Efficient Nutrient Management Practices for Sustainable Crop Productivity and Soil Fertility Maintenance Based on Permanent Manorial Experiments in Different Soil and Agro-Climatic Conditions

millet The was rotated Nine treatments tested in net 3.6 and row 45 for both The treatments tested were (i) Control; (ii) 40 + kg P/ha; (iii) 20 kg N (urea) + 10 kg P/ha; (iv) 20 kg N (crop residue)/ha; (v) 20 kg N (FYM)/ha; (vi) 20 kg N (crop residue) + 20 kg N (urea)/ha; (vii) 10 kg N (FYM) + 10 kg N (urea)/ha; (viii) 40 kg N (urea) + 20 kg P + kg ZnSO 4 /ha; and (ix) @ 5 t/ha. The crop residue contained 1.2% N, while FYM contained 0.5% significant only in FYM @ 10 t/ha, FYM @ 10 t/ha + 50% NPK and 100% NPK treatments. There was a decrease of soil K over years, in all treatments, but the decrease was significant only in FYM @ 10 t/ha ha + 100% NPK application. Based on the predictability of changes in soil nutrient status over years (R 2 ), the prediction (%) of yield ranged from 1 to 26% for soil N; 2 to 44% for soil P; and 1 to 26% for soil K for different treatments. The standard error based on a regression model ranged from 9.3 to 23.1 kg/ha for soil N; 4.5 to 12.3 kg/ha for soil P; and 11.4 to 19.2 kg/ha for soil K over years. The trends of changes in yield, and soil nutrients as affected by treatments over years indicated that in general, the soil P tended to increase, while soil N reflected the decreasing tendency over years. However, soil K decreased over years. Thus, the trends of soil fertility changes were similar for soil N and P nutrients (except in control), while it was the opposite trend for soil K over years. relationship with soil N in control, FYM @ 10 t/ha + 100% NPK and 100% NPK; soil K in all treatments except 100% NPK. The crop seasonal rainfall had a negative effect on finger millet yield in control and 100% NPK. The crop growing period had a positive correlation with grain yield attained by all treatments except FYM @ 10 t/ha and FYM @ 10 t/ha + 50% NPK. Among different treatments, the negative correlation of yield (in all treatments), soil N and soil K with time period indicated a decrease, while a positive correlation of soil P with time period indicated an increase with fertilizer application. yield attained by all treatments except control, while soil P had a negative effect on yield attained by all treatments. Soil K had a positive effect on yield of 50 kg N + 25 kg/ha, 25 kg N + 12.5 kg P/ha, 25 kg N/ha ( Leucaena ) and kg N/ha (FYM). standard error based on regression models was in the range of 315 kg/ha for control and Farmers practice (FYM @ 5 t/ha) to 397 kg/ha for 100% NPK (20–40–40 kg/ha). Among the effects of rainfall on yield, the rainfall received in July, September, October and November had a positive influence, while rainfall received in August had negative influence on the yield attained by all treatments. July rainfall had significant influence on yield attained by all treatments except 100% N (groundnut shells ~ 20 kg N/ha) and 100% N (groundnut shells ~ 20 kg N/ha) + 50% NPK (10–20–20 kg/ha); while November rainfall had significant influence on yield attained by 100% NPK (20–40–40 kg/ha), 50% NPK (10–20–20 kg/ha), 50% N (FYM ~ 10 kg N/ha) + 50% NPK (10–20–20 kg/ha), 100% NPK (20–40–40 kg/ha) + ZnSO 4 @ 25 kg/ha and Farmers practice (FYM @ 5 t/ha). The rate of change in yield of all treatments was positive and maximum for an unit change in November rainfall, followed by July, October and September, while it was negative for August. The analysis indicated that the model of farmers practice (FYM @ 5 t/ha) had maximum R 2 and minimum standard error, while the models of 100% NPK (20–40–40 kg/ha) and 100% N (groundnut shells ~ 20 kg N/ha) + 50% NPK (10–20–20 kg/ha) had minimum R 2 and maximum standard error (Table 14). the results obtained from this long term study incurring huge expenditure provide very good conjunctive nutrient use options with good conformity for different based on the study. The maximum pod yield was 1546 kg/ha under < 500 mm; 1541 kg/ha under 500-750 mm; 1329 kg/ha under 750-1000 mm rainfall situation. The SYI ranged from 26.6 for control to 33.6% for T6 under < 500 mm; 27.4% for control to 39.1% for T5 under 500-750 mm; 49.0 for control to 71.2% for T8 under 750-1000 mm rainfall situation.


