Find research datasets worth reusing

iSDAsoil: soil extractable Aluminium for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil extractable Aluminium (Al) log-transformed predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, <a href="https://landpotential.org/data-portal/">LandPKS</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_log.al_mehlich3_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil extractable Aluminium mean value,</li> <li>sol_log.al_mehlich3_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil extractable Aluminium model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: log.al_mehlich3 R-square: 0.881 Fitted values sd: 0.872 RMSE: 0.321 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -5.7042 -0.1036 0.0059 0.1189 3.3777 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.675492 2.771906 -0.244 0.807 regr.ranger 0.879567 0.005464 160.969 <2e-16 *** regr.xgboost 0.071537 0.005813 12.306 <2e-16 *** regr.cubist 0.150157 0.004553 32.979 <2e-16 *** regr.nnet 0.087603 0.431261 0.203 0.839 regr.cvglmnet -0.084440 0.003182 -26.534 <2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 0.3208 on 63551 degrees of freedom Multiple R-squared: 0.8808, Adjusted R-squared: 0.8808 F-statistic: 9.391e+04 on 5 and 63551 DF, p-value: < 2.2e-16</code></pre> To back-transform values (y) to ppm use the following formula: <pre><code>ppm = expm1( y / 10 )</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil extractable Potassium for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil extractable Potassium (K) log-transformed predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_log.k_mehlich3_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil extractable Potassium mean value,</li> <li>sol_log.k_mehlich3_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil extractable Potassium model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: log.k_mehlich3 R-square: 0.773 Fitted values sd: 0.938 RMSE: 0.509 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -4.3088 -0.2648 -0.0037 0.2639 6.8136 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 10.907726 6.134422 1.778 0.0754 . regr.ranger 1.004487 0.003878 259.026 <2e-16 *** regr.xgboost -0.004081 0.004739 -0.861 0.3892 regr.cubist 0.084556 0.004346 19.454 <2e-16 *** regr.nnet -2.205286 1.228586 -1.795 0.0727 . regr.cvglmnet -0.064510 0.003933 -16.401 <2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 0.5092 on 139122 degrees of freedom Multiple R-squared: 0.7725, Adjusted R-squared: 0.7725 F-statistic: 9.451e+04 on 5 and 139122 DF, p-value: < 2.2e-16 </code></pre> To back-transform values (y) to ppm use the following formula: <pre><code>ppm = expm1( y / 10 )</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil effective Cation Exchange Capacity (eCEC) for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil effective Cation Exchange Capacity (eCEC) log-transformed predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_log.ecec.f_tot_mehlich3_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil eCEC mean value,</li> <li>sol_log.ecec.f_mehlich3_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil eCEC model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: log.ecec.f R-square: 0.754 Fitted values sd: 0.729 RMSE: 0.417 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -3.2877 -0.1888 0.0097 0.2023 3.1494 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 2.807991 1.806781 1.554 0.1202 regr.ranger 1.046105 0.004845 215.911 < 2e-16 *** regr.xgboost -0.016558 0.005912 -2.801 0.0051 ** regr.cubist 0.031843 0.005063 6.289 3.21e-10 *** regr.nnet -1.142820 0.713071 -1.603 0.1090 regr.cvglmnet -0.027630 0.005607 -4.928 8.34e-07 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 0.4166 on 66380 degrees of freedom Multiple R-squared: 0.7538, Adjusted R-squared: 0.7538 F-statistic: 4.065e+04 on 5 and 66380 DF, p-value: < 2.2e-16 </code></pre> To back-transform values (y) to cmol(+)/kg use the following formula: <pre><code>cmol(+)/kg = expm1( y / 10 )</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil organic carbon for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil organic carbon (C) log-transformed predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_log.oc_mehlich3_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil organic Carbon mean value,</li> <li>sol_log.oc_mehlich3_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil organic Carbon model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: log.oc R-square: 0.791 Fitted values sd: 0.716 RMSE: 0.369 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -3.1517 -0.1900 -0.0060 0.1793 4.2621 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 1.821657 0.794000 2.294 0.0218 * regr.ranger 1.047507 0.005146 203.571 <2e-16 *** regr.xgboost -0.005943 0.005340 -1.113 0.2657 regr.cubist 0.052084 0.004884 10.664 <2e-16 *** regr.nnet -0.867384 0.359213 -2.415 0.0158 * regr.cvglmnet -0.050157 0.003863 -12.984 <2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 0.3687 on 122457 degrees of freedom Multiple R-squared: 0.7906, Adjusted R-squared: 0.7906 F-statistic: 9.248e+04 on 5 and 122457 DF, p-value: < 2.2e-16 </code></pre> To back-transform values (y) to g/kg use the following formula: <pre><code>g/kg = expm1( y / 10 )</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil extractable Zinc for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil extractable Zinc (Zn) log-transformed predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_log.zn_mehlich3_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil extractable Zinc mean value,</li> <li>sol_log.zn_mehlich3_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil extractable Zinc model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: log.zn_mehlich3 R-square: 0.711 Fitted values sd: 0.588 RMSE: 0.375 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -2.1382 -0.2038 -0.0274 0.1632 3.6353 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 1.836394 1.555108 1.181 0.23766 regr.ranger 0.823144 0.013982 58.871 < 2e-16 *** regr.xgboost 0.037861 0.013644 2.775 0.00552 ** regr.cubist 0.169653 0.010091 16.813 < 2e-16 *** regr.nnet -1.511745 1.244933 -1.214 0.22463 regr.cvglmnet 0.009770 0.008046 1.214 0.22467 --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 0.3747 on 39344 degrees of freedom Multiple R-squared: 0.7109, Adjusted R-squared: 0.7109 F-statistic: 1.935e+04 on 5 and 39344 DF, p-value: < 2.2e-16</code></pre> To back-transform values (y) to ppm use the following formula: <pre><code>ppm = expm1( y / 10 )</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil texture class (USDA system) for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil texture classes derived from sand, silt and clay fractions at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_texture.class_c_30m_*..*cm_2001..2017_v0.13_wgs84.tif = soil texture class,</li> </ul> Classes: <pre><code>Code,Name,Value,Color Cl,clay,1,#d5c36b SiCl,silty clay,2,#b96947 SaCl,sandy clay,3,#9d3706 ClLo,clay loam,4,#ae868f SiClLo,silty clay loam,5,#f86714 SaClLo,sandy clay loam,6,#46d143 Lo,loam,7,#368f20 SiLo,silt loam,8,#3e5a14 SaLo,sandy loam,9,#ffd557 Si,silt,10,#fff72e LoSa,loamy sand,11,#ff5a9d Sa,sand,12,#ff005b NODATA,,255,#ffffff </code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil sand content (USDA system) for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil sand content in % predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, <a href="https://landpotential.org/data-portal/">LandPKS</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_sand_tot_psa_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil sand content mean value,</li> <li>sol_sand_tot_psa_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil sand content model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: sand_tot_psa R-square: 0.736 Fitted values sd: 22.8 RMSE: 13.7 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -80.626 -5.321 0.221 6.071 88.686 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 6.687471 24.022001 0.278 0.780714 regr.ranger 1.060521 0.003503 302.742 < 2e-16 *** regr.xgboost -0.018718 0.004910 -3.812 0.000138 *** regr.cubist 0.031749 0.003922 8.096 5.73e-16 *** regr.nnet -0.161127 0.422746 -0.381 0.703098 regr.cvglmnet -0.028217 0.004462 -6.323 2.57e-10 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 13.65 on 122261 degrees of freedom Multiple R-squared: 0.736, Adjusted R-squared: 0.736 F-statistic: 6.818e+04 on 5 and 122261 DF, p-value: < 2.2e-16</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil total organic Nitrogen for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil total organic Nitrogen (N) log-transformed predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_log.n_tot_mehlich3_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil total N mean value,</li> <li>sol_log.n_tot_mehlich3_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil total N model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: log.n_tot_ncs R-square: 0.732 Fitted values sd: 0.326 RMSE: 0.197 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -1.87298 -0.09584 -0.00985 0.07613 3.14728 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.267429 0.493235 0.542 0.588 regr.ranger 1.128208 0.005766 195.669 < 2e-16 *** regr.xgboost -0.048780 0.006108 -7.987 1.4e-15 *** regr.cubist 0.143954 0.004424 32.539 < 2e-16 *** regr.nnet -0.482261 0.797938 -0.604 0.546 regr.cvglmnet -0.170889 0.004955 -34.489 < 2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 0.1972 on 99249 degrees of freedom Multiple R-squared: 0.7319, Adjusted R-squared: 0.7319 F-statistic: 5.419e+04 on 5 and 99249 DF, p-value: < 2.2e-16</code></pre> To back-transform values (y) to g/kg use the following formula: <pre><code>g/kg = expm1( y / 100 )</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

