Introduction
Tanzania has the third-largest cattle population in sub-Saharan Africa, estimated at 36.6 million head of cattle (World Bank, 2024). The dairy sector contributes 7% of the gross domestic product (GDP), although not meeting its potential due to low productivity and inadequate quality of the feeds (Maleko et al., Reference Maleko, Msalya, Mwilawa, Pasape and Mtei2018; United Republic of Tanzania, 2018). The dairy sector is experiencing rapid growth due to the increase in population, potentially boosting the demand for feed, including forages (Jones et al., Reference Jones, Nowak, Berglund, Grinnell, Temu, Paul, Renwick, Steward, Rosenstock and Kimaro2023), yet there is an insufficient amount of quality forage for dairy cows. Thus, increasing production of quality forages to enhance dairy production is essential for improving the income and livelihood of smallholder farmers and contributing to the country’s GDP.
Forages, which include herbaceous grasses and legumes as well as shrubs, are crops intentionally cultivated for consumption by livestock and are a crucial feed source for livestock productivity. Some of the widely adopted forages in East Africa include Napier (Pennisetum purpureum) and Rhodes grass (Chloris gayana) (Njarui et al., Reference Njarui, Gatheru, Ndubi, Gichangi and Murage2021). Improved forages, such as Urochloa species, are also gaining acceptance due to enhanced dairy production, particularly milk yield (Paul et al., Reference Paul, Koge, Maass, Notenbaert, Peters, Groot and Tittonell2020). The cultivation of improved perennial grasses like Urochloa cv. Cayman can play a critical role in regenerating soil health by increasing soil organic matter through soil organic carbon (SOC) sequestration, reducing erosion and enhancing soil fertility, and, ultimately, restoring degraded land and soil (Damene et al., Reference Damene, Bahir and Villamor2020; Horrocks et al., Reference Horrocks, Arango, Arévalo, Nuñez, Cardoso and Dungait2019; Maleko et al., Reference Maleko, Mwilawa, Msalya, Pasape and Mtei2019; Gichangi et al., Reference Gichangi, Njarui, Ghimire, Gatheru and Magiroi2016). Among the listed benefits, SOC sequestration is critical since it improves soil health, nutrient cycling, and water-holding capacity, thus improving biomass production that contributes to mitigation of the negative effects of climate change and existing climate variability. Increased forage yield can help bridge the gap between forage feed demand and supply, increase milk production, and ultimately improve smallholder farmers’ livelihoods (Djikeng et al., Reference Djikeng, Rao, Njarui, Mutimura, Caradus, Ghimire, Johnson, Cardoso, Ahonsi and Kelemu2014; Peters et al., Reference Peters, Herrero, Fisher, Erb, Rao, Subbarao, Castro, Arango, Chará, Murgueitio and Van Der Hoek2013; Thornton and Herrero, Reference Thornton and Herrero2010). In addition, improved forages can also reduce methane emission intensity from enteric fermentation through greater nutritional value and digestibility of the forage (Paul et al., Reference Paul, Koge, Maass, Notenbaert, Peters, Groot and Tittonell2020). To realize these co-benefits, farmers must implement improved management practices, including residue return, farmyard manure application, improved cultivars, irrigation, integrated nutrient management, conservation agriculture, and continuous vegetative cover (Notenbaert et al., Reference Notenbaert, Douxchamps, Villegas, Arango, Paul, Burkart, Rao, Kettle, Rudel, Vázquez and Teutscherova2021).
In addition, management practices that improve water availability at critical periods and increase soil organic matter are important. However, estimating medium and long-term forage productivity levels in East Africa is still an open problem with several challenges. First, there are only a few long-term field experiments that have a good record of forage management and herbage growth, as well as reported impacts on SOC (dos Santos et al., Reference dos Santos, Santos, Boote, Pequeno, Barioni, Cuadra and Hoogenboom2022; Kitonga, Reference Kitonga2019). Second, long-term monitoring of the performance of the improved forages is expensive and time-consuming. Thirdly, field experiments are site-specific and offer limited insight into the impacts of various management options. Process-based soil-crop models can be useful in overcoming these challenges because they can be used on broader time and spatial scales for making long-term projections and exploring additional management options that are typically not tested in experiments because of high costs and time associated with managing experiments (Tsuji et al., Reference Tsuji, Hoogenboom and Thornton1998).
