The Microwave Temperature Vegetation Drought Index

2 downloads 0 Views 10MB Size Report
Dec 11, 2016 - tion Drought Index (MTVDI) based on the Aqua satellite advanced mi- ... The C, N, NE, NW, QTP, S, SEC and SW are the abbreviation of central, northeast, northwest, south, southeast coast and ...... Tutorial on LM Algorithm.
Remote Sensing of Environment 199 (2017) 302–320

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

The Microwave Temperature Vegetation Drought Index (MTVDI) based on AMSR-E brightness temperatures for long-term drought assessment across China (2003–2010) Liyang Liu a,h, Jishan Liao a,1, Xiuzhi Chen a,b,⁎, Guoyi Zhou a, Yongxian Su b,c,⁎, Zhiying Xiang d, Zhe Wang e, Xiaodong Liu f, Yiyong Li g, Jianping Wu a,1, Xin Xiong a, Huaiyong Shao h,⁎ a Key Laboratory of Vegetation Restoration and Management of Degraded Ecosystems, Guangdong Provincial Key Laboratory of Applied Botany, South China Botanical Garden, Chinese Academy of Sciences, Guangzhou 510650, China b Laboratoire des Sciences du Climat et de l’Environnement, UMR 1572 CEA-CNRS UVSQ, 91191 Gif sur Yvette, France c Key Lab of Guangdong for Utilization of Remote Sensing and Geographical Information System, Guangdong Open Laboratory of Geospatial Information Technology and Application, Guangzhou Institute of Geography, Guangzhou 510070, China d School of Earth Sciences, Zhejiang University, Hangzhou 310027, China e School of Geographical Sciences and Urban Planning, Arizona State University, Tempe, AZ, USA f College of Forestry and Landscape Architecture, South China Agricultural University, Guangzhou 510642, China g College of Forestry and Landscape Architecture, Anhui Agricultural University, Hefei 230036, China h Key Laboratory of Geoscience Spatial Information Technology of Ministry of Land and Resources, College of Earth Science, Chengdu University of Technology, Chengdu 610059, China

a r t i c l e

i n f o

Article history: Received 27 September 2016 Received in revised form 26 May 2017 Accepted 15 July 2017 Available online xxxx Keywords: Temperature Vegetation Drought Index (TVDI) Microwave TVDI Drought monitoring Brightness temperatures (Tb) The Advanced Microwave Scanning Radiometer (AMSR‐E) Passive microwave remote sensing

a b s t r a c t Satellite-based drought indices have been proved to be effective and convenient in detecting drought conditions at regional and global scales. However, most current drought indices are based on the visible/near infrared/thermal remote sensing, which might be influenced greatly by cloud, atmospheric water content and rain-fall. Microwave sensors can overcome the shortages of visible/near infrared/thermal remote sensing and show to be another important approach for drought monitoring due to its all-weather working advantages. But to date, the application of microwave vegetation drought indices in drought monitoring has not been thoroughly investigated. Here, for the first time we constructed a microwave derived Temperature Vegetation Drought Index (TVDI) - MTVDI based on the theory of optical TVDI using the brightness temperatures (Tb) from the Advanced Microwave Scanning Radiometer (AMSR‐E) onboard Aqua satellite. Firstly, we built a new land surface temperature (Ts) inversion model based on the AMSR‐E 18.7 GHz horizontal, 23.8 GHz and 89.0 GHz vertical polarized Tb, and then developed the Microwave Normalized Difference Vegetation Index (MNDVI) from the AMSR ‐ E 23.8 GHz Microwave Polarization Difference Index (MPDI). After that, we constructed three versions of MTVDI: original MTVDI using Ts and MNDVI; Imp‐ MTVDI (Improved MTVDI) using the Ts‐ Tair (the difference between land surface temperature and air temperature) to replace the Ts; and NonL‐ MTVDI (Nonlinear MTVDI) using nonlinear equation to fit the dry and wet edges, respectively. Finally, we used precipitation, soil moisture (SM) and P/ PET (the ratio of precipitation to potential evapotranspiration) to validate the performances of MTVDI, Imp ‐ MTVDI, NonL ‐ MTVDI, MODIS derived TVDI and iTVDI (improved TVDI). The time-series drought assessments across China from 2003 to 2010 indicated that the trends of the proposed MTVDI showed the most negative correlations with the variations of precipitation, P/PET and SM, and showed best performances of significance test in most regions of China. Moreover, the MTVDI could better separate the drought levels in different degrees than MODIS-derived TVDI. However, the proposed MTVDI still has some uncertainties in regions widely covered by desert, Gobi and large water surfaces. In addition, this paper mainly focuses on large spatial scale and long term drought monitoring and only uses satellite data for model validation. Further studies are needed to develop a higher spatial- and temporal-resolution MTVDI for short-term and small spatial-scale drought monitoring. © 2017 Elsevier Inc. All rights reserved.

1. Introduction ⁎ Corresponding authors. E-mail addresses: [email protected], [email protected] (X. Chen), [email protected] (Y. Su), [email protected] (H. Shao). 1 Co-first authors.

http://dx.doi.org/10.1016/j.rse.2017.07.012 0034-4257/© 2017 Elsevier Inc. All rights reserved.

Droughts are broadly classified into four major types: meteorological (reduction of precipitation), agricultural (shortage of available water for plant growth), hydrological (deficiency of surface and

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

subsurface water supply), and socioeconomic drought (insufficient water supply to meet the demand of economic growth) (Wilhite, 2005; Zhang and Jia, 2013; AghaKouchak et al., 2015). It is the most costly natural disaster that poses significant water and food security concerns worldwide (Wilhite, 2005; Godfray et al., 2010; Zhang and Jia, 2013; AghaKouchak et al., 2015). It is also considered to be one of the most complex but least understood natural hazard (Hagman, 1984; Obasi, 1994; Wilhite, 2000). Given the serious impacts of droughts on economies, eco-environments and especially on the agricultures, it is vital and urgent to develop effective means for timely monitoring large spatial-scale drought events (Goddard et al., 2003; Tadesse et al., 2005). In recent years, drought has been widely assessed based on observed changes in vegetation health and land cover from remotely sensed data (Tucker and Choudhury, 1987; Silleos et al., 2006; Nemani et al., 2009). Many drought indices based on visible/near infrared/thermal remote sensing data have been proposed to detect drought at the regional or global scales (Qin et al., 2006; Rhee et al., 2010; Escorihuela and Quintana-Seguí, 2016). They are mainly classified into three types: vegetation, temperature, and temperature-vegetation indices (Table.1). The Normalized Difference Vegetation Index (NDVI) is the most frequently used vegetation index and the first visible/near infrared/thermal remote sensing-based measure used to monitor vegetation drought (Rouse et al., 1974; Jain et al., 2009; Ji and Peters, 2003; Karnieli et al., 2010; Bajgiran et al., 2008). Building on the original definition of NDVI, a number of vegetation indices were established in detecting droughts, such as the Transformed Vegetation Index (TVI) (Deering and Rouse, 1975; Tucker, 1979), Vegetation Condition Index (VCI) (Kogan, 1995a, 1995b) and Enhanced Vegetation Index (EVI) (Huete et al., 2002; Saleska et al., 2007). These indices describe the vegetation condition by combining spectral information from different parts of the electromagnetic spectrum that are sensitive to biophysical characteristics of vegetation, such as chlorophyll content, water content, and internal leaf structure (AghaKouchak et al., 2015). However, vegetation index is strongly correlated with the vegetation greenness (Sellers et al., 1992). It is often referred to a greenness index rather than a drought index (Jackson et al., 2004). The land surface temperature (Ts) computed from thermal infrared bands has been found to provide valuable information on surface moisture conditions (Gutman, 1990). The most used temperature indices are the Normalized Difference Temperature Index (NDTI) (McVicar and Jupp, 1998) and Temperature Condition Index (TCI) (Bhuiyan et al., 2006; Jain et al., 2009). Compared with the vegetation index, the temperature indexis more sensitive to soil water stress due to the relationship between leaf temperature and transpiration (Goetz, 1997; Wang et al., 2004). Therefore, the temperature index might ignore the different

303

drought-resistance of plants and thereby brought some inaccuracies in detecting the drought conditions of different vegetation types (Moran, 2004). In recent years, various ways of combining NDVI and Ts information have been explored for drought monitoring and impact assessment (e.g., the Temperature Vegetation Drought Index (TVDI) combining NDVI and Ts) (Price, 1990; Nemani et al., 1993; Moran et al., 1994; Carlson et al., 1995). The TVDI method is an index based on the empirical interpretation of the NDVI-Ts triangle space (Fig. 3, in Section 3.3), relying on the relationship (typically, negative correlation) between NDVI and Ts (Lambin and Ehrlich, 1996; McVicar and Bierwirth, 2001; McVicar and Jupp, 1998; Karnieli et al., 2010). It has been widely applied in drought monitoring at the drainage basin scale (Son et al., 2012; Wang et al., 2010), regional scale (Chen et al., 2011a; Li et al., 2010; Liu et al., 2008; Cao et al., 2016; Zhang et al., 2016, July; Gao et al., 2011) and national scale (Liang et al., 2014; Wang et al., 2004). Studies showed that the combination of vegetation and temperature indices provides a more powerful tool for monitoring the drought conditions of vegetation than the individual vegetation and temperature indices (Sandholt et al., 2002; Singh et al., 2003). Optical-based drought indicators are sensitive to cloud cover, atmospheric effects, aerosols, water vapor, and land cover condition and brings certain limitation for drought monitoring applications (Andela et al., 2013; Shi et al., 2008; Liu et al., 2011a). Unlike optical sensors, microwave sensors are less affected by atmospheric conditions and can penetrate into dense canopy, showed to be another important approach for drought monitoring due to its all weather working advantages (Zhang and Jia, 2013). Soil moisture derived from the microwave remote sensing is a sensitive index of drought, and was widely used to monitor water deficit (Andreadis et al., 2005; Cai et al., 2009; Chen et al., 2012; Yuan et al., 2015; Chen et al., 2016). But soil water deficit would much refer to the hydrological drought rather than the agricultural drought. Several studies investigated on microwave drought index for drought monitoring, such as the Microwave Integrated Drought Index (MIDI) (Zhang and Jia, 2013) and the Microwave Polarization Index (MPI) (Mao et al., 2010). But to date, the application of microwave vegetation drought indices in drought monitoring has not been thoroughly investigated (Zhang and Jia, 2013). To integrate the advantages of both TVDI and microwave remote sensing in drought monitoring, this paper proposed a new drought monitoring index called the Microwave Temperature Vegetation Drought Index (MTVDI) based on the Aqua satellite advanced microwave scanning radiometer (AMSR‐ E) brightness temperatures (Tb) (Ashcroft and Wentz, 2000) of 89.0 GHz, 23.8 GHz and 18.7 GHz. The main objectives of this study are: 1) to construct the NDVI-Ts triangle using AMSR ‐ E microwave Tb for drought monitoring, like the optical

Table 1 Drought indices based on visible/near infrared/thermal remote sensing data. α represented the weight of single index. NDVImax and NDVImin represent the maximum and minimum values of the Normalized Difference Vegetation Index, respectively. LST and T represent the Land Surface Temperature, Tmax, LSTmax and Tmin, LSTmin represent the maximum and minimum values of Land Surface Temperature, respectively. Ts represent the observed surface temperature, T∞ and T0 represent a modelled surface temperature with infinite and zero surface resistance, respectively. TRMM represent the precipitation based on Tropical Rainfall Measuring Mission satellite, and TRMMmax and TRMMmin represent the maximum and minimum values of the precipitation, respectively. Drought Indices

Formula

Citation

NDVI EVI VCI LSWI TCI NDWI NDDI NDTI NMDI VHI SDCIa

(ρ858 −ρ650)/(ρ858 +ρ650) 2.5 ∗ (ρ858 − ρ650)/(ρ858 + 6 ∗ ρ650 − 7 ∗ ρ469 + 1) (NDVIi − NDVImin)/(NDVImax − NDVImin) (ρ858 − ρ1640)/(ρ858 + ρ1640) (Tmax − Ti)/(Tmax − Tmin) (ρ858 − ρ1240)/(ρ858 + ρ1240) (NDVI − NDWI)/(NDVI + NDWI) (T∞ − Ts)/(T∞ − T0) (ρ860 − (ρ1640 − ρ2130))/(ρ860 + (ρ1640 − ρ2130)) α ∗ VCI + (1 − α)∗ TCI ð1=4Þ  scaled LST þ ð1=2Þ  scaled TRMM þð1=4Þ  scaled NDVI

Rouse et al., 1974; Tucker, 1979; Kogan, 1991, 1995a Huete et al., 2002; Saleska et al., 2007 Kogan, 1995a Bajgain et al., 2015 Kogan, 1995a Gao (1996); Chen et al. (2005) Gu et al., 2007 McVicar and Jupp, 1998 Wang and Qu, 2007 Kogan, 1995a, 1995b Rhee et al., 2010

a

Scaled LST=(LSTmax −LSTi)/(LSTmax −LSTmin), scaled TRMM=(TRMMi −TRMMmin)/(TRMMmax −TRMMmin), scaled NDVI=(NDVIi −NDVImin)/(NDVImax −NDVImin).