iSDAsoil: soil silt content (USDA system) for Africa predicted at 30 m resolution at 0-20 and 20-50 cm depths

iSDAsoil dataset soil silt content in % predicted at 30 m resolution for 0–20 and 20–50 cm depth intervals. Data has been projected in WGS84 coordinate system and compiled as <a href="https://gdal.org/drivers/raster/cog.html">COG</a>. Predictions have been generated using multi-scale Ensemble Machine Learning with 250 m (MODIS, PROBA-V, climatic variables and similar) and 30 m (DTM derivatives, Landsat, Sentinel-2 and similar) resolution covariates. For model training we use a pan-African compilations of soil samples and profiles (<a href="https://www.isda-africa.com/national-soil-services/">iSDA points</a>, <a href="https://www.isric.org/projects/africa-soil-profiles-database-afsp">AfSPDB</a>, <a href="https://landpotential.org/data-portal/">LandPKS</a>, and other national and regional soil datasets). Cite as: Hengl, T., Miller, M.A.E., Križan, J. et al. African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning. Sci Rep 11, 6130 (2021). <a href="https://doi.org/10.1038/s41598-021-85639-y">https://doi.org/10.1038/s41598-021-85639-y</a> To open the maps in QGIS and/or directly compute with them, please use the <a href="https://gitlab.com/openlandmap/africa-soil-and-agronomy-data-cube">Cloud-Optimized GeoTIFF version</a>. Layer description: <ul> <li>sol_silt_tot_psa_mehlich3_m_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil silt content mean value,</li> <li>sol_silt_tot_psa_mehlich3_md_30m_*..*cm_2001..2017_v0.13_wgs84.tif = predicted soil silt content model (prediction) errors,</li> </ul> Model errors were derived using bootstrapping: md is derived as standard deviation of individual learners from 5-fold cross-validation (using spatial blocking). The model 5-fold cross-validation (<a href="https://mlr.mlr-org.com/reference/makeStackedLearner.html">mlr::makeStackedLearner</a>) for this variable indicates: <pre><code>Variable: silt_tot_psa R-square: 0.64 Fitted values sd: 11.9 RMSE: 8.92 Random forest model: Call: stats::lm(formula = f, data = d) Residuals: Min 1Q Median 3Q Max -63.746 -3.631 -0.526 2.630 72.486 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -35.876865 36.887592 -0.973 0.331 regr.ranger 0.948111 0.003874 244.733 < 2e-16 *** regr.xgboost 0.062717 0.005506 11.391 < 2e-16 *** regr.cubist 0.025705 0.004747 5.415 6.14e-08 *** regr.nnet 1.902142 1.968248 0.966 0.334 regr.cvglmnet -0.028579 0.005799 -4.928 8.32e-07 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 8.915 on 122223 degrees of freedom Multiple R-squared: 0.6399, Adjusted R-squared: 0.6399 F-statistic: 4.344e+04 on 5 and 122223 DF, p-value: < 2.2e-16</code></pre> To submit an issue or request support please visit <a href="https://isda-africa.com/isdasoil">https://isda-africa.com/isdasoil</a>

opencc-zeroOct 2020View details →

Phase resolution of heterozygous sites in diploid genomes is important to phylogenomic analysis under the multispecies coalescent model