Introduction
Rainfed agriculture plays an important role in contributing to the food bowl of the world. Its importance varies regionally but produces food for poor communities in developing countries. In India, rainfed agriculture is in about 85 million hectare, constituting about 60 % of net cultivated areas supporting 40% of the population of the country. In Sub-Saharan Africa, more than 95 % farm land is rainfed, while the corresponding figure for Latin America is almost 90%, for South Asia about 60 %, for East Asia 65% and for the Near East and North Africa 75% (Wani, et al., 2009). Besides, the climatic constraints especially erratic and uncertain pattern of rainfall, soils in the rainfed areas are under severe grip of degradation in terms of their physical, chemical and biological properties.
In AICRPDA, permanent manorial experiments (PME) are conducted on different rainfed crops viz., upland rice, sorghum, finger millet, pearl millet, cotton, maize, soybean, groundnut crops under varying soil and agro-climatic conditions at different centers. They are conducted in alfisols, vertisols, inceptisols, entisols, aridisols and other soil types. Alfisols are most abundant soils in the semi-arid tropics and cover about 16% of tropics and 33% of semi-arid tropics (SAT). These soils are mostly found in the south Asia, west and central Africa, and many parts of South America, particularly north eastern Brazil (Cocheme and Franquin 1967). Mostly these soils are shallow with a compacted sub-surface layer that limits the root development and water percolation. The loamy sand texture of top soil and abundance of 1:1 type clay minerals viz., kaolinite, make them structurally inert (Charreau 1977). These soils are constrained by crusting and hard setting tendencies under erratic rainfall distribution and occurrence of dry spells (Bansal, Awadhwal, and Mayande 1987). Owing to less contribution of root biomass due to low crop intensity, high temperature mediated fast oxidation of organic matter, poor recycling back of crop residues, washing away of top soil, reckless tillage and imbalanced fertilizer use results in low organic carbon and low fertility of these soils (Kampen and Burford 1980;El-Swaify, Singh, and Pathak 1983). Often these soils encounter a diversity of soil physical, chemical and biological constraints and provide a low productivity of crops.
Vertisols are the predominant soil groups found across the world. The majority of the acreage of Vertisols and associated soils in the world is spread in Australia (70.5 million ha), India (70 million ha), Sudan (40 million ha), Chad (16.5 million ha), and Ethiopia (10 million ha). These five countries constitute over 80% of the total area (250 million ha) of Vertisols in the world (Dudal 1965). In India, substantial Vertisol areas are found in the states of Maharashtra, Madhya Pradesh, Gujarat, Andhra Pradesh, Karnataka, and Tamil Nadu (Murthy 1981). Most of these regions receive 500 to 1300 mm of annual rainfall, concentrated in a short period of 3 to 3½ rainy months interspersed with droughts. Crop yields in these areas are miserably low and may vary from year to year. Virmani et al., (1989) have comprehensively characterized Vertisols found in India. Their texture may vary from clay to clay loam, or silty clay loam, with the clay content generally varying from 40% to 60% or more. They have high bulk density when dry, with clod density values ranging from 1.5 to 1.8 g cm -3 ); high CEC (47 to 65 cmol kg soil -1 ); and pH values usually above 7.5. Tropical Vertisols are low in organic matter and available plant nutrients, particularly N, P, and Zinc. The dominant clay mineral is smectite. High clay content, better effective soil depth associated with other physical properties makes these soils to store higher amount of moisture. Low organic matter status accompanied by poor soil fertility is one of the predominant constraints in these Vertisol soils. Farmers of the rainfed SAT regions, being poor, are not able to use adequate amount of chemical fertilizers. Earlier researchers have established that the productivity of these soils can be enhanced by way of supplying adequate nutrient inputs (Virmani et al., 1989;Willey et al., 1989;Burford et al., 1989). Based on numerous agronomic experiments, it has been found that supplementation of N, P and Zinc through fertilizer is inevitable to ensure satisfactory crop production in SAT soils especially in Vertisols (Kanwar 1972;Randhawa and Tandon 1982). Despite many efforts, there is a low adoption of fertilizers in rainfed crops which could probably be attributed to many reasons viz., incapability of the farmers to purchase fertilizers, erratic and uncertain rainfall leading to risk of crop failures, uncertainty and variability in crop responses (Jha and Sarin, 1984;Kanwar et al., 1973).
The productivity of any rainfed crop is significantly influenced by the distribution of seasonal rainfall during cropping season, soil fertility status and amount of fertilizer nutrient applied . Research studies have shown that among different variables, the quantity of rainfall received during crop growing period would significantly influence the response of a crop to fertilizer application under rainfed conditions (Behera et al., 2007;Mohanty et al., 2008). Vikas et al., (2007) while optimizing the fertilizer requirement of rainfed maize in a dry sub-humid Inceptisol at Jammu in north India opined that if fertilizer doses are judiciously optimized considering the rainfall distribution pattern during the cropping season, higher productivity could be achieved in rainfed crops. Nema et al., (2008) examined the effects of crop seasonal rainfall and soil moisture availability at different days after sowing on yield and identified suitable tillage and fertilizer practices for attaining sustainable pearl millet yield in a semi-arid Inceptisol at Agra in north India. Further, to attain sustainable yield of crops in any soil and agro-climatic conditions and to save on fertilizers, it is important that while optimizing the fertilizer doses, changes in soil fertility also need to be periodically monitored (Maruthi Sankar 1986;Vittal et al., 2003). Long term effects of fertilizer on crop yield and soil properties have also been examined for different crops in order to suitably restore soil fertility and prescribe soil test based fertilizer recommendation for different crops (Prasad and Goswami 1992;Bhat et al., 1991;Dalal and Mayer 1986;Mathur 1997). Permanent manorial experiments are conducted at different research centers of AICRPDA with an objective to i) assess the response of rainfed crops and changes in soil fertility (with special emphasis on nitrogen, phosphorus, and potassium) due to long term application of organic and inorganic sources of nutrients under changing crop seasonal rainfall situations and ii) identify an efficient treatment for attaining sustainable yield over long-term basis under different soils and climatic conditions. The details of PMEs conducted on (i) finger millet at Bangalore; (ii) sorghum/pearl millet rotation at Kovilpatti; (iii) groundnut at Anantapur; (iv) cotton + green gram (1:1) at Akola; (v) soybean at Indore are discussed in this paper.

