Wheat is a crucial staple in South Asia, with India cultivating 73% of the region's 36.1 million hectares. Meeting projected domestic demands necessitates a substantial increase in wheat production by 2050 (from 110 to 140 million tons). Land expansion being limited, yield intensification is vital. The Indo-Gangetic Plain is the main wheat-growing area, with the rice-wheat (RW) system prevalent in eastern India. While western regions (Punjab and Haryana) have high productivity, challenges like groundwater depletion exist. Therefore, intensification efforts focus on the Eastern Ganges Plain (EGP), characterized by high yield gaps attributed to factors including late sowing, older cultivars, weed issues, irrigation limitations, and labor shortages. Late sowing exposes wheat to heat stress during the reproductive phase, reducing yields. While timely sowing is crucial, its contribution to yield gaps in South Asia isn't fully understood. The RW system involves interconnected management decisions, often with trade-offs. This study uses 'climate resilience' to describe planting date adjustments that enhance yields, potential, and stability by mitigating unfavorable growing conditions. The study's four objectives are: (1) quantifying the impact of planting dates on wheat yields, potential, and gaps; (2) assessing potential yield gains from planting date adjustments; (3) identifying factors influencing wheat planting timing; and (4) simulating system-level strategies to enhance climate resilience via planting date changes. The goal is to highlight the importance of cropping calendar management for RW system performance in the EGP, informing sustainable intensification and climate adaptation.

Existing research confirms the importance of timely wheat sowing to avoid heat stress-related yield losses. However, the extent of the contribution of current sowing date patterns to yield gaps remains insufficiently characterized in South Asia. Several studies have analyzed yield gaps in the context of specific factors, like late sowing, irrigation practices, and cultivar choices. However, a comprehensive assessment of the interplay of these factors within the coupled rice-wheat system is needed. The existing literature frequently focuses on single-commodity analyses, neglecting the complex interactions within cropping systems. This study addresses this gap by employing a systems-level approach that considers rice and wheat as an integrated unit, acknowledging potential trade-offs and synergies.

This research combines field and household survey data, remotely sensed information (MODIS), and dynamic crop simulation (APSIM). **Household Surveys:** Detailed household surveys (1000 in 2013, revisited in 2016) were conducted across six Bihar districts to understand wheat planting decisions and reasons for delays. These surveys captured information on planting and harvesting times, spanning 2010-2015 (5766 site-year observations from 961 households). **Landscape Diagnostic Surveys (LDS):** Data on wheat yields, practices, and site characteristics were gathered across Bihar and adjacent Uttar Pradesh districts (2012-2017, 6216 site-years). In 2018, a revised sampling strategy (7648 fields) and survey instrument were used to obtain a more representative sample. Crop cuts were conducted to determine yields. Data included site attributes (field area, landscape position, soil texture), agronomic practices (fertilizer use, planting/harvest dates, irrigation, cultivar), socio-economic factors (land tenure, household size, income), and self-reported yields. Digital soil maps were used for 2018 data. Boundary line analysis was employed to determine wheat yield potential (Yp) as a function of planting date. Machine learning (Random Forest) modeled yield and ranked factor importance. **Satellite Data (MODIS):** MODIS satellite data (2002-2017) provided wheat area and establishment dates at 250m resolution. Time series of Enhanced Vegetation Index (EVI) were analyzed to estimate phenological parameters (sowing, maturity, duration). A correction factor was applied to account for the delay in satellite detection of early growth stages. An area mask separated wheat pixels from other vegetation types. **APSIM Simulation:** APSIM v7.09 simulated RW planting date scenarios to assess aggregate yields, yield stability, irrigation needs, and economic productivity. A site in Patna, Bihar was used, employing common cultivars and silt loam soil characteristics. The model was calibrated using experimental site data. Simulations examined various rice transplanting dates (June-mid-September) with long-duration (MTU7029) and medium-duration (Arize6444) rice, and wheat (PBW343). 43 years of weather data and best agronomic practices were used.

Machine learning models consistently identified sowing date as the strongest predictor of wheat yield. The median sowing date in 2018 was November 27th. Early sowing (before November 20th) yielded 3.4 t/ha, significantly higher than average (2.9 t/ha) and late sowing (2.5 t/ha, a 36% difference). Boundary line analysis showed that before November 20th, yield potential (Yp) decreased by 25 kg/ha/day, doubling to 51 kg/ha/day after that date. Yp increased by 69% for early November sowing (5.4 t/ha) compared to late December (3.2 t/ha). Post-November 20th, Yp stability decreased, indicating reduced reliability with later planting. MODIS data revealed a multi-year mean sowing date of December 13th, with significant spatial variability. Household surveys showed late sowing (after November 15th) in 84% of cases, primarily due to previous crop (rice) occupancy (44%), knowledge gaps (26%), and waterlogging (21%). The share of late sowing due to knowledge gaps decreased over time, suggesting the impact of extension efforts. A strong relationship existed between rice harvest and wheat sowing date after November 1st. The median time lag between rice harvest and wheat planting was 15 days in 2018. For the study area, wheat Yp is estimated at 3.68 t/ha under current conditions (Scenario A). Interventions advancing planting dates (wheat-specific, rice-focused, or combined) resulted in Yp gains of 0.30, 0.58, and 0.84 t/ha, respectively (Scenarios B, C, and D). Scenario D would increase the yield gap from 21% to 36%, resulting in a 1.88 million-ton annual production increase and a US$421 million revenue gain. APSIM simulations showed that for long-duration rice, transplanting before July 13th maximized grain productivity and profitability. A broader transplanting window existed for medium-duration hybrid rice. Spatial predictions indicated the highest yield potential gains (>1 t/ha) in the northern part of the study area.

The findings demonstrate that timely wheat establishment, achievable through improved management, is crucial for enhancing wheat productivity and climate resilience in the EGP. The combined rice and wheat interventions (Scenario D) offer the most substantial yield gains, highlighting the need for a systems approach. The substantial increase in potential wheat yield, if realized, would significantly improve food security and income generation in the region, aligning with Indian agricultural development policies. The study emphasizes the importance of addressing the constraints to early planting, primarily the influence of rice management on wheat sowing timing. The results highlight opportunities for intensification while also acknowledging the geographic variability in the feasibility of different interventions. As climate change shrinks the favorable thermal window for wheat, timely planting will become even more critical.

This research underscores the significant role of timely wheat sowing in boosting productivity and climate resilience in the EGP's rice-wheat system. Transformative gains necessitate integrated rice and wheat management, addressing both biophysical and socio-economic constraints. Future research should focus on refining adaptation pathways, considering regional variability, and incorporating socio-technical innovation bundles to promote the adoption of improved practices. The need for further research into context-specific adaptation strategies and the scaling-up of successful interventions is crucial for sustainable agricultural intensification.

The APSIM simulations used a single site in Patna, Bihar, neglecting sub-regional variations in soil and climate. While the model has been validated in prior studies across diverse settings, additional model verification may enhance confidence in the findings. The study primarily focuses on the EGP, limiting the generalizability of results to other regions with different agro-ecological conditions. The analysis relies on self-reported yield data, which may be subject to biases; however, this was partially mitigated through independent crop cuts in a portion of the study. The analysis of MODIS data depends on the accuracy of the correction factors and classification algorithms. Finally, economic data relied on minimum support prices, potentially underestimating actual revenue gains if market prices exceed support prices.