Elsevier

Remote Sensing of Environment

Volume 139, December 2013, Pages 18-34
Remote Sensing of Environment

Intra- and inter-annual variations in snow–water storage in data sparse desert–mountain regions assessed from remote sensing

https://doi.org/10.1016/j.rse.2013.07.033Get rights and content

Highlights

  • SWE and SCA were modeled for two mostly ungauged watersheds in NW China.

  • Moderate-resolution precipitation surfaces were produced by enhancing TRMM images.

  • Modeled SCA matched MODIS- and AMSR-E-based SCA reasonably well.

  • In-watershed annual variations in SWE and SCA were analyzed.

  • Accumulation of SWE possessed specific seasonal patterns and varied with elevation.

Abstract

Seasonal water flow to semi-arid to arid lands worldwide is regularly controlled by precipitation and snowmelt processes in nearby mountain ranges, where few monitoring stations exist. In this paper we present a new approach to assess the seasonal and inter-annual development of precipitation and associated snow-cover area (SCA) and snow–water storage in the Qilian Mountains of NW China. The approach addresses the spatial enhancement and calibration of remote sensing (RS) data acquired from coarse-resolution TRMM (Tropical Rainfall Measuring Mission; ~ 26-km resolution) and MODIS images in the development of image-surfaces of in-mountain precipitation and surface air temperature at 250-m resolution for input to a new spatially-distributed monthly snow-accumulation and snowmelt model. Spatially-enhanced precipitation and air temperature surfaces were subsequently calibrated using either geographically-weighted regression or simple linear regression and point-data from climate stations. When point-extractions from the calibrated products were compared against a new set of independent climate-station data, their respective values were comparable, giving an overall r2 of 0.87 for precipitation (RMSE = 10 mm month 1) and 0.96 for surface air temperature. With input from calibrated surfaces, monthly SCA and snow–water equivalence (SWE) in the Qilian Mountains were subsequently modeled over a 10-year period (2000–2009). Seasonal values of SCA and SWE were compared with MODIS-based (optical) and AMSR-E passive microwave-based estimates of SCA and SWE. In most cases, comparisons revealed suitable agreement across the various evaluations of SCA. Minor discrepancy between MODIS- and AMSR-E-based estimates of SCA resulted in a mean overlap of 73% when modeled SCA was compared to MODIS-based SCA and 84% overlap, when compared to AMSR-E-based SCA. Modeled and AMSR-E-based estimates of SWE at lower- to upper-mountain elevations (≤ 3900 m above mean sea level; AMSL) compared fairly well. At elevations > 3900 m AMSL, discrepancies between estimates were largely attributed to an overestimation of local precipitation.

Introduction

Assessing snow–water equivalence (SWE; i.e., the amount of water stored in snow) in the mountains is important in determining seasonal water flow in snowmelt-dominated river systems in mountainous watersheds. In these watersheds, snow accumulated in the mountains during the snow season is an important source of water to river networks downslope during late spring to mid-summer, when air temperatures in the mountains exceed the melting point of snow.

Measurements of snow depth, density, and SWE are usually not included in the standard data capture of most climate stations (Braun, 1991). Automated stations have been established to observe snowpack conditions at fine temporal resolutions (hourly to daily) in some parts of the world (e.g., SNOTEL; Fassnacht, Dressler, & Bales, 2003) and some manual measurements have been collected on project bases, but these measurements are usually very difficult to acquire and when acquired, they are usually not very representative of conditions beyond the area in which they were collected. Empirical studies in mountainous regions show very high standard deviations (60–100%) from mean snow cover as a result of the broad range of snow conditions encountered in mountainous environments (Fassnacht et al., 2003, Marchand and Killingtveit, 2005). Moreno, Latron, and Lehmann (2010) analyzed the accuracy of different interpolation methods in predicting snow distribution and found that the grid and sample size greatly affected model accuracy. They found that for a small catchment of 6 km2, model accuracy diminished from 80% to 50% based on the D-index (Willmott, 1982), when sample locations were reduced from 991 to 100, suggesting that interpolation would fail to provide acceptable estimates of SWE for even larger unmonitored mountainous watersheds.

