African Crop Science Conference Proceedings, Vol. 7. pp. 1049-1056 Printed in Uganda. All rights reserved ISSN 1023-070X/2005 $ 4.00 © 2005, African Crop Science Society Validation of pedotransfer functions for soil bulk density estimation on a Lake Victoria Basin soilscape G. TAULYA, M.M. TENYWA 1 , M.J.G. MAJALIWA 1 , T.L. ODONG 2 , J. KAINGO 1 & A. KAKONE 1 International Institute of Tropical Agriculture, Uganda 1 Soil Science Department, Faculty of Agriculture, Makerere University, P. O. Box 7062, Kampala, Uganda 2 Crop Science Department, Faculty of Agriculture, Makerere University, P.O. Box 7062, Kampala, Uganda Abstract To validate selected Pedotransfer functions (PTFs) for soil bulk density (ñ b ) estimation on a Lake Victoria Basin (LVB) soilscape, 4 undisturbed soil cores were taken at each of 60 points in a regular grid. Point-wise, the oven-dry ñ b for 2 cores was separately determined and their arithmetic mean computed as the observed ñ b . The remaining 2 cores were bulked and analysed for organic carbon and texture; the out put was used to predict ñ b using 8 selected PTFs. PTF precision and bias were assessed using mean error (ME), unity plots and root mean square prediction difference (RMSPD). Two and six of the PTFs under- and over-estimated ñ b , respectively. The ME and RMSPD ranged from 0.004 and 0.089, respectively for Huntington PTF, to 0.231 and 0.248, respectively for Kaur PTF. The best PTFs were Huntington, Manrique-Jones-B (ME=-0.086; RMSPD=0.118) and Alexander-A (ME=-0.100; RMSPD=0.128); the worst was Kaur. However the R 2 values for all the PTFs were low (5.1-11.0 %). Recalibration of the Huntington PTF on no-till land use data reduced the model’s bias (ME=0.003) and improved its precision (RMSPD=0.009; R 2 =0.578). Key words: Model precision, soil organic carbon, texture, tillage, Uganda Résumé Valider les fonctions de Pédotransfert sélectionnées (PTFs) pour l’estimation de la densité volumétrique de sol (ò b ) sur un bassin versant du lac Victoria (LVB), 4 échantillons de sol intacts étaient pris à chacun des 60 points dans une grille régulière, et pour chaque point les 2 échantillons étaient séchés au four de séchage et ò b était de manière séparée déterminé et leur moyenne arithmétique calculée comme le ò b observé. Les 2 échantillons restant étaient estimés et analysés pour le carbone organique et la texture ; le résultat était utilisé pour prédire ò b en utilisant 8 PTFs sélectionnées. La précision de PTF et l’erreur préjugée étaient évaluée en utilisant l’erreur moyenne (ME), L’unité des terrains et la racine carrée de la moyenne de différences de prédiction (RMSPD). Deux et six des PTFs avaient sous- et sur- estimé ò b , respectivement. La ME et la RMSPD avaient varié de 0,004 et 0,089, respectivement pour la PTF Huntington, à 0,231 et 0,248, respectivement pour La PTF Kaur. Les meilleurs PTFs étaient Huntington, Manrique-Jones-B (ME=0.086 ; RMSPD=0,118) et Alexander-A (ME=0,100 ; RMSPD=0,128) ; le pire était Kaur. Cependant, les valeurs de R 2 pour toutes les PTFs étaient faibles (5,1-11,0%). La récalibration de la PTF Huntington sur les données de terre non cultivée a réduit l’erreur biaisée du model (ME=0,003) et amélioré sa précision (RMSPD=0,009 ; R 2 =0.578). Mots clés: Précision du model, carbone organique de sol, texture, Labour, Ouganda Introduction The Lake Victoria Basin (LVB) has until recently been the major food basket for Uganda (Tenywa et al., 1999), growing mainly banana (Musa spp.), which is the main staple food crop for the country (Speijer and Karamura, 2000). From around the 1970’s, there was a banana yield decline that threatened the food security of the nation, with soil degradation highly ranked among the causative factors (Bekunda and Woomer, 1996). Soil degradation in the LVB is largely due to soil erosion (Lufafa et al., 2003), which increases soil bulk density (ñ b ) through compaction (Islam and Weil, 2000). High ñ b restricts root growth, crop water and nutrient use efficiency (Ishaq et al., 2001a) resulting in yield decline (Ishaq et al., 2001b). This is especially true for bananas, whose fragile root systems have low penetrating power and high oxygen demand (Martin-Prevel, 1987). Blomme (2000) reported strong positive correlation between banana root system development and bunch weight. Root system development has been shown to be a function of soil factors, including ñ b , and genotype (Anguillar, 2000). Banana yield increase in response to use of organic amendments has been attributed to their ñ b -lowering effect (McIntyre et al., 2000). High ñ b reduces infiltration (Abu Hamdeh et al., 2000) leading to high runoff generation, which exacerbates soil erosion (Mwendera and Saleem, 1997). Since in addition to predisposing soil to erosion, ñ b also determines soil physical properties and processes that influence water and solute transport in soils, it has been widely used in environmental modeling (Sobieraj et al., 2001) for soil and water conservation. Unfortunately, ñ b is rarely determined in most soil characterisation programmes. This is mainly because determination of ñ b is laborious, time- consuming and expensive; very large numbers of samples are required to adequately represent a soil unit due to the high degree of spatial variability that ñ b commonly exhibits (Lindbo et al., 1994). In order to overcome this constraint, pedotransfer functions (PTF) have been derived and used to estimate ñ b from soil organic carbon (OC)/organic matter (OM)