Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing
Final Report
December 2008
Justin Sheffield, Eric Wood, Francisco Munoz-Arriola
Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Executive Summary
Evapotranspiration (ET) is a key variable in the terrestrial water cycle that provides the link between the land surface and the atmosphere. It is also potentially the largest component of the land water budget and is therefore central to quantifying and managing water resources. However, at the spatial scales necessary for making informed decisions about, for example, water allocation, reservoir management and irrigation scheduling, ground observations of ET are essentially non existent. An emerging potential source of data is from remote sensing, that has the advantage of having global and near temporally continuous coverage. This project seeks to develop a method for estimating ET from mostly remote sensing data sources for Mexico and provide a long-term and high resolution dataset that can help answer questions about water variability and change.
The method is based on a modified Penman-Monteith algorithm (RS-PM) which is generally considered the most accurate method for estimating potential evaporation (PE) and hence ET over large vegetated regions. Key to estimating ET is determining the value of canopy conductance which is scaled from stomatal conductance using the vegetation leaf area index (LAI). The method also makes adjustments to conductance to reflect the daily variation of environmental controls such as humidity and temperature. Evaporation directly from the soil surface is also calculated. Daily inputs of surface radiation and meteorology are taken from the WRCP International Satellite Cloud Climatology Project (ISCCP) database which are derived from satellite radiances and modeled data for 1984-2006. These data are downscaled from 280km resolution to 1/8 th degree resolution using the fine scale data from the North America Regional Reanalysis (NARR). Surface characteristics, including albedo and emissivity, are taken from the ISCCP, and vegetation cover and LAI are resampled from AVHRR remotely sensed datasets.
Using the RS-PM method and ISCCP input datasets, daily time series of PE and ET have been generated at 1/8 th degree resolution for 1984-2006 for the whole of Mexico. The data are evaluated against a number of other large scale estimates of ET from off-line land surface modeling (VIC) and atmospheric reanalysis (NARR), and station measurements of pan evaporation. These evaluations indicate good agreement in the seasonal cycle and spatial distribution of both PE and ET with a slight underestimation by the RS-PM. For PE, the NARR is higher on average by about 1mm/day that can be attributed to the higher NARR shortwave radiation. For ET, the RS-PM has a low bias in the summer/autumn relative to VIC and NARR and a slight high bias in the spring. Overall there is general agreement in the amplitude and phase of the seasonal cycle with the largest differences restricted to warmer and more humid regions. The monthly bias and rmse relative to VIC over the whole of Mexico are -0.13 and 0.39 mm/day respectively. One potential drawback of the RS-PM method is that it does not calculate direct evaporation of water lying on the vegetation canopy, which can be a significant part of the total evapotranspiration and thus may explain part of these differences.
To the authors‘ knowledge, this is the first regional, high resolution, long-term dataset of evapotranspiration derived from remote sensing data. Given the importance of understanding the variability of ET over large scales this presents a major advance in demonstrating the utility of using remote sensing to quantify the hydrologic cycle.
2 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Table of Contents 1. Overview...... 4 2. Methods...... 4 2.1. Remote-Sensing Penman-Monteith (RS-PM)...... 4 2.2. Spatial Downscaling ...... 5 2.3. Evaluation and Uncertainty Analysis...... 5 3. Datasets...... 5 3.1. ISCCP Remotely Sensed Radiation, Meteorology and Surface Characteristics...... 5 3.2. SRB Remotely Sensed Radiation...... 6 3.3. AVHRR Vegetation Distribution and LAI ...... 6 3.4. NARR Atmospheric Reanalysis ...... 6 3.4. VIC Hydrologic Simulation...... 7 3.5. SEMARNAT Station and Gridded Data...... 8 4. Results...... 9 4.1. Initial Evaluations and Downscaling ...... 10 4.1.1. Comparison of Input Data...... 10 4.1.2. Estimates of Low Resolution PE ...... 11 4.1.3. Downscaled Input Data...... 12 4.2. High Resolution Estimates of PE...... 14 4.3. High Resolution Estimates of Actual ET...... 21 4.4. Some Uncertainties and Caveats...... 26 5. Data Availability and Format...... 27 6. References...... 28
3 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
1. Overview The aim of the project is to reconstruct time series of ET over Mexico from primarily remotely sensed data and assess the uncertainty by comparison with other large scale estimates from reanalysis and off-line modeling, and local estimates derived using station observations. The tasks that were carried out are as follows:
