The GEWEX/ISLSCP Global Soil Wetness Project is an international initiative to enlist a spectrum of land-surface modeling centers and programs to use common global atmospheric forcing data sets to drive their respective land-surface models, in order to simulate and compare global soil wetness fields, and ultimately apply them in GCM sensitivity tests. As a participant in this project, we used the common ISLSCP atmospheric forcing data for 1987-1988 and our NCEP land-surface model (LSM) to generate global hydrological fields, including soil moisture, which cannot be obtained uniformly over large continental areas from scattered observations. We found that the simulated global hydrological cycle reasonably reflects the seasonal variation of those hydrological fields. Despite the relatively coarse 1×1 degree resolution of forcing data, the soil moisture field has a remarkable spatial variability, which is related not only to the spatial pattern of precipitation, but also to the spatial distribution of runoff and evaporation. Also, the spatial variability in vegetation and soil characteristic contributes to the variability in the surface hydrological cycle. We also compared our simulated global land-surface hydrological cycle to the 1987-1988 NCEP/NCAR reanalysis of land-surface products. It seems that, in most regions, the reanalysis of soil moisture has a larger annual cycle amplitude than the GSWP soil moisture. A more remarkable difference is that in the mid- and high-latitudes of the northern hemisphere, the reanalysis has a more uniformly distributed and wetter soil moisture than the corresponding GSWP soil moisture, which reflected the artificial climatological soil moisture damping field applied in the reanalysis system. Finally, we attempted to validate our soil moisture simulations with soil moisture measured by the Illinois Soil Moisture Network, and focused on the comparison of the area-averaged deep root zone soil moisture. It appears that the simulated soil moisture of 1987-1988 in the state of Illinois is able to reasonably capture not only the phase of the seasonal evolution, but also the amplitude of annual variation. The encouraging results indicate that a sound LSM can well simulate the natural soil moisture evolution, given accurate atmospheric forcing and reasonable specification of local vegetation and soil characteristics.
To produce a global soil moisture “data set, ” ten different land surface models were forced with meteorological observations for a two-year period (1987-88) for the Global Soil Wetness Project (GSWP). We compare observed plant-available soil moisture in the top 1-m soil layer to the same quantity generated by the models. Our soil moisture observations are from grasslands and agricultural regions in Russia, Illinois (USA), China, and Mongolia. None of the models does a good job of producing the actual soil moisture value for any of the regions. Thus, GSWP has not yet demonstrated the ability to produce global soil wetness data sets. Once the bias is removed, the models do a fairly good job of reproducing the seasonal cycle of soil moisture for the various areas. The model biases are different in different locations, so correcting them with a simple adjustment of the mean will not produce correct results. Better specification of parameters, or better representation of physical and biological processes, is still needed to improve these models. Future GSWP experiments should be conducted for a longer time period. They should emphasize catchment scale validation and higher time resolution of model output. Increased soil moisture observations, possibly incorporating satellite measurements, also would greatly improve a second project.
The site-averaged data, compiled by the First International Satellite Land Surface Climatology Project (ISLSCP) Field Experiment (FIFE) from 1987 to 1988, are compared with the output from a simple biosphere model introduced by the Japan Meteorological Agency (JMA-SiB), and with the atmospheric forcing archived in the ISLSCP Initiative I CD-ROM. In the JMA-SiB output, 1988 soil moisture below a depth of 5cm is systematically underestimated, due to budget calculation errors from autumn 1987 to spring 1988. The total precipitation during this period exhibits close agreement between the observed and the one used as the atmospheric forcing. In the observation, the precipitation directly leads to an increase in soil moisture, while in the JMA-SiB, most of the forcing precipitation is consumed by evapotranspiration. Reasons for this discrepancy in spring 1988 can be summarized as follows: (1) The near surface mixing ratio is systematically underestimated in winter, which does not reflect the evapotranspiration in the JMA-SiB. (2) The atmospheric forcing underestimates the intense precipitation concentrated over the last 11 days, just prior to the resumption of soil moisture observations. Since the characteristics of the observations and those of the atmospheric forcing are found to be fairly different, simulated soil moisture should not be directly compared with that observed. It appears necessary to carry out an another validation experiment using the site-averaged FIFE data as the atmospheric forcing for the JMA-SiB.
