River flow, the main tributary providing 70% of the lakes’ total inflow, and precipitation during 1960-2016 by calculating standardized indices for precipitation (SPI) and discharge (SDI). Also, we applied Normalized Difference Spectral Indices (NDSIs) using satellite images from 1987 to present to observe monthly areal change of the lakes. The transboundary water body is responding to changes in regional water management, which has severely reduced the lakes’ inflow. Upstream water regulation in Afghanistan coupled with reservoir construction on the Iranian side has caused nearly full desiccation of major parts of the lake system. There is a discernible shift in the relation between the Hirmand River flow at the international border and upstream precipitation over the lakes’ basin before and after 2004. From 1960 to 2003, high river flows were expected to feed the lakes due to high precipitation over the basin. However, the Hirmand River flow at the border declined after 2004 despite large amounts of upstream precipitation, including the largest recorded amounts, especially in the Hindu Kush mountains. Further, environmental water stress caused by anthropocentric water management in Iran by reservoir construction has impacted the area of the lakes. Although a long period of drought from 1998-2004, i.e. climatic driver, decreased the lakes’ area, the lake system is primarily falling victim to anthropogenic flow alterations in the transboundary river basin. The lakes’ shrinkage places socio-economic stress on an already-vulnerable region with important public health implications as the exposed lake beds turn into major sources of sand and dust storms.


9
of Adraskan, Farah and Khash rivers (Figure 1-b), as well as missing water withdrawal data in mid-152 basin in Iran (from international border to the lakes). Approximately 70% of the flow into the Hamun 153 Lakes is from Hirmand River while the rest is supplied mostly by Farah River based on USGS data 154 (more detail in Appendix A). The Pearson correlation coefficient for annual inflow between Farah and 155 Hirmand Rivers during 1955-1980 is 0.82, meaning Hirmand River is a reasonable indicator of total 156 inflow to Hamun Lakes ( Figure S 1). 157 Available rain gauge data in the study area in Afghanistan and Iran do not have good spatial and 158 temporal coverage. We used widely-used satellite-based rainfall products, namely GPCC (Schneider 159 et al., 2011), PERSIANN-CDR (Ashouri et al., 2015) and TRMM (Huffman et al., 2007). All of these 160 products show high amount of precipitation in the region in recent years (more detail in Appendix B). 161 We used the GPCC rainfall data to estimate precipitation in Hirmand River sub-basin from 162 which has high correlation with PERSIANN (0.84) and TRMM (0.94). We calculated annual 163 precipitation and flow data based on the Iranian water year from October to September. Digital 164 elevation model (DEM) data are from ALOS World 3D -30m (AW3D30) (Tadono et al., 2016). 165

Water body detection 167
The Normalized Difference Spectral Indices (NDSIs) are commonly used for surface water detection 168 (Akbari et al., 2020;Boschetti et al., 2014). Among different NDSIs, those using visible bands (such 169 as red, green, etc.), near-infrared band and short wave near infrared band have been shown to 170 outperform others (Boschetti et al., 2014). The Normalized Difference Vegetation Index (NDVI) and 171 the Normalized Difference Water Index (NDWI) are two examples of this class of NDSIs, which 172 facilitate water detection (Akbari et al., 2020;Chipman and Lillesand, 2007;Ouma and Tateishi, 2006;173 normalized difference of bands in the electromagnetic spectrum and vary between -1.0 to 1.0: 175 where NIR, RED and SWIR are reflections in the near-infrared, red visible and short wave near 176

infrared. 177
High NDWI and low NDVI values represent water and we need to define a specific threshold to 178 determine water from non-water. We have access to multispectral remotely sensed products from 179 different satellites, such as Sentinel, Landsat, and MODIS. We utilized MODIS images available after 180 year 2001 because daily temporal resolution of this product helps resolve the common cloud cover 181 issue by providing more images in each month. Furthermore, we used Landsat images available for 182 the study area from 1987-2001 to expand our temporal coverage. We used MODSI NDWI and Landsat 183 NDVI products due to their good quality in the study region to determine the monthly variation of 184 water bodies' area. Also, < 0 and > 0.1 were considered as water using Google Earth 185 Engine Java Script API (Gorelick et al., 2017)  The relationship between inflow to Hamun Lakes and the lakes' area was investigated based on NDWI 195 monthly images and monthly inflow of Hirmand River to Iran from Jan. 1987to Aug. 2013 monthly inflow data is available. We defined three classes of monthly inflow: 1) inflow less than 0.5 197 km 3 (275 cases), 2) inflow between 0.5 and 1 km 3 (30 cases) and 3) inflow more than 1 km 3 (15 cases). 198 We chose 0.5 km 3 as a threshold of runoff classes because this is approximately equal to the active 199 capacity of CNR4 and water demand in the Sistan region. This approach allowed an investigation of 200 how water demand and new water regulation capacity after CNR4 went into operation affected the 201 area of the lakes. 202