a) A detailed outline for reconstructing the times series of ET over Mexico was developed. This included required data sources, potential algorithmic approaches and possible approaches to assessing the uncertainty in the estimates. b) A review was made of available daily data over Mexico for surface meteorology (minimum and maximum air temperature, relative humidity, and precipitation), tower- based surface heat fluxes and surface radiation. SEMARNAT helped identify and made available station and gridded datasets. c) The ISCCP remotely sensed based surface radiation and surface meteorology data were obtained and downscaled from its available spatial scale (2.5-degree) to 25km using NARR products. This allowed the development of a high resolution ET dataset. d) A daily time series of ET (and PE) for 1984-2006 was developed using a version of the Penman-Monteith algorithm that adjusts the vegetation response to environmental conditions such as low humidity and minimum temperatures. e) The ET estimates were evaluated by carrying out an uncertainty analysis. This included comparing the estimates to available station data, evaluating the sensitivity of the estimated ET to assessed uncertainty in the inputs, and comparing the remotely sensed estimates to estimates from reanalysis and land surface modeling. 2. Methods ET is estimated using the Penman-Monteith equation with inputs primarily from remote sensing data, compiled as part of the WCRP‘s International Satellite Cloud Climatology Project (ISCCP). The Penman-Monteith equation is generally considered the most accurate method for calculating PE and hence actual ET (AET) over large vegetated regions. We use the modified Penman-Monteith algorithm of Mu et al. (2007) (RS-PM) which is designed specifically for use with remote sensing data. Next we describe the details of this model, the method for downscaling the input data, and the approach used for the uncertainty analysis.
2.1. Remote-Sensing Penman-Monteith (RS-PM) Remote sensing estimates of daily ET were derived following the revised RS-PM formulation presented in Mu et al. (2007) which is based on the Penman-Monteith equation (Monteith, 1965). This models the diffusion of energy from plants or soil against stomatal and aerodynamic resistance given inputs of net radiation, temperature and humidity. Stomatal conductance is upscaled to canopy conductance via the leaf area index (LAI) and the RS-PM invokes additional environmental constraints as a function of vapor pressure deficit and minimum temperature. Evaporation directly from the soil surface is also calculated. Required inputs are radiation (downward surface short- and longwave), surface meteorology (humidity, air and surface temperature, pressure and windspeed) and surface characteristics (vegetation distribution, LAI, emissivity, albedo). These were mostly obtained from the ISCCP database.
4 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
2.2. Spatial Downscaling To produce estimates at the spatial scales relevant to water resources planning, we developed a downscaling scheme to bring the ISCCP input data to 1/8 th degree resolution (~12km) based on the NARR (available at 25km resolution). The downscaling scheme works as follows. At the monthly time scale, the NARR data are aggregated to 2.5 degree. Then the ratios of the NARR to the ISCCP data are interpolated to 1/8 th degree resolution and smoothed in space. These ratios are then used to scale the NARR data at 1/8 th degree to produce a high resolution dataset that matches the ISCCP data when aggregated to 2.5 degree.
2.3. Evaluation and Uncertainty Analysis The ET estimates were evaluated by comparison with estimates taken from off-line land surface modeling (VIC) and reanalysis (NARR). The VIC model has been calibrated to measured streamflow at a number of basins across Mexico and so should produce reasonable estimates of ET when forced by observed precipitation and temperature. The NARR data is used to infer ET as a residual of the atmospheric water balance (ET = P œ MC œ dw/dt), where P is precipitation, MC is atmospheric total column water vapor convergence and dw/dt is the change in column water vapor. The NARR assimilates gauge precipitation and radiosonde data of atmospheric temperature and humidity and therefore is likely to do a reasonable job of replicating the variation in atmospheric water vapor and thus ET as a residual. We also evaluated the data with observations from pan evaporation data provided by SEMARNAT. Uncertainty in the RS-PM ET estimates are derived from uncertainties in the forcing data (radiation, meteorological data and surface characteristics) as well as variations in the parameterizations used in the algorithm (for example, the calculation of canopy conductance). The impacts of these are explored in section 4. We also discuss the likely impact of canopy evaporation, which is not modeled in the RS-PM algorithm, on the ET estimates. 3. Datasets The primary data sources for the Penman-Monteith estimates are the ISCCP satellite based product that provides radiation and other meteorological variables and surface parameters. This dataset is available at 2.5 deg lat-lon, 3-hourly, for 1983-2006. Wind speed, which is not measured from remote sensing over land, is taken from the NCEP/NCAR global reanalysis. To provide a higher spatial resolution product than the ISCCP data, we downscaled the ISCCP data (2.5 degree) to 1/8 th degree using the NARR through statistical relationships between the coarse and fine resolution data as explained above. The distribution of vegetation (including type, fractional area and LAI) was taken from the AVHRR remote sensing based dataset at 1/8 th degree also. Details of the various input datasets used are given below. We also describe i) the SRB remote sensing based radiation dataset that is used as an alternative source of input data to assess the uncertainties in the ET estimates, and ii) the evaluation datasets (VIC, NARR, pan data).