Soil moisture and water balance for global and regional scales have been calculated using a land-surface process model (SiB2) forced by observed and model assimilated data. The simulated runoff for each grid cell has been provided as input to a global river routing model, in order to simulate river discharge rates. The simulated soil moisture and water balance have been compared with available observations for their annual mean and seasonal cycles and for global, basin and grid point scales. The global distributions of the annual-mean soil moisture and wetness have been reasonably simulated. There were large inter-annual variations of soil moisture in both the simulations and observations at local stations. The simulated annual discharges for major river basins agree reasonably well with observations, but with some underestimates for large discharges and some overestimates for small discharges. The seasonal cycle of river discharges has been well simulated for specific basins in the tropics, midlatitudes, and high latitudes, although for some basins the annual mean is underestimated. In the tropics, the seasonal cycles of soil moisture and the surface water balance are dominated by the precipitation cycle. In mid- and high latitudes, soil moisture and the water balance are affected by both the temperature and precipitation cycles, and by the snow accumulation/melting cycle. The range of seasonal soil moisture variations becomes smaller with increasing latitude. The seasonal cycles of soil moisture for selected grid points have been compared with selected station observations. Even though there are differences in forcing and in some specific surface boundary parameters at the stations, the simulated soil moisture agrees well with multiyear observations at a majority of the stations. However, for almost all the selected grid cells, the seasonal variations are smaller, the snow melt and soil drying processes are late by about one month, and the soil is relatively wet in summer, compared with observations. These errors can be partly attributed to the unrealistically cool temperatures provided to the model as forcing data, favoring less surface evaporation and a later seasonal cycle, especially for mid- and high latitudes.
As one effort to estimate the global soil moisture distribution, the Global Soil Wetness Project (GSWP) was conceived. Under the GSWP, the global soil moisture distribution on 1°×1° mesh for 1987 and 1988 was simulated in an offline mode by 11 land surface models (LSMs). Even though the forcing conditions are mostly based on observations, validation studies are necessary because LSMs may not simulate accurately the partitioning of water at the surface of the earth between runoff, evaporation, and changes in soil moisture. A gridded 1°×1° global river channel network, named Total Runoff Integrating Pathways (TRIP) is used to calculate mean runoff estimated by the LSMs for drainage areas upstream of 250 operational gauging stations. Runoff observations from these stations in 150 major river basins of the world have been collected for 1987 and 1988, and were compared with the LSM products. It was found that LSMs estimated annual runoff fairly well, with a relative root mean square error of 40% for drainage areas with a fairly high density of raingauge observations (≥30/106km2), which was used to prepare the forcing precipitation. The error corresponds to approximately 18% of annual evapotranspiration. LSMs are also found to have a tendency to underestimate the annual runoff. This may be caused by underestimation of raingauges under strong wind conditions, especially for snow, because all of the LSMs underestimated the runoff for most of the drainage areas located in higher latitudes. A linear river routing model was applied for the global runoff products from the LSMs and analyzed at 250 gauging stations. The correlations between observed and simulated monthly runoff were improved for most of the LSMs by introducing the routing. River runoff information was found to be effective for the validation of water cycles on the continental scale.
The annual runoff rates generated by land surface models (LSMs) participating in the Global Soil Wetness Project (GSWP) are compared to observed rates in well-instrumented basins across the globe. Because such an offline evaluation can be clouded by an overwhelming influence of the atmospheric forcing itself (relative to the imposed land surface physics) on model output, we also estimate runoff rates with a decades-old climatological relation devised by M. I. Budyko, a relation that depends solely on annual precipitation and net radiation. The LSMs and the estimates derived with Budyko's relation are found to have standard errors of the same order (roughly 100mm/yr). Thus, we conclude that the complexities inherent in these state-of-the-art LSMs did not lead to increased accuracy in the simulated energy and water balances at the annual time scale. Proper LSM formulations are nevertheless recognized as essential for realistic land surface behavior at shorter (e. g., monthly or hourly) time scales, for which a simple Budyko-type equation would be in much greater error.