Forage models, including the CSM-CROPGRO-Perennial Forage Model (PFM) of DSSAT, APSIM, and ALMANAC, have been used in Brazil, the USA, and Thailand (Bosi et al., Reference Bosi, Sentelhas, Pezzopane and Santos2020; Pequeno et al., Reference Pequeno, Pedreira, Boote, Alderman and Faria2018; Andrade et al., Reference Andrade, Santos, Pezzopane, De Araujo, Pedreira, Pedreira, Marin and Lara2016; Norsuwan et al., Reference Norsuwan, Attachai and Carsten2015; Kiniry et al., Reference Kiniry, Cassida, Hussey, Muir, Ocumpaugh, Read, Reed, Sanderson, Venuto and Williams2005). However, applying such models in new regions requires calibration and evaluation with regional data for accurate simulation under those environmental conditions, including weather and soil. The DSSAT CROPGRO-Perennial-Forage-Model (PFM) version 4.8.0 (Hoogenboom et al., Reference Hoogenboom, Porter, Shelia, Boote, Singh, Pavan, Oliveira, Moreno-Cadena, Ferreira, White, Lizaso, Pequeno, Kimball, Alderman, Thorp, Cuadra, Vianna, Villalobos, Batchelor, Asseng, Jones, Hopf, Dias, Hunt and Jones2023, Reference Hoogenboom, Porter, Boote, Shelia, Wilkens, Singh, White, Asseng, Lizaso, Moreno, Pavan and Boote2019; Jones et al., Reference Jones, Hoogenboom, Porter, Boote, Batchelor, Hunt, Wilkens, Singh, Gijsman and Ritchie2003) was chosen for this study due to its successful application in modelling perennial forages in tropical regions for various forage species. Some of the forage species that have been previously modelled with the CROPGRO-PFM include Urochloa brizantha cv. Piatã (Bosi et al., Reference Bosi, Sentelhas, Pezzopane and Santos2020), Urochloa brizantha cv. Marandu (Santos et al., Reference Santos, Boote, Faria and Hoogenboom2019; Pequeno et al., Reference Pequeno, Pedreira, Boote, Alderman and Faria2018; Pequeno et al., Reference Pequeno, Pedreira and Boote2014), Panicum maximum Jacq. cv. Tanzânia, Cynodon spp. ‘Tifton 85’ Bermuda grass (Lara et al., Reference Lara, Pedreira, Boote, Pedreira, Moreno and Alderman2012; Pedreira et al., Reference Pedreira, Pedreira, Boote, Lara and Alderman2011), and Bahia grass (Paspalum notatum) (Smith et al., Reference Smith, Wilson, Rymph, Santos and Boote2023). Despite the wide application of CROPGRO-PFM with different tropical grasses, the model has not been applied and evaluated for perennial forages in East Africa.
Using data from six demonstration plots in different locations in the Southern highlands of Tanzania, the objectives of this study were to: (i) evaluate the ability of the DSSAT CROPGRO-PFM version 4.8.0 to simulate herbage yields and SOC under Urochloa hybrid cv. Cayman and (ii) apply the evaluated model to assess herbage productivity and SOC under manure application rates of 0, 1, 5 and 10 t ha-1. Sensitivity simulation for the various manure rates depicts the potential productivity and SOC sequestration.
Materials and methods
Site description and data
The field trials were conducted in three districts in Southern highlands of Tanzania with two study sites per district: Kichiwa and Ikuna ward (Njombe district), Mtwango and Igowole ward (Mufindi district), and Kiwira and Lufingo ward (Rungwe district) (Supplementary Material Fig. S1). The study sites are in different agroecological zones with the annual average temperature ranging from 13 to 25 0C and the annual mean rainfall ranging from 1300 to 1700 mm (Table 1, Fig. S2). The districts have an altitude ranging from 1300 to 2050 m above sea level, with coordinates listed in Table 1. The soil types for these districts range from moderate to well drained. In addition, the three districts have a long history of dairy farming, thus making significant contributions to the GDP of Tanzania (Mteta Reference Mteta2023).
Description of the location, annual rainfall, average temperature, baseline soil data, and land use history for all the study sites
The dataset that was used in this study was collated in a Urochloa hybrid cv. Cayman trial that was established in January/February 2018 across the six sites and was replicated three times. Urochloa hybrid cv. Cayman is an important forage option for livestock systems in tropical regions. This forage variety performs well in soils with low fertility, is resistant to pests, and is tolerant to drought conditions. At the time of establishment, both manure and diammonium phosphate were applied at the rate of 0.93 Mg ha-1 and 18 kg N ha-1, respectively. In January 2020, after the initial harvest, urea fertilizer was applied at the rate of 199 kg N ha-1. Manual weeding was conducted each time after harvesting to allow the forage to regrow without weed competition. The forage was managed by cutting at maturity.