3.1. ISCCP Remotely Sensed Radiation, Meteorology and Surface Characteristics The International Satellite Cloud Climatology Project (ISCCP) (http://isccp.giss.nasa.gov/products/products.html) was established in 1982 as part of the World Climate Research Programme (WCRP) to collect and analyze satellite radiance measurements to
5 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report infer the global distribution of clouds, their properties, and their diurnal, seasonal, and interannual variations (Rossow and Duenas, 2004). Data collection began on 1 July 1983 and is currently planned to continue through 30 June 2010. The resulting datasets and analysis products can be used to evaluate the role of clouds in radiation and studies of the hydrological cycle. We use the Climatological Summary Product (FD-SRF) RadFlux version of the dataset which contains surface radiative fluxes and a summary of the physical quantities used to calculate them. Shortwave and longwave radiative flux profiles are calculated using remotely sensed and modeled datasets to specify the properties of Earth's atmosphere and surface. The data are currently available from July 1983 to Dec 2006, with global coverage. The spatial resolution is 280km equal area (~2.5 degrees lat-lon at the equator). The temporal resolution is 3- hourly.
3.2. SRB Remotely Sensed Radiation The NASA/Langley Research Center product is available from 1983-2006. The primary data sources are satellite data from the ISCCP C1 data product and from the Earth Radiation Budget Experiment (ERBE). The C1 data provide cloud parameters derived from a network of geostationary satellites and NOAA‘s polar orbiters, along with temperature and humidity profiles from TOVS, on a 2.5 degree equal-area global grid and a 3-hourly time resolution. Two versions are available for short and longwave radiation. Firstly, the SRB-SW and SRB-LW products are derived using the algorithms of Pinker and Laszlo (1992) and Fu et al. (1997) respectively. Secondly, the SRB-QCSW and SRB-QCLW products are derived using the algorithms of Darnell et al. (1992) and Gupta et al. (1992), respectively. Comparison of these products with surface measurements has indicated that no one product is superior globally.
3.3. AVHRR Vegetation Distribution and LAI The distribution of vegetation cover is taken from the Advanced Very High Resolution Radiometer (AVHRR) based, 1km, global land cover dataset of Hansen et al. (2000), which uses the University of Maryland (UMD) classification scheme, by calculating the fractional area of each vegetation type within each 1/8 th degree grid cell. Values of LAI are specified for each vegetation type that exists in each grid cell by resampling the dataset of Myneni et al. (1997), which is based on AVHRR normalized difference vegetation index values. The LAI values are specified for each month but do not vary from year to year.
3.4. NARR Atmospheric Reanalysis Mesinger et al. (2006) describe the development of the NARR, including an overview of the improvements relative to the first NCEP/NCAR global reanalysis and initial evaluations. This reanalysis is from 1979 to present and covers the North American continental region and surrounding ocean area at spatial resolution of 32km with 45 atmospheric levels. The atmospheric model is the Eta model run within the Eta Data Assimilation System. The largest improvement is deemed to be the assimilated observed precipitation, and the knock-on effect to land-atmosphere coupling and land surface hydrology, which also benefited from improvements to the land surface model, Noah (Mitchell et al. 2004). This was also expected to give good representation of extreme events, such as droughts and floods. The precipitation is assimilated in terms of latent heating profiles that then force precipitation in the model. For the conterminous US, a daily rain gauge data analysis at 1/8 th degree is orographically adjusted using the Parameter-elevation Regressions on Independent Slopes Model (PRISM) approach (Daly et al.,
6 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
1994) and disaggregated to hourly from a 2.5 degree hourly gauge analysis. The version of the Noah land surface model used in the NARR, is similar to that used in recent coupled and uncoupled studies (Ek et al., 2003; Mitchell et al., 2004).