The Global Soil Wetness Project questioned the accuracy of the ISLSCP Initiative-1 downward longwave radiation data. The data appear to be offset by several hours from what is expected, such that the maximum downward longwave radiation consistently occurs late at night, when air temperatures are decreasing. This problem may result from the formula used to calculate the 6-hourly longwave radiation data. The Biosphere-Atmosphere Transfer Scheme (BATS) was used to test the sensitivity of one land-surface model to changes in the diurnal distribution of downward longwave radiation. Three simulations were performed: 1) a control run, in which the unmodified values were used; 2) a run in which all the longwave values were shifted forward six hours; and 3) a run in which the daily average value was used. Comparisons are made of both the large-scale monthly averages and of the monthly average diurnal cycles at six individual points. The two sensitivity simulations, using the altered downward longwave radiation, both increased total incoming radiation during the day and decreased it at night, relative to the control simulation. Consequently, upward longwave radiation increased during the day while decreasing by a greater amount at night, resulting in a net decrease in upward longwave radiation. The changes in upward longwave radiation must be balanced by changes in sensible, latent and soil heat fluxes. As a result of the changes to longwave radiation, simulated sensible heat flux in the sensitivity simulations increases during the day and decreases at night. The daytime increase is larger than the nighttime decrease, resulting in a net increase in sensible heat. The changes in sensible heat flux are usually within ±1W/m2 of the changes in upward longwave radiation, with small changes in latent and soil heat fluxes making up the difference. Water budget terms, and hence any of the GSWP soil wetness index calculations, generally experience only negligible changes.
A suite of sensitivity experiments focusing on leaf area index (LAI) was conducted within the GSWP framework using one land surface scheme. First, the LAI was changed from the original 1°×1° resolution provided by GSWP so that it was determined by vegetation type. This was then varied by plus and minus one standard deviation. Additional experiments explored sensitivity to the global use of a single LAI value, to aggregation of LAI to coarser resolution, and to changes in the seasonal variation in LAI. Results showed a sensitivity to LAI such that changes in total evaporation of 30W m-2 (July average) were common. Changes in soil wetness occurred which commonly exceeded 5%, and sometimes exceeded 10%, of the soil moisture at saturation. These changes in soil wetness are large enough to suggest that LAI needs to be provided with care within GSWP, and that any soil wetness product requires LAI to be provided with reasonable accuracy (perhaps within±1 and certainly within±2). Any problems in the LAI data (particularly in the magnitude rather than the precise seasonal variation) will lead to errors in the simulation of a soil wetness product.
A global two-dimensional implementation of the simplified Simple Biosphere (SSiB) land surface scheme is integrated offline for two years as part of the Global Soil Wetness Project (GSWP). A climatology of soil wetness and surface fluxes has been produced. This climatology is compared to a number of sensitivity studies that have been performed to investigate how the partitioning of precipitation between runoff and evapotranspiration is affected, when aspects of the soil parameterization and the treatment of convective precipitation are altered. The control integration has a reasonable spatial distribution of the surface hydrologic balance components, and shows realistic seasonal and interannual variations. Evaporation from the soil surface accounts for a majority of the water fluxes from the soil over all but heavily forested areas, where transpiration dominates. The sensitivity studies show that in general the most sensitive terms on seasonal time scales appear to be runoff, direct evaporation from the soil, and the seasonal change of water storage in the soil matrix. A realistic distribution of convective precipitation in space and time is necessary to simulate at the grid scale adequately high values of runoff, and to not over-represent direct evaporation of rainfall intercepted by the canopy. Sensitivity is found to the choice of thickness of the surface soil layer-a parameter often assigned arbitrarily in land surface models. Little sensitivity is found when imposing a vertical profile of soil porosity intended to account for surface soil aeration and deep soil compaction.