Experimental details
The permanent manorial experiments (PME) have been conducted on different crops with a set of organic and inorganic fertilizer treatments at different AICRPDA research centers for more than 20-25 years (Table 1). The PMEs were conducted on (i) finger millet at Bangalore (Karnataka); (ii) groundnut at Anantapur (Andhra Pradesh); (iii) soybean at Indore (Madhya Pradesh); (iv) cotton + green gram in 1:1 row ratio at Akola (Maharastra); (v) sorghum rotated with pearl millet (yearly) at Kovilpatti (Tamil Nadu); (vi) rice at Phulbani (Orissa); (vii) rice at Varanasi (Uttar Pradesh); (viii) rice at Ranchi (Jharkhand); (ix) pearl millet at Agra (Uttar Pradesh); (x) rabi sorghum at Solapur (Maharastra); (xi) rabi sorghum at Bijapur (Karnataka); (xii) pearl millet/castor/cluster bean rotation (yearly) at SK Nagar (Gujarat); (xiii) rice in kharif followed by wheat in rabi at Rewa (Madhya Pradesh). The treatments were replicated thrice and tested in a Randomized Block Design. The treatments were randomized only in the first year and were fixed and superimposed to the same plots every year. Before superimposing the fertilizer treatments, initial soil samples were collected from each plot at a soil depth of 0-30 cm and analyzed for soil organic carbon (Walkley and Black, 1934), available (easily oxidizable) N (Subbaiah and Asija, 1956), available P (Olsen et al., 1954) and available K (Jackson, 1973). Soil sulphur was estimated by turbidity method (Chesnin and Yien, 1950) in each season. Observations on daily rainfall, variety, date of sowing and harvest, crop growing period, length of dry spells and other related details were also recorded every year and used for analysis. Farmers practice (FYM @ 5 t/ha). The FYM contained 0.5% N, 1% P and 0.75% K on dry weight basis. All the improved agronomic practices prescribed for groundnut were adopted while conducting the trials.