3.4. VIC Hydrologic Simulation The Variable Infiltration Capacity (VIC) land surface model (Liang et al., 1994, 1996; Cherkauer et al., 2002) was used to generate spatially and temporally consistent fields of ET and other water budget flux and state variables for comparison with RS-PM estimates. The VIC model simulates the terrestrial water and energy balances and distinguishes itself from other land surface schemes through the representation of sub-grid variability in soil storage capacity as a spatial probability distribution, to which surface runoff is related (Zhao et al., 1980), and by modeling base flow from a lower soil moisture zone as a nonlinear recession (Dumenil and Todini, 1992). The VIC model has been applied extensively at regional (Abdulla et al., 1996; Maurer et al., 2002) and global scales (Nijssen et al., 2001; Sheffield et al., 2004b), including snow and ice dominated regions (Bowling et al., 2003; Su et al., 2006). Horizontally, VIC represents the land surface by a number of tiled land cover classes. The land cover (vegetation) classes are specified by the fraction of the grid cell which they occupy, with their leaf area index, canopy resistance, and relative fraction of roots in each of the soil layers. Evapotranspiration is calculated using a Penman-Monteith formulation with adjustments to canopy conductance to account for environmental factors. The subsurface is discretized into multiple soil layers. Movement of moisture between the soil layers is modeled as gravity drainage, with the unsaturated hydraulic conductivity a function of the degree of saturation of the soil. Cold land processes in the form of canopy and ground snow pack storage, seasonally and permanently frozen soils and sub-grid distribution of snow based on elevation banding are represented in the model. Seasonally and permanently frozen soils are represented in the VIC model according to the algorithm of Cherkauer and Lettenmaier (1999). Soil temperatures are calculated using a finite difference solution of the heat diffusion equation for a user-specified number of nodes that are independent of the soil moisture layers. Soil ice content is estimated based on node temperatures and infiltration and baseflow are restricted based on the reduced liquid soil moisture capacity. The VIC model was run at 1/8 th degree spatial resolution at daily time step for the period 1925-2004. This setup has been previously employed by Zhu and Lettenmaier (2007) to assess the long-term hydrologic cycle over Mexico. The simulation was forced by their gauge based dataset of precipitation, and temperature (maximum and minimum daily values), with windspeed taken from NCEP/NCAR reanalysis. The model was calibrated by Zhu and Lettenmaier (2007) against measured streamflow for a number of basins across Mexico (see Figures 1 and 2) and as a result should give reasonable estimates of ET.
7 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 1. Locations of 14 selected river basins. Shaded areas are the contributing regions to each identified gauge point. (From Zhu and Lettenmaier, 2007).
3.5. SEMARNAT Station and Gridded Data The SEMARNAT data are in the form of station and gridded meteorological data. The station data consists of daily precipitation, Tmax, Tmin, and pan evaporation for 5552 stations with time periods ranging from 1961-2007. These data were extracted and converted into binary station format for comparison with the VIC forcing datasets and the estimates from the RS-PM algorithm. The gridded data are the Maya V1 dataset which consists of daily precipitation, maximum and minimum temperature for 1961-2000 at 0.3 degrees lat-lon, which are based on the station data. These were also extracted into binary format and were compared to the similar dataset developed by Zhu and Lettenmaier (2007) which is available at 0.125deg resolution. The gridded data were not used in this project but could be used to force the VIC land surface model to assess the uncertainties in the ET output.
8 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 2. Monthly calibrated routed simulated runoff (solid lines) vs observed streamflows (dashed lines). Ordinate values are runoff in m 3 s -1, and the abscissa is one 10-yr period, the beginning of which varies by basin, depending on observed flow availability. (From Zhu and Lettenmaier, 2007).
4. Results For the remote sensing based evaporation retrievals we initially focused on developing estimates of PE using the RS-PM model and inputs (radiation, temperature, humidity, etc) from the ISCCP dataset, at 2.5 degree and monthly resolution. To produce estimates at the fine scale, we used the downscaling scheme described in Section 2 to bring the ISCCP input data to 1/8 th degree resolution (~12km) based on the NARR (available at 25km resolution). Using the downscaled data, PE and then ET were then calculated using the RS-PM model at daily, 1/8 th degree resolution for 1984-2006.
9 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
4.1. Initial Evaluations and Downscaling
4.1.1. Comparison of Input Data The main forcing of ET is net radiation as derived from the residual of downward and upward short- and longwave radiation fluxes. Comparisons were done between the SRB, NARR and ISCCP data to assess uncertainties in the ET estimates through uncertainties in radiation. The ISCCP and SRB fluxes share some common inputs but use different algorithms. The SRB data are available on a 1.0 degree equal area grid at a monthly resolution from 1984-2006. Comparisons of the SRB and ISCCP SW data show that they differ regionally by less than 40 W/m 2 on average in the summer (Figure 3) and are very similar in other seasons. Comparisons of 2 the SRB and NARR datasets show that the NARR is biased high regionally by up to 80 W/m on average in the summer (Figure 4). Differences in the longwave fluxes are minimal between the three datasets (not shown).