Baseline soil data were collected in January/February 2018 during the time of establishment of the experiment. Six composite samples were collected in each of the replicates at the depths of 0-20 cm and 20-50 cm with 3 samples for each depth using a soil auger. The samples were analysed at the International Institute of Tropical Agriculture laboratory in Dar es Salaam for SOC (Walkley-Black method), soil texture (hydrometer method), pH (water), and phosphorus (Olsen). Subsequent soil samplings in each replicate were conducted at depths of 0-20 cm and 20-50 cm in June/July 2019 and September/October 2020 and analysed for total soil C. Additional soil profile data needed for setting the soil water-holding characteristics were collected in January/February 2021 at 0 to 100 cm depth with incremental depths of 20 cm. The samples were analysed for texture, SOC, and bulk density (Supplementary Material Table S1).
Herbage samples were collected beginning in July 2018, with a total of twelve harvests at intervals of approximately 2 to 3 months when the height of the forages exceeded 30 cm. Prior to the first harvest, a ‘staging’ cut was done in May 2018 to allow all plots to have uniform starting conditions for regrowth. Biomass was harvested on a plot-by-plot basis for all plots across the sites, thus, a total of 10 harvests in Igowole, Mtwango, Ikuna, and Kichiwa and 12 harvests in Kiwira and Lufingo. For each plot, the outermost row was excluded, resulting in a net plot of 8 m2 for harvesting. The fresh weight of the biomass was measured using a spring balance. A sub-sample of approximately 500 grams from each plot was collected for dry matter (DM) determination. Fresh weights were recorded with a spring balance, oven-dried to a constant weight at 105 °C for 24 hours, and then reweighed to determine the DM content. This content was expressed as the proportion of dry weight to the original fresh weight. The obtained DM was then extrapolated to a per-hectare basis. The harvesting schedule was as follows: first - September 2018 for Kiwira and Lufingo, second - October 2018, third - January 2019, fourth - April 2019, fifth - August 2019, sixth - November 2019, seventh - January 2020, eighth - May 2020, ninth - September/October 2020, tenth harvest - January 2021, eleventh harvest - April 2021, and twelfth harvest - June/July 2021. The forages were harvested to a 5-cm stubble height; thus, the 5 cm was left above ground as stubble.
The CROPGRO-Perennial Forage Model
The Decision Support System for Agrotechnology Transfer (DSSAT) CROPGRO-PFM is a mechanistic model that is part of the DSSAT (version 4.8.0) software (Hoogenboom et al., Reference Hoogenboom, Porter, Shelia, Boote, Singh, Pavan, Oliveira, Moreno-Cadena, Ferreira, White, Lizaso, Pequeno, Kimball, Alderman, Thorp, Cuadra, Vianna, Villalobos, Batchelor, Asseng, Jones, Hopf, Dias, Hunt and Jones2023, Reference Hoogenboom, Porter, Boote, Shelia, Wilkens, Singh, White, Asseng, Lizaso, Moreno, Pavan and Boote2019). DSSAT broadly includes 42 crop models plus database tools for soil, weather, crop management, genetic information, observed experimental data, utilities, and application programs. The crop models simulate growth, development, and yield as a function of daily soil-plant-atmosphere dynamics.
Rymph (Reference Rymph2004) adapted the CROPGRO model to simulate the growth of tropical perennial grasses such as Bahia grass (Paspalum). The resulting CROPGRO-PFM model simulates rhizome and stolon growth, regrowth based on reserves, daylength-related dormancy, and accounts for freeze events that allow for progressive freeze damage. In addition, species and cultivar parameters affecting photosynthesis, partitioning of DM, carbon (C) and nitrogen (N) remobilization, growth, senescence, and plant phenology were calibrated to reflect Bahia grass growth and development based on literature values and parameter optimization (Smith et al., Reference Smith, Wilson, Rymph, Santos and Boote2023; Rymph, Reference Rymph2004). Separating parameter files from the main code makes CROPGRO a versatile model adaptable to new varieties and species upon estimating values for these parameters. These modules have been combined with other modules for crop, weather, and soil processes to form the Cropping Systems Model (CSM) (Jones et al., Reference Jones, Hoogenboom, Porter, Boote, Batchelor, Hunt, Wilkens, Singh, Gijsman and Ritchie2003).
Important adaptations and calibrations of the growth and development processes for other perennial grasses were based on data for Urochloa (Pequeno et al., Reference Pequeno, Pedreira and Boote2014; Pedreira et al., Reference Pedreira, Pedreira, Boote, Lara and Alderman2011) and Panicum species (Lara et al., Reference Lara, Pedreira, Boote, Pedreira, Moreno and Alderman2012).