304

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

NDVI-Ts relationship (see method); 2) to assess the proposed MTVDI using precipitation (Zhang et al., 2016; Hao et al., 2015; Du et al., 2013), soil moisture (SM) (Chen et al., 2016; Yuan et al., 2015) and P/ PET (the ratio of precipitation to potential evapotranspiration) (Zhou

et al., 2015; Chen et al., 2016); 3) to evaluate the time-series monthly drought conditions of China during the period 2003–2010. The study is expected to enrich existing drought monitoring approaches by using microwave remote sensing.

Fig. 1. The study area. (a) Provinces in China. The C, N, NE, NW, QTP, S, SEC and SW are the abbreviation of central, northeast, northwest, south, southeast coast and Qinghai-Tibet Plateau, respectively; (b) the land cover types (ESA&UCL, 2009) in China. The identification codes of land cover types are shown in the Appendix Table A1.

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

305

2. Area descriptions and data

2.2. Data and processing

2.1. Study area

The Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR ‐ E) was launched in May 2002 on NASA's Aqua satellite (Parkinson, 2003). The instrument was developed for NASA by the Japan Aerospace Exploration Agency (JAXA) (Kawanishi et al., 2003). It can provide global passive microwave measurements of terrestrial, oceanic, and atmospheric variables for the investigations of global water and energy cycles. Each AMSR‐ E Tb file contains images of six frequencies: 6.9 GHz, 10.7 GHz, 18.7 GHz, 23.8 GHz, 36.5 GHz, and 89.0 GHz. In this study, the AMSR ‐ E/Aqua daily global quarter-degree Tb data of 18.7 GHz, 23.8 GHz and 89.0 GHz bands from 2003 to 2009 were provided by the NSIDC (Snow and Ice Data Center, http://nsidc. org/). Because the transit time of the ascending is at 1:30 pm during

China feeds about 22% of the world's population with only 7%–8% of its arable land (Yuan et al., 2015). China's crop production and food security has been a major problem because in recent years China has experienced severe short-term (monthly to seasonal) drought (A. Wang et al., 2011; Feng et al., 2014), leading to an increasing threat of drought in cultivated areas (Piao et al., 2010). For better control the drought situation in China, it is necessary to carry out drought monitoring research in China. The mainland of China has 32 provinces (municipalities, autonomous regions) (Fig. 1a), including 21 land cover types (Arino et al., 2012; Di Gregorio, 2005) (Fig. 1b and Appendix Table A1).

Fig. 2. The flowchart of constructing MTVDI model.

306

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

the daytime (Rüdiger et al., 2009), it is best for constructing the surface temperature inversion model. Therefore, ascending microwave data were selected to develop the MTVDI model. It has been proved that radio frequency influence (RFI) has a significant impact on brightness temperatures at C band, and to a lesser extent at X band (Owe et al., 2008). V-polarized indices provided somewhat better RFI discrimination than the H-polarized indices (Njoku et al., 2005). Therefore, we use spectral difference statistics (means and standard deviations) to identify RFI-contaminated pixels. We calculated RFI as follows: 1) RFIC = Tb06V − Tb10V, 2) RFIX = Tb10V − Tb18V, 3) RFIK = Tb18V − Tb23V, and we simulated the RFI classification maps using thresholds of μ (means) = 3 K and σ (standard deviations) = 3 K for RFIC, μ = −0.25 K and σ = 3 for RFIX, and μ = 10 K for RFIK in each month, respectively (Njoku et al., 2005; McKague et al., 2010). For 36.5 GHz and 89.0 GHz we also used RFIK to screen RFI, due to lack of research on RFI with high frequency. According to the MYD11A2 MODIS LST and MOD13A2 MODIS NDVI product, we performed 8-day average and 16day maximal synthesis of the AMSR‐ E Tb respectively, when establishing the Ts and Microwave Normalized Difference Vegetation Index (MNDVI) inversion model. For establishing the relationship between Tb and NDVI, the 16-day MODIS NDVI product (MOD13A2) (1 km) was used. The starting time for MOD13A2 is consistent with MYD11A2, which is the first day of the year. Although the overpass time of MOD is not consistent with that of MYD, the state of vegetation is relatively stable in a short time and is not easily affected by overpass time. Some studies showed that the average retrieval errors of MODIS LST products under non-cloud weather conditions had reached 1.0 °C (Wan et al., 2002, 2004). Besides, both AMSR ‐ E and MODIS ‐ MYD are on the Aqua satellite at the same time. In addition, daily MODIS LST product has much invalid values which is not conducive to the construction of the Ts model. Therefore, the 8-day MODIS day LST product (MYD11A2) (1 km) with same overpass time to AMSR ‐ E, was used as the test data to validate the surface temperature inverse model. As we used the 16-day and 8-day products of MODIS for establishing the triangle implementation and proposed MTVDI, all results and discussions in this paper are based on these temporal accuracies. We also used the monthly MODIS NDVI (MYD13A3) (1 km) and monthly MODIS LST (MYD11C3) product (0.05°) to develop the monthly MODIS-derived TVDI and iTVDI (improved Temperature Vegetation Drought Index) (Rahimzadeh-Bajgiran et al., 2012). All of the MODIS data were downloaded from NASA (http://modis.gsfc.nasa.gov/). We used bilinear interpolation to resample the MODIS data to the same spatial resolution of AMSR‐ E (0.25 arc-degrees). In this paper, we masked out the cloud, ice and snow-cover regions and used the remaining MODIS Tsto develop AMSR ‐ E Ts algorithm. In order to eliminate effects of snow and ice, as well as mixed pixel, we developed daily mask imageries based on the daily 25 km NISE product (Near-Real-Time SSM/I-SSMIS EASE-Grid Daily Global Ice Concentration and Snow Extent) obtained from NSIDC. All pixels with NISE value not equal to 0 were removed. In addition, we also developed monthly masks to exclude the pixels covered by snow for N15 days per month. The 0.25 degree monthly precipitation data - Tropical Rainfall Measuring Mission (TRMM) product TRMM_3B43 was obtained from GESDISC (Goddard Earth Sciences Data and Information Services Center, http://disc.gsfc.nasa.gov). To calculate the P/PET and Ts ‐ Tair (the land surface temperature minus air temperature), we obtained the monthly precipitation (PRE), monthly potential evapotranspiration (PET) and monthly air mean temperature (Tair) data with resolution of 0.5° from CRU (Climatic Research Unit) of University of East Anglia (http://www.cru.uea.ac.uk/) (Harris et al., 2014; Mitchell et al., 2004; Mitchell and Jones, 2005; Rhee et al., 2010), and Tair were resampled to 0.25° by bilinear interpolation method. Besides, we obtained 0.25 degree daily ESA (European Space Agency) CCI (Climate Change Initiative) SSMV (Surface Soil Moisture Volumetric) soil moisture (SM) product (Liu et al., 2011b; Liu et al., 2012; Wagner et al., 2012) from ESA (European Space Agency) and

calculated the monthly average soil moisture to validate the performances of MTVDI and TVDI. 3. Methodology 3.1. The technical route Fig. 2 shows the flowchart of constructing MTVDI in this study. We firstly screened RFI, and built daily and monthly masks based on the NISE product to eliminate the effects of ice, snow and water vapors on AMSR‐ E Tb imageries (see in Section 2). We then retrieved land surface temperature using a new land surface temperature inversion model based on AMSR ‐ E 89.0 GHz, 23.8 GHz vertically polarized Tb and 18.7 GHz horizontally polarized Tb (see in Section 3.2). And then we constructed the MNDVI (Microwave Normalized Difference Vegetation Index) based on 23.8 GHz MPDI (see in Section 3.4). After establishing AMSR ‐ E-derived Ts and MNDVI, we built three different types of dryedge and wet-edge equations to construct the AMSR ‐ E-derived TVDI, namely MTVDI (see in Section 3.4), Imp ‐ MTVDI (Improved MTVDI, see in Section 3.5) and NonL ‐ MTVDI (Nonlinear MTVDI, see in Section 4.2). We used precipitation, SM and P/PET to validate the MTVDI, Imp ‐ MTVDI and NonL ‐ MTVDI. Finally, we selected the optimal AMSR ‐ E-derived MTVDI for monitoring the monthly drought conditions of China from 2003 to 2010. 3.2. The Ts retrieval model At present, the most commonly used passive microwave Ts inversion models are empirical inversion models (Gao et al., 2003; Lacava et al., 2005; Mallick et al., 2009) and physical-based inversion models (Fily et al., 2003; Wang et al., 2009; Li et al., 2011). Empirical inversion model is one of the simplest ways to obtain surface temperature, which often use multiple horizontal and polarized channels to build multiple linear regression model. The vertical polarization is more suitable for retrieving surface temperatures than horizontal polarization (Mao et al., 2007; Holmes et al., 2009; Chen et al., 2014b). In order to develop a simple drought index, we therefore used an empirical Ts inversion model to retrieve the Ts based on the AMSR ‐ E Tb by referring to previous studies (Mao et al., 2007; McFarland et al., 1990; Pan et al., 2001; Wu and Yang, 2007) (Formula 1). 2

2

Ts ¼ A  ðTb1 þ M Þ þ B  ðTb2 þ N Þ þ C  Tb3 þ D

ð1Þ

Fig. 3. The NDVI-Ts triangle space and definition of the TVDI (reproduced from Sandholt et al., 2002). M and N are the lines for estimating TVDI for a given pixel in the NDVI‐Ts feature space. Ts is land surface temperature. NDVI is the Normalized Difference Vegetation Index. The a and b are the parameters defining the dry edge, which are estimated from an area large enough to represent the entire range of surface moisture content.