Figure 3. Difference between SRB and ISCCP downward surface shortwave radiation over Mexico for January, April, July and October for 1985-2006.
10 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 4. 6. Difference (left) Downward between surface NARR shortwave and SRB downwardradiation for surface the NARR shortwave (top), radiation downscaled over MexicoISCCP (middle) for January, and April,their difference July and October (bottom). for (right) 1985-2006. Downward surface longwave radiation.
4.1.2. Estimates of Low Resolution PE We initially focused on developing low resolution estimates of PE at 2.5 degree and monthly resolution using the ISCCP input data at its native spatial resolution. Figure 5 gives the annual map of PE and shows a reasonable distribution of evaporation across the country despite the coarse spatial resolution (2.5 degree).
11 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 5. Average annual PE for 1984-2000 derived using the Penman-Montieth equation with 2.5 degree monthly ISCCP surface radiation and near-surface meteorology.
4.1.3. Downscaled Input Data
Figure 6: (left) Downward surface shortwave12 radiation for the NARR (top), downscaled ISCCP (middle) and their difference (bottom). (right) Downward surface longwave radiation.
Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
To produce estimates at the fine scale, we used the downscaling scheme to bring the ISCCP input data to 1/8 th degree resolution (~12km). This was done for each 3-hourly ISCCP data field for 1984-2006. Figures 6-8 show examples of the original and downscaled radiation and meteorological fields.
Figure 7. As figure 6, but for (left) 2m air temperature and (right) 2m specific humidity.
13 Figure 8. As figure 6, but for (left) surface pressure and (right) 10m windspeed. Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
4.2. High Resolution Estimates of PE Using the downscaled data, PE was calculated using the RS-PM model and is shown in Figures 9 and 10 compared to the NARR. The spatial patterns and annual cycle are similar although the NARR data are higher by about 1 mm/day on average. Figure 11 shows the seasonal cycle of PE averaged over Mexico as well as the meteorological forcing data for ISCCP and the NARR. The high bias in the NARR shortwave radiation may be responsible for much of the difference in the two PE estimates.
Figure 9. Maps of average annual PE (1984-2000), from the NARR (top), ISCCP forced RS-PM estimate (middle) and their difference (NARR-ISCCP).
14 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 10. Time series of monthly PE (1984-200) from the NARR (black) and ISCCP forced RS-PM estimate (green).
We next show comparisons of potential and reference crop evaporation (ETrc) against the large database of pan evaporation measurements obtained from SEMARNAT and described in Section 3 that are available back to 1961 for over 5000 stations across Mexico. Reference crop evaporation is an idealized estimate defined as the rate of evaporation from an idealized grass crop of height 0.12m, an albedo of 0.23 and a surface resistance of 69 sm -1. It can be related to pan evaporation with a pan coefficient which varies by the type of pan, surrounding ground cover (e.g. short grass), and mean humidity and windspeed/fetch conditions. It is generally of the order of 0.4 - 0.85 for US class A pans and up to 1.1 for sunken Colorado-type pans under low wind conditions. Figure 12 shows the mean seasonal cycle averaged over Mexico for the RS-PM PE and ET rc compared to the pan measurements and NARR PE. The daily pan measurements are averaged to monthly values and then sampled on the same 1/8 th degree grid as the other datasets. Where more than one station lies within a grid cell, their average value is used. The NARR and RS-PM PE values diverge in the spring (likely caused by the higher shortwave radiation in the NARR as shown above) but are otherwise similar at these large scales. The RS-PM ET rc follows the RS-PM PE but is consistently lower because of the canopy resistance term used in its calculation. The RS-PM ET rc and pan measurements are quite similar. Figure 13 shows the monthly time series of the same datasets.
15 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 11. Mean seasonal cycle for 1984-2000 of monthly PE, radiation and meteorology averaged over Mexico from the NCEP North American Regional Reanalysis (NARR) (black) and ISCCP forced Penman-Monteith estimate (green).
16 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 12. Mean monthly seasonal cycle of PE from the RS-PM and pan evaporation for 1948-2000.
17 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 13. Monthly timeseries of PE from the RS-PM and pan evaporation for 1984-2000.
Figure 14 shows seasonal maps of the RS-PM and NARR PE and confirms the high bias in the NARR data in the spring and in the north during the summer. Figure 15 shows the same thing but for the RS-PM ETrc and station pan data. Again, there is a tendency for the RS-PM data to be lower in the spring, especially in the highlands.