CROPGRO-PFM simulates daily forage growth and productivity in response to soil, weather, and management. It simulates general biological processes such as photosynthesis and nitrogen (N) uptake using parameters unique to each species and cultivar in order to predict crop growth and development under a variety of conditions. The CROPGRO-PFM within the DSSAT system is linked to the CENTURY soil carbon module. The CENTURY module was adapted to a daily time-step in DSSAT for simulating SOC dynamics (Gijsman et al., Reference Gijsman, Hoogenboom, Parton and Kerridge2002; Parton et al., Reference Parton, Stewart and Cole1988). The CENTURY module includes three pools of organic matter: easily decomposable (SOC1), recalcitrant (SOC2), and almost inert ‘stable’ (SOC3), and allows for initialization of stable carbon pools.
The data required to run CROPGRO-PFM include soil surface and soil profile properties (i.e., clay and silt percentage, organic carbon, bulk density, and pH), daily weather data (i.e., solar radiation, maximum and minimum temperature, and rainfall), and crop management (i.e., crop species and cultivar, initial conditions (ICs), planting details, inorganic and organic fertilizer rates, and harvest details) (Hoogenboom et al., Reference Hoogenboom, Jones, Wilkens, Porter, Boote, Singh, Hunt and Tsuji2011). Weather data for the experimental period were obtained from the Climate Hazards Group InfraRed Precipitation with Station data at a resolution of 0.05 arc degrees (Funk et al., Reference Funk, Peterson, Landsfeld, Pedreros, Verdin, Shukla, Husak, Rowland, Harrison, Hoell and Michaelsen2015), and National Aeronautics and Space Administration (NASA) Langley Research Center (LaRC) Prediction Of Worldwide Energy Resources Project funded through the NASA Earth Science/Applied Science Program using the latitude and longitude Global Positioning System (GPS) coordinates (Fig. S4-S6). Satellite data were used after comparison with real-ground measurements because the measured weather data had many missing values. Soil texture was used to estimate the volumetric water content (lower limit, upper limit, and saturation) using the pedotransfer function in the SBuild programme in DSSAT because there were no real-ground measurements of water-holding attributes.
Model modification and evaluation
The CROPGRO-PFM version 4.8 developed for Urochloa brizantha cv. Marandu by Pequeno et al. (Reference Pequeno, Pedreira and Boote2014) was used for our study for a Urochloa hybrid cv. Cayman considering the two species are similar in terms of genetics and ecotype. There were no modifications of the species, ecotype, and cultivar parameters of the released Marandu adaptation, which was accepted as previously calibrated. However, modifications were made to the CENTURY soil carbon pool parameters, particularly to mimic the N mineralization needed to reproduce the yields of the modestly fertilized forages. Additionally, the soil water holding traits and rooting depth were adjusted.
The model was run using daily weather, soil, and management data. The weather data include daily rainfall, maximum temperature, minimum temperature, and solar radiation. The soil data includes texture, bulk density, SOC, saturated water content, drained upper limit, and lower limit of plant extractable water. The crop and management data were described in the management file and include location, ICs, fertilizer input, planting, and harvesting dates. The ICs require stable organic carbon fraction and nitrate and ammonium. The initial nitrate and ammonium were achieved from a spin-up simulation. The spin-up simulation was done as follows: we ran the model with the prior year of weather with a typical prior crop (maize or other crop) with its typical management that was recorded from the farms. We then take the ammonium and nitrate values for the respective soil layers from the SOILNI.OUT at the end of that simulation (same date as we start the PFM). We then put those into the ICs section of File X. The stable soil organic carbon (SOM3) for the DSSAT-Century module linked to CROPGRO-PFM model is not a truly-measurable quantity, but was hypothetically estimated by varying the SOM3 percentage within the range of 80 to 97 % of the total SOC, with the goal to achieve a reasonable N mineralization to support non-fertilized grass productivity. The high percentages of SOM3 are typical of inherently low soil fertility in tropical soils. Simulated herbage values were compared with observed values. The amount of stubble mass (MOW), leaf fraction (RSPLF), and stubble height after harvest were defined in the experimental file. Since there were no field measurements of the stubble mass, hypothetical values within the range of 1000, 1200, 1400, and 1600 kg ha-1 were evaluated in a sensitivity analysis to determine reasonable effect of the amount of stubble mass remaining in the field, which has small effects on productivity. This is relevant because the regrowth rate is partially dependent on residual mass and residual leaf area index (LAI). The final value accepted was 1200 kg ha-1 since the model produced realistic herbage results with this value. This value is also close to that used with other perennial forages in CROPGRO-PFM. With leaf fraction (RSPLF) equal to 0.20, this corresponds to approximate residual LAI of 0.11 to 0.42. The soil water holding traits and rooting depth were modified because they are important to herbage growth. The pedo-transfer functions in SBUILD of DSSAT often underestimate water-holding capacity of soils that have substantial clay (K.J. Boote, personal communication). In addition, the SBUILD by default gives the same shallow, exponentially-declining rooting shape that applies to all crops and soils. In reality, we know that perennial forages are deeply rooted. Harvested herbage mass over time and in relation to water deficit periods was the target while modifying soil water holding traits and fraction stable soil C (SOM3).
