Ninety-four silages were made over 5 years from predominantly perennial ryegrass swards using a range of cutting dates (19 May to 18 September), wilting periods (0 to 48 h) and additives (none, acids, inoculants, sugar, sugar + acids, sugar + inoculants). A wide range of silage composition was achieved (CV for dry matter (DM), crude protein (CP), digestible organic matter (DOMD), lactic acid, total volatile fatty acids (VFA) and sugar were 0-22, 0-19, 0-07, 0A3, 0-84 and 0-69 respectively). Silage dry-matter intake (SDMI) was measured for 88 silages using lambs (mean live weight (M) 28 kg) given silage as their sole diet in four incomplete block design experiments using four lambs per silage and a standard hay given every third period for covariance correction. Thirty-four of the silages were also evaluated using early lactation cows (M, 561 kg and milk yield 27 kg/day) with 7 kg/day of concentrate in eight incomplete block change-over experiments each using 12 cows. Intakes (SDMI mean, range, s.d. glkg M 075 ) were 56, 25 to 84, 13-7 for lambs and 90, 64 to 119, 13-4 for cows. Scaling lamb SDMI by M 1 ' 47 accounted best for the effect of lamb weight on intake (mean, range 5-07, 2-43 to 7-68). (NIRSdry) or fresh samples (NIRSwetl using a vertical transport mechanism and NIRSwet2 using a rotating cup). The most useful predictors within each group were firstly identified by step-wise multiple linear regression and models were then derived by partial least squares. Standard errors of cross validation (SECV) obtained by the 'leave one out' method were for lamb SDMI (g/M 1 ' 47 ) BASAL + HPLC, BASAL + ET, NIRSdry,[5][6][7][8][9][2][3][4][5]. Inclusion of fermentation measurements made by ET, but not by HPLC, improved SDMI prediction over that obtained from the BASAL set. However, NIRSdry and NIRSwet2 were the most accurate methods giving values for s.d. (reference population) I SECV of 2-27 and 2-13 for lambs and 2-65 and 5-28 for dairy cows. Use of these methods in advisory silage evaluation should substantially reduce the errors of predicting the intake potential of grass silages. Silage predictors were grouped as follows: traditional values (BASAL) -DM, CP, organic matter (OM), DOMD, neutral-detergent fibre (NDF), acid-detergent fibre (ADF), ammonia nitrogen (NH 3 N), pH, acid hydrolysed ether extract (AHEE); silage fermentation values obtained by high-performance liquid chromatography (HPLC); or by electrometric titration (ET); and near infra-red reflectance spectra (NIRS) obtained on either 100°C dried

The greatest error in formulating rations is due to the inaccuracy of prediction of silage dry matter intake (SDMI). Until recently, predictions have been based on die method of Lewis (1981) which predicts intake from traditional silage analysis :- dry matter (DM), crude protein (CP), digestible organic matter in the dry matter (DOMD) and ammonia N. Recently, the incorporation of new feed characterisation data, obtained from electrometric titration (ET), has unproved predictions (Offer et al., 199S). A 4 year study has yielded data to evaluate alternative methods for the prediction of SDMI using traditional, ET and HPLC data and spectral information obtained by near infra-red reflectance spectroscopy (NIRS) of fresh and dried samples.

Eight change-over design experiments (each a duplicated 3X3 Latin square design using six rumen-fistulated wether sheep, live weight 50 to 60 kg) measured rumen fermentation patterns for 24 perennial ryegrass silages. Sheep were offered 800 g dry matter (DM) per day of each silage in two equal meals at 09.00 and 17.00 h. Samples of rumen liquor were taken on days 19 and 21 of each 21-day period, at 08.50 h and at 1-5-h intervals until 16.30 h. Rumen samples were analysed by gas chromatography; silages by high-performance liquid chromatography and by near infra-red reflectance spectroscopy (NIRS) using samples scanned after drying at 100°C (NIRSdry) or in the fresh state (NIRSwet). Mean intake of DM was 737 glday. The range of silage composition was as follows (mean, range, s.d., glkg DM unless specified): metabolizable energy (ME MJ/kg DM) 11-1, 8-8 to 12-6, 0-81; pH 4-0, 3-6 to 5-0, 0-34; lactic acid 86, 4 to 139, 42-6; butyric acid 4-7, 0-1 to 46-7,10-2. Rumen measurements varied substantially both diurnally and between silages. Mean diurnal rumen values for the 24 silages (mean, range, s.d.) were: pH 6-76, 6-55 to 7-09, 0-155; ammonia (mg/l) 132, 70 to 247, 47-7; total volatile fatty acids (TVFA mmolll) 58-2, 45-8 to 72-0, 8-97; (acetate+butyrate)/propionate (ABP) 3-2, 2-2 to 4-8, 0-56. Partial least-square models were developed to predict rumen fermentation (means for six sampling times) using either the silage chemical composition (CHEM glkg DM unless specified : DM, ME (MJ/kg DM), crude protein (CP), ammonia (NH 3 , g N per kg total N), neutralizing value (meq per kg DM), sugar, lactic, formic, acetic, propionic and butyric acids and ethanol) or silage NIRSdry or NIRSwet. Prediction performance was assessed comparing values for R 2 , standard error of cross validation (SECV) and SD/SECV (s.d. of reference population! SECV) obtained by the 'leave one out' cross validation method. NIRSwet gave slightly better prediction accuracy overall than NIRSdry but both were superior to prediction from chemical composition. Values for R 2 , SECV and SD/SECV for pH were 0-The silage predictors with greatest influence in the CHEM model for rumen ABP ratio were sugar, CP and lactic acid (negative) and butyrate and ethanol (positive). NIRS shows considerable promise as a means of predicting rumen fermentation of animals given grass silage diets.

The greatest error in formulating rations is due to the inaccuracy of prediction of silage dry matter intake (SDMI). Until recently, predictions have been based on die method of Lewis (1981) which predicts intake from traditional silage analysis :- dry matter (DM), crude protein (CP), digestible organic matter in the dry matter (DOMD) and ammonia N. Recently, the incorporation of new feed characterisation data, obtained from electrometric titration (ET), has unproved predictions (Offer et al., 199S). A 4 year study has yielded data to evaluate alternative methods for the prediction of SDMI using traditional, ET and HPLC data and spectral information obtained by near infra-red reflectance spectroscopy (NIRS) of fresh and dried samples.