18 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 14. Mean seasonal PE from the RS-PM method (left column), the NARR (middle) and their difference (right).
19 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 15. Mean seasonal ET rc from the RS-PM method (left column), pan measurements (middle) and their difference (right).
We investigate the relationship between the ET rc and pan data further in Figure 16, which shows their annual values and their ratio (pan coefficient). There is a delineation in the ratio values between the drier highland/inland and more humid coastal regions, with the latter being within the generally accepted range of pan coefficients. In the drier inland regions the ratios approach one or above, approaching the upper limit of the accepted range.
20 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 16. Comparison of annual RS-PM ET rc and pan evaporation (top row), the calculated pan coefficients (RS-PM / pan) and the distribution of specific humidity (bottom).
4.3. High Resolution Estimates of Actual ET Next the calculated RS-PM ET is compared to estimates from the VIC model, the NARR and inferred values from the NARR atmospheric water budget. Their seasonal cycles over Mexico are compared in Figure 17 and the full time series in Figure 18. There is a slight low bias in the summer and autumn relative to VIC and the NARR and a slight high bias in the Spring. Overall there is general agreement in the amplitude and phase of the seasonal cycle. Also shown in Figure 17 is the mean seasonal cycle derived from the atmospheric water balance using NARR precipitation, moisture convergence and change in atmospheric moisture. This comparison shows good agreement during the peak months and end of year, but the inferred values are much lower during the spring which requires further investigation.
21 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 17. Mean monthly seasonal cycle of ET from the RS-PM, VIC, NARR and inferred from the atmospheric water budget for 1984-2000.
22 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 18. Monthly timeseries of ET from the RS-PM, VIC and NARR for 1984-2000.
Figure 19 gives maps of the RS-PM and VIC simulated ET, which shows that the spatial distribution is similar. The tendency for the RS-PM data to be lower in the summer and autumn, as seen in the time series in Figures 17 and 18, is generally confined to the warmer and more humid coastal regions. The maps in Figure 20 show that the differences of the RS-PM with the NARR are more randomly distributed, although the RS-PM is again generally lower in coastal regions. Note that the NARR is at a coarser resolution (25 km) compared to the RS-PM data, as shown by its much smoother spatial distribution.
23 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 19. Mean seasonal maps of ET from RS-PM and VIC for 1984-2000.
24 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 20. Mean seasonal maps of ET from RS-PM and NARR for 1984-2000.
A summary of the error statistics for the RS-PM ET is given in Table 1. The overall bias for 1984-2006, over Mexico is of the order 0.1-0.2 mm/day, with RMSE values of about 0.4 and 0.25 mm/day at monthly and annual time scales respectively. These values are well within the uncertainty in the true values of ET at these large scales.
Table 1. Error statistics for the RS-PM relative to the VIC and NARR datasets, for 1984-2006. Bias RMSE Monthly Annual VIC -0.18 0.39 0.23 NARR -0.13 0.45 0.27
25 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Figure 21. Sensitivity of monthly averaged ET over Mexico to (left) the method for calculating the stomatal resistance and (right) to different input parameter configurations.
4.4. Some Uncertainties and Caveats A key part of the Penman-Monteith method is the specification of the stomatal and canopy conductance that adjusts the energy fluxes to simulate the response of the vegetation to variations in environmental conditions and the seasonal cycle of phenological development. A number of different parameterizations of varying complexity exist for simulating these effects and Figure 21 (left panel) shows the sensitivity of ET to some of these. The RS-PM method (black line) scales a specified minimum stomatal resistance for each vegetation type to a canopy conductance value based on the current LAI (based on Cleugh et al., 2007). It also makes adjustments for the effect of high humidity and low minimum daily temperatures that both act to increases resistance and decrease ET (see Mu et al., 2007 for details). The ET values based on only the LAI adjustment of Cleugh et al. (2007) is given by the green line, and the ET with no adjustment (fixed resistance) is shown in yellow. At these large scales, there is very little difference between the RS-PM and the Cleugh method, but making no adjustment to the resistances increases the ET considerably as expected. The right panel of Figure 21 shows the sensitivity of the calculated ET to some of the input surface characteristics, such as albedo, LAI and surface temperature. The RS-PM method uses spatially varying LAI derived from AVHRR data, spatially varying albedo from ISCCP and Tsurf for the canopy temperature. The sensitivity experiments show the impact of i) using a spatially uniform (but seasonally varying) LAI value for each vegetation type, irrespective of its location; ii) a spatially uniform albedo value (again seasonally varying) for each vegetation type; and iii) air temperature instead of Tsurf. The largest impact is when using uniform LAI values (which are based on literature values for the different vegetation types), which are obviously higher than those derived from the AVHRR data as they give a much higher ET response. One potential drawback of the RS-PM method is that it does not calculate direct evaporation of water lying on the vegetation canopy, which may explain part of these differences. Figure 22 shows scatter plots of annual mean evaporation of RS-PM versus VIC (black circles) and RS-PM versus VIC minus canopy evaporation (red circles). The latter is perhaps a fairer comparison with the RS-PM data as it compares transpiration plus soil evaporation only. We show the raw data for each grid cell (left panel) and when averaged to 1.0 degree (~ 100km) resolution (right panel).