The model evaluation was based on the agreement between simulated and observed values for herbage and SOC. Statistical analysis used to assess how well the model simulated the observed data included means, root-mean-square error (RMSE), and the Willmott agreement index (d-statistic) (Willmott et al., Reference Willmott, Ackleson, Davis, Feddema, Klink, Legates, O’donnell and Rowe1985). The equation for RMSE is
$${\sqrt {{1 \over N}}\sum\nolimits_{i = 1}^N {{{\left( {{y_1} - {{\hat Y}_i}} \right)}^2}} } $$
where N is the total number of data points for comparison, Y
i is a given observed value, and
${\hat Y_i}$
is the corresponding value predicted by the model. A better model prediction will produce a smaller RMSE. The equation for Willmott agreement index (d-statistic) is:
$${{\sum\nolimits_{i = 1}^N {{{\left( {{Y_i} - {{\hat Y}_i}} \right)}^2}} } \over {\sum\nolimits_{i = 1}^N {\left( {{{\hat Y}_i} - {{\overline Y}_i}} \right)} \left| + \right|{{\left( {{Y_i} - {{\overline Y}_i}} \right)}^2}}}$$
where N is the number of observed data points, Y
i is a given observed value,
${\hat Y_i}$
is the corresponding value predicted by the model, and
${\overline Y_i}$
is the mean of the observed data. D-statistic values range from 0 to 1, with values near 1 indicating good model predictions.
Model application simulations
In order to verify cases where the model simulated low herbage values compared to observations during water-deficit periods, we conducted further sensitivity simulations. The evaluated model was set to automatic irrigation, which re-filled a given soil layer (in this case, 40 cm deep) when the remaining extractable available soil water in the 40 cm depth had fallen to 50% of available. This means that when soil moisture in the upper 40 cm depth dropped below 50%, the model triggered an irrigation event. This maximized the potential of herbage production that allowed for a comparison between typical systems with no irrigation (our study) and systems with adequate irrigation.
To evaluate the benefits of fertilization in a scenario analysis, the model was applied to assess the long-term impacts of different manure rates on herbage production and SOC from the year 1990 to 2021. Manure was applied at the rate of 0, 1, 5, and 10 t ha-1 from 1990 to 2021, once a year with a nitrogen concentration of 2.1 %. This range of manure rates was selected for sensitivity simulation to realize the potential herbage production and its consequence on SOC sequestration.
Results
Model performance in simulating herbage and SOC
The CROPGRO-PFM simulated the herbage production for Urochloa cv. Cayman reasonably well with a d-Stat. of 0.83 for Igowole, 0.74 for Mtwango, 0.85 for Ikuna, 0.73 for Kichiwa, 0.84 for Kiwira, and 0.82 for Lufingo. This shows that the model performed well except for Mtwango and Kichiwa. The RMSE was 1.6 t ha-1 for Igowole, 0.9 t ha-1 for Mtwango, 1.5 t ha-1 for Ikuna, 2.8 t ha-1 for Kichiwa, 1.3 t ha-1 for Kiwira, and 1.9 t ha-1 for Lufingo (Fig. 1). Across all the sites, Kichiwa had the highest observed mean periodic herbage yield of 3.1 t ha-1 compared to a simulated mean yield of 2.8 t ha-1; thus, the model underestimated the herbage for this site.
The model captured the temporal variability for all the sites, particularly for Igowole and Lufingo, where the simulated herbage matched the observed herbage, especially the higher peaks in simulated herbage coinciding with observed herbage values mostly during the wet season (Fig. 1, Fig. 2). The higher biomass across the sites in 2018 could be attributed to fewer cutting events compared to other years as well as initial soil surface residue and inorganic nitrogen in the soil profile. During the months of June to November, when low rainfall was recorded, the model simulated low values of herbage, which matched with the periods where the simulated water stress was high. However, in other cases, the model simulated lower herbage compared to the observed herbage, and in rare cases, the model simulated no herbage harvest at all in the month of November 2019 for Ikuna, Kichiwa, Igowole, and Mtwango, despite the data indicating that there was some harvested herbage (Fig. 2). To understand these results better, additional simulation outputs were evaluated for the factors that affect herbage growth. The analysis showed that these low-yielding points were due to the drought stress for photosynthesis that was simulated by the model (see Fig. S3 for details of simulated water stress). The occurrence of reduced herbage matched with the simulated occurrence of water stress, and this was captured correctly by the model. For some sites, for example, Kichiwa and Lufingo, the effects of drought stress on herbage were more evident in the observed data than the drought stress effects on simulated herbage. In Igowole, Kichiwa, Kiwira, and Lufingo, the reduction in simulated herbage production was attributed to water stress. However, for all the sites, drought stress was evident, which is typical of the dry periods for this climatic region.