Statistical analysis
The differences in effects of treatments in influencing soil fertility of N, P and K nutrients and crop yield were tested based on the standard Analysis of Variance (ANOVA) procedure. The treatments with a significantly higher effect on soil nutrients and yield were identified based on Least Significant Difference (LSD) criteria (Gomez and Gomez, 1984). Based on correlation coefficients measured between pairs of variables, the type (positive or negative) and extent of relation between yield, crop seasonal rainfall, and soil N, P and K nutrients were assessed for each treatment over years. Regression models of yield attained by each treatment were calibrated for assessing the influence of crop seasonal rainfall, soil N, P and K nutrients on yield of a crop over years as suggested by Draper and Smith (1998). The regression model through crop seasonal rainfall, soil N, P and K could be postulated as In model (1), α is intercept and β1 to β9 are regression coefficients measuring effects of variables on yield. The variables of monthly rainfall are retained depending on the dates of sowing and harvest and crop growing period. Soil sulphur was also included in the model calibrated for soybean at Indore. The usefulness of a regression model for yield prediction could be assessed based on the coefficient of determination (R 2 ) and unexplained variation measured by the prediction error. The sustainability yield index (SYI) of a fertilizer treatment could be derived as a ratio of the 'difference between mean yield and prediction error' and 'maximum mean yield' attained by any treatment in the study period (Behera et al., 2007;Nema et al., 2008;Maruthi Sankar et al., 2011, 2012a, 2012b. At Kovilpatti, observations were recorded on daily rainfall (mm) and Pan Evaporation (EP, in mm) during 1987 to 2005. Accordingly, the daily soil water balance computational procedure of Rijtema and Aboukhaled (1975) was used to calculate the Water Requirement (WR, mm), Potential Evapotranspiration (PET, mm) and Actual Evapotranspiration (AET, mm) for sorghum and pearl millet. The Crop Water Stress (CWS) was estimated by using the procedure as discussed by Hiler and Clark (1971). The crop coefficient values were determined by interpolating the values given by Doorenbos and Kassam (1979). The CWS ranged from 0. 1% in 1987, 1993 and 1997 to 60.5% in 1995 with mean of 15.6% and variation of 119.9% for sorghum. In pearl millet, it ranged from 0.1% in 1996 to 71.5% in 1994 with mean of 31.7% and variation of 71.8% for pearl millet.
At Kovilpatti, the treatment-wise regression models of yield were developed using different variables of soil N, P, and K, crop seasonal rainfall, crop growing period, crop water stress measured under each treatment (Maruthi Sankar, 1986). The regression model of yield could be postulated as In (2), α is intercept and β1 to β6 are regression coefficients of variables considered in the model.

Semi-arid alfisols at Bangalore
At Bangalore, the earliest date of sowing of finger millet was on 14 Four crop seasonal rainfall situations viz., < 500, 500-750, 750-1000 and 1000-1250 mm were observed during 1984 to 2008. The crop seasonal rainfall was < 500 mm in 3 years, 500-750 mm in 11 years, 750-1000 mm in 8 years and 1000-1250 mm in 3 years. June received a mean rainfall of 81 mm with a variation of 77.4%; while July received 98 mm with variation of 59.1%. August received a mean rainfall of 139 mm with a variation of 61.2%, while September received a mean rainfall of 200 mm with variation of 50.3%. October received a mean rainfall of 188 mm with a variation of 66.7%, while November received 50 mm with variation of 95.5% over the 25 years of study. The mean rainfall in a month increased from < 500 mm to 1000-1250 mm crop seasonal rainfall group. Under < 500 mm crop seasonal rainfall situation occurred for 3 years (1990, 2002 and 2006), the mean monthly rainfall ranged from 54 mm with a variation of 51.1% in July to 105 mm with a variation of 62.9% in October. Under 500-750 mm crop seasonal rainfall situation for 11 years (1984, 1985, 1986, 1987, 1989, 1994, 1995, 1996, 2001, 2003 and 2007), the mean monthly rainfall ranged from 38 mm with a variation of 83.2% in November to 199 mm with a variation of 47.7% in September. Under 750-1000 mm crop seasonal rainfall situation for 8 years (1988, 1992, 1993, 1997, 1999, 2000, 2004 and 2008), the mean monthly rainfall ranged from 51 mm with a variation of 125.2% in November to 263 mm with a variation of 31.0% in September. Under 1000-1250 mm crop seasonal rainfall situation for 3 years (1991, 1998 and 2005), the mean monthly rainfall ranged from 77 mm with a variation of 84.8% in November to 435 mm with a variation of 38.6% in October. The mean crop growing period was 121 days with variation of 17.9% under < 500 mm; 131 days with variation of 8.8% under 500-750 mm rainfall; 122 days with variation of 7.1% under 750-1000 mm rainfall; and 125 days with variation of 2.9% under 1000-1250 mm rainfall situation. The details of crop growing period, rainfall, date of sowing and harvest of finger millet under different crop seasonal rainfall situations during 1984 to 2008 are given in Table 2.

