Integrating Reanalysis and Satellite Cloud Information to Estimate Surface Downward Long-Wave Radiation

Abstract

The estimation of downward long-wave radiation (DLR) at the surface is very important for the understanding of the Earth’s radiative budget with implications in surface–atmosphere exchanges, climate variability, and global warming. Theoretical radiative transfer and observationally based studies identify the crucial role of clouds in modulating the temporal and spatial variability of DLR. In this study, a new machine learning algorithm that uses multivariate adaptive regression splines (MARS) and the combination of near-surface meteorological data with satellite cloud information is proposed. The new algorithm is compared with the current operational formulation used by the European Organization for the Exploitation of Meteorological Satellites (EUMETSAT) Satellite Application Facility on Land Surface Analysis (LSA-SAF). Both algorithms use near-surface temperature and dewpoint temperature along with total column water vapor from the latest European Centre for Medium-range Weather Forecasts (ECMWF) reanalysis ERA5 and satellite cloud information from the Meteosat Second Generation. The algorithms are trained and validated using both ECMWF-ERA5 and DLR acquired from 23 ground stations as part of the Baseline Surface Radiation Network (BSRN) and the Atmospheric Radiation Measurement (ARM) user facility. Results show that the MARS algorithm generally improves DLR estimation in comparison with other model estimates, particularly when trained with observations. When considering all the validation data, root mean square errors (RMSEs) of 18.76, 23.55, and 22.08 W·m−2 are obtained for MARS, operational LSA-SAF, and ERA5, respectively. The added value of using the satellite cloud information is accessed by comparing with estimates driven by ERA5 total cloud cover, showing an increase of 17% of the RMSE. The consistency of MARS estimate is also tested against an independent dataset of 52 ground stations (from FLUXNET2015), further supporting the good performance of the proposed model.

Evaluation of Downward Surface Longwave Flux Estimates Using Meteosat Cloud Observations

“…Subsequent to the training procedure, the least effective terms are then removed within the algorithm at each step to avoid model overfitting. Here, we use the previously calibrated MARS algorithm, considering a random selection of 6 months of input data (or 40% of the available information) from each of the 23 ground stations used, corresponding to a DSLF sample range of about 5.87 years (Lopes et al., 2022). ERA5 hourly total column water vapor, 2‐m air and dewpoint temperatures, together with MSG cloud fraction, were used as predictors during the calibration procedure.…”

“…All selected stations are within the MSG disc (i.e., longitude/latitude ± 75°E/N), being mostly located within the European region. More details concerning each station information are available from Lopes et al (2022), in which both observational networks were used for an initial hourly assessment of DSLF estimates produced with the MARS algorithm.…”

“…In recent decades, there has been an increasing use of this approach to derive DSLF, namely based on multiple regression models (e.g., Trigo et al, 2010;Zeng et al, 2020) and, more recently, machine learning models (e.g., Cao et al, 2022;Feng, Ye, & Zou, 2020;Feng, Zhang, et al, 2020;Feng et al, 2021;Jiang et al, 2022;Lopes et al, 2022;Shao et al, 2023;Wang et al, 2012Wang et al, , 2020Zhou et al, 2018Zhou et al, , 2019Jung et al, 2019), which are able to account for complex nonlinear interactions that occur between DSLF and respective predictors. The latter can be separated into different branches, including artificial neural network (e.g., Feng, Ye, & Zou, 2020;Feng, Zhang, et al, 2020;Jiang et al, 2022;Wang et al, 2012), random forest (e.g., Shao et al, 2023;Wang et al, 2020), extremely randomized trees (e.g., Cao et al, 2022), gradient boosting regression tree (e.g., Feng et al, 2021), and multivariate adaptive regression splines (MARS, e.g., Jung et al, 2019;Lopes et al, 2022;Zhou et al, 2018Zhou et al, , 2019. For instance, Jiang et al (2022) used a generic algorithmartificial neural network to create a model that combines DSLF data simulated by the Moderate Resolution Atmospheric Transmittance and Radiance Code (MODTRAN-5) with satellite information (radiance at top-ofatmosphere, cloud mask, and water vapor content) from the Moderate-resolution Imaging Spectroradiometer (MODIS) and ERA5 reanalysis (cloud base height and temperature, total cloud cover, 2-m air and dew point temperatures) to produce estimates of DSLF under cloudy-sky conditions.…”

“…The validation results yield a comparable accuracy of the MARS model to other existing products, being attained using an independent group of observational sites. More recent efforts toward the use of a high spatial resolution algorithm that enables the combination of cloud information from remotely sensed data with ground observations and reanalysis data have been achieved by Lopes et al (2022), being the starting point of the present study.…”

“…In particular, we aim to produce instantaneous values (at 30-min frequency) of DSLF for the whole disc observed by the Meteosat Second Generation (MSG) satellite series, within the framework of the European Organization for the Exploitation of Meteorological Satellites (EUMETSAT) Satellite Application Facility on Land Surface Analysis (LSA SAF) operations. For that purpose, we use the MARS algorithm, previously calibrated with in situ observations, to estimate DSLF, which, as described by Lopes et al (2022), provided a good performance with very satisfactory results. The MARS algorithm was trained with BSRN observations of DSLF from 23 ground stations, air and dewpoint temperatures and water vapor gathered from the European Centre for Medium-range Weather Forecasts (ECMWF) latest reanalysis (ERA5), while cloud information was obtained from the MSG satellite to compute separately DSLF for clear and cloudy-sky conditions.…”