Observed versus simulated herbage (t ha-1) for (a) Igowole, (b) Mtwango, (c) Ikuna, (d) Kichiwa, (e) Kiwira, and (d) Lufingo. The number of represented harvests is 12 for Kiwira and Lufingo, and 10 for Igowole, Mtwango, Ikuna, and Kichiwa. The dashed line is the 1:1 line and the grey line is the regression line.
Simulated and observed herbage for (a) Igowole, (b) Mtwango, (c) Ikuna, (d) Kichiwa, (e) Kiwira, and (f) Lufingo from 2018 to 2021. Blue points represent the observed herbage and the bar is the standard deviation over the three replicates, while the black line and points represent the simulated herbage.
There was large variability in the observed SOC, and as a result the CROPGRO-PFM did not appear to simulate the SOC well in comparison with the observed data (Fig. 3). The simulated SOC between subsequent sample dates was anticipated to vary slowly, and there should be no expectation that the model can reproduce the sudden increases or decreases in observed SOC that occurred in less than 3 years, which is mostly due to the variability of the soil sampling. The d-Statistic ranged from 0.19 to 0.35 and 0.40 to 0.53 for the 0-20 cm and 20-50 cm soil depths, respectively. According to the d-Statistic, the model simulated SOC better at 20–50 cm depth compared to the top 0–20 cm depth. In general, no significant trends in observed or simulated SOC were evident, considering the large variability of observed SOC (Fig. 3).
Time series soil organic carbon (SOC) (g C kg-1) for (a) Igowole, (b) Mtwango, (c) Ikuna, (d) Kichiwa, (e) Kiwira, and (f) Lufingo from 2018 to 2021 for the 0–20 cm soil depth. Blue points and error bars are observed data, and black line is model-simulated beginning at the initial conditions of first date.
Response of herbage to automatic irrigation
To evaluate the impact of drought stress on simulated herbage production, we used the automatic irrigation option of the model. In addition to understanding the effects of water or drought stress, the model was used to simulate the application of automatic irrigation under the assumption that water resources and irrigation technology are available or can be made available in the future. The DSSAT has an automatic irrigation scheduling tool that schedules irrigation when the extractable soil moisture in the top given depth (here 40 cm) of the soil profile drops below a set threshold value (here 50%). The automatic irrigation increased herbage production (Fig. 4), and the model did not simulate any drought stress. For the case where the model was simulating low or zero herbage under no irrigation, especially during the month of November 2019, automatic irrigation greatly increased herbage production. Thus, we can confirm that the low herbage production can be attributed to a small amount of rainfall. Igowole showed the greatest increase in mean herbage at 3.16 t ha-1, while Kiwira had the lowest increase in mean herbage at 1.13 t ha-1. Thus, at optimum irrigation, herbage production is higher. However, there is variability of herbage between years and sites. The sites have different soil types and climate conditions, thus allowing for yearly differences in responses to irrigation. Years have varied rainfall events, thus causing differences in benefits of irrigation events. Irrigation had the most impact in drier sites, with greater increase in biomass reported compared to sites where there was more rainfall.
Simulated herbage production (t/ha) under no irrigation and automatic irrigation for (a) Igowole, (b) Mtwango, (c) Ikuna, (d) Kichiwa, (e) Kiwira, and (f) Lufingo from 2018 to 2021. The black line indicates simulated total biomass under no irrigation, black point shows observed herbage and blue line indicates simulated total biomass under automatic irrigation.
Impact of long-term simulations on soil organic carbon stocks
There was SOC loss over a 30-year period for all the sites when no manure was applied (0 t ha-1). In addition, for Kiwira and Lufingo, there was a decline in SOC stocks, for low application rates of 0, 1 and 5 t ha-1 of manure (Fig. 5). The decline in SOC stocks was simulated for those sites that had initial high SOC stocks (Fig. 5). The model simulated SOC sequestration at all sites with manure application rates of 10 t ha-1 for the 30-year period. Except for two sites, the SOC stocks generally increased as the manure rates increased (Fig. 5) because manuring increased herbage production and carbon is contributed with the manure itself. The annual simulated SOC sequestration rate ranged from 0 to 0.7 Mg C ha-1 yr-1, with the highest sequestration in Igowole. The model projected an annual SOC loss of 0.1 to 0.4 Mg C ha-1 yr-1 for manure rate at 0 t ha-1 and 0.1 to 0.3 Mg C ha-1 yr-1 for manure rate at 1 t ha-1 (Fig. 5). It is important to note that for forages, herbage harvests are removed, so there is less return of crop residue.