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

where Tb1 and Tb2 are vertically polarized Tb, Tb3 is horizontally polarized Tb; A, B, C, D, M and N are constants, which can be calculated by the Levenberg-Marquardt algorithm (see Section 3.7). Because 89.0 GHz vertical polarization is the best single band for surface temperature inversion (Mao et al., 2007), we also took it as a key variable in Formula 1. Then, the remaining 11 channels were included into Formula 1 in turn. We used the Levenberg-Marquardt algorithm to obtain the optimal solution of model coefficients. For model validation, we used 10-fold cross-validation to validate the optimal model. 3.3. Temperature Vegetation Dryness Index (TVDI) A number of studies have documented that two-dimensional scatterplot of Ts and NDVI has a triangular or trapezoidal shape (Petropoulos et al., 2009; Carlson, 2007). The scatter plot of NDVI and Ts exhibits a triangular distribution when the study area contains a full range of land cover types (i.e. dry bare soil, saturated bare soil, waterstressed vegetation and well-watered vegetation) (Mallick et al., 2009; Carlson et al., 1994; Price, 1990). In this triangle (Fig. 3), the point A and point B denotes the dry bare and moist bare soil, respectively. As surface vegetation cover increases, surface temperature decreases. The point C indicates that the area is completely covered by vegetation, and the soil moisture is sufficient. At this time, the evapotranspiration resistance is the least. Therefore, the AC edge indicates that the effectiveness of soil moisture is low and the surface evapotranspiration is small, which is considered as the dry edge of NDVI ‐ Ts feature space. The BC edge is considered as the wet edge of the NDVI ‐ Ts feature space, which indicates that the soil moisture is sufficient, the vegetation growth is not limited by water, and the surface evapotranspiration is equal to the potential evapotranspiration. Sandholt et al. (2002) found that there are many contour lines (Fig. 3) in the NDVI ‐ Ts feature space when studying soil moisture. Therefore, the Temperature Vegetation Dryness Index (TVDI) was proposed. The TVDI is defined as follows:

307

MNDVI. Then, the dry and wet edges of both MTVDI and Imp ‐ MTVDI in each month during January 2003 to December 2010 were fitted using the linear regression method. Neale et al. (1990) and Wang et al. (2006) proved that MPDI has a significant negative correlation with NDVI. Here, we retrieved MNDVI using the natural logarithm of MPDI (Ln(MPDI)) at the 23.8 GHz channel (see Fig. 5, in Section 4.1). We tried three methods to build relationship between NDVI and Ln(MPDI) using simple linear regression, quadratic and cubic polynomial regression. However, the fitting accuracy did not increase significantly with the increase of the complexity of the fitting function. Therefore, we chose the linear fitting model of Ln(MPDI) to retrieve NDVI. The Levenberg-Marquardt algorithm (see Section 3.7) was used to obtain the optimal solution of model coefficient. We then used the 10-fold cross-validation to validate the fitting results. The MNDVI is defined as follows, MNDVI ¼ e  LnðMPDIÞ þ f

ð8Þ

Here, MPDI is microwave polarization difference index. e and f are constants, which can be calculated by the Levenberg-Marquardt algorithm. The definition of MPDI is similar to the definition of NDVI, given as

MPDI ¼

TbV −TbH TbV þ TbH

ð9Þ

Tsmax ¼ a1 þ b1  NDVI

ð3Þ

where TbVis vertically polarized Tb, TbH is horizontally polarized Tb. Artificial land cover types (i.e. agricultural lands, urban area, and roads) could affect the performance of MTVDI (Fan et al., 2015). Therefore, we firstly removed the pixels of artificial land cover types based on the land cover data of ESA (European Space Agency) GlobCover (Global Cover) 2009 product before constructing the triangular space. We also calculated the percentage of land cover types for all the remaining pixels in each period (see Appendix Table A2, taking 2003 as an example). The selected samples included a full range of land cover types, and the triangular space were fully determined (Mallick et al., 2009; Sandholt et al., 2002).

Tsmin ¼ a2 þ b2  NDVI

ð4Þ

3.5. Improved TVDI and improved M-TVDI

TVDI ¼

Ts−Tsmin Tsmax −Tsmin

ð2Þ

where Ts is the observed surface temperature at a given pixel; NDVI is the observed vegetation index; a1 and b1are the intercept and slope of the dry edge; a2 and b2 are the intercept and slope of the wet edge; where Tsmax and Tsminare the maximum and minimum surface temperature observation for a given NDVI. 3.4. Microwave Temperature Vegetation Dryness Index (MTVDI) On basis of the TVDI theory, we developed MTVDI using AMSR − E microwave Tb, MTVDI ¼

Ts−Tsmin Tsmax −Tsmin

Some scholars have found that TVDI model may be improved to characterize larger areas or higher NDVI values if the air temperature is taken into account (Sandholt et al., 2002; Ran et al., 2005; Hassan et al., 2007). Rahimzadeh-Bajgiran et al. (2012) proposed the improved TVDI (iTVDI), which used the difference between surface temperature and air temperature (Ts ‐ Tair) to replace the Ts used in the triangular space. The results indicated that the performance of iTVDI are better than TVDI in the mountainous area. Therefore, we used monthly air mean temperature acquired from CRU to calculate the Ts ‐ Tair and iTVDI using Formula 10. The correspondingly improved MTVDI (Imp ‐ MTVDI) is defined as Formula 11.

ð5Þ

Tsmin ¼ c1  MNDVI þ d1

ð6Þ

Tsmax ¼ c2  MNDVI þ d2

ð7Þ

where MNDVI is the Microwave Normalized Difference Vegetation index;c1 and d1are the slope and intercept of the dry edge; c2 and d2 are the slope and intercept of the wet edge; Tsmaxand Tsmin are the maximum and minimum surface temperature observation for a given MNDVI. In this paper, we firstly calculated a series of maximum and minimum Ts corresponding to the MNDVI, taking 0.01 as the interval of

iTVDI ¼

T obs −T min T max −T min

ð10Þ

where Tobs is observed Ts‐ Tair and Ts here is based on MODIS, Tmax is dry edge and Tmin is wet edge. Imp‐MTVDI ¼

ΔT−ΔT min ΔT max −ΔT min

ð11Þ

where ΔT is observed Ts ‐ Tair and Ts here is derived by AMSR ‐ E Tb, ΔTmax is dry edge and ΔTmin is wet edge.

308

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

Fig. 4. The scatter diagrams between MODIS Ts and AMSR‐E Tb.

(in total 96 months from 2003 to 2010) correlation analysis between MTVDI, Imp‐MTVDI, NonL‐MTVDI and precipitation and P/PET. Due to the influence of snow, ice and water vapors, some pixels are absent between November and April. Pixels with N80% months from January 2003 to December 2010 having data were selected for time-series validation.

3.6. Model validation We used precipitation, P/PET and to SM validate the performances of the proposed MTVDI, MODIS-derived TVDI models, Imp ‐ MTVDI, and NonL ‐ MTVDI (Nonlinear MTVDI). We conducted pixel level time-series

Table 2 The simulated optimal values of the parameters in the proposed Ts and MNDVI inversion model. Variable

A

B

C

D

M

N

e

f

Value

6.134 ∗ 10−3

9.934 ∗ 10−3

−0.353

349.582

−278.818

−216.029

−0.231

−0.578

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

309

Fig. 5. The correlation between Ln(MPDI) and NDVI. (The Ln(MPDI) is logarithmically transformed from MPDI).

Fig. 6. The MNDVI‐Ts space for the MTVDI and NonL‐MTVDI models. (a–c) linear fitting for MTVDI; (d–f) linear and nonlinear fitting for NonL‐MTVDI; (g–i) linear fitting for Imp‐MTVDI.

310

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

In addition, to compare the performances between AMSR‐E-derived TVDI and MODIS-derived TVDI, we included TVDI and iTVDI into the analysis as well. Because the spatial resolution of P/PET is 0.5°, we resampled the MTVDI, Imp‐ MTVDI, NonL ‐MTVDI, TVDI and iTVDI to 0.5 arc-degrees in order to conduct the correlation analysis with P/PET. To eliminate the effect of seasonal variations on validation, anomalies of these indices, which were calculated by subtracting the mean of the corresponding months, were used in validations. It is difficult to validate the MTVDI and MODISderived TVDI models using ESA CCI SM products, which exist a large number of pixel missing. As a result, we selected 24 sites (Fig. 1a, Table 5) that

are not affected by missing pixels based on the agriculture meteorology stations over China from the National Meteorological Information Center at China Meteorological Administration (Yuan et al., 2015). 3.7. Levenberg-Marquardt algorithm The Levenberg-Marquardt algorithm (LMA) (Ranganathan, 2004) was proposed by Levenberg (1944) and was rediscovered by Marquardt (1963). It is the most widely used optimization algorithm (Chen et al., 2014a; Chen et al., 2011b; Lagacé, 2012). LMA is a

Fig. 7. The correlation between precipitation and (a) MODIS TVDI (b) MODIS iTVDI (c) MTVDI (d) NonL‐MTVDI (e) Imp‐MTVDI. NonL‐MTVDI is a AMSR‐E derived TVDI with the nonlinear wet and dry edge. a1, b1, c1, d1 and e1 are the spatial distribution of the correlation coefficient, a2, b2, c2, d2 and e2 are histograms of the correlation coefficients, a3, b3, c3, d3 and e3 are the spatial distribution of P-values.

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

combination of steepest descent and the Gauss-Newton method (Lourakis, 2005). In this study, the LMA was driven by First Optimization software manufactured by 7D–Soft High Technology Inc. 4. Results 4.1. Simulating the model parameters, MNDVI, and dry and wet edges of MTVDI The MODIS Ts and AMSR ‐ E Tb in eight periods (DOY009–016, DOY065–072, DOY105–112, DOY153–160, DOY196–200, DOY241–

311

248, DOY289–296 and DOY337–344) in 2009 are chosen to develop optimize optimal Ts inversion model. The scatter diagrams of MODIS Ts and AMSR‐ E Tb are shown in Fig. 4. Compared with the horizontal polarization channel, the Tb of the vertical polarization channel has stronger correlation with MODIS Ts, which is consistent with the previous studies (Mao et al., 2007; Holmes et al., 2009; Chen et al., 2014a). The final optimal Ts inversion model included 18.7 GHz horizontal polarization channel, 23.8 GHz vertical polarization channel and 89.0 GHz vertical polarization channel. The optimal simulation values of the parameters in the Ts inversion model are shown in the Table 2. The average RMSE

Fig. 8. The correlation between P/PET and (a) MODIS TVDI (b) MODIS iTVDI (c) MTVDI (d) NonL‐MTVDI (e) Imp‐MTVDI. NonL‐MTVDI is the AMSR‐E derived TVDI with the nonlinear wet and dry edge. a1, b1, c1, d1 and e1 are the spatial distribution of the correlation coefficient, a2, b2, c2, d2 and e2 are histograms the correlation coefficients, a3, b3, c3, d3 and e3 are the spatial distribution of P-values.

312

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

(Root Mean Square Error) of the proposed Ts inversion model is 6.492 K and the determination coefficient (R2) is 0.855. To keep consistent with the period of the data used in the Ts inversion model, we select the MODIS NDVI in four periods (DOY033–048, DOY129–144, DOY209–224 and DOY305–320) in 2009 to simulate the optimal parameters of MNDVI model. A total number of 44,631 samples were retained after removing the pixels with low quality flag based on the NDVI quality imageries. The scatter diagrams between MODIS NDVI and the natural logarithm of AMSR ‐ E MPDI (Ln(MPDI)) are presented in Fig. 5. Results showed a strong negative relationship between NDVI and Ln(MPDI). The Ln(MPDI) of 23.8 GHz channel had the highest correlation with NDVI (R2 = 0.642). The optimal simulation values of the parameters in the proposed MNDVI inversion model are shown in the Table 2. The RMSE (Root Mean Square Error) and coefficient of determination (R2) are 0.156 and 0.642, respectively. The MNDVI ‐ Ts spaces showed that most of the dry edges and wet edges were well fitted using linear regression (Fig. 6a–c). However, there were still some MNDVI ‐ Ts spaces of which the dry and wet edges were obviously nonlinear. For this type of data, we used both linear and nonlinear regressions to fit the dry and wet edges, and established linear MTVDI and NonL ‐ MTVDI for drought assessment,

respectively (Fig. 6d–e). As for the Imp ‐ MTVDI, of which the Ts was replaced by Ts ‐ Tair, the dry and wet edges of MNDVI ‐ Ts spaces (Fig. 6g (dry edge: R2 = 0.762; wet edge: R2 = 0.033), Fig. 6h (dry edge: R2 = 0.895; wet edge: R2 = 0.280), Fig. 6i (dry edge: R2 = 0.855; wet edge: R2 = 0.049)) were better fitted with linear regression compared with those of NonL ‐ MTVDI (Fig. 6d (dry edge: R2 = 0.001; wet edge: R2 = 0.475), Fig. 6e (dry edge: R2 = 0.589; wet edge: R2 = 0.304), Fig. 6f (dry edge: R2 = 0.043; wet edge: R2 = 0.318)) (Fig. 6). 4.2. Evaluating the proposed MTVDI The time-series correlation analysis between MTVDI, Imp ‐ MTVDI, NonL ‐ MTVDI, MODIS-derived TVDI, iTVDI and precipitation are shown in Fig. 7. The time-series correlation analysis between the MTVDI, Imp ‐ MTVDI, NonL ‐ MTVDI and P/PET are shown in Fig. 8. Results showed that the percentages of pixels that have positive correlation with precipitation follow the orders as (Table 3): TVDI (3.43%) b MTVDI (8.71%) b NonL ‐ MTVDI (12.01%) b Imp ‐ MTVDI (16.98%) b iTVDI (18.36%) (Fig. 7a1–e1). The percentages of pixels with P values of corresponding significance test b0.05 follow the orders as: MTVDI (38.21%) N Imp ‐ MTVDI (29.82%) N NonL ‐ MTVDI (19.74%) N TVDI (16.28%) N iTVDI (5.19%) (Fig.