Genome sequencing projects routinely generate haploid consensus sequences from diploid genomes, which are effectively chimeric sequences with the phase at heterozygous sites resolved at random. The impact of phasing errors on phylogenomic analyses under the multispecies coalescent (MSC) model is largely unknown. Here we conduct a computer simulation to evaluate the performance of four phase-resolution strategies (the true phase resolution, the diploid analytical integration algorithm which averages over all phase resolutions, computational phase resolution using the program PHASE, and random resolution) on estimation of the species tree and evolutionary parameters in analysis of multi-locus genomic data under the MSC model. We found that species tree estimation is robust to phasing errors when species divergences were much older than average coalescent times but may be affected by phasing errors when the species tree is shallow. Estimation of parameters under the MSC model with and without introgression is affected by phasing errors. In particular, random phase resolution causes serious overestimation of population sizes for modern species and biased estimation of cross-species introgression probability. In general the impact of phasing errors is greater when the mutation rate is higher, the data include more samples per species, and the species tree is shallower with recent divergences. Use of phased sequences inferred by the PHASE program produced small biases in parameter estimates. We analyze two real datasets, one of East Asian brown frogs and another of Rocky Mountains chipmunks, to demonstrate that heterozygote phase-resolution strategies have similar impacts on practical data analyses. We suggest that genome sequencing projects should produce unphased diploid genotype sequences if fully phased data are too challenging to generate, and avoid haploid consensus sequences, which have heterozygous sites phased at random. In case the analytical integration algorithm is computationally unfeasible, computational phasing prior to population genomic analyses is an acceptable alternative.

opencc-by-4.0Sep 2020View details →

Corrected super-resolution microscopy enables nanoscale imaging of auto-fluorescent lung macrophages

Observing the cell surface and underlying cytoskeleton at nanoscale resolution using super-resolution microscopy has enabled many insights into cell signalling and function. However, the nanoscale dynamics of tissue-specific immune cells have been relatively little studied. Tissue macrophages, for example, are highly auto-fluorescent, severely limiting the utility of light microscopy. Here, we report a correction technique to remove auto-fluorescent noise from Stochastic Optical Reconstruction Microscopy (STORM) datasets. Simulations identified a moving median filter as an accurate and robust correction technique. Using this, we were able to visualise lung macrophages activated through Fc receptors by antibody-coated glass slides. Accurate, nanoscale quantification of macrophage morphology revealed that activation induced the formation of cellular protrusions tipped with MHC class I protein. These data are consistent with a role for lung macrophage protrusions in antigen presentation. We further show that the tetraspanin and extracellular vesicle (EV) marker CD81 appears in ring-shaped structures (mean diameter 93 ± 50 nm) at the surface of activated lung macrophages, likely marking the secretion of extracellular vesicles. Moreover, this correction method for super-resolution microscopy is widely applicable to other challenging biological samples.

opencc-by-4.0Nov 2020View details →

Monitoring recent changes in the Beaufort Sea coast using very high resolution remote sensing

Arctic permafrost coasts are major carbon (Schuur et al., 2015) and mercury pools (Schuster et al., 2018). They represent about 34% of the Earth’s coastline, with long sections affected by high erosion rates (Fritz et al, 2017), increasingly threatening coastal communities. Year-round reduction in Arctic sea ice is forecasted and by the end of the 21st century, models indicate a decrease in sea ice area from 43 to 94% in September and from 8 to 34% in February (IPCC, 2014). An increase of the sea-ice free season leads to a longer exposure of coasts to wave action. Further, climate warming is also expected to modify the contribution of terrestrial erosion (Fritz et al., 2015, Ramage et al., 2018, Irrgang et al., 2018). Within the project EU Horizon2020 project NUNATARYUK, we are updating the mapping of the Arctic coast, with the Canadian Beaufort coast as a case-study. The surveying methodology includes: i. a high resolution update of the coastline mapping and change rates using Pleiades (CNES) satellite acquisitions from 2018, ii. a survey using RTK-UAV aerial imagery of long-term monitoring sites from the Canada-US border to King Point, and iii. the experimental use of TerraSAR-X staring spotlight scenes and PAZ at key sites to monitor intraseasonal dynamics of cliff edge retreat. This research is funded by the EC H2020 Project NUNATARYUK. Support on remote sensing imagery access by the WMO Polar Space Task Group.