ANOVA of soil test values and yield in different seasons
The mean and coefficient of variation of soil fertility of nutrients and yield of crops attained under each rainfall situation at Bangalore, Akola, Kovilpatti and Indore are given in Table 3.  At Bangalore, the changes in soil N, P and K nutrients over years were assessed. The trends indicated that the soil N decreased in all treatments, however, the decrease was significant only in control. There was a build-up of soil P in all treatments, but, the increase was significant only in FYM @ 10 t/ha, FYM @ 10 t/ha + 50% NPK and 100% NPK treatments. There was a decrease of soil K over years, in all treatments, but the decrease was significant only in FYM @ 10 t/ha ha + 100% NPK application. Based on the predictability of changes in soil nutrient status over years (R 2 ), the prediction (%) of yield ranged from 1 to 26% for soil N; 2 to 44% for soil P; and 1 to 26% for soil K for different treatments. The standard error based on a regression model ranged from 9.3 to 23.1 kg/ha for soil N; 4.5 to 12.3 kg/ha for soil P; and 11.4 to 19.2 kg/ha for soil K over years. The trends of changes in yield, and soil nutrients as affected by treatments over years indicated that in general, the soil P tended to increase, while soil N reflected the decreasing tendency over years. However, soil K decreased over years. Thus, the trends of soil fertility changes were similar for soil N and P nutrients (except in control), while it was the opposite trend for soil K over years.

Semi-arid vertisols at Akola
At Akola, the ANOVA indicated that fertilizer treatments differed significantly in influencing soil fertility of nutrients and yield in all years. They were also significantly different when pooled over years under each rainfall situation (Gomez and Gomez, 1985). A minimum mean yield of 360 kg/ha (variation of 43.7%) and 492 kg/ha (variation of 52.5%) was attained under control, while a maximum of 527 kg/ha (variation of 33.9%) and 807 kg/ha (variation of 56.1%) was attained under 25 kg N (FYM) + 25 kg N (urea) + 25 kg P/ha in case of green gram and cotton respectively. Application of 25 kg N (FYM) + 25 kg N (urea) + 25 kg P/ha was also superior for enhancing soil fertility status by providing a maximum mean soil N of 251.8 kg/ha (variation of 22.8%), soil P of 33.5 kg/ha (variation of 19.5%) and soil K of 368.6 kg/ha (variation of 20.4%) over years. A build-up of soil N and a depletion of soil K were observed under all treatments over years. A build-up of soil P was observed under control and 25 kg N + 12.5 kg P/ha, while there was a depletion under all the remaining treatments. Comparison of pairs of treatments for differences in yield, soil N, P and K nutrients indicated that 25 kg N (FYM) + 25 kg N (urea) + 25 kg P/ha was superior to all other treatments by attaining a significantly higher yield of cotton and green gram and maintaining higher soil fertility status over years.