Optical RS has been previously used in the past to generate satisfactory calculations of snow-cover area (SCA; Braun, 1991) in the production of moderate to coarse resolution SWE distributions for input in hydrological models (Martinec, Seidel, Burkart, & Baumann, 1991). Recent introduction of the Moderate Resolution Imaging Spectroradiometer (MODIS) and Advanced Microwave Scanning Radiometer-Earth Observing System (AMSR-E) sensors have improved our capabilities to map SCA and SWE for regional and global applications (Kuchment, Romanov, Gelfan, & Demidov, 2010). However, sensor resolution and accuracy has caused some problems with the detection of snow, particularly when snow cover is shallow and discontinuous.

Accurate determination of spatiotemporal patterns in precipitation and surface air temperature is basic to modeling SCA and SWE. Historically, spatially-distributed estimates of precipitation and temperature were acquired by interpolating climate-station observations. Inherent in this approach is the assumption that precipitation and surface air temperature at locations not directly measured are adequately represented by values from surrounding monitoring stations. As a general rule, interpolation is recommended for areas where the number of climate stations is high with respect to target spatial resolution and application scale; whereas, spatial enhancement (statistical downscaling) of remote sensing (RS)-based products is recommended for areas where climate stations are sparse and landform is particularly complex (Lanza, Ramirez, & Todini, 2001).

RS-based assessments of precipitation, unlike assessments of land surface temperature, are usually indirect and are mostly modeled from remote measurements of related variables. Accuracy of RS-based estimates of instantaneous precipitation, thus far has generally been low, but reasonable estimates have been generated at longer timescales (Fiorucci et al., 2001, Hu et al., 2013). For example, Ebert and Manton (1998) compared the results of the evaluation of 50 precipitation algorithms with precipitation data acquired with Doppler radar and concluded that satellite-based estimates of precipitation corresponded reasonably well at monthly timescales, with coefficients of correlation of 0.86–0.90 for geostationary satellites and 0.69–0.86 for AVHRR and SSM/I sensors. Lower accuracies were reported for instantaneous precipitation measurements with coefficients of correlation of 0.43–0.58 and 0.6–0.78 for geostationary satellites and AVHRR or SSM/I sensors, respectively. Tropical Rainfall Measuring Mission (TRMM; i.e., product 3B43), available at 0.25° × 0.25° spatial resolution (~ 26 km × ~ 26 km), provides one of the best possible satellite-based product of total precipitation at broad temporal scales (e.g., monthly or annually; GCMD, 2010, Huffman et al., 1997, Huffman et al., 2007). Validation of TRMM-precipitation for major climatic regions of Africa has shown that the 3B43-data closely matched raingauge data, with r2-values (coefficients of determination) of 0.84–0.97 and normalized root mean squared error (RMSE) < 50% for all regions, except for the arid-lands of northern Africa (Adeyewa & Nakamura, 2003). In another validation exercise (Fleming, Awange, Kuhn, & Featherstone, 2011), 3B43-precipitation data from Australia was compared with gridded monthly data generated from raingauge observations and found strong agreement between the two datasets, with coefficient of correlation of 0.93 (r2 = 0.86).

An important limitation of TRMM-image data resides in the data's coarse spatial resolution. On account of the data's resolution, unrefined TRMM-data are usually not recommended for regional application. Spatial enhancement of TRMM-image products at scales appropriate for regional application is the subject of active research, mostly centered on the usage of a coarse relationship between normalized difference vegetation index (NDVI) and precipitation (Duan and Bastiaanssen, 2013, Immerzeel et al., 2009, Jia et al., 2011). Although the relationship is useful at broader timescales (e.g., annually), the relationship fails to capture a significant portion of the variability at shorter time intervals, including at monthly timesteps (Duan and Bastiaanssen, 2013, Verlinde, 2011). Duan and Bastiaanssen (2013) proposed the generation of 1-km resolution monthly precipitation product from TRMM by initially enhancing the original TRMM-images with a NDVI-to-TRMM-precipitation-type relationship at an annual level and disaggregating annual values into monthly estimates based on proportions derived from the original TRMM-image series.

