Comment on egusphere-2023-1826 DOI Creative Commons

Piersilvio De Bartolomeis,

Alexandru Meterez, Zixin Shu

и другие.

Опубликована: Ноя. 6, 2023

Abstract. Accurate predictions of environmental controls on ecosystem photosynthesis are essential for understanding the impacts climate change and extreme events carbon cycle provisioning services. Using time-series measurements fluxes paired with meteorological variables from a network globally distributed sites remotely sensed vegetation indices, we train recurrent deep neural (Long-Short-Term Memory, LSTM), simple (DNN), mechanistic, theory-based model aim to predict gross primary production (GPP). We test these models' ability spatially temporally generalise across wide range conditions. Both models outperform considering leave-site-out cross-validation (LSOCV). The LSTM performs best achieves mean R2 0.78 in LSOCV an average 0.82 relatively moist temperate boreal sites. This suggests that networks provide basis robust data-driven modelling respective biomes. However, limits global upscaling identified using by types continents. In particular, our performance is weakest at arid where unknown exposure water limitation reliability.

Язык: Английский

An effective machine learning approach for predicting ecosystem CO2 assimilation across space and time DOI Creative Commons

Piersilvio De Bartolomeis,

Alexandru Meterez, Zixin Shu

и другие.

Опубликована: Сен. 5, 2023

Abstract. Accurate predictions of environmental controls on ecosystem photosynthesis are essential for understanding the impacts climate change and extreme events carbon cycle provisioning services. Using time-series measurements fluxes paired with meteorological variables from a network globally distributed sites remotely sensed vegetation indices, we train recurrent deep neural (Long-Short-Term Memory, LSTM), simple (DNN), mechanistic, theory-based model aim to predict gross primary production (GPP). We test these models' ability spatially temporally generalise across wide range conditions. Both models outperform considering leave-site-out cross-validation (LSOCV). The LSTM performs best achieves mean R2 0.78 in LSOCV an average 0.82 relatively moist temperate boreal sites. This suggests that networks provide basis robust data-driven modelling respective biomes. However, limits global upscaling identified using by types continents. In particular, our performance is weakest at arid where unknown exposure water limitation reliability.

Язык: Английский

Процитировано

2
Comment on egusphere-2024-276 DOI Creative Commons
Peter M. van Bodegom

Опубликована: Апрель 2, 2024

Abstract. Biomes and their biogeographic patterns have been derived from a large variety of variables including species distributions, bioclimate or remote sensing products. Yet, whether plant trait data are suitable for biome classification has rarely tested. Here, we aimed to assess systematically which traits most classification. We 33 different by combining crowd-sourced distribution the TRY database. Using supervised cluster analyses, developed schemes using these 31 maps. A sensitivity analysis with randomly sampled combinations was performed identify maps that appropriate achieved highest data-model agreement. Due gaps in data, models were used obtain at global scale. showed can be conduit density, rooting depth, height, leaf traits, specific area nitrogen. Data-model agreement maximized when inform analyses based on zonation contrast optical reflectance. The availability is heterogeneous prevalent. Nonetheless, it possible derive predict good Filling essential further improve trait-based

Язык: Английский

Процитировано

0
Comment on egusphere-2024-276 DOI Creative Commons
Simon Scheiter, Sophie Wolf, Teja Kattenborn

и другие.

Опубликована: Апрель 17, 2024

Abstract. Biomes and their biogeographic patterns have been derived from a large variety of variables including species distributions, bioclimate or remote sensing products. Yet, whether plant trait data are suitable for biome classification has rarely tested. Here, we aimed to assess systematically which traits most classification. We 33 different by combining crowd-sourced distribution the TRY database. Using supervised cluster analyses, developed schemes using these 31 maps. A sensitivity analysis with randomly sampled combinations was performed identify maps that appropriate achieved highest data-model agreement. Due gaps in data, models were used obtain at global scale. showed can be conduit density, rooting depth, height, leaf traits, specific area nitrogen. Data-model agreement maximized when inform analyses based on zonation contrast optical reflectance. The availability is heterogeneous prevalent. Nonetheless, it possible derive predict good Filling essential further improve trait-based

Язык: Английский

Процитировано

0
Comment on egusphere-2024-276 DOI Creative Commons

Bianca Rius

Опубликована: Апрель 18, 2024

Abstract. Biomes and their biogeographic patterns have been derived from a large variety of variables including species distributions, bioclimate or remote sensing products. Yet, whether plant trait data are suitable for biome classification has rarely tested. Here, we aimed to assess systematically which traits most classification. We 33 different by combining crowd-sourced distribution the TRY database. Using supervised cluster analyses, developed schemes using these 31 maps. A sensitivity analysis with randomly sampled combinations was performed identify maps that appropriate achieved highest data-model agreement. Due gaps in data, models were used obtain at global scale. showed can be conduit density, rooting depth, height, leaf traits, specific area nitrogen. Data-model agreement maximized when inform analyses based on zonation contrast optical reflectance. The availability is heterogeneous prevalent. Nonetheless, it possible derive predict good Filling essential further improve trait-based