Semi-arid vertic inceptisols at Kovilpatti
At Kovilpatti, the ANOVA of sorghum data indicated a significant difference among treatments in individual years and also when pooled over years in influencing soil nutrients and grain yield. .8% under FYM @ 5 t/ha. A higher mean soil N was observed in sorghum trials, while higher mean soil P and K were observed in pearl millet trials under all treatments. A higher variation of yield was observed in sorghum compared to pearl millet in all treatments. In sorghum, soil K had maximum variation in 8 treatments compared to soil P in only one treatment, while soil N had minimum variation in all treatments. In pearl millet, soil P had maximum variation in 4 treatments, followed by soil K in 3 treatments and soil N in 2 treatments.

Semi-arid vertisols at Indore
The F-test indicated that the organic and inorganic treatment combinations were significantly different in both individual years and also when pooled over years in influencing the soybean yield and soil nutrients. The mean soybean yield ranged from 1275 kg/ha with variation of 31.1% under control to 2095 kg/ha with a variation of 25.3% under 20 kg N (urea) + 13 kg P + FYM @ 6 t/ha. The superior treatment also gave a maximum potential yield of 3247 kg/ha in 2006. Application of 20 kg N (urea) + 13 kg P + FYM @ 6 t/ha was also superior with a maximum mean soil N (274 kg/ha), soil P (22.2 kg/ha), soil K (741 kg/ha), and soil Sulphur (18.1 kg/ha). The control gave a minimum soil N of 178 kg/ha, soil P of 11.4 kg/ha, soil K of 540 kg/ha, and soil Sulphur of 13.6 kg/ha. 20 kg N (urea) + 13 kg P/ha had a minimum variation of 7% for soil N and 18.1% for soil K. The control had a maximum variation of 57.5% for soil P and 43.2% for soil S. FYM @ 6 t/ha had a maximum variation of 26.2% for soil N, while crop residue @ 5 t/ha had a maximum of 31.7% for soil K.

Finger millet experiments at Bangalore
The estimates of correlation between finger millet yield, soil fertility of nutrients and monthly rainfall are given in Table 5. At Bangalore, with application of 100% NPK over years, the grain yield had a significant negative correlation with soil P. It had a positive relationship with soil N in control, FYM @ 10 t/ha + 100% NPK and 100% NPK; soil K in all treatments except 100% NPK. The crop seasonal rainfall had a negative effect on finger millet yield in control and 100% NPK. The crop growing period had a positive correlation with grain yield attained by all treatments except FYM @ 10 t/ha and FYM @ 10 t/ha + 50% NPK. Among different treatments, the negative correlation of yield (in all treatments), soil N and soil K with time period indicated a decrease, while a positive correlation of soil P with time period indicated an increase with fertilizer application.

Cotton and green gram experiments at Akola
The estimates of correlation of cotton and green gram yield with soil fertility of nutrients and monthly rainfall are given in Table 6. At Akola, June rainfall had a significant positive correlation with green gram yield of all treatments except 25 kg N/ha (FYM) compared to August rainfall with yield attained by control, 25 kg N + 12.5 kg P/ha, 25 kg N (Leucaena) + 25 kg N (urea) + 25 kg P/ha and 25 kg N (FYM) + 25 kg N (urea) + 25 kg P/ha. The yield had a significant negative correlation with soil N under all treatments except 25 kg N (FYM) + 25 kg N (urea) + 25 kg P/ha; while a significant positive correlation with soil K under 25 kg N (Leucaena) + 25 kg N (urea) + 25 kg P/ha. There was no significant correlation between any pair of variables in case of cotton. The analysis indicated that rainfall of June, July, August and September, soil P, and soil K had a positive correlation, while soil N have a negative correlation with green gram yield over years. Similarly, the monthly rainfall of June to October, and soil K have a positive correlation, while November rainfall, soil N and P have a negative correlation with cotton yield over years.

Soybean experiments at Indore
The estimates of correlation between soybean yield, soil fertility of nutrients and monthly rainfall are given in Table 8. The soybean yield had a significant and positive correlation with uptake N under all the 9 treatments. It ranged from 0.90** for application of 20 kg N (urea) + 13 kg P + FYM @ 6 t/ha to 0.97** under control plot. The soybean yield was found to have a significant negative correlation with soil N (-0.56*) observed under FYM @ 6 t/ha over years. Similarly, the control yield had a significant negative correlation with crop growing period (-0.54*). T1  T2  T3  T4  T5  T6  T7  T8

