Multi-Step Forecasting of Grape Berry Temperature based on LSTM Model with Feed-Forward Attention

Srikanth Gorthi; L. G. Divyanth; Dattatray Bhalekar; Markus Keller; Lav Khot

cs.AIAug 29, 2026

Multi-Step Forecasting of Grape Berry Temperature based on LSTM Model with Feed-Forward Attention

Authors: Srikanth Gorthi, L. G. Divyanth, Dattatray Bhalekar, Markus Keller, Lav Khot

Organizations: Center for Precision and Automated Agricultural Systems & Department of Biological Systems Engineering, Washington State University, Prosser, WA, USA · AgWeatherNet, Washington State University, Prosser, WA, USA · Department of Biological & Environmental Engineering, Cornell University, Ithaca, NY, USA · Department of Viticulture and Enology, Washington State University, Prosser, WA, USA

Abstract

Accurate forecasting of grape berry temperature (Tb) is essential for enabling timely heat stress management in vineyards. In this study, a feed-forward attention mechanism integrated with a Long Short-Term Memory network (FAM-LSTM) was developed and evaluated for multi-step, high-resolution Tb prediction. Models were trained using environmental data from 2023 and 2024 at Prosser, WA, USA, and validated on 2025 summer data. FAM-LSTM was benchmarked against LSTM, GRU, RNN, and Random Forest (RF) across horizons ranging from 15 minutes to 72 hours (288 time steps). Two input scenarios were evaluated: nearest open-field weather station observations and in-vineyard microclimate measurements. FAM-LSTM consistently outperformed all benchmark models across all horizons and input scenarios. Incorporating in-vineyard microclimate data significantly improved forecasting accuracy at longer horizons. Using open-field data, FAM-LSTM achieved MAE and RMSE ranges of 0.58 to 1.70 deg C and 0.65 to 2.07 deg C, respectively. In-vineyard observations further improved performance, with MAE and RMSE in the ranges of 0.51 to 1.55 deg C and 0.71 to 1.87 deg C. Error analysis showed prediction uncertainty was highest during peak daytime periods (11:00 to 18:00) and increased progressively with forecast horizon. Overall, the FAM-LSTM framework offers robust Tb forecasting to support precision heat stress management in vineyards.

Figures & tables

Figure 1: The experimental setup to collect berry temperature (thermistor insertion, highlighted in red) and in-vineyard microclimate (air temperature and humidity and soil moisture probe) data using LoRaWAN sensing node.

Model	Metric	$t+1$	$t+4$	$t+24$	$t+48$	$t+96$	$t+192$	$t+288$
		15 min	1 h	6 h	12 h	24 h	48 h	72 h
RF	MAE	1.05	1.59	1.76	1.86	2.36	2.43	2.45
	RMSE	1.44	2.05	2.25	2.41	3.01	3.11	3.14
	MAPE	5.35	7.78	8.43	9.21	11.99	12.04	12.12
RNN	MAE	0.94	1.07	1.58	1.71	1.75	1.86	1.95
	RMSE	1.14	1.36	1.98	2.07	2.14	2.26	2.36

Table 1: Prediction performance of evaluated models using open-field weather inputs across multiple forecast time horizons.

Figure 3: Berry temperature ( $T_{b}$ ) forecasts with feed-forward attention mechanism and long-short term memory (FAM-LSTM) using open-field weather station dataset for a) 12-h, b) 24-h, c) 48-h, and d) 72-h.

Model	Metric	$t+1$	$t+4$	$t+24$	$t+48$	$t+96$	$t+192$	$t+288$
		15 min	1 h	6 h	12 h	24 h	48 h	72 h
RF	MAE	1.01	1.40	1.70	1.78	1.95	2.02	2.17
	RMSE	1.39	1.81	2.08	2.15	2.40	2.34	2.84
	MAPE	5.02	6.68	7.98	8.11	9.45	9.56	10.35
RNN	MAE	0.85	1.10	1.54	1.68	1.72	1.69	1.71
	RMSE	1.09	1.37	1.85	2.01	2.10	2.15	2.15

Table 2: Prediction performance of evaluated models using in-vineyard weather inputs across multiple forecast horizons.

Figure 4: Berry temperature ( $T_{b}$ ) forecasts with feed-forward attention mechanism and long-short term memory (FAM-LSTM) using in-vineyard dataset for a) 12-h, b) 24-h, c) 48-h, and d) 72-h.