Global soil moisture data of high quality and resolution are not available by direct observation, but are useful as boundary and initial conditions in comprehensive climate models. In the framework of the Global Soil Wetness Project (GSWP), the ISBA land-surface scheme of Météo-France has been forced with meteorological observations and analyses in order to study the feasibility of producing a global soil wetness climatology at a 1°×1° horizontal resolution between January 1987 and December 1988. A control experiment and several sensitivity tests have been performed, suggesting that soil moisture remains one of the most difficult climatological parameters to model and that any computed climatology must be considered with great caution. The prescription of the soil depth is particularly critical, showing the relevance of the absolute value of the soil water content and the interest for land surface schemes to include a deep layer beyond the rooting depth. Compared to a river flow climatology, the runoff simulated over large river basins seems to be underestimated because of deficiencies in both the ISBA scheme and the GSWP experiment design. In order to obtain a more reliable climatology, a global reanalysis of soil moisture has been attempted, using a sequential optimal interpolation technique, in which soil moisture is corrected by iterative comparison between simulated and observed near-surface air temperature and relative humidity. Preliminary tests have been performed for July 1987, showing the potential of this method in idealized conditions. In practice, many uncertainties, either in the observations, the land surface properties or the atmospheric forcing, are liable to jeopardize the quality of the reanalysis, suggesting the need for more consistent data within the GSWP framework. Some outlooks are presented for improving the robustness of the assimilation technique, which lead to encouraging results.
On global atmospheric climate model spatial scales, water budget variables (evapotranspiration, soil moisture and runoff) can vary nonlinearly within a typical grid box primarily due to soil moisture heterogeneity. A good deal of this variability results from subgrid variability of soil texture. For such scales, consideration of the variability of the parameters used to characterize the soil hydrology is warranted. A simple approach, amenable to climate modeling, for characterizing subgrid soil parameter variability is proposed in which several parallel noninteracting soil columns are configured beneath a single soil/vegetation surface. The hydrological parameter mean values and statistical moments, which must be defined for each column, are generated using simple regression relationships which relate the parameters to the grid box mean soil texture (sand and clay composition). This simple approach is used because subgrid heterogeneity parameter data is somewhat limited on a global scale. The Parameterization for Land-Atmosphere-Cloud Exchange (PLACE) model is used within the Global Soil Wetness Project (GSWP) experimental design to generate global soil moisture fields using the soil heterogeneity model. Grid box average evapotranspiration (used in the solution of the surface energy budget), soil moisture, and runoff represent the three soil columns surface-weighted totals: Results show a profound effect on the primary water budget variables due to consideration of the parameter variability: globally-averaged evapotranspiration is reduced by 17%, and total runoff is increased by 48% compared to a control run assuming a homogeneous soil texture distribution within each grid box. The global mean runoff ratio is increased by 12%. Soil wetness (SW) increases by 19%, while the soil wetness index (SWI) increases by 49%. It is suggested that future land-surface global data sets contain information regarding subgrid variability of the soil for further testing of methods for modeling sub-grid heterogeneity.