The research described here is motivated by the need to further investigate the eco-hydrometeorological processes involved in the long term self-support of desert oases in NW China. In earlier studies (e.g., Bourque and Hassan, 2009, Bourque and Mir, 2012, Zhang et al., 2010), precipitation was calculated spatially with Genetic Algorithms trained on 10-years of monthly climate-station data and an assumed point-trend in upper-mountain precipitation. In general, the method reproduced monthly recorded precipitation at lower elevations (i.e., ≤ 3400 m above mean sea level, AMSL) reasonably well (r2  0.75; Bourque and Hassan, 2009, Zhang et al., 2010). Greatest discrepancies between monthly values (modeled and recorded) normally occurred in the dry, lowlands of the Shiyang River watershed in the Hexi Corridor of westcentral Gansu (r2 = 0.75), where local precipitation was observed to be lower (~ 115 mm yr 1) and generally more intermittent because of its convective origin (Bourque and Hassan, 2009, Zhang et al., 2010). With further analysis of generated precipitation surfaces, it was determined that the quantities of precipitation calculated in the mountains may have been too high and, as a result, may have caused snow accumulation and associated calculations of SWE in the mountains to have been overestimated (Bourque & Mir, 2012; their Fig. 6). There is an anticipation that improved estimates of precipitation at a finer spatial resolution may help ameliorate computations of SWE in the mountains.

The specific objectives of this paper are to: (i) develop a new monthly snow accumulation and snowmelt model (“snowmelt model”, hereafter) that addresses the melting of snow as a function of melting degree-days (utilizing surface air temperature; Dou et al., 2011, Li and Williams, 2008, Martinec and Rango, 1986, Wang and Li, 2001) and precipitation inputs based on spatially-enhanced MODIS-temperature and TRMM images; (ii) apply the snowmelt model to calculate SCA and SWE in the Qilian Mountains (NW China) from a decade of monthly data (2000–2009); and (iii) compare modeled outputs of SCA and SWE with their counterparts derived from MODIS and AMSR-E image-products in an analysis of intra- and inter-annual variation in SCA and SWE.

Section snippets

Study area

The study area comprises of the Shiyang and Hei River watersheds in westcentral Gansu Province, NW China (Fig. 1). The Shiyang River originates from the Qilian Mountains and flows northeastward before terminating in the Minqin-lake district (Li, Xiao, He, Chen, & Song, 2007). The watershed area is ≈ 49,500 km2. Elevation in the Shiyang River watershed varies from 1284 to 5161 m AMSL, with an average elevation of 1871 m AMSL. The Hei River watershed, covering ≈ 128,000 km2, is the second largest

Snowmelt model

Separating precipitation into snow and rain is important for snow accumulation and melt modeling. The phase of precipitation depends on many factors, including temperature, thickness of the atmospheric layer, types of cloud, movement of the air column, and relative humidity (Kienzle, 2008). To accommodate all these factors into a single snowmelt model would be highly impractical. To simplify the modeling process, empirical relationship between the snow–rain phase with several meteorological and

Results

Fig. 3 provides comparisons of fine-resolution MODIS-based and climate-station T and Td (representative of 2006–2009) for three climate stations in each of the Shiyang and Hei River watersheds (Fig. 1). Validation shows very high correlation between datasets (Fig. 3), giving r2's and RMSE's of 0.96, 2.56 mm month 1 and 0.86, 2.39 mm month 1 for T and Td, respectively.