opencc-by-4.0Dec 2020View details →

High Resolution Fundus Image Database for Monomodal Single-Channel Image Registration of Thin Features

A high resolution image database of 42 image pairs (related by elastic deformations) created from original images from the High Resolution Fundus Image Database. Consists of thin linear structures that lack sufficient overlap to pose a challenge for classic similarity measures based on overlapping pixels commonly used in image registration. The dataset contains the intensity grayscale images, as well as binary masks of the retinal area, and labels of the vessels, segmented by expert annotators.

opencc-by-4.0Dec 2019View details →

An automatically generated high-resolution earthquake catalogue for the 2016-2017 Central Italy seismic sequence, including P and S phase arrival times

Catalog of 440,697 earthquakes of the 2016-2017 Central Italy seismic sequence semi-automatically generated by Spallarossa et al. (2020). The catalogue covers one year of aftershocks following the first mainshock of the sequence (from 08242016 to 08312017). The catalog has been generated using the Complete Automatic Seismic Processor (CASP) procedure (Scafidi et al., 2019) to detect the events and an advanced picker engine (RSNI-Picker2; Scafidi et al., 2018; Spallarossa et al., 2014) to determine their phase arrival times. The final set of about 7 million P- and 10 million S-wave arrival times have been used to locate the events using a non-linear location algorithm (NonLinLoc; Lomax et al. 2000), with a 1D velocity model calibrated for the area (De Luca et al., 2009) and station corrections. For each event, also local magnitudes (ML) has been calculated as well as a locations quality. Earthquake locations quality has been classified by means of the procedure proposed by Michele et al., (2019) consisting of the combination of diverse uncertainty parameters provided by the NonLinLoc location code. Locations quality is provided in terms of a unique numeric normalized value, named quality factor, varying between qf=0 (best quality location) and qf=1 (worst quality location). Then locations have been assigned to a quality class depending on the qf parameter value according to the following scheme: A-class (0 < qf ≤ 0.25), B-class (0.25 < qf ≤ 0.50), C-class (0.50 < qf ≤ 0.75), and D-class (0.75 < qf < 1.00). The earthquake locations are distributed between the quality classes as A-30.6%, B-31.4%, C-18.6%, and D-19.4% (details in Spallarossa et al., 2020). We accompanied the catalogue with the 30 events with M>3.5 missed by our procedure (bring the total number of events to 440,727), including the first Amatrice mainshock (MW6.0; see Spallarossa et al., 2020). These 30 missing events recognisable by the ID starting with ISI), have been taken from INGV bulletin (<a href="http://terremoti.ingv.it">http://terremoti.ingv.it</a>; ISIDe Working Group., 2007), manually generated. These additional events report INGV locations and magnitude parameters while are missing related quality factors and quality class, being generated by a different procedure. We added to the larger events, the available moment magnitudes (MW) from Time Domain Moment Tensor catalogue (<a href="http://terremoti.ingv.it/tdmt">http://terremoti.ingv.it/tdmt</a>; Scognamiglio