Table 3 The statistical results of correlation between precipitation and MODIS TVDI, MODIS iTVDI, MTVDI, NonL‐MTVDI and Imp‐MTVDI.The C, N, NE, NW, QTP, S, SEC and SW are the abbreviation of central, northeast, northwest, south, southeast coast and Qinghai-Tibet Plateau, respectively. Region

Drought index

Correlation coefficient b0

N0

Mean

Median

b0.05

0.05–0.1

N0.1

Whole

MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI

91.29% 83.02% 88.00% 96.57% 81.64% 99.92% 99.83% 99.92% 99.10% 80.31% 100.00% 100.00% 99.41% 100.00% 99.80% 100.00% 100.00% 100.00% 100.00% 99.88% 75.16% 45.49% 65.77% 95.34% 71.90% 84.47% 81.87% 81.87% 92.53% 89.99% 99.87% 97.47% 98.99% 96.87% 73.46% 100.00% 99.86% 97.99% 98.90% 68.73% 99.89% 99.89% 97.79% 95.98% 92.46%

8.71% 16.98% 12.01% 3.43% 18.36% 0.08% 0.17% 0.08% 0.90% 19.69% 0.00% 0.00% 0.59% 0.00% 0.20% 0.00% 0.00% 0.00% 0.00% 0.12% 24.84% 54.51% 34.23% 4.66% 28.10% 15.53% 18.13% 18.13% 7.47% 10.01% 0.13% 2.53% 1.01% 3.13% 26.54% 0.00% 0.14% 2.01% 1.10% 31.27% 0.11% 0.11% 2.21% 4.02% 7.54%

−0.15 −0.12 −0.12 −0.13 −0.07 −0.20 −0.18 −0.15 −0.12 −0.06 −0.20 −0.22 −0.16 −0.18 −0.12 −0.22 −0.23 −0.21 −0.18 −0.15 −0.08 0.02 −0.05 −0.14 −0.05 −0.09 −0.10 −0.08 −0.12 −0.11 −0.22 −0.15 −0.15 −0.12 −0.03 −0.24 −0.16 −0.18 −0.15 −0.02 −0.15 −0.14 −0.12 −0.10 −0.07

−0.17 −0.14 −0.13 −0.13 −0.06 −0.20 −0.19 −0.15 −0.12 −0.06 −0.20 −0.21 −0.15 −0.17 −0.12 −0.22 −0.24 −0.21 −0.18 −0.15 −0.07 0.02 −0.04 −0.12 −0.06 −0.10 −0.11 −0.09 −0.12 −0.11 −0.22 −0.16 −0.16 −0.12 −0.04 −0.25 −0.16 −0.18 −0.14 −0.03 −0.14 −0.14 −0.12 −0.10 −0.07

38.21% 29.82% 19.74% 16.28% 5.19% 47.55% 42.58% 16.27% 5.70% 0.16% 46.65% 52.17% 25.79% 32.60% 7.85% 64.95% 66.09% 52.69% 25.12% 11.27% 19.63% 19.47% 14.76% 18.81% 4.84% 14.23% 24.11% 7.41% 17.64% 16.20% 62.81% 25.49% 18.72% 13.74% 0.06% 66.48% 25.50% 41.69% 22.73% 0.00% 22.71% 15.04% 9.04% 4.23% 0.21%

12.78% 13.14% 14.08% 12.18% 4.52% 18.72% 13.41% 19.73% 10.33% 1.06% 17.13% 16.73% 13.98% 16.70% 11.27% 20.05% 12.94% 21.31% 31.55% 17.67% 6.75% 9.39% 7.56% 9.73% 3.16% 8.84% 10.07% 7.54% 9.47% 8.46% 14.55% 17.58% 23.34% 9.78% 0.38% 13.32% 19.34% 17.05% 14.88% 0.14% 15.04% 13.77% 10.30% 8.03% 0.96%

49.00% 57.04% 64.96% 71.54% 90.28% 33.73% 44.01% 64.00% 83.97% 98.78% 36.22% 31.10% 60.24% 50.70% 80.89% 15.01% 20.96% 26.00% 43.33% 71.06% 73.62% 71.14% 77.68% 71.45% 92.00% 76.93% 65.82% 85.06% 72.89% 75.34% 22.64% 56.93% 57.94% 76.49% 99.55% 20.20% 55.16% 41.26% 62.40% 99.86% 62.25% 71.19% 80.65% 87.74% 98.83%

C

N

NE

NW

QTP

S

SEC

SW

P value

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

313

Table 4 The statistical results of correlation between P/PET and MODIS TVDI, MODIS iTVDI, MTVDI, NonL‐MTVDI and Imp‐MTVDI. The C, N, NE, NW, QTP, S, SEC and SW are the abbreviation of central, northeast, northwest, south, southeast coast and Qinghai-Tibet Plateau, respectively. Region

Whole

C

N

NE

NW

QTP

S

SEC

SW

Drought index

MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI MTVDI Imp-MTVDI NonL-MTVDI TVDI iTVDI

Correlation coefficient

P value

b0

N0

Mean

Median

b0.05

0.05–0.1

N0.1

91.97% 78.76% 89.26% 95.40% 71.15% 100.00% 99.34% 99.67% 100.00% 75.25% 100.00% 100.00% 100.00% 99.22% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 80.06% 49.59% 72.61% 94.61% 69.80% 79.73% 61.33% 77.07% 87.57% 78.90% 100.00% 84.36% 99.49% 95.12% 35.22% 100.00% 98.87% 97.74% 100.00% 53.41% 100.00% 98.36% 98.36% 94.78% 89.13%

8.03% 21.28% 10.74% 4.56% 28.85% 0.00% 0.66% 0.33% 0.00% 24.75% 0.00% 0.00% 0.00% 0.78% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 19.94% 50.41% 27.39% 5.39% 30.20% 20.27% 38.67% 22.93% 12.43% 21.10% 0.00% 15.64% 0.51% 4.88% 64.78% 0.00% 1.13% 2.26% 0.00% 46.59% 0.00% 1.64% 1.64% 5.22% 10.87%

−0.21 −0.11 −0.16 −0.17 −0.07 −0.29 −0.21 −0.20 −0.18 −0.06 −0.28 −0.26 −0.23 −0.24 −0.18 −0.32 −0.28 −0.28 −0.25 −0.20 −0.15 0.01 −0.08 −0.19 −0.08 −0.09 −0.03 −0.07 −0.11 −0.07 −0.29 −0.09 −0.21 −0.14 0.04 −0.30 −0.14 −0.22 −0.19 0.00 −0.17 −0.14 −0.14 −0.14 −0.08

−0.23 −0.12 −0.17 −0.17 −0.06 −0.28 −0.19 −0.20 −0.18 −0.06 −0.28 −0.26 −0.22 −0.24 −0.17 −0.31 −0.28 −0.28 −0.24 −0.19 −0.15 0.00 −0.09 −0.17 −0.06 −0.09 −0.03 −0.09 −0.10 −0.06 −0.29 −0.10 −0.21 −0.14 0.04 −0.30 −0.12 −0.24 −0.20 −0.01 −0.18 −0.15 −0.14 −0.15 −0.08

57.48% 29.83% 38.06% 37.36% 11.76% 84.82% 45.54% 48.51% 39.60% 3.96% 79.84% 70.54% 59.69% 71.88% 32.81% 95.28% 81.97% 87.55% 74.55% 40.35% 34.85% 25.45% 20.91% 39.54% 16.17% 11.73% 8.27% 5.60% 18.11% 5.48% 87.69% 11.54% 56.67% 20.82% 4.63% 85.88% 20.34% 64.97% 46.02% 0.00% 34.02% 19.26% 10.25% 25.65% 1.74%

8.88% 9.16% 12.93% 12.74% 5.71% 4.95% 10.89% 17.82% 16.50% 3.96% 10.08% 9.30% 16.28% 13.28% 51.56% 3.43% 9.01% 8.15% 13.84% 14.91% 10.37% 8.10% 9.08% 9.48% 5.45% 10.67% 5.07% 5.87% 6.49% 3.84% 3.85% 5.38% 18.72% 17.22% 3.34% 6.78% 11.30% 9.04% 18.75% 1.70% 21.31% 20.49% 23.77% 13.04% 4.35%

33.64% 61.01% 49.01% 49.90% 82.53% 10.23% 43.56% 33.66% 43.89% 92.08% 10.08% 20.16% 24.03% 14.84% 15.63% 1.29% 9.01% 4.29% 11.61% 44.74% 54.78% 66.45% 70.02% 50.98% 78.38% 77.60% 86.67% 88.53% 75.41% 90.68% 8.46% 83.08% 24.62% 61.95% 92.03% 7.34% 68.36% 25.99% 35.23% 98.30% 44.67% 60.25% 65.98% 61.30% 93.91%

7a3–e3). Similarly, the percentages of pixels that have positive correlation with P/PET follow the orders as (Table 4): TVDI (4.56%) b MTVDI (8.03%) b NonL‐MTVDI (10.74%) b Imp‐MTVDI (21.28%) b iTVDI (28.85%) (Fig. 8a1– e1). The percentages of pixels with P values of significance test b0.05 follow the orders as: MTVDI (57.48%) N NonL ‐ MTVDI (38.06%) N TVDI (37.36%) N Imp‐MTVDI (29.83%) N iTVDI (11.76%) (Fig. 8a3–e3). Obviously, the trends of MODIS-derived TVDI and the proposed MTVDI showed the most consistent with the negative variation of precipitation and P/PET than those of Imp ‐ MTVDI, NonL ‐ MTVDI, and iTVDI. The proposed MTVDI always had higher negative correlations with precipitation and P/PET, and showed better performances of significance test than the MODIS-derived TVDI in central, north, northeast, south, southeast coast and southwest China, except for northwest and Qinghai-Tibet Plateau region of China (Table 3 and Table 4). In addition, the histograms showed that the correlation coefficient of TVDI (Precipitation: Stdev (Standard Deviation) = 0.078; P/PET: Stdev = 0.110) and iTVDI (Precipitation: Stdev = 0.082; P/PET: Stdev = 0.115) are distributed more concentratedly (Figs. 71a2–b2 and 8a2–b2), while the distribution of MTVDI (Precipitation: Stdev = 0.118; P/PET: Stdev = 0.139), Imp ‐ MTVDI (Precipitation: Stdev = 0.190; P/PET: Stdev = 0.155) and NonL ‐ MTVDI (Precipitation: Stdev = 0.104; P/PET: Stdev = 0.127) are distributed discretely (Figs. 7c2–e2 and 8c2–e2). It might indicate that

the MTVDI could better separate the drought levels in different degrees than MODIS-derived TVDI. The performances of MTVDI were better than those of NonL MTVDI model. It is probably because that the nonlinear fitting method of dry and wet edges might lead to uncertain errors at both ends of the scatter plot where there is the less samples than those of linear models. This might result in inconsistent between the model simulation results and actual drought situation. The correlation coefficients between Imp ‐ MTVDI and precipitation, P/PET were also much lower compared with MTVDI. The data sources of land surface temperature that is mainly obtained using remote sensing inversion model are significant different from the air temperature, which is obtained using in-situ meteorological data. This differences might lead to poor reliability of the Imp ‐ MTVDI model when applied for longterm large-scale applications. To further validate the performances of MTVDI and TVDI in drought monitoring, the temporal variation trends of soil moisture (SM), MTVDI and TVDI are analyzed and compared (Fig. 9). There are obvious negative correlations between MTVDI and SM (maximum R = −0.27; average R = − 0.52; minimum R = − 0.83), as well as TVDI and SM (maximum R = 0.14; average R = −0.37; minimum R = −0.83). The negative correlations between MTVDI and SM (Central China:

314

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

315

Table 5 The correlation coefficient (R) and P value (P) between MTVDI, MODIS TVDI and soil moisture. The C, N, NE, NW, QTP, S, SEC and SW are the abbreviation of central, northeast, northwest, south, southeast coast and Qinghai-Tibet Plateau, respectively. Index

b1 h1 g3 a3 Mean of C b3 Mean of N c2 a2 c3 h2 Mean of NE b2 c1 Mean of NW g1 a1 Mean of QTP f2 e1 e2 f1 e3 Mean of S d3 d1 f3 d2 g2 Mean of SEC h3 Mean of SW