Groundnut experiments at Anantapur
The estimates of correlation between groundnut pod yield, soil fertility of nutrients and monthly rainfall are given in Table 9. The groundnut pod yield had a better correlation with July rainfall in the range of 0.22 to 0.53 compared to other months in different years. The pod yield attained by all the 9 fertilizer treatments was found to decrease over years as indicated by the negative correlation. The correlations were found to be non-significant over years although they indicated the likely positive or negative trends or effects on the yield. It is observed that yield had a better correlation with July and September rainfall in T9; August rainfall in T8; November rainfall in T3. Maximum pod yield decrease was observed under T4 over years.  Table 9. Relation between yield, monthly rainfall, and CGP over years at Anantapur

Regression model of yield through soil nutrients and rainfall
Multiple regression models for yield attained by each treatment owing to simultaneous influence of crop seasonal rainfall, soil N, P and K nutrients were calibrated and the regression coefficients of variables along with coefficient of determination (R 2 ) and standard error (SE) are given in Table 10 for Bangalore, Table 11 for Akola and Table 12 for Kovilpatti.

Regression model of finger millet yield at Bangalore
At Bangalore, the yield predictability (R 2 ) was in the range of 11% for FYM @ 10 t/ha to 52% for the yield attained with application of 100% NPK over years based on the model. The standard error ranged from 435 to 717 kg/ha in the 25 year study. Based on the model, the crop seasonal rainfall had a positive effect under application of FYM @ 10 t/ha, FYM @ 10 t/ha + 50% NPK and FYM @ 10 t/ha + 100% NPK. The effect of soil N under control, FYM @ 10 t/ha + 50% NPK and 100% NPK; soil K under FYM @ 10 t/ha + 50% NPK and FYM @ 10 t/ha + 100% NPK were positive. The analysis indicated that soil P had a significant negative effect on finger millet yield attained by FYM @ 10 t/ha + 50% NPK. Similarly, soil K had a significant negative effect on the yield attained by 100% NPK application as given in Table 10.

Regression models of cotton and green gram yield at Akola
At Akola, the model of green gram yield through rainfall of June to September, crop duration, soil N, P and K gave a predictability in the range of 0.53 for 25 kg N (FYM) + 25 kg N (urea) + 25 kg P/ha to 0.73 for control (Table 11) In case of cotton at Akola (

Regression model of soybean yield at Indore
Based on the regression model of soybean yield as a function of monthly rainfall received during June to October, crop growing period, soil N, P, K and S, crop growing period had a significant effect on yield attained by all treatments except control and crop residue @ 5 t/ha. July rainfall had a significant effect on yield attained by 20 kg N (urea) + 13 kg P + FYM @ 6 t/ha and 20 kg N (urea) + 13 kg P + FYM @ 5 t/ha, while August rainfall had a significant effect on yield attained by all treatments except control and crop residue @ 5 t/ha. September rainfall had a significant effect on yield attained by all treatments except control based on the model. Among soil nutrients, soil N and P had a significant influence on yield attained by 20 kg N (urea) + 13 kg P + FYM @ 5 t/ha. The R 2 ranged from 0.81 for control to 0.98 for 20 kg N (urea) + 13 kg P + FYM @ 5 t/ha, while standard error ranged from 130 kg/ha for 20 kg N (urea) + 13 kg P + FYM @ 5 t/ha to 320 kg/ha for control. The regression model indicated that maximum rate of change in yield of 31.07 for a unit change in soil N and 1.81 for soil K occurred in control compared to a minimum of -21.23 in 30 kg N (urea) + 20 kg P/ha and -2.16 in 40 kg N (urea) + 26 kg P/ha for the two soil nutrients respectively. The yield attained by 20 kg N (urea) + 13 kg P + FYM @ 6 t/ha had a minimum rate of change for soil P and maximum rate of change for soil Sulphur, while 20 kg N (urea) + 13 kg P/ha had maximum rate of change for soil P and control had a minimum rate of change for soil S (Table 13).

Sustainability yield index of treatments under different rainfall situations
Using the mean yield of treatments over years under different crop seasonal rainfall situations, standard error, and maximum mean yield attained by any treatment over years, the estimates of sustainability yield index of treatments were derived for different crop seasonal rainfall situations and are given in Table 15.