Drought Indices (SPI and SDI) 203
Using annual (Oct. to Sep.) precipitation and runoff, we calculated Standardized Precipitation Index 204 (SPI) and Standardized Discharge Index (SDI). To analyse the temporal hydro-climatological status 205 of the Hirmand River sub-basin, the trend, the variation, and the average value of rainfall and discharge 206 were calculated. Temporal climate variability was characterized using SPI, which is designed to 207 evaluate metrological drought (McKee, 1995) and has been widely used for evaluating climate 208 variability (Hao et al., 2014;Irannezhad et al., 2015). SPI requires fitting a probability density function 209 (McKee, 1995;Thom, 1966) to the frequency distribution of precipitation at a given station for a 210 particular timescale (e.g. 3 months and 6 months). In this study, annual SPI was estimated as 211 (Farahmand and AghaKouchak, 2015): 212 where ∅ is the standardized normal distribution function and p is the corresponding empirical 213 probability when the precipitation in Hirmand River sub-basin are sorted in ascending order. Based on 214 SPI, climate conditions can be divided into eight categories as classified in Table 1. 215 SDI calculation is like SPI, but we used Hirmand River flow to Iran instead of precipitation. SPI 216 and SDI are used to describe various drought categories. Over time, increased water consumption 217 typically occurs in the upstream part of many basins. Increasing upstream water withdrawal or land-218 use change which can significantly alter river flow, and subsequently downstream water delivery. We 219 compared SPI and SDI to evaluate the possible link between rainfall and discharge variation (Shukla 220 and Wood, 2008;Torabi Haghighi et al., 2020). 221

Areal change of the lakes in the Sistan region 223
The monthly area of CNRs ( Figure 2-e to h) shows that these reservoirs did not experience complete 224 desiccation in all operating years except at the beginning of the 2000s due to extremely dry conditions 225 (Table 1) December because of conveying water to CNR4 (i.e., the rising limb of the boxplot shown in Figure  239 2-h). This operation strategy prepares CNR1, CNR2 and CNR3 to capture inflow in April-May by 240 lowering the water level in CNR1 to maximize the discharge of the Feeder Canal (Figure 1-d), which 241 is why the area of CNR4 is the highest during this period. 242 After April-May, the lakes' area gradually decreased to less than 5% of maximum value due to 243 low inflow (falling limb of monthly inflow hydrograph in Figure  -2). 254

Rate of desiccation of Hamun Lakes 255
At the beginning of March 1999, Hamun-i Hirmand was 950 km 2 , i.e. half the maximum area based 256 on available satellite images since 1987. Hamun-i Hirmand dried out over the next 8 months when 257 inflow to Hamun Lakes was zero (Figure 3). The slope of the desiccation line was lower when the 258 lake's area was between 950-700 km 2 compared with when the area ranged from 700-0 km 2 , which 259 means the shrinkage process accelerates as the water body becomes smaller ( Figure 3). Hamun-i Puzak 260 14 and Hamun-i Sabari took 17 months to dry up. The rates of desiccation (slope of lines shown in Figure  261 3) in Hamun-i Puzak, Hamun-i Sabari, and Gaud-i Zirreh are higher, but they become smaller. Gaud-262 i Zirreh is more resistant to inflow cut--Shile Canal inflow was zero after 2000--and its complete 263 desiccation takes about 6 years to happen (70 months) due to higher depth of this lake compared to 264 other Hamun Lakes. 265

Water flow through Hamun Lakes 266
We In March 1991 maximum recorded inflow (4.5 km 3 ) of the Hirmand River entered Iran. Max area 286 for Hamun-i Puzak (1300 km 2 ) and Hamun-i Sabari (1500 km 2 ) was observed in this month but the 287 maximum area of Hamun-i Hirmand (1800 km 2 ) occurred one month later (April 1991). The area of 288 Hamun-i Hirmand in March 1991 was 1700 km 2 . The largest area for Gaud-i Zirreh in 1991 was 2600 289 km 2 observed four months later in July. Therefore, the time lag for water conveyance from Hamun-i 290 Puzak and Hamun-i Sabari to Hamun-i Hirmand and finally to Gaud-i Zirreh was almost one and four 291 months, respectively. Additionally, the maximum area of Gaud-i Zirreh was 3000 km 2 , which occurred 292 more than 25 months later (in June 1993) because of accumulating inflow volume in preceding years. 293