Longitude

Latitude

Name

Province

Location

106.48 109.23 111.76 115.56

37.28 34.4 37.25 37.55

Weizhou Lintong Fenyang Jixian

Ningxia Shaanxi Shanxi Hebei

C C C C

111.45

41.1

Wuchuan

Inner Mongolia

N

116.12 123.97 119.4 123.32

43.95 44.25 43.83 41.42

Xilinhaote Changling Balin Dengta

Inner Mongolia Jilin Inner Mongolia Liaoning

NE NE NE NE

85.25 75.95

44.85 39.15

Paotai Keketao

Xinjiang Xinjiang

NW NW

100.3 91.13

29.05 29.67

Daocheng Lasa

Sichuan Tibet

QTP QTP

114.7 117.2 114.97 111.73 112.58

25.68 29.3 31.18 27.23 33.03

Nankang Jingdezhen Macheng Shaodong Nanyang

Jiangxi Jiangxi Hubei Hunan Henan

S S S S S

113.58 118.11 109.61 114.68 119.7

24.8 27.33 23.11 23.73 30.23

Shaoguan Jiangyang Guigang Heyuan Lin'an

Guangdong Fujian Guangxi Guangdong Zhejiang

SEC SEC SEC SEC SEC

103.66

25.03

Luliang

Yunnan

SW

R = − 0.60; North China: R = − 0.3, Northeast China: R = − 0.36, Northwest China: R = − 0.41, Qinghai-Tibet Plateau: R = − 0.81, South China: R = − 0.6, Southeast coastal area: R = − 0.5, Southwest China: R = − 0.32) are relatively higher than those of TVDI and SM (Central China: R = − 0.21; North China: R = − 0.01, Northeast China: R = − 0.17, Northwest China: R = − 0.59, Qinghai-Tibet Plateau: R = − 0.80, South China: R = − 0.41, Southeast coastal area: R = − 0.48, Southwest China: R = − 0.18) in most regions of China (Table 5), which is consistent with previous validation based on precipitation and P/PET. 4.3. Applying MTVDI in drought monitoring across China In view of the above good performances of the proposed MTVDI, this paper used AMSR‐ E Tb data to derive the MTVDI for assessing the longterm drought variations in China from 2003 to 2010. Former studies divided the TVDI into five ranges in intervals of 0.2 for categorizing different drought levels (Son et al., 2012; Wang et al., 2004): very wet (0.0 b TVDI ≤ 0.2), wet (0.2 b TVDI ≤ 0.4), balanced (0.4 b TVDI ≤ 0.6), dry (0.6 b TVDI ≤ 0.8), and very dry (0.8 b TVDI ≤ 1.0). In order to better differ the drought levels in China, the proposed MTVDI was categorized into seven classes in this paper: extremely wet (0 ≤ MTVDI b 0.1), moderately wet (0.1 ≤ MTVDI b 0.2), slightly wet (0.2 ≤ MTVDI b 0.4), balanced (0.4 ≤ MTVDI b 0.6), slightly drought (0.6 ≤ MTVDI b 0.8), moderately drought (0.8 ≤ MTVDI b 0.9), extremely drought (0.9 ≤ MTVDI ≤ 1). The time-series drought monitoring results were listed in Supplementary Fig. S1. Take the spatiotemporal variations of monthly MTVDI in 2010 as an example (Fig. 10), from January to April, the drought occurred mainly in

MTVDI

TVDI

R

P

R

P

−0.55 −0.73 −0.35 −0.77 −0.6 −0.3 −0.3 −0.41 −0.37 −0.27 −0.37 −0.36 −0.54 −0.27 −0.41 −0.78 −0.83 −0.81 −0.74 −0.61 −0.41 −0.63 −0.61 −0.6 −0.66 −0.72 −0.41 −0.42 −0.29 −0.5 −0.32 −0.32

b0.01 b0.01 b0.01 b0.01

−0.42 −0.3 −0.2 0.07 −0.21 −0.01 −0.01 −0.3 −0.18 −0.05 −0.15 −0.17 −0.58 −0.6 −0.59 −0.83 −0.76 −0.80 −0.65 −0.62 −0.47 −0.44 0.14 −0.41 −0.58 −0.64 −0.48 −0.39 −0.29 −0.48 −0.18 −0.18

b0.01 b0.01 0.05 0.49

b0.05 b0.01 b0.01 b0.05 b0.01 b0.01 b0.05 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01

0.91 b0.01 0.09 0.62 0.17 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 b0.01 0.18 b0.01 b0.01 b0.01 b0.01 b0.01 0.08

the southwest China (such as Yunnan and western of Guangxi), which was also detected in former studies (Mao et al., 2012; Wu et al., 2015). It was mainly because that during this period the southwest China encountered with high temperature and large evaporation and there would largely contributed to the regional water deficit for plant and thereby induced the occurrence of severe drought (S. Wang et al., 2011). From April to June, the drought conditions in the southern regions of China were significantly eased, due to arrival of the rainy season (May to September) in southern China (Zhou et al., 2010). During wet-season period, the total rainfall was about 80% of the annual rainfall (Yan et al., 2015). The abundant rainfall could effectively mitigate the drought levels (Fig. 11b). On the contrary, Xinjiang and Inner Mongolia regions in northern China at this period were experiencing significant drought events (Fig. 11b), possibly due to rising surface temperatures, and the less rainfall. From September to November, south China (Guangdong, Fujian and eastern of Guangxi) were in the dry-season period, during which the precipitation was significantly reduced (340 mm, 20% of annual) (Fig. 11b) (Yan et al., 2015). However, the air temperature did not show a significant decline, which still cause a lot of water consumption due to vegetation transpiration and evaporation (Chen et al., 2017). It was also worth mentioning that, in the northeastern China such as the northeastern of Inner Mongolia, severe droughts occurred mainly during the period from June to August (Zhang and Jia, 2013), while the Qinghai-Tibet Plateau becomes more and more moist during May to August due to an increased precipitation (Fig. 11b) and glacier (frozen soil) melting (Pritchard, 2017). In the Xinjiang and northwest of Inner Mongolia have maintained a very dry state, because it is located in the arid zone of China (Xu et al., 2015), which is perennial in a dry state.

Fig. 9. Temporal trends of SM, MTVDI and MODIS TVDI. Horizontal axis 1 to 96 represent date from January 2003 to December 2010. The left vertical axis represents TVDI and MTVDI, and the right vertical axis represents the soil moisture. The breakpoint of the curve indicates that the corresponding month data is missing. The time-series MTVDI and MODIS TVDI imageries from 2003 to 2010 were shown in Figs. S1 and S2, respectively. The details of sites are shown in Table 5.

316

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

Fig. 10. The spatiotemporal variation of monthly MTVDI in 2010.

5. Discussion 5.1. Compared with former drought monitoring in China Our proposed MTVDI revealed significant seasonal droughts in different regions of China, which were consistent with previous studies based on a variety of data sources such as satellite remote sensing indices, meteorological data and soil moisture data (Lai et al., 2015; Huang et al., 2010, 2013; S. Wang et al., 2011). South China is a particular region where has sufficient rainfall in wet seasons but suffers frequent droughts in dry seasons. In recent years, there was a significant drying trend (Ma and Ren, 2007) in south China. The frequency of dry-season droughts (December–May) also increased significantly (He et al., 2011; A. Wang et al., 2011). Huang et al. (2013) investigated the longterm drought conditions of South China using the meteorological data of 268 meteorological stations from 1959 to 2008. They also found that droughts in Guangxi and Guangdong provinces mainly occurred in winter and autumn seasons (September–November), and in Fujian province droughts mainly happened in autumn and summer seasons (June–August), while in Hubei, Hunan, Anhui and Jiangxi provinces

drought mainly occurred in the autumn season, which is in good agreement with the main findings of our study (Supplementary Fig. S1). In the eastern part of Sichuan and Chongqing provinces, the drought mainly occurred in the summer. A severe drought was detected in 2006 during the period from July to August (Pang and Qin, 2013). Another study conducted by Hao et al. (2015) proposed the OMDI (Optimized Meteorological Drought Index) and the OVDI (Optimized Vegetation Drought Index) based on multi-source satellite data for monitoring the drought variations in southwestern China from 2005 to 2009. The time-series imageries of OMDI and OVDI showed highly consistent with the spatial pattern and temporal trend of our MTVDI (Supplementary Fig. S1c–g). North China was also experiencing an obvious increasing trend of drought in the past decade. Ma et al. (2012) used SPI (standardized precipitation index) and relative moisture index to analyze drought conditions in the Northeast China from 1961 to 2009. The results agreed well with the results of MTVDI that Northeast China tend to encounter with more drought events in the period from May to September. The frequency of droughts in Northeast China is higher from 2001 to 2009. Xuan et al. (2016) used meteorological data to analyze drought variations in Xinjiang from 1963 to 2012 and found that Xinjiang maintain

Fig. 11. The variation of monthly average (a) P/PET and (b) Precipitation in 2010. The value of P/PET and precipitation in the seven provinces were derived using the spatial statistical analysis of ArcGIS software.

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

the drying conditions throughout the whole year. This is also clearly presented in the MTVDI imageries (Supplementary Fig. S1), where always showed the persistent droughts. There is another thing worth discussion here. Our results indicated that soil moisture deficit and low precipitation does not always coincide with the plant drought due to the different drought resistance capacities of plant (Jones, 2007). A useful comparison is that, for regions with similar levels of MTVDI such as the Wuchuan (SM = 0.207, MTVDI = 0.835, Fig. 9b3, July 2009) and Jianyang (SM = 0.348, MTVDI = 0.840, Fig. 9d1, July 2010), Paotai (SM = 0.072, MTVDI = 1, Fig. 9b2, July 2005) and Jingdezhen (SM = 0.316, MTVDI = 1, Fig. 9e1, July 2005), Xilingele (SM = 0.102, MTVDI = 0.991, Fig. 9c2, September 2005) and Shaoguan (SM = 0.239, MTVDI = 0.986, Fig. 9d3, September 2005), and Lin'an (SM = 0.354, MTVDI = 0.861, Fig. 9g2, September 2010) and Fenyang (SM = 0.144, MTVDI = 0.862, Fig. 9g3, May 2003), the soil moisture obviously differed greatly. On the contrary, under the same soil water content, the MTVDI of Daocheng (SM = 0.305, MTVDI = 0.236, Fig. 9g1, July 2008) and Lasa (SM = 0.211, MTVDI = 0.342, Fig. 9a1, September 2010) are much lower than those of Guigang (SM = 0.305, MTVDI = 0.792, Fig. 9f3, August 2010) and Jixian (SM = 0.210, MTVDI = 0.607, Fig. 9a3, September 2006), respectively. Similarly, the south China in dry season that has obvious more precipitation than those in northwestern China, however, tends to encountered with higher-frequency and more severe level of drought (Fig. 10). The conclusion is also supported by various case studies, which got poor correlations between soil moisture, precipitation and TVDI at large scale applications (Wang et al., 2004; Wang et al., 2010; Sun et al., 2012). Plant in different regions or under different climates conditions might suffer obviously different levels of drought under the same soil moisture/precipitation levels. It is mainly due to differences in background climate (Zheng et al., 2010) and drought resistance capacities of vegetation (Hao et al., 2017). Soil water deficit and low precipitation would much refer to the hydrological and a meteorological drought rather than the plant drought. Therefore, it is greatly limited when using soil moisture and precipitation to characterize the drought of vegetation if the regional background and vegetation type are not considered. 5.2. Some uncertainness In the Taklimakan Desert, northern Tibet and northwestern Qinghai province, the proposed MTVDI, as well as all five TVDI models, showed positive correlations with precipitation (Fig. 7) and P/PET (Fig. 8). It was believed that precipitation in the regions was extremely low and the meltwater of snow and glaciers from the northern slope of the Himalaya Mountains and mountains in Qinghai-Tibetan Plateau was perhaps the main water supply (Chen et al., 2017). Besides, in some areas such as the Yangtze River Basin Delta, the correlation between precipitation, P/PET and MTVDI are slightly weaker (Figs. 7 and 8). It is probably because that there are a greater number