et al., 2006). The catalog is in csv format, semicolon separator, ordered by origin time and the header content is the following: <ul> <li>Id-event – ID</li> <li>Latitude (°) expressed in decimal degrees - LAT</li> <li>Longitude (°) expressed in decimal degrees - LON</li> <li>Depth(km) hypocentral depth expressed in kilometres - DEP</li> <li>Year of origin time in the format yyyy - YR</li> <li>Month of origin time in the format mo - MON</li> <li>Day of origin time in the format dd - DY</li> <li>Hour of origin time in the format hh - HR</li> <li>Minute of origin time in the format mi - MIN</li> <li>Second of origin time in the format XX.XXX s - SEC</li> <li>Local Magnitude - ML</li> <li>Standard deviation of the Local Magnitude – STD</li> <li>Moment Magnitude – Mw (from TDMT)</li> <li>Horizontal Error (from NLL output) (km) expressed in kilometres - ERH</li> <li>Vertical Error (from NLL output) (km) expressed in kilometres - ERZ</li> <li>RMS (from NLL output) (s) expressed in seconds - RMS</li> <li>Number of Phases – NPHS</li> <li>Stations Azimuthal GAP (°) expressed in decimal degrees - GAP</li> <li>Quality factor - Qf</li> <li>Quality class - Qc</li> </ul>   De Luca G., M. Cattaneo, G. Monachesi and A, Amato (2009). Seismicity in the Umbria-Marche region from the integration of national and regional seismic networks. Tectonophysics, 476(1), 219-231.  doi: 10.1016/j.tecto.2008.11.032. ISIDe Working Group. (2007). Italian Seismological Instrumental and Parametric Database (ISIDe). Istituto Nazionale di Geofisica e Vulcanologia (INGV); https://doi.org/10.13127/ISIDE. Lomax, A., J. Virieux, P. Volant, and C. Berge-Thierry (2000). Probabilistic earthquake location in 3D and layered models: introduction of a Metropolis–Gibbs method and comparison with linear locations. In: Advances in seismic event location, ed. C. H. Thurber and N. Rabinowitz, 101–134. Dordrecht and Boston: Kluwer Academic Publishers. Michele, M., Latorre, D., Emolo, A. (2019). An Empirical Formula to Classify the Quality of Earthquake Locations. Bulletin of the Seismological Society of America. Vol. 109, No. 6, pp. 2755–2761, December 2019, doi: 10.1785/0120190144. Scafidi, D., Viganò A., Ferretti G., and Spallarossa D. (2018). Robust picking and accurate location with RSNI-Picker2: real-time automatic monitoring of earthquakes and non-tectonic events, Seismol. Res. Lett, Vol. 89 (4), pp. 1478-1487, doi: 10.1785/0220170206. Scafidi D, Spallarossa D, Ferretti G, Barani S, Castello B, Margheriti L (2019). A complete automatic procedure to compile reliable seismic catalogs and travel-time and strong-motion parameters datasets. Seismol Res Lett 90(3):1308–1317. Scognamiglio, L., Tinti, E., Quintiliani, M. (2006). Time Domain Moment Tensor [Data set]. Istituto Nazionale di Geofisica e Vulcanologia (INGV). https://doi.org/10.13127/TDMT. Spallarossa, D., G. Ferretti, D. Scafidi, C. Turino, and M. Pasta (2014). Performance of the RSNI-Picker, Seismol. Res. Lett. 85, 1243–1254. Spallarossa D., Cattaneo M., Scafidi D., Michele M., Chiaraluce L., Segou M. and I. G. Main (2020). An automatically generated high-resolution earthquake catalogue for the 2016-2017 Central Italy seismic sequence, including P and S phase arrival times. Geophys. J. Int. doi: 10.1093/gji/ggaa604.

