A convolutional neural network for spatial downscaling of satellite-based solar-induced chlorophyll fluorescence (SIFnet)

Abstract

Abstract. Gross primary productivity (GPP) is the sum of leaf photosynthesis and represents a crucial component of the global carbon cycle. Space-borne estimates of GPP typically rely on observable quantities that co-vary with GPP such as vegetation indices using reflectance measurements (e.g., normalized difference vegetation index, NDVI, near-infrared reflectance of terrestrial vegetation, NIRv, and kernel normalized difference vegetation index, kNDVI). Recent work has also utilized measurements of solar-induced chlorophyll fluorescence (SIF) as a proxy for GPP. However, these SIF measurements are typically coarse resolution, while many processes influencing GPP occur at fine spatial scales. Here, we develop a convolutional neural network (CNN), named SIFnet, that increases the resolution of SIF from the TROPOspheric Monitoring Instrument (TROPOMI) on board of the satellite Sentinel-5P by a factor of 10 to a spatial resolution of 500 m. SIFnet utilizes coarse SIF observations together with high-resolution auxiliary data. The auxiliary data used here may carry information related to GPP and SIF. We use training data from non-US regions between April 2018 until March 2021 and evaluate our CNN over the conterminous United States (CONUS). We show that SIFnet is able to increase the resolution of TROPOMI SIF by a factor of 10 with a r2 and RMSE metrics of 0.92 and 0.17 mW m−2 sr−1 nm−1, respectively. We further compare SIFnet against a recently developed downscaling approach and evaluate both methods against independent SIF measurements from Orbiting Carbon Observatory 2 and 3 (together OCO-2/3). SIFnet performs systematically better than the downscaling approach (r=0.78 for SIFnet, r=0.72 for downscaling), indicating that it is picking up on key features related to SIF and GPP. Examination of the feature importance in the neural network indicates a few key parameters and the spatial regions in which these parameters matter. Namely, the CNN finds low-resolution SIF data to be the most significant parameter with the NIRv vegetation index as the second most important parameter. NIRv consistently outperforms the recently proposed kNDVI vegetation index. Advantages and limitations of SIFnet are investigated and presented through a series of case studies across the United States. SIFnet represents a robust method to infer continuous, high-spatial-resolution SIF data.

From remotely‐sensed solar‐induced chlorophyll fluorescence to ecosystem structure, function, and service: Part II—Harnessing data

“…To overcome these limitations, a number of “value‐added” SIF products have been derived based on native SIF retrievals (summarized in Table S1b). These products include RSIF (Gentine & Alemohammad, 2018), SIF 005 (Wen et al, 2020), SIF oco2_005 (Yu et al, 2019), GOSIF (Li & Xiao, 2019b), CSIF (Zhang, Joiner, Alemohammad, et al, 2018), LT_SIF c * (Wang, Zhang, et al, 2022), and other fine‐resolution SIF products downscaled from GOME‐2 (Duveiller et al, 2020; Duveiller & Cescatti, 2016) or TROPOMI (Gensheimer et al, 2022; Turner et al, 2020). These products are derived from different native SIF products, and have disparate spatial and temporal resolutions as well as temporal coverage (Figure 1; Table S1b).…”

“…Exploration along this line can already be started with platforms like OCO‐3 or synthetic simulations with observing system simulation experiments (OSSEs)‐type systems (Somkuti et al, 2021). To concurrently alleviate the issues of coarse spatial resolution (which is the case for geostationary satellite), data fusion with other types of spaceborne observations available at fine resolutions (e.g., reflectance, thermal, radar) with state‐of‐the‐art ML techniques (Gensheimer et al, 2022) are worth research exploration. Elucidating the mechanisms in response to stress, especially co‐occurring events, requires effective synergy of different sensing techniques (e.g., SIF, thermal, hyperspectral, lidar) along with mechanistic models/understanding.…”

Investigating the spatio-temporal pattern evolution characteristics of vegetation change in Shendong coal mining area based on kNDVI and intensity analysis

“…This study highlighted the distinct benefits of utilizing kNDVI in mitigating saturation effects, managing intricate phenological cycles, and accounting for seasonal variations. The suitability of the index for effectively representing the status of vegetation coverage in both natural and agricultural systems has been demonstrated in several studies (Liu et al, 2021b;Forzieri et al, 2022;Gensheimer et al, 2022;Wang et al, 2022b). Furthermore, the suitability of kNDVI for evaluating the growth conditions and temporal variations of vegetation in the mining region has been well-recognized (Wang et al, 2023).…”

Sun-induced fluorescence as a proxy for primary productivity across vegetation types and climates

“…Efforts have been made to improve the resolution and coverage of SIF datasets by combining SIF data with other highresolution remote sensing data (Gentine and Alemohammad, 2018;Li and Xiao, 2019;Zhang et al, 2018a;Yu et al, 2018;Gensheimer et al, 2022). These approaches generally rely on statistical inference, through machine learning methods.…”