317

of large water bodies such as Dongting and Poyang Lakes, which might lead to more obvious mixed-pixel problems due to coarse spatial resolution of the AMSR ‐ E imageries (Verhoeye and De Wulf, 2002). Large water bodies could bring lower brightness temperatures and higher polarization differences (McFarland et al., 1990). Some studies also found that the high atmospheric water vapor of water surfaces could greatly influence the high frequency Tb (especially in 89.0 GHz) of AMSR‐ E imageries (Li and Ma, 2015). The atmospheric transmittance of 89 GHz could even not be N0.7 due to the serious impacts by water vapor (Li and Ma, 2015). Finally, the proposed MTVDI model was constructed based on the 8day MODIS Ts and 16-day MODIS NDVI products. Two different temporal resolutions might bring incompletely matching and thereby lead to some uncertainties. 6. Conclusion This paper for the first time developed a microwave TVDI (MTVDI) using passive microwave remote sensing for drought monitoring. The MTVDI not only utilized the advantage of TVDI which combine Ts with vegetation index, but also overcame the shortage of visible/near infrared/thermal remote sensing, which are greatly influenced by weather conditions. Our results indicated that MTVDI are highly correlated with precipitation, SM and P/PET, suggesting a great potential of MTVDI to monitor drought at large spatial scale. However, the proposed MTVDI still has some uncertainties in regions such as the Yangtze River Basin Delta with large water bodies and Xinjiang and Inner Mongolia regions, where are widely covered with desert and Gobi. In addition, this paper mainly focused on large spatial scale and long term drought monitoring and only uses satellite data for model validation. Further studies are needed to develop the higher spatial and temporal resolutions of MTVDI for short-term and small spatial-scale drought monitoring. Acknowledgment The authors thank the editor and the anonymous reviewers for their thorough reviews and constructive comments. This study was supported by the National Natural Science Foundation of China [grant numbers 41430529, 31500357, 41401055, 31600353], the Natural Science Foundation of Guangdong Province, China [grant numbers 2014A030310233, 2015A030313809, 2015A030313811, 2016A030310450], the Pearl River S&T Nova Program of Guangzhou [grant number 201610010134], Special Plan Project of Guangdong Province [grant number 2016TQ03Z354], and the Key Programs of the Chinese Academy of Sciences [grant numbers KFZD-SW-312, QYZDJ-SSW-DQC003], the Water Resource Science and Technology Innovation Program of Guangdong Province [grant numbers 2016-16], and the Special Funds for Innovation Development and Capacity Building of GDAS [grant numbers 2017GDASCX-0501].

Appendix A Table A1 The land surface type in study area. Codes

Label

Proportion (%)

11 14 20 30 40 50 60 70 90 100 110

Post-flooding or irrigated croplands (or aquatic) Rained croplands Mosaic cropland (50–70%)/vegetation (grassland/shrubland/forest) (20–50%) Mosaic vegetation (grassland/shrubland/forest) (50–70%)/cropland (20–50%) Closed to open (N15%) broadleaved evergreen or semi-deciduous forest (N5 m) Closed (N40%) broadleaved deciduous forest (N5 m) Open (15–40%) broadleaved deciduous forest/woodland (N5 m) Closed (N40%) needleleaved evergreen forest (N5 m) Open (15–40%) needleleaved deciduous or evergreen forest (N5 m) Closed to open (N15%) mixed broadleaved and needleleaved forest (N5 m) Mosaic forest or shrubland (50–70%)/grassland (20–50%)

5.132 14.440 5.302 8.727 2.336 0.902 0.115 5.715 3.402 2.934 1.923 (continued on next page)

318

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

Table A1 (continued) Codes

Label

Proportion (%)

120 130 140 150 170 180 190 200 210 220

Mosaic grassland (50–70%)/forest or shrubland (20–50%) Closed to open (N15%) (broadleaved or needleleaved, evergreen or deciduous) shrubland (b5 m) Closed to open (N15%) herbaceous vegetation (grassland, savannas or lichens/mosses) Sparse (b15%) vegetation Closed (N40%) broadleaved forest or shrubland permanently flooded - saline or brackish water Closed to open (N15%) grassland or woody vegetation on regularly flooded or waterlogged soil - fresh, brackish or saline water Artificial surfaces and associated areas (urban areas N50%) Bare areas Water bodies Permanent snow and ice

1.348 3.079 9.993 3.809 b0.001 0.005 0.655 27.830 1.162 1.190

Table A2 The proportion of land surface types corresponding to points in the triangular space of each month in 2003. Codesa

11 14 20 30 40 50 60 70 90 100 110 120 130 140 150 200 210 220 Sum a

Proportion (%) Jan.

Feb.

Mar.

Apr.

May.

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

6.507 18.425 5.571 9.532 4.189 0.285 0.217 8.231 0.034 3.584 0.479 0.696 5.354 6.975 2.237 26.027 0.936 0.034 100

6.476 19.228 5.800 9.647 4.081 0.299 0.211 8.339 0.022 3.648 0.532 0.787 5.201 7.053 1.907 25.771 0.954 0.044 100

5.970 19.129 6.000 9.525 3.684 0.369 0.190 7.618 0.220 3.494 0.729 0.948 4.692 7.368 2.216 26.847 0.938 0.060 100

5.397 16.387 5.586 8.968 2.971 0.735 0.150 6.369 3.018 3.255 1.881 1.304 3.737 8.360 3.326 27.568 0.806 0.182 100

5.201 15.882 5.540 8.870 2.879 0.707 0.143 6.472 2.946 3.172 1.917 1.368 3.570 9.050 3.270 27.766 0.797 0.451 100

5.005 15.303 5.550 8.813 2.782 0.696 0.136 6.447 2.818 3.069 1.864 1.334 3.421 10.018 3.227 27.745 0.803 0.968 100

4.997 15.271 5.534 8.794 2.774 0.694 0.136 6.427 2.810 3.067 1.859 1.330 3.410 10.131 3.232 27.697 0.801 1.037 100

4.987 15.232 5.522 8.768 2.768 0.692 0.136 6.414 2.804 3.061 1.855 1.327 3.403 10.230 3.218 27.709 0.799 1.077 100

4.978 15.203 5.519 8.759 2.763 0.691 0.135 6.402 2.799 3.055 1.851 1.325 3.397 10.276 3.212 27.729 0.805 1.104 100

5.069 15.504 5.578 8.879 2.822 0.705 0.138 6.494 2.851 3.113 1.883 1.353 3.469 9.956 3.192 27.387 0.785 0.822 100

5.395 16.338 5.524 8.918 3.090 0.752 0.152 6.764 1.665 3.218 1.729 1.321 3.802 8.598 3.410 28.018 0.849 0.456 100

5.735 17.382 5.354 9.017 3.504 0.409 0.177 7.408 0.158 3.263 0.688 0.920 4.387 8.701 2.649 29.020 0.930 0.297 100

The land cover types of the identification codes are the same as in Table A1.

Appendix B. Supplementary data Supplementary data to this article can be found online at http://dx.doi. org/10.1016/j.rse.2017.07.012.

References AghaKouchak, A., Farahmand, A., Melton, F.S., Teixeira, J., Anderson, M.C., Wardlow, B.D., Hain, C.R., 2015. Remote sensing of drought: progress, challenges and opportunities. Rev. Geophys. 53 (2), 452–480. Andela, N., Liu, Y.Y., Van Dijk, A.I.J.M., De Jeu, R.A.M., McVicar, T.R., 2013. Global changes in dryland vegetation dynamics (1988–2008) assessed by satellite remote sensing: comparing a new passive microwave vegetation density record with reflective greenness data. Biogeosciences 10 (10), 6657. Andreadis, K.M., Clark, E.A., Wood, A.W., Hamlet, A.F., Lettenmaier, D.P., 2005. Twentiethcentury drought in the conterminous United States. J. Hydrometeorol. 6 (6), 985–1001. Arino, O., Perez, J.J.R., Kalogirou, V., Bontemps, S., Defourny, P., Van Bogaert, E., 2012. Global Land Cover map for 2009 (GlobCover 2009). Ashcroft, P., Wentz, F., 2000. Algorithm Theoretical Basis Document, AMSR Level 2A Algorithm. RSS, Santa Rosa, CA, p. 121 (Tech. Rep). Bajgain, R., Xiao, X., Wagle, P., Basara, J., Zhou, Y., 2015. Sensitivity analysis of vegetation indices to drought over two tallgrass prairie sites. ISPRS J. Photogramm. Remote Sens. 108, 151–160. Bajgiran, P.R., Darvishsefat, A.A., Khalili, A., Makhdoum, M.F., 2008. Using AVHRR-based vegetation indices for drought monitoring in the Northwest of Iran. J. Arid Environ. 72 (6), 1086–1096. Bhuiyan, C., Singh, R.P., Kogan, F.N., 2006. Monitoring drought dynamics in the Aravalli region (India) using different indices based on ground and remote sensing data. Int. J. Appl. Earth Obs. Geoinf. 8 (4), 289–302. Cai, W., Cowan, T., Briggs, P., Raupach, M., 2009. Rising temperature depletes soil moisture and exacerbates severe drought conditions across southeast Australia. Geophys. Res. Lett. 36 (21). Cao, X., Feng, Y., Wang, J., 2016. An improvement of the Ts-NDVI space drought monitoring method and its applications in the Mongolian plateau with MODIS, 2000–2012. Arab. J. Geosci. 9 (6), 1–14.

Carlson, T., 2007. An overview of the “triangle method” for estimating surface evapotranspiration and soil moisture from satellite imagery. Sensors 7 (8), 1612–1629. Carlson, T.N., Gillies, R.R., Perry, E.M., 1994. A method to make use of thermal infrared temperature and NDVI measurements to infer surface soil water content and fractional vegetation cover. Remote Sens. Rev. 9 (1–2), 161–173. Carlson, T.N., Gillies, R.R., Schmugge, T.J., 1995. An interpretation of methodologies for indirect measurement of soil water content. Agric. For. Meteorol. 77 (3), 191–205. Chen, D., Huang, J., Jackson, T.J., 2005. Vegetation water content estimation for corn and soybeans using spectral indices derived from MODIS near-and short-wave infrared bands. Remote Sens. Environ. 98 (2), 225–236. Chen, S.S., Chen, X.Z., Chen, W.Q., Su, Y.X., Li, D., 2011a. A simple retrieval method of land surface temperature from AMSR-E passive microwave data—a case study over Southern China during the strong snow disaster of 2008. Int. J. Appl. Earth Obs. Geoinf. 13 (1), 140–151. Chen, J., Wang, C., Jiang, H., Mao, L., Yu, Z., 2011b. Estimating soil moisture using Temperature–Vegetation Dryness Index (TVDI) in the Huang-huai-hai (HHH) plain. Int. J. Remote Sens. 32 (4), 1165–1177. Chen, X.Z., Chen, S.S., Zhong, R.F., Su, Y.X., Liao, J.S., Li, D., Li, X., 2012. A semi-empirical inversion model for assessing surface soil moisture using AMSR-E brightness temperatures. J. Hydrol. 456, 1–11. Chen, X.Z., Li, Y., Su, Y.X., Han, L.S., Liao, J.S., Yang, S.B., 2014a. Mapping global surface roughness using AMSR-E passive microwave remote sensing. Geoderma 235, 308–315. Chen, X., Su, Y., Li, Y., Han, L., Liao, J., Yang, S., 2014b. Retrieving China's surface soil moisture and land surface temperature using AMSR-E brightness temperatures. Remote Sens. Lett. 5 (7), 662–671. Chen, X., Su, Y., Liao, J., Shang, J., Dong, T., Wang, C., et al., 2016. Detecting significant decreasing trends of land surface soil moisture in eastern china during the past three decades (1979–2010). J. Geophys. Res.-Atmos. 121 (10), 5177–5192. Chen, X., Liu, X., Liu, Z., Zhou, P., Zhou, G., Liao, J., Liu, L., 2017. Spatial clusters and temporal trends of seasonal surface soil moisture across China in responses to regional climate and land cover changes. Ecohydrology 10 (2). Deering, D.W., Rouse, J.W., 1975. Measuring ‘forage production’ of grazing units from Landsat MSS data. International Symposium on Remote Sensing of Environment, 10th, pp. 1169–1178 (Ann Arbor, Mich). Di Gregorio, A., 2005. Land Cover Classification System: Classification Concepts and User Manual: LCCS (No. 8). Food & Agriculture Org.