Correlation analysis (Table 2) shows that TRMM-precipitation is reasonably correlated to RH (r = 0.7), and less so to TPW, H, and EVI (

Discussion

In the past, spatial-enhancement of TRMM-images depended on relating the patterns of vegetation growth and precipitation (Immerzeel et al., 2009, Jia et al., 2011), as a result of the strong positive correlation between variables at annual and seasonal timescales. However, the strength of these correlations usually varied for different vegetation, soil, and topographic and precipitation characteristics (Davenport and Nicholson, 1993, Nicholson et al., 1990, Nightingale and Phinn, 2003). At

Conclusion

This paper developed a methodology to produce spatially-distributed monthly precipitation, SCA, and SWE at moderate spatial resolutions (i.e., 250-m resolution) using remote sensing-based products (Table 1) and a simple snowmelt model. Digitally-enhanced precipitation surfaces after calibration matched climate-station data (at lower elevations) reasonably well, both in terms of their magnitude and seasonality. Discrepancies in magnitude are likely to be greatest in the high mountains of the

Acknowledgment

This study was jointly funded by the: (i) China Meteorological Administration; (ii) Lanzhou Regional Climate Centre of the Gansu Provincial Meteorological Bureau (GMB), Lanzhou, China (Natural Science Foundation of China under Grant 40830957); and (iii) the Faculty of Forestry and Environmental Management, University of New Brunswick (UNB), New Brunswick, Canada. We are grateful to Fang Feng of GMB for his assistance with retrieving the meteorological data used in this paper. We would also like

References (122)

  • A.R. Huete et al.

    A comparison of vegetation indices over a global set of TM images for EOS-MODIS

    Remote Sensing of Environment

    (1997)
  • W.W. Immerzeel et al.

    Spatial downscaling of TRMM precipitation using vegetative response on the Iberian Peninsula

    Remote Sensing of Environment

    (2009)
  • S.F. Jia et al.

    A statistical spatial downscaling algorithm of TRMM precipitation based on NDVI and DEM in the Qaidam Basin of China

    Remote Sensing of Environment

    (2011)
  • Y. Kamarianakis et al.

    Evaluating remotely sensed rainfall estimates using nonlinear mixed models and geographically weighted regression

    Environmental Modelling & Software

    (2008)
  • A.G. Klein et al.

    Validation of daily MODIS snow cover maps of the Upper Rio Grande River Basin for the 2000–2001 snow year

    Remote Sensing of Environment

    (2003)
  • W.P. Kustas et al.

    Estimating subpixel surface temperatures and energy fluxes from the vegetation index-radiometric temperature relationship

    Remote Sensing of Environment

    (2003)
  • G. Lindstrom et al.

    Development and test of the distributed HBV-96 hydrological model

    Journal of Hydrology

    (1997)
  • J. Martinec et al.

    Parameter values for snowmelt runoff modeling

    Journal of Hydrology

    (1986)
  • T.R. McVicar et al.

    Global review and synthesis of trends in observed terrestrial near-surface wind speeds: Implications for evaporation

    Journal of Hydrology

    (2012)
  • J.C. Price

    Combining panchromatic and multispectral imagery from dual resolution satellite instruments

    Remote Sensing of Environment

    (1987)
  • Z.D. Adeyewa et al.

    Validation of TRMM radar rainfall data over major climatic regions in Africa

    Journal of Applied Meteorology

    (2003)
  • T. Akiyama et al.

    Surfacewater–groundwater interaction in the Heihe River Basin, Northwest China

    Bulletin of Glaciological Research

    (2007)
  • I. Amro et al.

