TL;DR
This paper presents a novel method for improving land surface models in the Sahel by assimilating coarse-resolution ERS wind scatterometer (WSC) radar data. By using an evolutionary strategy algorithm, the study successfully constrains key vegetation and soil parameters, leading to more accurate simulations of herbaceous mass and water fluxes, crucial for understanding regional climate dynamics.
What Problem Does This Paper Address?
Accurate modeling of land surface processes is vital for understanding and predicting climate, especially in regions like the Sahel where water and energy exchanges are critical. Traditional land surface models often struggle due to limitations in forcing data (like rainfall variability) and the difficulty of representing complex surface processes at large scales. While remote sensing offers a way to monitor surface characteristics, integrating this data into models, particularly with coarse-resolution radar, presents a significant challenge. This paper tackles the problem of how to effectively use ERS WSC radar data to improve the simulation of vegetation dynamics and water budgets in a Sahelian land surface model, addressing the gap in utilizing such data for parameter estimation and model calibration.
Methods
The core of this research lies in the development and application of a data assimilation methodology. The authors first conducted a sensitivity analysis using a Monte Carlo approach to identify which parameters of their Sahelian land surface model (STEP) were most sensitive to WSC radar observations. This led to the selection of 10 key parameters.
The assimilation itself employed an Evolutionary Strategy (ES) algorithm, a type of evolutionary computation. This algorithm is well-suited for complex, nonlinear inverse problems like parameter identification. It works by evolving a population of potential parameter sets over generations, favoring those that minimize a cost function. The cost function quantifies the difference between the observed WSC backscattering coefficient and the simulated coefficient from the RTM-STEP model. Importantly, the ES algorithm doesn't just find a single optimal solution but identifies a set of "acceptable" solutions that meet a defined criterion, providing a more robust estimate of parameter uncertainty.
The STEP model itself simulates vegetation growth and soil water dynamics daily. It's coupled with a Radiative Transfer Model (RTM) to simulate the backscattering coefficient as observed by the WSC instrument. This RTM links surface properties (soil moisture, vegetation water content, biomass, canopy height) to the radar signal.
Datasets
- ERS Wind Scatterometer (WSC) Data: Coarse-resolution (approx. 50 km) backscattering coefficient measurements at 5.3 GHz (VV polarization) from ERS-1 and ERS-2 satellites, collected between August 1991 and December 2000. Data were filtered for incidence angles between 40° and 55° and normalized to 45°.
- Ground Measurements of Herbaceous Mass: Collected at four 25x25 km² sites in the Gourma region of Mali during the growing seasons of 1992, 1993, 1999, and 2000. These measurements are expressed in kg DM/ha.
- Meteorological Data: Rainfall data from four meteorological stations, and other climatic variables (global radiation, temperature, air humidity, wind speed) extracted from the climatological atlas of Morel (1992).
Key Insights and Ideas
- WSC Radar as a Constraint: Coarse-resolution WSC radar data, despite its limitations, can effectively constrain key parameters related to vegetation biomass and soil moisture in a land surface model.
- Evolutionary Strategies for Inverse Problems: ES algorithms are powerful tools for tackling complex parameter identification problems in environmental modeling, offering a robust way to handle nonlinearities and uncertainties.
- Interplay of Soil and Vegetation: The assimilation process highlights the strong interplay between soil moisture dynamics and vegetation growth. Changes in soil resistance to water transfer, influenced by WSC data, directly impact simulated evaporation and transpiration.
- Importance of Rainfall Uncertainty: The study underscores that uncertainties in rainfall data are a major driver of uncertainty in simulated herbaceous mass, even after WSC data assimilation.
- Water Flux Dynamics: Assimilation leads to more realistic simulations of water fluxes, particularly a higher day-to-day variability in soil evaporation and values closer to potential evapotranspiration, aligning better with previous field campaign measurements.
Key Results
- Improved Herbaceous Mass Simulation: After WSC data assimilation (STEP-WSC), the simulated herbaceous mass showed an average error of 187 kg DM/ha compared to ground measurements, a significant improvement over uncalibrated simulations (STEP-free, 615 kg DM/ha AAE).
- Realistic Water Fluxes: Simulated water fluxes (evaporation and transpiration) after assimilation exhibited a more realistic behavior, with higher evaporation rates and greater day-to-day variability, consistent with field observations.
- Parameter Sensitivity Identified: A sensitivity analysis identified 10 key STEP model parameters that could be constrained by WSC data.
- Uncertainty Estimates: The study provided quantitative estimates of uncertainty for herbaceous mass (139-268 kg DM/ha) and water fluxes, primarily driven by uncertainties in meteorological forcing.
- Spatial Variability Captured: The assimilation method showed good performance across different sites along a north-south bioclimatic gradient, capturing spatial variability in land surface responses.
Limitations and Future Work
- Coarse Resolution of WSC Data: The ~50 km spatial resolution of WSC data limits the ability to precisely capture local measurement variations and can lead to spatial representativeness issues for rainfall data.
- Underestimation at Early Growth Stages: STEP-WSC simulations showed a slight underestimation of herbaceous mass at the very beginning of the growing season, attributed to a time shift in vegetation development compared to ground-calibrated models.
- Water Flux Uncertainty: While improved, the uncertainty in simulated water fluxes remains higher, particularly for yearly transpiration. The authors suggest that incorporating additional satellite information from different wavelengths would be beneficial.
- Need for Multispectral Assimilation: The paper concludes by emphasizing the need for a multispectral assimilation approach to better constrain water flux estimations.
- Future Field Campaigns: The authors look forward to further validation of their approach with data from upcoming AMMA field campaigns, specifically eddy covariance measurements of water fluxes.
- Underdetermination of Parameters: The study highlights the potential for underdetermination of parameters when using a single data source for assimilation in nonlinear models, suggesting the danger of using only one data set.
Summary updated