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320 Du, L., Tian, Q., Yu, T., Meng, Q., Jancso, T., Udvardy, P., Huang, Y., 2013. A comprehensive drought monitoring method integrating MODIS and TRMM data. Int. J. Appl. Earth Obs. Geoinf. 23, 245–253. Escorihuela, M.J., Quintana-Seguí, P., 2016. Comparison of remote sensing and simulated soil moisture datasets in Mediterranean landscapes. Remote Sens. Environ. 180, 99–114. Fan, L., Xiao, Q., Wen, J., Liu, Q., Tang, Y., You, D., ... Li, X., 2015. Evaluation of the airborne CASI/TASI Ts-VI space method for estimating near-surface soil moisture. Remote Sens. 7 (3), 3114. Feng, L., Li, T., Yu, W., 2014. Cause of severe droughts in southwest China during 1951– 2010. Clim. Dyn. 43, 2033–2042. Fily, M., Royer, A., Goıta, K., Prigent, C., 2003. A simple retrieval method for land surface temperature and fraction of water surface determination from satellite microwave brightness temperatures in sub-arctic areas. Remote Sens. Environ. 85 (3), 328–338. Gao, B.C., 1996. NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens. Environ. 58 (3), 257–266. Gao, F., Wang, F., Ma, Y., 2003. Retrieval of land surface temperatures on the Tibetan Plateau using passive microwave data. Geoscience and Remote Sensing Symposium, 2003. IGARSS'03. Proceedings. 2003 IEEE International. vol. 5. IEEE, pp. 3145–3147. Gao, Z., Gao, W., Chang, N.B., 2011. Integrating temperature vegetation dryness index (TVDI) and regional water stress index (RWSI) for drought assessment with the aid of LANDSAT TM/ETM+ images. Int. J. Appl. Earth Obs. Geoinf. 13 (3), 495–503. Goddard, S., Harms, S.K., Reichenbach, S.E., Tadesse, T., Waltman, W.J., 2003. Geospatial decision support for drought risk management. Commun. ACM 46 (1), 35–37. Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., ... Toulmin, C., 2010. Food security: the challenge of feeding 9 billion people. Science 327 (5967), 812–818. Goetz, S.J., 1997. Multi-sensor analysis of NDVI, surface temperature and biophysical variables at a mixed grassland site. Int. J. Remote Sens. 18 (1), 71–94. Gu, Y., Brown, J.F., Verdin, J.P., Wardlow, B., 2007. A five–year analysis of MODIS NDVI and NDWI for grassland drought assessment over the central Great Plains of the United States. Geophys. Res. Lett. 34 (6). Gutman, G.G., 1990. Towards monitoring droughts from space. J. Clim. 3 (2), 282–295. Hagman, G., 1984. Prevention Better than Cure: Report on Human and Natural Disasters in the Third World. Swedish Red Cross, Stockholm. Hao, C., Zhang, J., Yao, F., 2015. Combination of multi-sensor remote sensing data for drought monitoring over Southwest China. Int. J. Appl. Earth Obs. Geoinf. 35, 270–283. Hao, Y.Y., Luo, X.B., Zhong, B., Yang, A.X., 2017. Methods of the national vegetation classification based on vegetation partition. Remote Sens. Technol. Appl. 32 (2), 315–323. Harris, I.P.D.J., Jones, P.D., Osborn, T.J., Lister, D.H., 2014. Updated high–resolution grids of monthly climatic observations–the CRU TS3. 10 dataset. Int. J. Climatol. 34 (3), 623–642. Hassan, Q.K., Bourque, C.P.A., Meng, F.R., Cox, R.M., 2007. A wetness index using terraincorrected surface temperature and normalized difference vegetation index derived from standard MODIS products: an evaluation of its use in a humid forest-dominated region of eastern Canada. Sensors 7 (10), 2028–2048. He, J., Zhang, M., Wang, P., Wang, S., Wang, X., 2011. Climate characteristics of the extreme drought events in Southwest China during recent 50 years. Acta Geograph. Sin. 9 (006). Holmes, T.R.H., De Jeu, R.A.M., Owe, M., Dolman, A.J., 2009. Land surface temperature from Ka band (37 GHz) passive microwave observations. J. Geophys. Res.-Atmos. 114 (D4). Huang, W., Yang, X., Li, M., Zhang, X., Wang, M., Dai, S., Ma, J., 2010. Evolution characteristics of seasonal drought in the south of China during the past 58 years based on standardized precipitation index. Trans. Chin. Soc. Agric. Eng. 26 (7), 50–59. Huang, W.H., Sui, Y., Yang, X.G., Dai, S.W., Li, M.S., 2013. Characteristics and adaptation of seasonal drought in southern china under the background of climate change. v. seasonal drought characteristics division and assessment in southern China. Ying Yong Sheng Tai Xue Bao 24 (10), 2917–2925. Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 83 (1), 195–213. Jackson, T.J., Chen, D., Cosh, M., Li, F., Anderson, M., Walthall, C., ... Hunt, E.R., 2004. Vegetation water content mapping using Landsat data derived normalized difference water index for corn and soybeans. Remote Sens. Environ. 92 (4), 475–482. Jain, S.K., Keshri, R., Goswami, A., Sarkar, A., Chaudhry, A., 2009. Identification of drought–vulnerable areas using NOAA AVHRR data. Int. J. Remote Sens. 30 (10), 2653–2668. Ji, L., Peters, A.J., 2003. Assessing vegetation response to drought in the northern Great Plains using vegetation and drought indices. Remote Sens. Environ. 87 (1), 85–98. Jones, H.G., 2007. Monitoring plant and soil water status: established and novel methods revisited and their relevance to studies of drought tolerance. J. Exp. Bot. 58 (2), 119–130. Karnieli, A., Agam, N., Pinker, R.T., Anderson, M., Imhoff, M.L., Gutman, G.G., ... Goldberg, A., 2010. Use of NDVI and land surface temperature for drought assessment: merits and limitations. J. Clim. 23 (3), 618–633. Kawanishi, T., Sezai, T., Ito, Y., Imaoka, K., Takeshima, T., Ishido, Y., ... Spencer, R.W., 2003. The Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), NASDA's contribution to the EOS for global energy and water cycle studies. IEEE Trans. Geosci. Remote Sens. 41 (2), 184–194. Kogan, F.N., 1991. Observations of the 1990 US drought from the NOAA-11 polar-orbiting satellite. Drought Network News. 3(2), pp. 7–11. Kogan, F.N., 1995a. Application of vegetation index and brightness temperature for drought detection. Adv. Space Res. 15 (11), 91–100. Kogan, F.N., 1995b. Droughts of the late 1980s in the United States as derived from NOAA polar-orbiting satellite data. Bull. Am. Meteorol. Soc. 76 (5), 655–668.

319

Lacava, T., Cuomo, V., Di Leo, E.V., Pergola, N., Romano, F., Tramutoli, V., 2005. Improving soil wetness variations monitoring from passive microwave satellite data: the case of April 2000 Hungary flood. Remote Sens. Environ. 96 (2), 135–148. Lagacé, P.J., 2012. Power flow methods for improving convergence. IECON 2012-38th Annual Conference on IEEE Industrial Electronics Society. IEEE, pp. 1387–1392 (October). Lai, X., Wen, J., Cen, S., Huang, X., Tian, H., Shi, X., 2015. Spatial and temporal soil moisture variations over China from simulations and observations. Adv. Meteorol. 2016. Lambin, E.F., Ehrlich, D., 1996. The surface temperature-vegetation index space for land cover and land-cover change analysis. Int. J. Remote Sens. 17 (3), 463–487. Levenberg, K., 1944. A method for the solution of certain problems in least-squares. Q. Appl. Math. 2, 164–168. Li, M., Ma, Z., 2015. Soil moisture drought detection and multi-temporal variability across China. Sci. China Earth Sci. 58 (10), 1798. Li, H., Li, C., Lin, Y., Lei, Y., 2010. Surface temperature correction in TVDI to evaluate soil moisture over a large area. J. Food Agric. Environ. 8 (3–4), 1141. Li, Q., Zhong, R., Huang, J., Gong, H., 2011. Comparison of two retrieval methods with combined passive and active microwave remote sensing observations for soil moisture. Math. Comput. Model. 54 (3), 1181–1193. Liang, L., Zhao, S.H., Qin, Z.H., He, K.X., Chong, C.H.E.N., Luo, Y.X., Zhou, X.D., 2014. Drought change trend using MODIS TVDI and its relationship with climate factors in China from 2001 to 2010. J. Integr. Agric. 13 (7), 1501–1508. Liu, Z., Shi, X., Warner, E., Ge, Y., Yu, D., Ni, S., Wang, H., 2008. Relationship between oriental migratory locust plague and soil moisture extracted from MODIS data. Int. J. Appl. Earth Obs. Geoinf. 10 (1), 84–91. Liu, Y.Y., de Jeu, R.A., McCabe, M.F., Evans, J.P., van Dijk, A.I., 2011a. Global long–term passive microwave satellite–based retrievals of vegetation optical depth. Geophys. Res. Lett. 38 (18). Liu, Y.Y., Parinussa, R.M., Dorigo, W.A., De Jeu, R.A., Wagner, W., Van Dijk, A.I.J.M., ... Evans, J.P., 2011b. Developing an improved soil moisture dataset by blending passive and active microwave satellite-based retrievals. Hydrol. Earth Syst. Sci. 15 (2), 425. Liu, Y.Y., Dorigo, W.A., Parinussa, R.M., de Jeu, R.A., Wagner, W., McCabe, M.F., ... Van Dijk, A.I.J.M., 2012. Trend-preserving blending of passive and active microwave soil moisture retrievals. Remote Sens. Environ. 123, 280–297. Lourakis, M.I.A., 2005. A Brief Description of the Levenberg-Marquardt Algorithm Implemented by Levmar. Foundation of Research & Technology. Ma, Z., Ren, X., 2007. Drying trend over China from 1951 to 2006. Adv. Clim. Chang. Res. 3 (4), 195–201. Ma, J., Xu, Y., Pan, J., 2012. Drought tendency based on standardized precipitation index (SPI) and relative moisture index over Northeast China from May to September during 1961–2009. J. Meteorol. Environ. 28 (3), 90–95. Mallick, K., Bhattacharya, B.K., Patel, N.K., 2009. Estimating volumetric surface moisture content for cropped soils using a soil wetness index based on surface temperature and NDVI. Agric. For. Meteorol. 149 (8), 1327–1342. Mao, K., Shi, J., Li, Z., Qin, Z., Li, M., Xu, B., 2007. A physics-based statistical algorithm for retrieving land surface temperature from AMSR-E passive microwave data. Sci. China Ser. D Earth Sci. 50 (7), 1115–1120. Mao, K.B., Gao, C.Y., Han, L.J., Zhang, W., 2010. The drought monitoring in China by using AMSR-E data. Second IITA International Conference on Geoscience and Remote Sensing. vol. 1. IEEE, pp. 181–184. Mao, K.B., Ying, M.A., Lang, X.I.A., Tang, H.J., Han, L.J., 2012. The monitoring analysis for the drought in China by using an improved MPI method. J. Integr. Agric. 11 (6), 1048–1058. Marquardt, D.W., 1963. An algorithm for least-squares estimation of nonlinear parameters. J. Soc. Ind. Appl. Math. 11 (2), 431–441. McFarland, M.J., Miller, R.L., Neale, C.M., 1990. Land surface temperature derived from the SSM/I passive microwave brightness temperatures. IEEE Trans. Geosci. Remote Sens. 28 (5), 839–845. McKague, D., Puckett, J.J., Ruf, C., 2010. Characterization of K-band radio frequency interference from AMSR-E, WindSat and SSM/I. Geoscience and Remote Sensing Symposium (IGARSS), 2010 IEEE International. IEEE, pp. 2492–2494 (July). McVicar, T.R., Bierwirth, P.N., 2001. Rapidly assessing the 1997 drought in Papua New Guinea using composite AVHRR imagery. Int. J. Remote Sens. 22 (11), 2109–2128. McVicar, T.R., Jupp, D.L., 1998. The current and potential operational uses of remote sensing to aid decisions on drought exceptional circumstances in Australia: a review. Agric. Syst. 57 (3), 399–468. Mitchell, T.D., Jones, P.D., 2005. An improved method of constructing a database of monthly climate observations and associated high–resolution grids. Int. J. Climatol. 25 (6), 693–712. Mitchell, T.D., Carter, T.R., Jones, P.D., Hulme, M., New, M., 2004. A Comprehensive Set of Climate Scenarios for Europe and the Globe: The Observed Record (1900–2000) and 16 Scenarios (2000−2100). University of East Anglia, p. 30. Moran, M.S., 2004. Thermal infrared measurement as an indicator of planet ecosystem health. Thermal Remote Sensing in Land Surface Processes, pp. 257–282. Moran, M.S., Clarke, T.R., Inoue, Y., Vidal, A., 1994. Estimating crop water deficit using the relation between surface-air temperature and spectral vegetation index. Remote Sens. Environ. 49 (3), 246–263. Neale, C.M., Mcfarland, M.J., Chang, K., 1990. Land-surface-type classification using microwave brightness temperatures from the special sensor microwave/imager. IEEE Trans. Geosci. Remote Sens. 28 (5), 829–838. Nemani, R., Pierce, L., Running, S., Goward, S., 1993. Developing satellite-derived estimates of surface moisture status. J. Appl. Meteorol. 32 (3), 548–557. Nemani, R., Hashimoto, H., Votava, P., Melton, F., Wang, W., Michaelis, A., ... White, M., 2009. Monitoring and forecasting ecosystem dynamics using the Terrestrial Observation and Prediction System (TOPS). Remote Sens. Environ. 113 (7), 1497–1509.