    A survey of classical methods and new trends in pansharpening of multispectral images

    EURASIP Journal on Advances in Signal Processing

    (2011)
  • E. Anderson

    Point energy and mass balance model of a snow cover: A technical report, NWS-19

    (1976)
  • A.H. Auer

    The rain versus snow threshold temperatures

    Weatherwise

    (1974)
  • A.K. Bagchi

    Orographic variation of precipitation in a high-rise Himalayan Basin

    Hydrological Sciences Journal-Journal Des Sciences Hydrologiques

    (2009)
  • S. Bergström

    Development and application of a conceptual runoff model for Scandinavian catchments

    (1976)
  • A. Boudhar et al.

    Evaluation of the snowmelt runoff model in the Moroccan High Atlas Mountains using two snow-cover estimates

    Hydrological Sciences Journal-Journal Des Sciences Hydrologiques

    (2009)
  • C.P.-A. Bourque et al.

    Vegetation control in the long-term self-stabilization of the Liangzhou Oasis of the upper Shiyang River watershed of westcentral Gansu, Northwest China

    Earth Interactions

    (2009)
  • L.N. Braun

    Modeling of the snow water equivalent in the mountain environment

    Snow, Hydrology and Forests in High Alpine Areas

    (1991)
  • R.D. Brown et al.

    Gridded North American monthly snow depth and snow water equivalent for GCM evaluation

    Atmosphere-Ocean

    (2003)
  • C. Brunsdon et al.

    Geographically weighted regression: A method for exploring spatial nonstationarity

    Geographical Analysis

    (1996)
  • C. Brunsdon et al.

    Spatial variations in the average rainfall–altitude relationship in Great Britain: An approach using geographically weighted regression

    International Journal of Climatology

    (2001)
  • C. Carpenter

    Alashan Plateau semi-desert (PA1302)

  • C. Carpenter

    Montane grasslands and shrublands

  • CGIAR-CSI

    SRTM 90-m digital elevation data

  • G. Cliche et al.

    Integration of the spot panchromatic channel into its multispectral mode for image sharpness enhancement

    Photogrammetric Engineering and Remote Sensing

    (1985)
  • D. Cline

    Snow surface energy exchanges and snowmelt at a continental alpine site

    Biogeochemistry of Seasonally Snow-Covered Catchments

    (1995)
  • COE

    Snow hydrology, summary report of snow investigations

    US Army Corps of Engineers, North Pacific Division, Portland, Oregon

    (1956)
  • A. Dai

    Temperature and pressure dependence of the rain–snow phase transition over land and ocean

    Geophysical Research Letters

    (2008)
  • M.L. Davenport et al.

    On the relation between rainfall and the normalized difference vegetation index for diverse vegetation types in East-Africa

    International Journal of Remote Sensing

    (1993)
  • R.J. Donohue et al.

    On the importance of including vegetation dynamics in Budyko's hydrological model

    Hydrology and Earth System Sciences

    (2007)
  • Y. Dou et al.

    The simulation of snowmelt runoff in the ungauged Kaidu River Basin of TianShan Mountains, China

    Environmental Earth Sciences

    (2011)
  • E.E. Ebert et al.

    Performance of satellite rainfall estimation algorithms during TOGA COARE

    Journal of the Atmospheric Sciences

    (1998)
  • S.R. Fassnacht et al.

    Snow water equivalent interpolation for the Colorado River Basin from snow telemetry (SNOTEL) data

    Water Resources Research

    (2003)
  • J. Feiccabrino et al.

    Precipitation phase discrimination in Sweden

    (2008)
  • R.I. Ferguson

    Snowmelt runoff models

    Progress in Physical Geography

    (1999)
  • P. Fiorucci et al.

    A geostatistical approach to multisensor rain field reconstruction and downscaling

    Hydrology and Earth System Sciences

    (2001)
  • K. Fleming et al.

    Evaluating the TRMM 3B43 monthly precipitation product using gridded raingauge data over Australia

    Australian Meteorological and Oceanographic Journal

    (2011)
  • J.L. Foster et al.

    A blended global snow product using visible, passive microwave and scatterometer satellite data

    International Journal of Remote Sensing

    (2011)
  • Cited by (6)

    View full text