Simulated long-term soil organic carbon (SOC) stocks for a) Igowole, (b) Mtwango, (c) Ikuna, (d) Kichiwa, (e) Kiwira, and (f) Lufingo as affected by the rate of manure that was applied once a year.
Benefits of manuring on herbage production
The increased rate of manuring increased herbage production as well as SOC stocks (Fig. 6). Response of herbage production to manure at 1 t ha-1 was relatively modest at 16% compared to no manuring. Herbage increases under 5 and 10 t ha-1 of manure were considerably greater up to 70 and 135%, respectively (Fig. 6). Part of the productivity response to long-term manuring could be attributed to the increase in SOC and soil quality over time in the 30-year simulations which supports the value of continued addition of manure or mineral N fertilizer to sustain soil quality and improve forage productivity.
The simulated change in soil organic carbon (SOC) (A) and yield (B) versus different manure application rates (1, 5 and 10 t ha-1) compared to the zero application of manure that was applied once a year, averaged for 30 years, for each study site.
Discussion
Simulation of herbage production and soil organic carbon
The model was successful in simulating herbage production and SOC time-course in Tanzania after calibration of the relevant local soils-related parameters, soil water-holding traits, and stable SOC pools, while the genetic coefficients were not modified. The justification for not altering genetic coefficients is that proper calibration for a new species/cultivar requires intensive time-series data on growth dynamics under conditions not limited by N fertility and water deficit. Our prior experience with non-legumes is that the N supplying status and soil water dynamics of the soil must be looked at first to avoid falsely modifying genetic factors to account for nitrogen stress or water deficit. The model with crop parameters previously calibrated for Marandu (Urochloa) in Brazil, predicted adequately the herbage production in Southern Tanzania. However, the model performed better for some sites compared to other sites, and this may be attributed to the different soils for these sites and the agroecological zones. Comparing model performance in Brazil and Tanzania, the model calibrated for Brazilian conditions showed a higher D-statistic up to 0.96 (Pequeno et al., Reference Pequeno, Pedreira and Boote2014), and this may be attributed to more frequently measured time-course data used for calibration. Furthermore, those growth parameters were calibrated to data that were obtained for good N fertilization conditions. By contrast, the N fertilization was not optimum for the conditions in Tanzania and the soil water and N-supplying capacities were relatively unknown. Nevertheless, the model predicted well the underlying factors that affect forage growth, such as drought and nitrogen stress. According to Santos et al. (Reference Santos, Boote, Faria and Hoogenboom2019), Pequeno et al. (Reference Pequeno, Pedreira, Boote, Alderman and Faria2018), Pequeno et al. (Reference Pequeno, Pedreira and Boote2014), and Pedreira et al. (Reference Pedreira, Pedreira, Boote, Lara and Alderman2011), occasions of reduced biomass growth found in their experiments could be attributed to nitrogen stress. A study by Santos et al. (Reference Santos, Boote, Faria and Hoogenboom2019) showed that nitrogen stress lowered the photosynthetic rate and ultimately the regrowth after harvest. However, our study was limited by relatively less frequent growth analysis data that did not allow for calibrating the species-related growth parameters. In this study, we only modified the soil water holding characteristics and SOM3 because water availability and N supply are critical to herbage growth.
The model performance in simulating SOC appeared to be acceptable, but accuracy statements are not warranted or useful in view of the high variability of SOC measurements and the short span of years for which observed data were collected (Fig. 3). True observed SOC increase as well as model simulated increase over time will require at least 5 to 10 years to determine a detectable difference (Don et al., Reference Don, Schumacher and Freibauer2011).
Although the model was simulated for a different cultivar than the one we used, the model performed well. A comparison is needed between the improved forage cultivars and the current forages that farmers typically grow. This would help contribute to making better recommendations to stakeholders.
Annual herbage production response to full irrigation
Herbage availability in the dry season is crucial to maintain high milk and yield production. The presented results show that the yields in the study sites can be improved through irrigation where water is available. The high increase in herbage production of up to 102 % in Igowole highlights the potential productivity in drought-stressed sites. However, as expected, the response to irrigation varies depending on the type of soil. These scenarios indicate that the application of water during drought conditions provides significant increases in herbage productivity, but detailed experimental data on soil characterization are critical for further research. Similar studies indicate an increase in herbage under full irrigation conditions for forages (Lopes et al., Reference Lopes, Torres, Fanaya Júnior, Silva Neto, Margatto and Kraeski2016; Pequeno, Reference Pequeno2014). Root senescence under severe drought may also be a cause for less simulated soil water extraction or delayed recovery, which also requires further research. Previous studies have recommended that the root senescence response in the model needs to be improved, especially in extreme drought conditions, where the recovery is quite slow (Santos et al., Reference Santos, Boote, Faria and Hoogenboom2019; Boote et al., Reference Boote, Sau, Hoogenboom, Jones, Ahuja, Reddy, Saseendran and Yu2008). Evaluating the model response to automatic irrigation provided knowledge concerning the reduction in yield caused by water stress, and thus has the potential to be used by decision makers with knowledge on how to maximize yield while optimizing available water, especially under unpredictable weather conditions.