opencc-by-4.0Jul 2020View details →

Supplementary Information for "Resolution and the Detection of Cultural Dispersals: development and application of spatiotemporal methods in Lowland South America"

Data and code to reproduce the analyses in "Resolution and the Detection of Cultural Dispersals: development and application of spatiotemporal methods in Lowland South America"   Abstract Inferring episodes of expansion, admixture, diffusion, and/or migration in prehistory is at present undergoing a resurgence in macro-scale archaeological interpretation. In parallel to this renewed popularity, expanding access to computational tools and datasets has seen the use of aggregated radiocarbon datasets for the study of dispersals also increasing. This paper advocates for developing reflexive practice in the application of radiocarbon dates to prehistoric dispersals, by reflecting on the quality and qualities of the underlying data, particularly chronometric uncertainty, and framing dispersals explicitly in terms of hypothesis testing. This paper draws on cultural expansions within South America and employs two emblematic examples, the Arauquinoid and Tupiguarani traditions, to develop an analytical solution that not only incorporates chronometric uncertainty in bivariate regression but, importantly, tests whether the datasets provide statistically significant evidence for a dispersal process. The analysis, which the paper provides the means to replicate, identifies fundamental issues with resolution and data quality that impede identification of pre-Columbian cultural dispersals through simple spatial gradients of radiocarbon data. The results suggest that reflexivity must be fed back into theoretical frameworks of prehistoric mobility for the study of dispersals, in turn informing the construction of more critical statistical null models. As a first step, alternative models of cultural expansion should be formally considered alongside demographic models.Inferring episodes of expansion, admixture, diffusion, and/or migration in prehistory is at present undergoing a resurgence in macro-scale archaeological interpretation. In parallel to this renewed popularity, expanding access to computational tools and datasets has seen the use of aggregated radiocarbon datasets for the study of dispersals also increasing. This paper advocates for developing reflexive practice in the application of radiocarbon dates to prehistoric dispersals, by reflecting on the quality and qualities of the underlying data, particularly chronometric uncertainty, and framing dispersals explicitly in terms of hypothesis testing. This paper draws on cultural expansions within South America and employs two emblematic examples, the Arauquinoid and Tupiguarani traditions, to develop an analytical solution that not only incorporates chronometric uncertainty in bivariate regression but, importantly, tests whether the datasets provide statistically significant evidence for a dispersal process. The analysis, which the paper provides the means to replicate, identifies fundamental issues with resolution and data quality that impede identification of pre-Columbian cultural dispersals through simple spatial gradients of radiocarbon data. The results suggest that reflexivity must be fed back into theoretical frameworks of prehistoric mobility for the study of dispersals, in turn informing the construction of more critical statistical null models. As a first step, alternative models of cultural expansion should be formally considered alongside demographic models.

opencc-zeroDec 2013View details →

Data from: Multi-modal ultra-high resolution structural 7-Tesla MRI data repository

Structural brain data is key for the understanding of brain function and networks, i.e., connectomics. Here we present data sets available from the 'atlasing of the basal ganglia (ATAG)' project, which provides ultra-high resolution 7Tesla (T) magnetic resonance imaging (MRI) scans from young, middle-aged, and elderly participants. The ATAG data set includes whole-brain and reduced field-of-view MP2RAGE and T2*-weighted scans of the subcortex and brainstem with ultra-high resolution at a sub-millimeter scale. The data can be used to develop new algorithms that help building high-resolution atlases both relevant for the basic and clinical neurosciences. Importantly, the present data repository may also be used to inform the exact positioning of electrodes used for deep-brain-stimulation in patients with Parkinson's disease and neuropsychiatric diseases.

opencc-zeroJan 2020View details →

Anatomy of the neural endocranium and stapes of Diadectes absitus (Diadectomorpha) from the early Permian of Germany based on the high‐resolution X‐ray microcomputed tomography

A detailed anatomy of the braincase and stapes of the subadult specimen of Diadectes absitus from early Permian sediments of Germany based on the high-resolution X-ray microcomputed tomography are described for the first time. In contrast to previous studies of Diadectes, the bones of the braincase (opisthotic, prootic, supraoccipital, basioccipital, exoccipital, basisphenoid, sphenethmoid), and parasphenoid of D. absitus are not co-ossified, but suturally defined. This has allowed for a reconstruction of a complete braincase with all sutures between the individual bones. The opisthotic, prootic, and supraocciptal contain a well-preserved endosseous labyrinth. The 3D-reconstruction of its cavities shows a well-preserved vestibule, three semicircular canals, and well-developed cochlear recess. In addition, a shallow subarcuate fossa is present on the ventral surface of the supraoccipital, which lies medial to the anterior semicircular canal. A typical feature of the diadectid braincase is the presence of the otic tube leading from the fenestra vestibuli to the vestibule. A revision of the topology of this structure is presented here. Here we describe new structures of the stapes, especially in its proximal portion, as well as its position to the fenestra vestibuli. These structures are described for the first time not only in D. absitus, but for the genus.

opencc-zeroSep 2018View details →

Data from: Impacts of inference method and dataset filtering on phylogenomic resolution in a rapid radiation of ground squirrels (Xerinae: Marmotini)