320

L. Liu et al. / Remote Sensing of Environment 199 (2017) 302–320

Njoku, E.G., Ashcroft, P., Chan, T.K., Li, L., 2005. Global survey and statistics of radio-frequency interference in AMSR-E land observations. IEEE Trans. Geosci. Remote Sens. 43 (5), 938–947. Obasi, G.O.P., 1994. WMO's role in the international decade for natural disaster reduction. Bull. Am. Meteorol. Soc. 75 (9), 1655–1661. Owe, M., Jeu, R.D., Holmes, T., 2008. Multisensor historical climatology of satellite-derived global land surface moisture. J. Geophys. Res. Earth Surf. 113 (F1), 196–199. Pan, G.D., Wang, C., Tian, G.L., 2001. Land surface temperature retrieval with SSM/I data. J. Remote Sens. Beijing 5 (4), 258–265. Pang, J., Qin, J., 2013. Advances in characteristics and causes of drought research in Southwest China. J. Nanjing Univ. Inf. Sci. Technol. 5 (2), 127–134. Parkinson, C.L., 2003. Aqua: an Earth-observing satellite mission to examine water and other climate variables. IEEE Trans. Geosci. Remote Sens. 41 (2), 173–183. Petropoulos, G., Carlson, T., Wooster, M., Islam, S., 2009. A review of Ts/VI remote sensing based methods for the retrieval of land surface energy fluxes and soil surface moisture. Prog. Phys. Geogr. 33, 224–250. Piao, S., Ciais, P., Huang, Y., Shen, Z., Peng, S., Li, J., ... Friedlingstein, P., 2010. The impacts of climate change on water resources and agriculture in China. Nature 467 (7311), 43–51. Price, J.C., 1990. Using spatial context in satellite data to infer regional scale evapotranspiration. IEEE Trans. Geosci. Remote Sens. 28 (5), 940–948. Pritchard, H.D., 2017. Asia's glaciers are a regionally important buffer against drought. Nature 545 (7653), 169–174. Qin, Z., Li, W., Gao, M., Zhang, H.O., 2006. An algorithm to retrieve land surface temperature from ASTER thermal band data for agricultural drought monitoring. Remote Sensing. International Society for Optics and Photonics, p. 63591F (September). Rahimzadeh-Bajgiran, P., Omasa, K., Shimizu, Y., 2012. Comparative evaluation of the Vegetation Dryness Index (VDI), the Temperature Vegetation Dryness Index (TVDI) and the improved TVDI (iTVDI) for water stress detection in semi-arid regions of Iran. ISPRS J. Photogramm. Remote Sens. 68, 1–12. Ran, Q., Zhang, Z.X., Zhou, Q.B., Wang, Q., 2005. Soil moisture derivation in China using AVHRR data and analysis of its affecting factors. Geoscience and Remote Sensing Symposium, 2005, IGARSS'05. Proceedings, pp. 4497–4500. Ranganathan, A., 2004. The Levenberg-Marquardt Algorithm. Tutorial on LM Algorithm. pp. 1–5. Rhee, J., Im, J., Carbone, G.J., 2010. Monitoring agricultural drought for arid and humid regions using multi-sensor remote sensing data. Remote Sens. Environ. 114 (12), 2875–2887. Rouse Jr., J., Haas, R.H., Schell, J.A., Deering, D.W., 1974. Monitoring vegetation systems in the Great Plains with ERTS. NASA Special Publication. 351, p. 309. Rüdiger, C., Calvet, J.C., Gruhier, C., Holmes, T.R., De Jeu, R.A., Wagner, W., 2009. An intercomparison of ERS-Scat and AMSR-E soil moisture observations with model simulations over France. J. Hydrometeorol. 10 (2), 431–447. Saleska, S.R., Didan, K., Huete, A.R., Da Rocha, H.R., 2007. Amazon forests green-up during 2005 drought. Science 318 (5850), 612. Sandholt, I., Rasmussen, K., Andersen, J., 2002. A simple interpretation of the surface temperature/vegetation index space for assessment of surface moisture status. Remote Sens. Environ. 79 (2), 213–224. Sellers, P.J., Berry, J.A., Collatz, G.J., Field, C.B., Hall, F.G., 1992. Canopy reflectance, photosynthesis, and transpiration. III. A reanalysis using improved leaf models and a new canopy integration scheme. Remote Sens. Environ. 42 (3), 187–216. Shi, J., Jackson, T., Tao, J., Du, J., Bindlish, R., Lu, L., Chen, K.S., 2008. Microwave vegetation indices for short vegetation covers from satellite passive microwave sensor AMSR-E. Remote Sens. Environ. 112 (12), 4285–4300. Silleos, N.G., Alexandridis, T.K., Gitas, I.Z., Perakis, K., 2006. Vegetation indices: advances made in biomass estimation and vegetation monitoring in the last 30 years. Geocarto Int. 21 (4), 21–28. Singh, R.P., Roy, S., Kogan, F., 2003. Vegetation and temperature condition indices from NOAA AVHRR data for drought monitoring over India. Int. J. Remote Sens. 24 (22), 4393–4402. Son, N.T., Chen, C.F., Chen, C.R., Chang, L.Y., Minh, V.Q., 2012. Monitoring agricultural drought in the Lower Mekong Basin using MODIS NDVI and land surface temperature data. Int. J. Appl. Earth Obs. Geoinf. 18, 417–427. Sun, L., Sun, R., Li, X., Liang, S., Zhang, R., 2012. Monitoring surface soil moisture status based on remotely sensed surface temperature and vegetation index information. Agric. For. Meteorol. s 166–167 (s 166–167), 175–187.

Tadesse, T., Brown, J.F., Hayes, M.J., 2005. A new approach for predicting drought-related vegetation stress: integrating satellite, climate, and biophysical data over the US central plains. ISPRS J. Photogramm. Remote Sens. 59 (4), 244–253. Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8 (2), 127–150. Tucker, C.J., Choudhury, B.J., 1987. Satellite remote sensing of drought conditions. Remote Sens. Environ. 23 (2), 243–251. Verhoeye, J., De Wulf, R., 2002. Land cover mapping at sub-pixel scales using linear optimization techniques. Remote Sens. Environ. 79 (1), 96–104. Wagner, W., Dorigo, W., de Jeu, R., Fernandez, D., Benveniste, J., Haas, E., Ertl, M., 2012. Fusion of active and passive microwave observations to create an essential climate variable data record on soil moisture. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences (ISPRS Annals). 7, pp. 315–321. Wan, Z., Zhang, Y., Zhang, Q., Li, Z.L., 2002. Validation of the land-surface temperature products retrieved from Terra Moderate Resolution Imaging Spectroradiometer data. Remote Sens. Environ. 83 (1), 163–180. Wan, Z., Zhang, Y., Zhang, Q., Li, Z.L., 2004. Quality assessment and validation of the MODIS global land surface temperature. Int. J. Remote Sens. 25 (1), 261–274. Wang, L., Qu, J.J., 2007. NMDI: a normalized multi–band drought index for monitoring soil and vegetation moisture with satellite remote sensing. Geophys. Res. Lett. 34 (20). Wang, C., Qi, S., Niu, Z., Wang, J., 2004. Evaluating soil moisture status in China using the temperature–vegetation dryness index (TVDI). Can. J. Remote. Sens. 30 (5), 671–679. Wang, L., Li, Z., Chen, Q., 2006. The applications of MPDI during the soil moisture retrieval from radiometer in the region with vegetation cover. J. Remote Sens. Beijing 10 (1), 34. Wang, L., Wen, J., Zhang, T., Zhao, Y., Tian, H., Shi, X., ... Lu, S., 2009. Surface soil moisture estimates from AMSR-E observations over an arid area, Northwest China. Hydrol. Earth Syst. Sci. Discuss. 6 (1), 1055–1087. Wang, H., Li, X., Long, H., Xu, X., Bao, Y., 2010. Monitoring the effects of land use and cover type changes on soil moisture using remote-sensing data: a case study in China's Yongding River basin. Catena 82 (3), 135–145. Wang, A., Lettenmaier, D.P., Sheffield, J., 2011. Soil moisture drought in China, 1950–2006. J. Clim. 24, 3257–3271. Wang, S., Zhong, R., Li, Q., Qin, M., 2011. Drought monitoring in Yunnan province with AMSR-E-based data. Proc. of SPIE Vol. vol. 8203, p. 820312-1. Wilhite, D.A., 2000. Drought as a Natural Hazard: Concepts and Definitions. Wilhite, D.A., 2005. Drought as hazard: Understanding the natural and social context. In: Wilhite, D.A. (Ed.), Drought and Water Crisis: Science, Technology, and Management Issues. Taylor and Francis Group, Boca Raton, pp. 5–10. Wu, S.L., Yang, H., 2007. Global land surface temperature retrieval with AMSR-E brightness temperature and MODIS land cover type products. Remote Sens. Technol. Appl. 22 (2), 234–237. Wu, J., Zhou, L., Mo, X., Zhou, H., Zhang, J., Jia, R., 2015. Drought monitoring and analysis in China based on the Integrated Surface Drought Index (ISDI). Int. J. Appl. Earth Obs. Geoinf. 41, 23–33. Xu, K., Yang, D., Yang, H., Li, Z., Qin, Y., Shen, Y., 2015. Spatio-temporal variation of drought in China during 1961–2012: a climatic perspective. J. Hydrol. 526, 253–264. Xuan, H., Zheng, J., Liu, Z., 2016. SPEI-based spatiotemporal variation of drought in Xinjiang. Arid Zone Res. 33 (2), 338–344. Yan, J., Li, K., Wang, W., Zhang, D., Zhou, G., 2015. Changes in dissolved organic carbon and total dissolved nitrogen fluxes across subtropical forest ecosystems at different successional stages. Water Resour. Res. 51 (5), 3681–3694. Yuan, X., Ma, Z., Pan, M., Shi, C., 2015. Microwave remote sensing of short-term droughts during crop growing seasons. Geophys. Res. Lett. 42 (11), 4394–4401. Zhang, A., Jia, G., 2013. Monitoring meteorological drought in semiarid regions using multi-sensor microwave remote sensing data. Remote Sens. Environ. 134, 12–23. Zhang, J., Mu, Q., Huang, J., 2016. Assessing the remotely sensed Drought Severity Index for agricultural drought monitoring and impact analysis in North China. Ecol. Indic. 63, 296–309. Zheng, J., Yin, Y., Li, B., 2010. A new scheme for climate regionalization in China. Acta Geograph. Sin. 65 (1), 3–12. Zhou, G., Wei, X., Luo, Y., Zhang, M., Li, Y., Qiao, Y., et al., 2010. Forest recovery and river discharge at the regional scale of Guangdong province, China. Water Resour. Res. 46 (9), 5109–5115. Zhou, G., Wei, X., Chen, X., Zhou, P., Liu, X., Xiao, Y., et al., 2015. Global pattern for the effect of climate and land cover on water yield. Nat. Commun. 6 (3), 5918.