Язык: Английский

Процитировано

0
Reply on CC1 DOI Creative Commons
Simon Scheiter

Опубликована: Май 7, 2024

Abstract. Biomes and their biogeographic patterns have been derived from a large variety of variables including species distributions, bioclimate or remote sensing products. Yet, whether plant trait data are suitable for biome classification has rarely tested. Here, we aimed to assess systematically which traits most classification. We 33 different by combining crowd-sourced distribution the TRY database. Using supervised cluster analyses, developed schemes using these 31 maps. A sensitivity analysis with randomly sampled combinations was performed identify maps that appropriate achieved highest data-model agreement. Due gaps in data, models were used obtain at global scale. showed can be conduit density, rooting depth, height, leaf traits, specific area nitrogen. Data-model agreement maximized when inform analyses based on zonation contrast optical reflectance. The availability is heterogeneous prevalent. Nonetheless, it possible derive predict good Filling essential further improve trait-based

Язык: Английский

Процитировано

0
Reply on RC1 DOI Creative Commons
Simon Scheiter

Опубликована: Май 7, 2024

Abstract. Biomes and their biogeographic patterns have been derived from a large variety of variables including species distributions, bioclimate or remote sensing products. Yet, whether plant trait data are suitable for biome classification has rarely tested. Here, we aimed to assess systematically which traits most classification. We 33 different by combining crowd-sourced distribution the TRY database. Using supervised cluster analyses, developed schemes using these 31 maps. A sensitivity analysis with randomly sampled combinations was performed identify maps that appropriate achieved highest data-model agreement. Due gaps in data, models were used obtain at global scale. showed can be conduit density, rooting depth, height, leaf traits, specific area nitrogen. Data-model agreement maximized when inform analyses based on zonation contrast optical reflectance. The availability is heterogeneous prevalent. Nonetheless, it possible derive predict good Filling essential further improve trait-based

Язык: Английский

Процитировано

0
Reply on RC2 DOI Creative Commons
Simon Scheiter

Опубликована: Май 7, 2024

Abstract. Biomes and their biogeographic patterns have been derived from a large variety of variables including species distributions, bioclimate or remote sensing products. Yet, whether plant trait data are suitable for biome classification has rarely tested. Here, we aimed to assess systematically which traits most classification. We 33 different by combining crowd-sourced distribution the TRY database. Using supervised cluster analyses, developed schemes using these 31 maps. A sensitivity analysis with randomly sampled combinations was performed identify maps that appropriate achieved highest data-model agreement. Due gaps in data, models were used obtain at global scale. showed can be conduit density, rooting depth, height, leaf traits, specific area nitrogen. Data-model agreement maximized when inform analyses based on zonation contrast optical reflectance. The availability is heterogeneous prevalent. Nonetheless, it possible derive predict good Filling essential further improve trait-based

Язык: Английский

Процитировано

0
Comment on egusphere-2023-1826 - Peer review halted DOI Creative Commons
Jens‐Arne Subke

Опубликована: Дек. 14, 2023

Abstract. Accurate predictions of environmental controls on ecosystem photosynthesis are essential for understanding the impacts climate change and extreme events carbon cycle provisioning services. Using time-series measurements fluxes paired with meteorological variables from a network globally distributed sites remotely sensed vegetation indices, we train recurrent deep neural (Long-Short-Term Memory, LSTM), simple (DNN), mechanistic, theory-based model aim to predict gross primary production (GPP). We test these models' ability spatially temporally generalise across wide range conditions. Both models outperform considering leave-site-out cross-validation (LSOCV). The LSTM performs best achieves mean R2 0.78 in LSOCV an average 0.82 relatively moist temperate boreal sites. This suggests that networks provide basis robust data-driven modelling respective biomes. However, limits global upscaling identified using by types continents. In particular, our performance is weakest at arid where unknown exposure water limitation reliability.

Язык: Английский

Процитировано

0
Comment on egusphere-2023-1826 DOI Creative Commons

Piersilvio De Bartolomeis,

Alexandru Meterez, Zixin Shu

и другие.

Опубликована: Ноя. 6, 2023

Abstract. Accurate predictions of environmental controls on ecosystem photosynthesis are essential for understanding the impacts climate change and extreme events carbon cycle provisioning services. Using time-series measurements fluxes paired with meteorological variables from a network globally distributed sites remotely sensed vegetation indices, we train recurrent deep neural (Long-Short-Term Memory, LSTM), simple (DNN), mechanistic, theory-based model aim to predict gross primary production (GPP). We test these models' ability spatially temporally generalise across wide range conditions. Both models outperform considering leave-site-out cross-validation (LSOCV). The LSTM performs best achieves mean R2 0.78 in LSOCV an average 0.82 relatively moist temperate boreal sites. This suggests that networks provide basis robust data-driven modelling respective biomes. However, limits global upscaling identified using by types continents. In particular, our performance is weakest at arid where unknown exposure water limitation reliability.

Язык: Английский

Процитировано

0