Simulation of long-term soil organic carbon for different manure application rates
Soil organic carbon is important to soil health and crop productivity. The simulated long-term SOC stocks give an indication of changes in SOC stocks under different weather and manuring conditions. The addition of manure increased SOC stocks rather consistently across the different wards in these simulations and to some extent reached a point of equilibrium. The increase in SOC stocks could be attributed to the concurrent increase in herbage production and root biomass, which was driven by the N release from the manure applied. An additional component of the SOC increase could be from the residual C from the applied manure. The findings are similar to studies that have been conducted in the tropics, such as western Kenya and Colombia, where the application of manure reduced SOC losses. However, this was mainly affected by the different land use history in terms of tillage and inputs applied (Nyawira et al., Reference Nyawira, Hartman, Nguyen, Margenot, Kihara, Paul, Williams, Bolo and Sommer2021; Sommer et al., Reference Sommer, Paul, Mukalama and Kihara2018). Generally improved forages have been reported to increase SOC stocks with sequestration values ranging from 0.3 to 3.5 Mg C ha-1 yr-1 (Dondini et al., Reference Dondini, Martin, De Camillis, Uwizeye, Soussana, Robinson and Steinfeld2023; Conant et al., Reference Conant, Cerri, Osborne and Paustian2017). For SOC to continue to increase, forage cultivation may need to be complemented with management such as application of mineral fertilizer, if manuring is insufficiently available. In a study that used the DayCent model, losses of SOC stocks were avoided and SOC increased by up to 0.22 Mg C ha-1 yr-1 with application of manure, reduced tillage, and fertilizer addition, a phenomenon attributed to reduced decomposition of the carbon pool (Nyawira et al., Reference Nyawira, Hartman, Nguyen, Margenot, Kihara, Paul, Williams, Bolo and Sommer2021). In Colombia, Urochloa (Syn. Brachiaria) humidicola was reported to increase SOC stocks up to 2.0 Mg C ha-1 yr-1 due to an increase in yield and root biomass (Costa et al., Reference Costa, Villegas, Bastidas, Matiz-Rubio, Rao and Arango2022). An increase in SOC stocks for specific sites indicates potential of these systems for soil carbon sequestration and, if well managed, enhanced soil organic matter will lead to increased forage productivity and consequently improved soil health. Thus, long-term application of manure has huge potential for increasing herbage production and SOC stocks, especially for sites that have an initial low SOC. However, land-use history in terms of tillage and inputs applied remains a critical factor in determining the impact of different management practices.
Conclusion
The findings of this study demonstrate the potential applicability of the CROPGRO-PFM for simulating herbage production and SOC dynamics in Southern Tanzania. The CROPGRO-PFM model has good potential to evaluate management and climatic impacts on long-term SOC, soil health, and forage production that are not easily assessed with short-term experiments. Despite the challenges in capturing the observed variability in biomass production, the model captured key factors influencing forage growth and responded effectively to irrigation interventions and long-term manure application. These results underscore the importance of tailored calibration efforts to enhance the model’s accuracy and applicability within East Africa. In addition, continued research is crucial for evaluating optimal irrigation strategies, addressing model limitations, and exploring the broader implications of large-scale adoption of improved forages, such as a Urochloa hybrid cv. Cayman, alongside accompanying agronomic management practices on agricultural sustainability and soil health in tropical agroecosystems.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0014479725100161
Acknowledgements
The data used in the modelling work were collected under auspices of the International Fund for Agricultural Development funded project on ‘Climate-smart Dairy systems in East Africa through Improved Forage and Feeding Strategies: Enhancing Productivity and Adaptive Capacity while Mitigating Greenhouse Gas Emissions’ implemented in Tanzania and Rwanda. The modelling work and preparation of the manuscript were done through the CGIAR Initiative on Livestock and Climate. We thank all donors who globally support our work through their contributions to the CGIAR System.
Author contributions
Mercy Korir contributed to formal analysis, investigation, methodology, and writing. Sylvia Nyawira, Ken Boote, and Gerrit Hoogenboom contributed to conceptualization, investigation, methodology, supervision, and writing. Leonardo Ordonez contributed to methodology and validation. Birthe Paul contributed to conceptualization. An Notenbaert contributed to funding acquisition, supervision, and writing.
Funding statement
Financial support was provided by International Centre for Tropical Agriculture.
Competing interests
The authors declare none.
Open access