Drought in the Hirmand River sub-basin 294
The analysis of the correlation between SPI and SDI in the years before and after 2004, which are 0.66  Table 1, a difference 300 equals to -2 between these two indices will considerably affect the classification of the year to dry or 301 wet state. In other words, when SPI is about 2, indicating an extremely wet year, the SDI can be less 302 than 0, which is characteristic of a dry year. 303 The mean annual precipitation in the whole Hirmand sub-basin (Figure 1-a)   the highest inflow to Iran is more frequent to occur with some time lag in April or May (more detail 316 in Appendix C). 317

Impact of reservoir construction in Iran 318
The capacity of CNR4 is more than 0.8 km 3 , i.e. 40% of annual Hirmand River flow into Iran in the 319 last 10 years. This reservoir has more than doubled the water regulation capacity in the region from 320 0.65 to 1.45 km 3 , which can affect the area of the Hamun Lakes. In this regard, we determined the 321 effect of CN4 on the state of the lakes' area by comparing similar hydrological conditions in different 322 years when Hirmand River deliveries to Iran were almost the same before and after operation of CNR4, 323 i.e. 1992/2016, 1996/2009and 1993/2011. 324 In 1992 ( ≈ 4 km 3 ), the area of Hamun-i Hirmand, Hamun-i Sabari and Hamun-i Puzak 325 were 1600, 1500 and 1000 km 2 , respectively, which decreased by 55, 43 and 87% in a similar condition 326 in 2016 after the construction of CNR4 (Figure 6-a). This demonstrates that the construction of CNR4 327 in Iran has worsened the situation of the Hamun Lakes in addition to the upstream water regulation in 328 Afghanistan. In 2009 ( ≈ 2.5 km 3 ) Hamun-i Hirmand lost 86% of its area compared to 1996 329 17 (Figure 6-b). The areal loss of this lake in 2011 ( ≈ 2 km 3 ) was 95% compared to 1993 (Figure  330 6-c). Likewise, the area of Hamun-i Puzak and Hamun-i Sabari decreased 57 and 45% in 2011 331 compared to similar conditions in 1993 (Figure 6-c). In 2009, the area of Hamun-i Sabari and Hamun-332 i Puzak were less affected by CNR4. In this year, only 19 and 11% of these lake's areas, respectively, 333 were lost compared to 1996 (Figure 6-b). There are no streamflow data available for other rivers that 334 feed to Hamun Lakes but water presence in a wetland in the northeast of Hamun-i Puzak (Figure 6-b) 335 shows considerable inflow from Khash River and likely other rivers in the north (Figure 1-b). Since 336 2009, more than 70% of this wetland was full (even more than its area in 1996). However, the wetland 337 was dry in all other years after CNR4 operation and 2009 is an exception. Based on our investigation, 338 Hamun-i Hirmand is more sensitive to the impact of CNR4 than Hamun-i Sabari and Hamun-i Puzak. 339

Monthly response of Hamun Lakes area to Hirmand River flow 340
Monthly inflow can largely affect the area of Hamun-i Puzak and Hamun-i Sabari in the same month 341 (Figure 7-a and b) since water retention time in connected Hamun Lakes above Shile Canal is small 342 due to their low depth (shown in Figure 3 and Figure 4). When monthly Hirmand flow to Iran is less 343 than 0.5 km 3 , the area of Hamun-i Puzak is most likely to be less than 500 km 2 (Figure 7-a). Also, 344 when inflow increases from 0.5 to between 0.5-1 km 3 , the area is more probable to exceed 500 km 2 . 345 Also, the area of this lake increases to more than 1250 km 2 when the Hirmand River delivery to Iran 346 exceeds 1 km 3 (Figure 7-a). Hamun-i Sabari is expected to be larger than 500 km 2 when inflow to Iran 347 rise from below 0.5 to above 0.5 km 3 . Likewise, greater areas than 500 km 2 are expected for Hamun-i 348  (Figure 1-a) with a collective water storage capacity of more than 3 km 3 heavily regulate the flow. The 376 lakes' area is declining although Hirmand River inflow to Iran has averaged 1.9 km 3 in recent years, 377 more than double the designated WP delivery. This indicates that temporal patterns of deliveries from 378 Afghanistan and human water demand on the Iranian side make water deliveries to the lakes 379

challenging. 380
The high correlation between SPI andSDI from 1960-2003 shows that high precipitation naturally 381 will lead to high runoff in the Hirmand River sub-basin. The large discrepancy between SPI and SDI 382 after 2004 is a strong evidence about the effects of recent anthropogenic modifications on the Hamun 383 Lakes. Water regulation upstream of the Hirmand River in Afghanistan has become more intensive, 384 decreasing the deliveries to Iran as a new phenomenon in the basin. Before 2010, the maximum 385 recorded annual precipitation of the Hirmand River sub-basin was 330 mm in 1990, which led to the 386