26 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
First note that the RS-PM and VIC data (black circles) are well correlated (r = 0.81), which reflects the spatial agreement in Figure 8, but there is some spread in the values (rmse = 1.41 mm/day). The slight low bias in the RS-PM is -0.16 mm/day. When averaged to 1.0 degree (right panel) the spread reduces considerably (r = 0.88, rmse = 0.44 mm/day) as the noise from the fine spatial scale variability in meteorological forcings and vegetation characteristics is averaged out. Looking at the comparison with the VIC data minus canopy evaporation (red circles), we see a large reduction in the VIC data, indicating a significant contribution of canopy evaporation to total evaporation. The correlation does not change much (r = 0.82), but the RS- PM is now larger than the VIC data (bias = 0.87 mm/day) and the spread is somewhat reduced (rmse = 0.94 mm/day). This suggests that the RS-PM method is actually over-estimating transpiration plus soil evaporation, but the fact that it lies somewhere in between the two VIC datasets is somewhat encouraging.
Figure 22. Annual scatter plots of RS-PM ET versus VIC ET (black circles) and VIC ET minus canopy evaporation (red circles) for all grid cells (left) and after spatial averaging to 1.0 degree resolution (right).
5. Data Availability and Format The data are available at the following resolutions and formats:
Temporal resolution: daily Temporal domain: 1984-2006 Spatial resolution: 1/8 th degree (~12km) Spatial domain: Mexico Variables: ET, ET ref , PE, Rnet Format: binary 4 byte float images (244 cols X 146 rows) with GraDS style header file containing domain and resolution definitions Data size = 4 bytes * 244 cols * 146 rows * 365 days * 23 years = ~1.1 Gb per variable
27 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
6. References Abdulla, F. A., D. P. Lettenmaier, E. F. Wood, and J. A. Smith, 1996: Application of a macroscale hydrologic model to estimate the water balance of the Arkansas-Red river basin, J. Geophys. Res., 101(D3), 7449œ7459. Bowling, L.C., D.P. Lettenmaier, B. Nijssen, L.P. Graham, et al. 2003, Simulation of high latitude hydrological processes in the Torne-Kalix basin: PILPS Phase 2(e) 1: Experiment description and summary intercomparisons, Journal of Global and Planetary Change, 38(1- 2), 1-30. Cherkauer, K. A., and D. P. Lettenmaier, 1999: Hydrologic effects of frozen soils in the upper Mississippi River basin, J. Geophys. Res., 104(D16), 19,599-19,610. Cherkauer, K. A., L. C. Bowling, and D. P. Lettenmaier, 2002: Variable Infiltration Capacity (VIC) cold land process model updates, Global Planet. Change , 38 , 151œ159. Cleugh, H. A., R. Leuning, Q. Mu and S. W. Running, 2007: Regional evaporation estimates from flux tower and MODIS satellite data. Remote Sensing of Environment, Volume 106, page 285œ304 - 2007 (doi: 10.1016/j.rse.2006.07.007) Daly, C., R.P. Neilson, and D.L. Phillips, 1994: A Statistical-Topographic Model for Mapping Climatological Precipitation over Mountanious Terrain. J. Appl. Meteor., 33, 140-158. Darnell, W. L., W. F. Staylor, S. K. Gupta, N. A. Ritchey, and A. C. Wilber, Seasonal variation of surface radiation budget derived from ISCCP-C1 data, J. Geophys. Res., 97, 15741, 1992. Dumenil, L., and E. Todini, 1992: A rainfall-runoff scheme for use in the Hamburg climate model, in Advances in theoretical hydrology, A tribute to James Dooge. P. O'Kane, ed., European Geophysical Society Series on Hydrological Sciences 1, Elsevier, Amsterdam. Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren, G. Gayno, and J. D. Tarpley (2003), Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model, J. Geophys. Res., 108(D22), 8851, doi:10.1029/2002JD003296. Fu, Q., K. N. Liou, and A. Grossman, 1997: Multiple scattering parameterization in thermal infrared radiative transfer. J. Atmos. Sci ., 54 , 2799-2814. Gupta, S. K., W. L. Darnell, and A. C. Wilber, A parameterization for longwave surface radiation from satellite data: Recent improvements, J. Appl. Meteorol., 31, 1361, 1992. Hansen, M. C., R. S. Defries, J. R. G. Townshend, and R. Sohlberg, Global land cover classification at 1 km spatial resolution using a classification tree approach, Int. J. Remote Sensing , 21 , 1331-1364, 2000. Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A simple hydrologically based model of land surface water and energy fluxes for GCMs, J. Geophys. Res., 99(D7), 14,415œ 14,428. Liang, X., E. F. Wood, and D. P. Lettenmaier, 1996: Surface soil moisture parameterization of the VIC-2L model: Evaluation and modifications, Global Planet. Change, 13, 195œ 206. Maurer, E. P., A. W. Wood, J. C. Adam, D. P. Lettenmaier, and B. Nijssen, 2002: A long-term hydrologically-based data set of land surface fluxes and states for the conterminous United States, J. Climate, 15, 3237-3251. Mesinger, F., G. DiMego, E. Kalnay, K. Mitchell, P. Shafran, W. Ebisuzaki, D. Jovic, J. Woollen, E. Rogers, E. Berbery, M. Ek, Y. Fan, R. Grum-bine, W. Higgins, H. Li, Y. Lin, G. Manikin, D. Parrish, and W. Shi, 2006: North American Re-gional Reanalysis. Bull. Amer. Meteor. Soc., 87 (3), 343-360.
28 Development of a Long-Term Evapotranspiration Product for Mexico from Remote Sensing - Final Report
Mitchell, K. E., D. Lohmann, P. R. Houser, E. F. Wood, J. C. Schaake, A. Robock, B. A. Cosgrove, J. Sheffield, Q. Duan, L. Luo, R. W. Higgins, R. T. Pinker, J. D. Tarpley, D. P. Lettenmaier, C. H. Marshall, J. K. Entin, M. Pan, W. Shi, V. Koren, J. Meng, B. H. Ramsay, and A. A. Bailey (2004), The multi-institution North American Land Data Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a continental distributed hydrological modeling system, J. Geophys. Res., 109, D07S90, doi:10.1029/2003JD003823. Monteith, J. L., 1965: Evaporation and environment, Symp. Soc. Exp. Biol. 19 , pp. 205œ234. Mu, Q., M. Zhao, F. A. Heinsch, M. Liu, H. Tian and S. W. Running, 2007: Development of a global evapotranspiration algorithm based on MODIS and global meteorology data. Journal of Geophysical Research, Volume 112, Number G01012 (doi: 10.1029/2006JG000179) Myneni, R. B., R. R. Nemani, and S. W. Running, 1997: Estimation of global leaf area index and absorbed par using radiative transfer models. Ieee Transactions on Geoscience and Remote Sensing , 35, 1380-1393. Nijssen, B., R. Schnur, and D. P. Lettenmaier, 2001: Global retrospective estimation of soil moisture using the Variable Infiltration Capacity land surface model, 1980-1993. J. Climate, 14, 1790-1808 Pinker, R. T. and I. Laszlo. 1992. Modeling surface solar irradiance for satellite applications on a global scale. J. Appl. Meteor., February issue. Rossow, W. B., and E. N. Duenas, 2004: The International Satellite Cloud Climatology Project (ISCCP) Web site - An online resource for research. Bull. Amer. Met. Soc. , 85, 167-172. Sheffield, J., A. D. Ziegler, E. F. Wood, and Y. Chen, 2004: Correction of the high-latitude rain day anomaly in the NCEP/NCAR reanalysis for land surface hydrological modeling, J. Climate, 17(19), 3814-3828. Su, F., J. C. Adam, K. E. Trenberth, and D. P. Lettenmaier, 2006: Evaluation of surface water fluxes of the pan-Arctic land region with a land surface model and ERA-40 reanalysis, J. Geophys. Res., 111, D05110, doi:10.1029/2005JD006387. Zhao, R. J., Y. L. Zhang, L. R. Fang, X. R. Liu, and Q. S. Zhang, 1980: The Xinanjiang model, in Hydrological forecasting proceedings Oxford Symposium, IASH 129, 351-356. Zhu C.M. and D.P. Lettenmaier, 2007:Long-term climate and derived surface hydrology and energy flux data for Mexico,1925-2004, Journal of Climate , 20, 1936-1946.
29