Phylogenomic datasets are illuminating many areas of the Tree of Life. However, the large size of these datasets alone may be insufficient to resolve problematic nodes in the most rapid evolutionary radiations, because inferences in zones of extraordinarily low phylogenetic signal can be sensitive to the model and method of inference, as well as the information content of loci employed. We used a dataset of >3,950 ultraconserved element (UCE) loci from a classic mammalian radiation, ground-dwelling squirrels of the tribe Marmotini (Sciuridae: Xerinae), to assess sensitivity of phylogenetic estimates to varying per-locus information content across 4 different inference methods (RAxML, ASTRAL, NJst, SVDquartets). Persistent discordance was found in topology and bootstrap support between concatenation- and coalescent-based inferences; among methods within the coalescent framework; and within all methods in response to different filtering scenarios. Contrary to some recent empirical UCE-based studies, filtering by information content did not promote complete among-method concordance. Nevertheless, filtering did improve concordance relative to randomly selected locus sets, largely via improved consistency of two-step summary methods (particularly NJst) under conditions of higher average per-locus variation (and thus increasing gene tree precision). The benefits of dataset filtering are notably variable among classes of inference methods and across different evolutionary scenarios, reiterating the complexities of resolving rapid radiations, even with robust taxon and character sampling.

opencc-by-4.0Jun 2021View details →

Impact of increased resolution on long-standing biases in HighResMIP-PRIMAVERA climate models

This contains the data and plot scripts to reproduce the figures of the manuscript: Impact of increased resolution on long-standing biases in HighResMIP-PRIMAVERA climate models. Authors: Eduardo Moreno-Chamarro1*, Louis-Philippe Caron1,2, Saskia Loosveldt Tomas1, Oliver Gutjahr3,4 , Marie-Pierre Moine5, Dian Putrasahan3 , Christopher D. Roberts6, Malcolm J. Roberts7, Retish Senan6, Laurent Terray5, Etienne Tourigny1, Pier Luigi Vidale8 1 Barcelona Supercomputing Center (BSC), Barcelona, Spain. 2 Ouranos, Montreal, H3A 1B9, Canada. 3 Max Planck Institute for Meteorology. Hamburg, Germany. 4 Now at Institut für Meereskunde, Universität Hamburg, Hamburg, Germany. 5 CECI, Université de Toulouse, CERFACS/CNRS, Toulouse, France. 6 ECMWF European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom. 7 Met Office, Exeter EX1 3PB, United Kingdom. 8 NCAS-Climate, Department of Meteorology, University of Reading, Reading, United Kingdom. Correspondence to: Eduardo Moreno-Chamarro (eduardo.moreno@bsc.es) Abstract. We examine the impacts of increased resolution on four long-standing biases using five different climate models developed within the PRIMAVERA project. Atmospheric resolution is increased from ~100–200 km to ~25–50 km, and ocean resolution is increased from ~1° (i.e., eddy-parametrized) to ~0.25° (i.e., eddy-present). For one model, ocean resolution is also increased to 1/12° (i.e., eddy-rich). Fully-coupled general circulation models and their atmosphere-only versions are compared with observations and reanalysis of near-surface temperature, precipitation, cloud cover, net cloud radiative effect, and zonal wind over the period 1980–2014. Both the ensemble mean and the individual models are analyzed. Increased resolution especially in the atmosphere helps reduce the surface warm bias over the tropical upwelling regions in the coupled models, with further improvements in the cloud cover and precipitation biases over the tropical Atlantic particularly. Related to this and to the improvement in the precipitation distribution over the western tropical Pacific, the double ITCZ bias also weakens with resolution. Overall, increased ocean resolution from 1° to 0.25° offers limited improvements or even bias degradation in some models, although an eddy-rich ocean resolution seems beneficial to reduce the North Atlantic cold bias and the Gulf Stream path. Despite the improvements, however, large biases in precipitation and cloud cover persist over the whole tropics as well as in the upper-troposphere zonal winds at mid-latitudes in both coupled and atmosphere-only models at higher resolutions. The SO warm bias also worsens or persists in some coupled models. And a new warm bias emerges in the Labrador Sea in all the high-resolution coupled models. The analysis of the PRIMAVERA models therefore suggests that, to reduce biases, i) increased atmosphere resolution up to ~25–50 km alone might not be sufficient and ii) an eddy-rich ocean resolution might be needed. The study thus adds to evidence that further improved model physics and tuning might be necessary in addition to increased resolution to mitigate biases.