In evaluation (s) as a part of the Global Soil Wetness Project using ISLSCP Initiative-I data, the snow-melt in SSiB in the Russian Wheat Belt region (RWB) was found to be substantially delayed, with very deficient meltwater infiltration as compared to observations. Furthermore, most of the meltwater emerged as runoff, as opposed to soil moisture recharge. The deficiency emanated from the crudeness of snow-physics of the combined snow and ground layer of SSiB. In the current work, a new snow model employing a separate snow-layer was included. The snow-pack absorbs and transmits the incoming solar flux, thereby affecting the snow and ground temperatures through the winter and snow-melt periods. In the ISLSCP Initiative-I data evaluations, the snow-melt over the RWB region occurs 2-3 weeks sooner in the new model, and its soil thaws quite early in the snow-melt duration, which helps to infiltrate more meltwater into the soil. The new-model produces a more realistic simulation of soil-moisture, as well as Volga river runoff in RWB evaluations. Some residual delay in the snow-melt (varying from 1-4 weeks) seems to be related to the following: (1) inaccuracies in the satellite retrievals of snow under dense forest canopies; (2) the modeling assumptions, e. g., neglecting the influence of snow aging on its thermal diffusivity, and simplifications in absorption of solar-flux in the snow cover, leading to an inadequate distinction between snow-pack surface and mean temperatures; and, (3) possible cold bias of the ISLSCP surface air temperature data.
The SSiB model, which was forced with GSWP ISLSCP Initiative I surface data, was modified to include more realistic snow physics and snow-melt infiltration. The new snow model in SSiB was again integrated with the GSWP data. The new SSiB simulation produced wetter and warmer soil, with more realistic snow-melt timing and runoff, in regions of significant snow-melt. The simulation was used for initializing land-surface temperature, soil moisture, and snow cover for the GEOS II GCM, which was integrated for JJA to generate an ensemble of runs for both 1987 and 1988. Each ensemble contained six cases starting from an ECMWF analysis for each day starting from 29 May through 3 June of each year. As compared to the old SSiB GCM, the new SSiB GCM with new initial hydrologic conditions significantly improved the prediction of precipitation in mid-to-high-latitude regions of Canada and Russia. Evapotranspiration, soil moisture, and runoff also compared more favorably in the new SSiB GCM simulation than the old SSiB GCM. Also, the 1988-1987 difference in northern India precipitation was more pronounced in the new SSiB GCM. In the U. S., where the old SSiB GCM had failed to simulate the 1988 drought circulation, the new SSiB GCM performed only slightly better. This was also evident in the 1988-1987 differences. In this region, the influence of initial conditions was mostly lost in about one month's time to the evolution of an unrealistic circulation.
A new data set of global high-resolution soil wetness for 1987-1988 has been prepared as part of the Global Soil Wetness Project (GSWP). To produce this data, the Simplified Simple Biosphere (SSiB) land surface process model (LSP) has been integrated offline, driven by observed and assimilated meteorological data to produce a two-year global climatology of soil wetness at 1°×1° resolution. GSWP data set has potentially higher quality data than those previously available. We are testing the impact of the GSWP data for climate simulations using the Center for Ocean-Land-Atmosphere Studies (COLA) general circulation model (GCM), coupled to the SSiB LSP. There are two principle questions which we will address with our preliminary GCM/LSP sensitivity experiments. First, does the inclusion of presumably more realistic GSWP soil wetness significantly improve the simulation and predictability of summer season climate? We use the 1987-1988 GSWP product as a specified boundary condition in seasonal simulations (June-August), and compared to existing GCM/LSP integrations, where soil wetness is initialized from operational analyses and allowed to evolve freely in the coupled system. In both sets of integrations, identical observed sea surface temperatures are specified. Results show that the GSWP soil wetness is significantly different from that of the coupled model's own climatology, and produces a better simulation of precipitation anomaly patterns over monsoon regions and the summer hemisphere extratropics. However, there is little improvement in the systematic error of the coupled model. Improvements can be attributed to changes in surface fluxes induced by the different soil wetness. Second, does the interannual variability in a multi-year soil wetness data set contribute to interannual variability in climate simulations? A parallel set of GCM/LSP integrations have been produced using specified GSWP soil wetness from the “wrong” (other) year (i. e., 1988 soil wetness applied in 1987 integrations, and vice versa). The use of soil wetness data from the wrong year significantly degrades the simulation of precipitation anomaly patterns. This indicates that interannual variability in soil wetness is important to climate. However, differences in precipitation due to SST variability generally dominated those apparently caused by soil wetness variations.