Forecast period	Zone	MAE	RMSE	MAPE (%)
0–24 h	Peak Temp (max)	1.66	1.66	3.85
	11:00–18:00	1.66	1.68	4.12
	18:00–11:00 (next)	0.86	1.01	3.44
24–48 h	Peak Temp (max)	1.67	1.67	3.93
	11:00–18:00	1.78	1.82	4.29
	18:00–11:00 (next)	1.17	1.30	4.17

Table 3: Diurnal zone error analysis for the open-field FAM-LSTM model across 24-h forecast sub-periods. MAE and RMSE are expressed in ∘ C, and MAPE is expressed in %.

Forecast period	Zone	MAE	RMSE	MAPE (%)
0–24 h	Peak Temp (max)	1.17	1.17	2.71
	11:00–18:00	1.04	1.07	2.69
	18:00–11:00 (next)	0.55	0.64	2.66
24–48 h	Peak Temp (max)	1.35	1.35	3.16
	11:00–18:00	1.30	1.32	3.32
	18:00–11:00 (next)	0.79	0.86	3.54

Table 4: Diurnal zone error analysis for the in-vineyard FAM-LSTM model across 24-h forecast sub-periods. MAE and RMSE are expressed in ∘ C, and MAPE is expressed in %.

Explore similar work

A Hybrid LSTM--Vision Transformer Architecture for Predicting HRRR Forecast Errors
Jun 17, 2026David Aaron Evans, Jay C. Rothenberger, Kara J. Sulia +2Multivariate Time Series Forecasting Recurrent Transformers
Atmospheric Sciences Research Center, University at Albany, SUNY, Albany, NY, USA · University of Oklahoma, Norman, OK, USA · State Weather Risk Communication Center, University at Albany, SUNY, Albany, NY, USA
Calibrating Biophysical Models for Grape Phenology Prediction via Multi-Task Learning
Aug 5, 2025William Solow, Sandhya SaisubramanianMulti-Task Learning Model Calibration
Oregon State University
Real-World Deployment and Performance Characterisation of Fog-Based Deep Learning for Cold-Chain Temperature Prediction over LoRaWAN
Sep 12, 2026Jeremiah Taguta, Jean Frederic Isingizwe Nturambirwe, Clement Nthambazale NyirendaTime Series Forecasting Edge Computing
Department of Computer Science University of the Western Cape Cape Town, South Africa · eResearch Office University of the Western Cape Cape Town, South Africa

Jun 17, 2026, cs.LGJ/K move · Enter open · S save

A Hybrid LSTM--Vision Transformer Architecture for Predicting HRRR Forecast Errors

David Aaron Evans, Jay C. Rothenberger, Kara J. Sulia, Nick P. Bassill+1

Atmospheric Sciences Research Center, University at Albany, SUNY, Albany, NY, USA · University of Oklahoma, Norman, OK, USA · State Weather Risk Communication Center, University at Albany, SUNY, Albany, NY, USA

Multivariate Time Series Forecasting Recurrent Transformers Weather Forecasting

Open paper AskPDF arXiv

Forecast errors in high-resolution numerical weather prediction (NWP) systems are often linked to unresolved planetary boundary layer (PBL) processes, convection, terrain-induced circulations, and other vertically structured atmospheric phenomena. Previous work demonstrated that Long Short-Term Memory (LSTM) networks can successfully predict forecast errors in the High-Resolution Rapid Refresh (HRRR) model using mesonet observations, but we believe performance degradation is linked to periods of complex vertical atmospheric evolution. To address this limitation, we develop a hybrid LSTM-Vision Transformer (LSTM-ViT) framework that combines temporal sequence learning from surface observations with atmospheric profiles from the New York State Mesonet profiler network. The LSTM-ViT framework is trained to predict HRRR hourly precipitation, 10 m wind speed, and 2 m temperature forecast errors at individual mesonet stations. Across all three predictors, incorporation of profiler-derived atmospheric structure improves forecast error prediction skill relative to the baseline LSTM architecture, with the largest gains occurring at shorter forecast lead times and during periods of enhanced PBL activity. Improvements are particularly pronounced for precipitation forecast error, where the LSTM-ViT framework achieves approximately a twofold increase in predictive skill relative to the baseline LSTM while better capturing convectively driven error evolution and reducing degradation associated with PBL processes. These results demonstrate that combining temporal sequence learning with vertically informed attention mechanisms provides a physically meaningful pathway for improving forecast error prediction in operational NWP systems. Our research offers forecasters enhanced guidance regarding model bias and forecast confidence.

Abstract

Explore similar work

A Hybrid LSTM--Vision Transformer Architecture for Predicting HRRR Forecast Errors

Calibrating Biophysical Models for Grape Phenology Prediction via Multi-Task Learning

Real-World Deployment and Performance Characterisation of Fog-Based Deep Learning for Cold-Chain Temperature Prediction over LoRaWAN